Report No: AUS11077 . South Asia Investment decision making in hydropower Decision Tree Case Study of the Upper Arun Hydropower Project and Koshi Basin Hydropower Development in Nepal . October 15, 2015 . GEEDR SOUTH ASIA . . Standard Disclaimer This volume is a product of the staff of the International Bank for Reconstruction and Development/ The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Copyright Statement The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/ The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, http://www.copyright.com/. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e- mail pubrights@worldbank.org. Currency Equivalents All currency values are in constant 2015 US$ unless otherwise specified. Weight and Measures Systeme Internationale (SI Units) Vice President: Laura Tuck Practice Managers: Julia Bucknall Task Team Leader: Pravin Karki Acknowledgments The report was sponsored by the World Bank Group, and prepared by a team consisting of Laura Bonzanigo, Casey Brown, Julien Harou, Anthony Hurford, and Patrick Ray. Significant modeling and analytical contributions were made by Ethan Yang and Sungwook Wi. The team was led by Pravin Karki and Julia Bucknall. The work was supported by the Water Global Practice and the Water Partnership Program. The World Bank Group peer reviewers were Rafaello Cervigni, Divas Basnet, Nathan Engle, Johan Grijsen, Rikard Liden, Peter Meier, Jie Tang, Michael Toman, and David Viner. i Table of Contents Acknowledgments................................................................................................................... i Table of Contents .................................................................................................................... ii Table of Figures...................................................................................................................... iv Table of Tables .....................................................................................................................viii Summary Findings ................................................................................................................. 1 Chapter 1. Programmatic Approach To Assessing Climate Risks to Hydropower Investments in South Asia ..................................................................................................... 3 Chapter 2. The Methodology ................................................................................................ 7 Chapter 3. Background to Case Study: Nepal’s Power Infrastructure ........................ 13 Nepal’s Power Infrastructure Context ........................................................................ 13 The Upper Arun Hydropower Project (UAHP) ........................................................ 17 Technical features .................................................................................................... 20 Chapter 4. Case Study Application: The Upper Arun Hydropower Project ............... 22 Phase 1 of the Decision Tree – Project Screening....................................................... 22 Phase 2 of the Decision Tree – Initial Analysis .......................................................... 22 Phase 3 of the Decision Tree – The Climate Stress Test ............................................ 29 Effects of Plausible Climate Changes ................................................................... 31 Results: Climate Change Effects ............................................................................ 36 Full Climate Stress Test including Non-Climate Factors ................................... 45 Phase 4 of the Decision Tree – Risk Management ..................................................... 52 Additional Design Options .................................................................................... 52 Opportunity Metric: Regret ................................................................................... 54 Results: Sensitivity and vulnerability to climate of different design options . 55 Results: Comparing all options’ performances under multiple uncertainties 61 Conclusions .............................................................................................................. 64 Chapter 5. Basin-level assessment of hydropower investment portfolios under uncertainty............................................................................................................................. 66 Introduction .................................................................................................................... 66 Proposed hydropower system investment appraisal approach.............................. 66 General problem ...................................................................................................... 66 Proposed basin-level hydropower investment approach .................................. 67 Phase I: System characterization ........................................................................... 68 ii Phase II: Vulnerability assessment ........................................................................ 69 Phase III: Automated search .................................................................................. 69 Phase IV: Stress test ................................................................................................. 69 Application to the Koshi basin ..................................................................................... 70 Phase I: System characterization ........................................................................... 70 Phase II: Vulnerability assessment ........................................................................ 73 River flows................................................................................................................ 73 Abstraction demand increases ............................................................................... 73 Environmental flow releases.................................................................................. 74 Phase III: Automated search for efficient hydropower interventions ............. 74 Uncertainty cases ..................................................................................................... 75 Results ....................................................................................................................... 75 Robustness analysis................................................................................................. 78 Phase IV: Basin-scale stress test ............................................................................. 86 5.5 Recommendations and discussion ........................................................................ 87 Limitations and future work ........................................................................................ 89 Conclusions ..................................................................................................................... 90 Chapter 6. Conclusions ........................................................................................................ 91 References .............................................................................................................................. 94 Appendix A: Climate Data Analysis and Supplementary Model Results ................... 96 Climate Data Analysis ................................................................................................... 96 Supplementary Model Results ................................................................................... 105 Appendix B: Scenario Discovery Methods and Results ................................................ 110 Appendix C. Stakeholders’ involvement during the project........................................ 114 The Project Team in Kathmandu, 8-13 September 2014 ......................................... 114 Monday ................................................................................................................... 114 Tuesday ................................................................................................................... 115 Wednesday ............................................................................................................. 115 Thursday ................................................................................................................. 115 Friday ...................................................................................................................... 115 Next Steps ............................................................................................................... 116 iii Table of Figures Figure 1. Decision Tree Schematic (Water Partnership Program, 2015) ......................... 9 Figure 2. National Electricity Authority (NEA) of Nepal statistics ............................... 14 Figure 3. Seasonality mismatch between system demand and generation .................. 15 Figure 4. FAO landuse in the Koshi basin ........................................................................ 17 Figure 5. Elevation and glacier cover within the Koshi basin ........................................ 18 Figure 6. Location and plan view of the UAHP (NEA 2014) ......................................... 19 Figure 7. Uwagaon hydrology (MCM/month) ................................................................ 24 Figure 8. Uwagaon streamflow exceedance probability (Apr 1986-Dec 2006) ............ 25 Figure 9. Theoretical potential (uncapped) hydropower exceedance probability ...... 26 Figure 10. UAHP, 335 MW PROR @ Q70, max daily GWhr/day generation relative to daily timeseries of GWhr potential (red line) ............................................... 27 Figure 11. Elasticity of NPV to changes in precipitation (left) and temperature (right) ............................................................................................................................. 29 Figure 12. Schematic representation of the Phase 3 process........................................... 32 Figure 13. Diagram of the UMass hydrologic model with glacial component ............ 34 Figure 14. Calibration and validation of the hydrologic model .................................... 35 Figure 15. Hydrologic model calibration results showing percent contribution to streamflow ......................................................................................................... 35 Figure 16. Response of streamflow to changes in climate by annual total, dry season (lower left) and wet season (lower right). Changes in precipitation are shown on the x axis and changes in temperature are shown on the y axis. Contour colors represent increasing streamflow, blue in excess of historic mean, and red less than historic mean. .......................................................... 37 Figure 17. Glacier area response surface - Year 2050 Glacier Area (as % Year 2014 Glacier Area) ...................................................................................................... 38 Figure 18. Seasonality shift in streamflow (MCM/month) with increasing temperature (no change to precipitation). ..................................................... 39 Figure 19. Climate change response surfaces – NPV (average values from 2020-2050) ............................................................................................................................. 40 Figure 20 Hindcast of 48 GCMs for Annual Mean Precipitation – Upper Arun River Basin .................................................................................................................... 41 Figure 21 Hindcast of 48 GCMs for Annual Mean Temperature – Upper Arun River Basin .................................................................................................................... 41 Figure 22 Hindcast (1971-2000) and Forecast (2036-2065) of 48 CMIP5 GCMs (1971- 2000) - Upper Arun River Basin ...................................................................... 42 iv Figure 23. CMIP5 precipitation change projections: 1971-2000 vs 2036-2065 .............. 43 Figure 24. CMIP5 temperature change projections: 1971-2000 vs 2036-2065 ............... 43 Figure 25. NPV of the 335 MW design with CMIP5 climate change projections (centered 2050) superimposed. ....................................................................... 44 Figure 26. Uncertainty space for the analysis of building a new hydropower plant on the Upper Arun. The black dots represent the initial assumptions. .......... 48 Figure 27 Sensitivity of NPV to changes in each factor holding all others at baseline values .................................................................................................................. 50 Figure 28. UAHP max daily GWhr/day generation relative to daily time series of GWhr potential: red = 335 MW; blue = 750 MW; green = 1000 MW; purple = 1355 MW; orange = 2000 MW. ..................................................................... 53 Figure 29. Changes in annual electricity production, 335 MW and 2000 MW options. Dots locate average values of temperature and precipitation projected by the CMIP5 generation of IPCC GCMs over the approximately lifetime of the project, 2020-2050 (Green: RCP 2.6; Blue: RCP 4.5; Yellow: RCP 6.0; Purple: RCP 8.5). ............................................................................................... 56 Figure 30. Wet season electricity production in the wet season, 335 MW and 2000 MW options ....................................................................................................... 56 Figure 31. Dry season electricity production, 335 MW and 2000 MW options ........... 56 Figure 32. Minimum monthly electricity production, 335 MW and 2000 MW options ............................................................................................................................. 57 Figure 33. NPV response to changes in precipitation and temperature ....................... 57 Figure 34. Economic performances across different climate scenarios, maintaining all other variables constant. The shaded area indicates a negative NPV. ...... 58 Figure 35. 335 MW performance under 81 different scenarios of climate change ...... 59 Figure 36. 2000 MW performance under 81 different scenarios of climate change .... 60 Figure 37. Tradeoffs between maximum regret and production in the dry season (note that decision makers were not interested in 1355 MW and 2000MW, as they are sensitive to changes in climate). .................................................. 62 Figure 38. Schematic of the IRAS-2010 Koshi Basin model showing the 5 new hydropower reservoirs being considered in this hydropower investment assessment. Existing reservoirs are also displayed ...................................... 71 Figure 39. Shaded area shows extent of IWMI Koshi basin model used as a data source for the system simulation model built for this study. ..................... 72 Figure 40. Performance of all portfolios identified by the search under average case conditions, in terms of investment and dry season generation.................. 76 Figure 41. The same plot from Figure 40 with coloring used to show which UAHP capacity option is used in each portfolio ....................................................... 76 v Figure 42. The most efficient set of interventions for increasing Koshi Basin dry season electricity generation alone with portfolios numbered and detailed for reference. ...................................................................................................... 77 Figure 43. Comparison of the best performing interventions for three types of energy generation and for each of three uncertainty cases. Variations between the uncertainty case portfolios (indicated by the UAHP option used) show that robustness to extreme conditions does not necessarily result from good average performance. ............................................................................. 78 Figure 44 Relationship between maximum regrets for the three uncertainty cases for infrastructure portfolios from Table 11 when they are operated to maximize efficiency across either investment and energy generation objectives alone, or across all objectives. All-objective efficiency results in lower total maximum regrets for portfolios 1, 5 and 7, showing the increased robustness available by changing the objectives of dam operation, even with the same infrastructure. .............................................. 80 Figure 45 Relationship between maximum regrets as in Figure 44 with 17 other additional infrastructure portfolios added which are efficient across all objectives. Eleven of these result in lower total maximum regrets, showing the further benefits of maximizing efficiency in relation to all objectives, rather than a restricted set of investment and energy objectives. ............................................................................................................................. 83 Figure 46. Average case performances of efficient and robust portfolios of hydropower infrastructure investments considering different sets of objectives. The extent of the red points shows that maximum dry season generation is reduced by around 300GWh/year to balance other objectives efficiently and robustly. Three portfolios from Table 12 are labelled according to their robustness rank and composition. Two are implementations of the same 4th ranked portfolio with operating rules which maximise generation but differ in their environmental performance. This difference depends on how environmental performance is balance with other objectives not shown. The 6th ranked portfolio is shown here operated to trade-off dry season generation to maximise environmental performance. ...................................................................................................... 84 Figure 47. Performance of a high performing portfolio across 10,000 futures ............ 86 Figure 48 Gridded Climate Data: 0.5 degree grids of APHRODITE daily temperature and precipitation dataset. Blue basin is for Uwagaon streamgage (600.1); green is for Turkighat (604.5); pink is for Simle (606) Circle points are climate stations in the Koshi basin. X’s are climate stations without usable data...................................................................................................................... 96 Figure 49 APHRODITE and GPCC are very precise (coeff of corr), but not particularly accurate (Nash Sutcliffe). However, gages tend to be at lower elevation, and topography of most cells includes very high elevation. Gridded data algorithms include topography in spatial averaging. It is vi therefore difficult to draw conclusions regarding the quality of gridded data...................................................................................................................... 96 Figure 50 APHRODITE is the only dataset with daily precip and temperature. It is fairly precise, but not very accurate. It consistently underestimates precipitation in some cells. .............................................................................. 97 Figure 51 GPCC typically performs well in the Himalayas (see Brahmaputra analysis), and its record is longest (monthly data from 1900-2010), but contains only monthly precipitation data, and at 2.5o grids (as opposed to Aphrodite’s 0.5o grids). .................................................................................... 97 Figure 52 Cell with the most observations (though also fairly low elevation) ............ 98 Figure 53 APHRODITE, when converted to streamflow, is much lower than other gridded data. GPCC most closely follows TRMM, which uses satellite observations, and tends to be most trusted (with record only from 1998). The choice is then to use GPCC data, “downscale” it using ratios from APHRODITE data (and distribute it into daily values). ............................. 99 Figure 54 Wavelet Generator Diagnosis: Wavelet based on GPCC area-averaged precipitation data for the Upper Arun basin from (water year) 1901-2010 ........................................................................................................................... 100 Figure 55 Autocorrelation at between 10 and 12 years ................................................. 101 Figure 56 The area-averaged precipitation data are not normal, and are not made normal by transformations using log or Box-Cox ...................................... 102 Figure 57 The WARM model (with 11-year signal) therefore is not a good fit for the data.................................................................................................................... 103 Figure 58 Neither is the simple ARIMA model.............................................................. 104 Figure 59 NPV ..................................................................................................................... 105 Figure 60 Annual hydropower production .................................................................... 106 Figure 61 Dry season hydropower production .............................................................. 107 Figure 62 Wet season hydropower production .............................................................. 108 Figure 63 Minimum month hydropower production ................................................... 109 Figure 64 Example of Scenario Discovery. ...................................................................... 111 Figure 65. Koshi Basin Modeling Project Workflow...................................................... 114 Figure 66. Iterative Application of Decision Tree Phase 4 for Climate Change Risk Management .................................................................................................... 116 vii Table of Tables Table 1. Design parameters for UAHP prefeasibility design – 335 MW PROR........... 20 Table 2 Design 1, 335 MW PROR, baselines ..................................................................... 28 Table 3. The Decision Matrix developed during the initial consultations with Nepalese experts and NEA.............................................................................. 30 Table 4 Internal variability across 30 traces of stochastic climate variability .............. 45 Table 5. The values utilized in the original analysis and in this study’s uncertainty analysis. .............................................................................................................. 46 Table 6. Sample from the database with the 6,500 futures ............................................. 49 Table 7. Updated uncertainty table, with all options ...................................................... 54 Table 8. Installed capacity of existing grid connected hydropower dams in the Koshi Basin .................................................................................................................... 70 Table 9 Proposed hydropower projects included as options in the basin-scale analysis ............................................................................................................................. 72 Table 10. Three uncertainty cases used in the search process and the circumstances under which performance for all metrics is evaluated ................................ 75 Table 11. Portfolios of infrastructure which are efficient and robust for investment and energy generation objectives under all 3 uncertainty cases. Ranking is performed on the basis of minimizing the total maximum regret for the 3 uncertainty cases. Performance metric values are each portfolio’s best investment and energy performance result under the average uncertainty case (i.e. averages across 20 scenarios)........................................................... 81 Table 12. Portfolios of infrastructure which are efficient for all objectives under all 3 uncertainty cases and lead to lower total maximum regret than the highest ranking infrastructure combination and its operation from Table 11. Ranking is performed on the basis of minimizing the total of maximum regret for all 3 uncertainty cases. Performance metric values are each portfolio’s best performance result under the average uncertainty case. ..................................................................................................................... 85 Table 13 Conditions defining scenarios in which each project has a negative NPV, and the coverage and density for each scenario. ........................................ 112 viii Decision Tree Case Study of the Upper Arun Hydropower Project and Koshi Basin hydropower development in Nepal Summary Findings  Hydropower development faces a number of risks, including climate change, high sediment loads, and environmental, social and financial risks. Investments in hydropower development must therefore be supported by a systematic approach to climate risk assessment. In fact, IDA 17 demands a rigorous yet workable approach to consideration of these multiple challenges in World Bank South Asia hydropower and reservoir projects.  Hydropower developers, including government planners and the private sector, need guidance for assessing climate risks. The most common current approaches, such as climate science-led analyses dependent on GCM downscaling, can be improved, as the Independent Evaluation Group concluded. A systematic approach to assessing climate risks, and that puts climate risks in the context of other risks, is needed. Without it, important investments that are needed for continued human development will be stalled.  The project level study employs the Decision Tree Framework, which was developed for the systematic assessment of climate risks to infrastructure investment. The Decision Tree is an implementation of general Decision Making under Uncertainty (DMU) approaches conceived by the Water Global Practice and supported by the Water Partnership Program. This application represents the culmination of learning in these areas, bringing together advanced physical and economic modeling to assess climate and non-climatic risks. A related DMU framework is proposed and applied at basin scale. This focuses on ways to identify and select robust and balanced portfolios of hydropower infrastructure at system level.  The analysis presented in this report assessed climate change risks of the pre- feasibility 335 MW design of the Upper Arun Hydropower Project (UAHP). In addition to climate change risks, other non-climate factors were identified through stakeholder discussions and included in the analysis. Stakeholders identified performance metrics for project evaluation, namely the economic value of the project (Net Present Value) and the total and dry season hydropower production.  The application of the Decision Tree to the UAHP development demonstrated that the project is robust to climate change and other risks. This document concentrates mostly on an exploration of the risks and potential benefits of investment in the UAHP, based on the understanding that it is a “preferred” investment alternative in Nepal’s hydropower sector. The basin-scale analysis confirms UAHP as a promising investment, with the most appropriate capacity option dependent on decision maker preferences.  The general hydrologic response of the Upper Arun River to changes in climate were increases in streamflow with warmer temperatures up to about +3C, after which streamflow decreases moderately due to declining contributions from 1 glacial melt. The streamflow during the low flow season was found to decline slightly with warmer temperatures but the effect was small. Precipitation effects were as expected, with increases leading to increased streamflow, and the effects were much larger than the temperature effects. Projections for the region, of unknown credibility, indicate warmer future temperatures and no clear signal for precipitation.  The economic value of the proposed 335 MW design is robust to the wide range of climate changes considered. The climate stress test sampled a wide range of changes in climate and variability to reveal the vulnerability of the design to plausible climate changes. In this case, climate change itself posed little risk to the 335 MW design. On the contrary, the analysis revealed the opportunity to consider larger capacities.  The assessment also considered possible alternative (larger capacities) designs for UAHP. The results indicated that each of the capacities considered was robust to plausible climate changes. Only the largest capacity exhibited vulnerabilities to climate changes but these were fairly extreme (e.g., 20% reduction in precipitation). Among non-climatic factors, the price of electricity and capital costs emerged as key risk factors, specifically if the former remains as low as it is today (approximately 0.05 USD/KWh) and the latter increases more than 50% due to delays in implementation or other implementation problems.  The analysis concluded that the original pre-feasibility design of 335 MW was robust to the range of uncertainties considered, with few problematic scenarios. However, it was not able to exploit much wet season flow. The design capacity of 1000 MW emerged as an attractive alternative, providing the best combination of robustness and opportunity, including dry season production, but is also more sensitive to increases in capital costs and low electricity prices. These risks should be carefully addressed if this design were selected.  The basin-scale analysis of hydropower infrastructure investment considered five proposed hydropower dams, including UAHP, alongside the Koshi’s five existing dams. The assessment identified robust portfolios of investments (those that perform acceptably well over a range of plausible futures) considering firm, dry season, and total annual electricity generation as well as environmental flow impacts, urban and agricultural water supply and flood alleviation (for storage schemes).  UAHP schemes (of various capacities) were members of the best (i.e. most efficient) investment portfolios, although achieving different balances of benefits. As shown by the project-level analysis, larger UAHP schemes are difficult to justify for dry season electricity generation alone, as flows are the limiting factor. Smaller UAHP capacities are able to remain members of efficient portfolios despite unfavorable conditions for hydropower generation while larger capacity options are better able to take advantage of increasing flows, which may result in the short term from climate change.  A stress test similar to the one applied in the project-level analysis was applied to one of the promising Koshi basin portfolios of hydropower investments. This confirmed that non-climate uncertainties of construction cost and electricity price are the key risk factors for the achievement of expected gains from hydropower investments. 2  Both project- and basin-level studies acknowledge the limited capacity of a run- of-river hydropower scheme such as the 335 MW UAHP to address dry season electricity shortage, even in a basin with relatively reliable flows. Larger schemes reduce the problem of load shedding during the dry season, but they are efficient only where their increased wet season generation can be used or sold.  Although this work explored hydropower investment options at the project and basin-scale, there could be benefits to undertaking such work at national scale. Such a study could borrow elements of the analyses presented here. Chapter 1. Programmatic Approach To Assessing Climate Risks to Hydropower Investments in South Asia The South Asia region requires major new energy resources to support its development goals. Hydropower development, especially in mountainous areas, is a largely untapped resource for low carbon energy. In countries like Nepal and Bhutan, it represents their most valuable natural resource, their key natural endowment to use for economic growth and human development. The link between electrification and development is clear and in this region there is a low carbon option. While the need for more energy investments is pressing and the logic of tapping hydropower is straightforward, investments in hydropower development have many potential risks, especially in mountainous regions like the Himalayas. These include climate change, sedimentation, uncertain data, and environmental, economic and financial risks. However, while a hydropower development may face many risks, that does not imply that the investment itself is risky. In fact, the risks may be quite low and easily managed. Nonetheless, just the existence of potential risks can impede needed energy investments due to the complicated political and institutional hurdles that hydropower developments must overcome. Considerations related to climate change highlight these issues. Potential impacts of climate change represent real concerns for energy development in South Asia. Scientific articles and the media regularly publish sensational papers about possible troubling hydrologic futures in the Himalaya especially. Yet the uncertainties of climate modeling in this region, and the general difficulty in predicting the future evolution of climate change anywhere, defy the usual approaches to assessing such concerns. In fact, the IEG found that the Bank’s efforts on climate change generally and especially on climate modeling and downscaling have not led to improved information for decision-making (IEG, 20121). 1Independent Evaluation Group (2012), Adapting to Climate Change: Assessing World Bank Group Experience, 193 pp. 3 Nonetheless, there is clear need for change to the traditional methods of planning water infrastructure in an age of climate change and recognition of our limited abilities to anticipate the future. The so-called “death of stationarity” implies that the historical record may no longer be a useful basis for designing infrastructure. Alternative approaches focus on identifying vulnerabilities and managing them pragmatically. Thus the proponents of hydropower development, including government planners and the private sector, face a dilemma. They must assess the risks of climate change to their proposed infrastructure investments. Yet there is no clear guidance on how to do so, and the most common approach (GCM-derived, top-down) has left many decision makers seeking improved approaches. Most worrying is that this dilemma may result in the stalling of important investments, or inaccurate and inadequate assessment of climate risks using ill-considered methods, or worse, the avoidance of these investments all together. This is a serious concern in countries like Nepal where hydropower provides over 90% of the current electricity production and which have ambitious plans to develop hydropower resources in the future. Hydropower production is affected by climate variability, and local governments are not able to provide clear guidance on screening and mitigation measures to hydropower developers. External expert guidance is required. 4 Box 1: Climate Change in Nepal, and Climate Scenarios Used in this Report The result is a strong need for a clear approach to the systematic appraisal of climate risks to infrastructure development in South Asia. At its best such an approach will provide a common framework that can be generally applied to infrastructure development. It will also assess other risks in addition to climate and as a result position climate risks within a broader context. Most important, the approach will lead to more robust and resilient projects that perform well over their lifetime. The articulation of a programmatic approach to assessing and managing climate risks in the context of other risks for hydropower investments is the goal of this document. The approach described here is drawn from the Decision Tree developed by the Water Global Practice with the support of the Water Partnership Program and the Decision Making under Uncertainty initiative led by the Office of Lead Economist, Climate Change. It represents the culmination of years of experience in assessing climate and other risks to infrastructure investments and the practice of decision-making under uncertainty generally. The approach is specifically designed for application to hydropower development in South Asia. Most important, this methodology is designed to lead to unambiguous results for assessing and 5 quantifying risks and clearly evaluating the benefits resulting from risk management and adaptation. The logic of the approach is simple: instead of attempting to use downscaling to predict climate change, which is inherently unpredictable, this approach identifies vulnerabilities to climate change and other factors by exploring the effects of a systematically sampled wide range of changes in climate and other non-climate factors. Since even prescribing the likely range of deeply uncertainty variables can be problematic (and stall needed analyses), the approach using ranges of these variables that go well beyond what could be argued as plausible. If vulnerabilities emerge, their level of concern is judged and managed pragmatically based on the specific variables and values that cause the vulnerability. A conceptual framework guides the project planner through application of DMU approaches to climate and other risks. This report demonstrates this process. This report describes the programmatic approach and illustrates its application with a case study of a proposed hydropower development in Nepal. Chapter 2 describes the methodology for the programmatic approach. Chapter 3 introduces the context for the case study, the energy sector of Nepal. Chapter 4 presents the application of the approach to the Upper Arun Hydropower Project (UAHP). Chapter 5 presents the basin-scale analysis. Finally, the conclusions from the application are discussed in Chapter 6. 6 Chapter 2. The Methodology The Decision Tree Framework is a process for evaluating climate risks to development projects. The process was designed in response to the need for climate change risk assessments of all projects according to IDA 172. While the requirement was enacted, no clear guidance was provided on the process for assessing climate risks, and past experience revealed the need for a clear, well-defined process, as IEG (2012) details. The process was designed with the needs of program managers and project developers in mind. It is conservative in terms of time and resources needed and escalates the level of analysis in proportion to the level of concerns that are identified. Most important, this method is based on the theory of decision making under uncertainty, using a “bottom-up” approach that focuses on identifying vulnerabilities of a given project to climate change, in the context of other uncertainties (e.g., financial, social, etc.), rather than focusing on prediction of future climate and the various climate modeling and downscaling approaches that prediction entails when designing an investment. It also is designed to identify projects that are robust to uncertainties rather than seeking optimal solutions that may be fragile as conditions change. The Decision Tree approach reflects the growing consensus that robustness-based approaches are needed to address the deep uncertainties associated with, but not only, climate change and their potential impacts on infrastructure planning. Deep uncertainties are defined as those where parties to a decision do not know—or do not agree on— the likelihood of future events, the best model for relating actions to outcomes, or the value of potential outcomes (Lempert et al., 2003). Climate change represents a prominent example of deep uncertainty, although not the only example. 2 As part of the IDA 17 Replenishment, these policy commitments related to climate and disaster risk screening were agreed: 1) “All IDA country partnership frameworks [would] incorporate climate and disaster risk considerations into the analysis of the coun try’s development challenges and priorities and, when agreed with the country, incorporate such considerations in the content of the programs and results framework”; and 2) “[Project planners would] screen all new IDA operations for short- and long-term climate change and disaster risks and, where risks exist, integrate appropriate resilience measures” (IDA, 2014). World Bank project teams are encouraged to conduct climate and disaster risk screening during the concept stage, in order to use the screening results as an input into project design. The World Bank has developed Climate and Disaster Risk Screening Tools to help project teams and other development practitioners identify these risks at early stages of project/program design. In addition, in its 201 3 report “Building Resilience: Integrating Climate and Disaster Risk into Development” IBRD stressed that building climate resilience is critical for achieving the World Bank Group’s goals to end extreme poverty and build shared prosperity. The report called for the international development community to build long-term resilience, reduce risk, and avoid escalating future costs. 7 Typical infrastructure planning process also face uncertainties about future benefits realized from an investment, future economic and population growth, and changes in prices of relevant products and commodities (e.g., electricity, water, food). And these uncertainties can stall project development or lead to poor performance if they are not addressed or addressed in an appropriate fashion. Common approaches to addressing uncertainty include seeking predictions of most likely futures or ignoring it. Prediction-based approaches, sometimes called “Predict- then-Act”, hinge on our accurately predicting and then reaching consensus on what the future will bring (Hallegatte et al., 2012; Lempert and Kalra, 2011). This is a difficult prospect. In the case of climate change, experience in attempts to better predict the evolution of future climate, in the hope that uncertainty might be reduced, have been found ineffective (IEG, 2010, 2012; The World Bank, 2013). They often result in major climate science efforts including focusing on emissions scenarios, modeling choices and downscaling methods, instead of focusing on the key uncertainties that effect a specific project at a specific location. Because better understanding of climate change often leads to better recognition of the true range of uncertainty, these efforts can actually backfire by increasing the climate uncertainty that planners face, instead of reducing it. But climate change is not the only deep uncertainty that may make a project fail. For instance, electricity prices, economic changes, population growth, and so forth during the project’s lifetime are difficult to predict at the time when we make the decision to implement it. The Decision Tree employs a different approach. It is designed to provide a path forward for project planners who face decisions potentially affected by climate change uncertainty by diagnosing the effects that plausible climate and other changes can have on a project. The process is focused on the project and begins with outlining key uncertain factors that may affect a project. It then identifies the potential vulnerabilities that a project incurs if one or a combination of these factors falls into problematic ranges. This is accomplished using a “climate stress test” which is a carefully structured sensitivity analysis. It does not restrict the range of climate changes considered to simply the range that global climate models produce, since it is well known that those models do not delimit the true range of uncertainty. Climate change projections are used to provide insight if the climate stress test reveals climate changes that pose significant risks to a project. In that case, the climate change analysis, including downscaling methods, can be specifically tailored to the problematic climate changes that are identified in the climate stress test to guide judgments regarding whether such changes are likely. In addition to addressing the fundamental science issues, the Decision Tree was also designed with the economic use of human and financial resources in mind. The Decision Tree is described in detail in Ray and Brown (2015). It equips project planners to: (i) assess the climate risks to proposed projects and understand the relative significance of those risks compared to risks of other types (e.g., 8 economic, demographic, political) (ii) incorporate climate information in a targeted and tailored way into a broader (all-uncertainty) assessment of a project’s potential performance (iii) make intelligent modifications to the project design to reduce its vulnerabilities to failure Because water projects are diverse (including hydropower), not all require the same level of effort for climate change risk assessment and risk management. The Decision Tree is designed to be applicable to all projects and as such it leads a planner to allocate the climate risk assessment effort in proportion to each project’s potential sensitivity to climate risk. To do so, the process is hierarchical—with different stages or phases of analysis triggered or not, depending on the findings of the previous phase. As shown in Figure 1, the procedure consists of four successive phases— Phase 1: Project Screening; Phase 2: Initial Analysis; Phase 3: Climate Stress Test; and Phase 4: Risk Management. The project moves through only as many phases of the decision tree as are justified by need. Though not emphasized in Figure 1, the overall procedure includes a feedback loop that addresses monitoring and evaluation, essential aspects of periodic project performance review in a changing climate. For more details on each phase, see Ray and Brown (2015). Figure 1. Decision Tree Schematic (Water Partnership Program, 2015) Phases 1 through 3 of the Decision Tree concentrate on assessment of risks to a 9 proposed (or pre-existing) project. Phase 4 shifts to management of those risks, as well as consideration the opportunities available for improved performance through modifications to the design or operation of the project. Great progress has been made recently in the advancement of tools and techniques for decision making under uncertainty (e.g., multi-objective robust optimization, regret minimization, trade-off analysis, real options analysis), meaning that for the accomplishment of Phase 4 of the Decision Tree, there are a variety of tools available. One particular aim of the decision tree is to help the project planner choose the Phase 4 tool that would be best suited to the problem at hand. This document describes the application of the Decision Tree to the UAHP. Though the Decision Tree was developed for water resources projects generally, and not hydropower projects in particular, the UAHP is a fitting first application of the Decision Tree. In the course of application, hydrologic and water resources infrastructure (hydropower simulation) models were used, a climate stress test was designed and implemented, and information science algorithms employed. These are all described in detail in the following chapters. This report accomplishes the dual purposes of: 1) carefully and thoroughly exploring the climate and non-climate risks to the proposed project, and 2) demonstrating the decision tree through practical application. Subsequent applications of the Decision Tree will be to non-hydropower projects. In order to accomplish the dual purposes, the study steps through all four phases of the Decision Tree. The choice to continue into later phases of the Decision Tree, after climate risks were shown in Phase 2 to be low, was made in response to the requests of the investors and stakeholders. As articulated by the stakeholders, for the UAHP the most critical practical challenge faced at the current stage is the choice of project options in terms of installed capacity. The project team therefore intended from the beginning to go beyond an evaluation of climate risks to the pre-feasibility design (phases 1-3), and evaluate the cost-benefit-risk tradeoffs of alternative design sizes. This further satisfied the desire of the project team to evaluate the results of the Decision Tree at all phases in order to explore the relative merit of increased modeling effort. In this case, as described briefly here. The choice to extend the analysis beyond Phase 2 is somewhat counter to the stated goal of the Decision Tree to provide a framework for the efficient and economical evaluation of climate risks, and in this way this particular application of the Decision Tree should not be taken as a template for other studies. However, in general the Decision Tree can be applied flexibly to meet stakeholder needs. Typically, in the first phase, Project Screening, the project would be screened using a worksheet to learn whether it can be excluded from a detailed analysis. For example, a training project does not require further climate change analysis and so the worksheet leads the planner to conclude that climate change risks as low and the project exits the process. This is not the case for UAHP, and so the project proceeds to phase 2. In phase 2, Rapid Project Scoping, a guided desktop analysis is used to decide whether more in-depth quantitative analysis is needed. From this analysis it 10 was shown that climate change was unlikely to result in very significant loss of electricity production and profit. The analysis could have been concluded at this point, but was continued for reasons just described. In Phase 3, a thorough structured analysis is used to identify the climate and non- climate factors that cause undesirable outcomes for the project. This includes the use of stochastic climate/weather generator, hydrologic modeling, hydropower simulation modeling and data mining. The results indicated that while climate risks were notably low for the original design, climate change may present opportunities for greater hydropower production. Thus the analysis proceeded to Phase 4, not due to concerns related to risk so much as concerns that greater opportunities for hydropower development might be missed in the current design (regret). Phase 4 evaluated alternative designs using the same analytical techniques as Phase 3 to identify promising investments that perform well over the wide range of possible futures. Robustness is a criterion that can be used to compare alternative decisions in terms of performance over a wide range of futures. One way to express the robustness of any decision is as a function of the relative number of possible futures (or size of the projected future domain) over which it performs well. Another way, which we use in this analysis, is the minimization of the maximum regret. Regret is the difference between a project’s performance in a given future and the best performance for that future. Throughout this analysis of the UAHP (including its place in a basin-wide development plan), tradeoffs are presented between system performance and risk. Risk is typically defined as a function of probability and impact, and in this analysis, particular attention is given to the risks of: 1) low (or negative) Net Present Value (NPV) of investment; and 2) low dry season hydropower generation. In this case, the use of probabilities was avoided, except in the calculation of the summary metric expected NPV. All climate response surfaces, as well as regret metrics, provide probability-neutral descriptions of risk and opportunity. 11 Box 2: The Decision Tree Method 12 Chapter 3. Background to Case Study: Nepal’s Power Infrastructure The chapter presents background on the UAHP case study and provides a brief review of the energy context in Nepal. It provides motivation for the need for assessment of new hydropower in Nepal. Nepal is a land-locked country that is facing major development challenges. With 27.8 million people, Nepal had a per capita income of US$730 in year 2013. Of the population, 25.2 percent live on less than US$1.25 per day and 82 percent live in rural areas. Endowed with rich hydropower resources, Nepal views hydropower development as its key opportunity for economic growth and human development, as was clearly evident from recent consultations with people at different levels of society at various places across Nepal. The Upper Arun Hydropower Project (UAHP) represents one of several hydropower projects that are in various phases of development, from planning, to design, to construction. Climate change is a prominent concern to all involved in the planning process, as are other concerns related to seismic activity, sedimentation, and the state of future electricity markets. In addition, while benefits from off-grid small and on- grid large hydropower development are expected in terms of accessing modern energy services, generating revenues, creating jobs, spurring economic growth and improving quality of living, the macroeconomic impacts of large scale hydropower investments are yet to be clearly understood. Nepal’s Power Infrastructure Context Nepal is actively exploring the expanded development of its hydropower resources, which are a significant portion of its current national energy generation portfolio (see Figure 2, top left). Because of the large reliance on hydropower, the energy- generating capacity of Nepal is particularly low in the dry season when monsoon flows and glacier melt are not abundant. Demand is fairly steady throughout the year, resulting in a mismatch in the seasonality of energy supply and demand in Nepal (Figure 3). 13 Figure 2. National Electricity Authority (NEA) of Nepal statistics Natural resources available in Nepal for power generation. Nepal’s hydropower potential is estimated at about 84,000 MW theoretically and 43,000 MW economically viable spread across the seven river basins – Kankai Mai, Koshi, Bagmati, Narayani, West Rapti, Karnali, and Mahakali. Hydropower remains the least-cost option for power generation to meet domestic demand and has the potential to make Nepal a power house of the South Asia region. 14 Mismatch Between System Monthly Peak Load Balance in INPS Demand and Generation 250.0 (For 200 MW Demand) Peak Load (MW) 200.0 Power from RoR Plant 150.0 Power Demand of INPS • Excess Capacity During Wet 100.0 Season where as Capacity 50.0 Deficit during Dry Season. 0.0 Shrawan Kartik Magh Bhadra Aswin Mansir Baisakh Poush Falgun Chaitra Jestha Asar • Huge Spill Energy During Wet Month Season with Very Limited Monthly Energy Balance in INPS Energy Production During Dry 160000 Season resulting in Heavy Load 140000 120000 Energy from ROR Plants Energy Demand in INPS Shedding. Energy (MWh) 100000 80000 60000 40000 • Can it be Balanced by PROR & 20000 Storage Projects ??? 0 Shrawan Bhadra Kartik Magh Aswin Mansir Baisakh Poush Falgun Chaitra Jestha Asar Month *from NEA [2014] Figure 3. Seasonality mismatch between system demand and generation Access to electricity services. According to the national census published in 2013, about 75 percent of the population in Nepal is estimated to have connections to grid (about 50 percent) and off-grid (about 25 percent) electricity. Although off-grid connections provide relatively reliable electricity supply in the rural areas, access to the grid in rural and urban areas does not necessarily mean access to electricity due to continuing load shedding. Lack of access to reliable grid-supplied electricity is one of the key obstacles to lifting out of poverty those still living below the poverty line. While Nepal has achieved remarkable progress in off-grid electrification, coordination with grid extension needs to be enhanced through planning future rural electrification to avoid stranded off-grid assets when the grid is extended to the off-grid areas. Energy crisis as a major constraint to growth. The electricity supply and demand gap was about 410 MW in November 2013, when peak demand reached 1,201 MW, resulting in load shedding of up to 14 hours a day. The lack of grid-supplied electricity is a major barrier for Nepal to expand access to quality electricity services, improve living standards, raise agriculture productivity and incomes, and help its youth transit from farming to non-farm employment. Commercial and industrials consumers run captive generators using expensive imported diesel fuel at a very high cost, ranging from US$0.35 to US$1.20 per kWh. This high cost has severely weakened their productivity, competitiveness and ability to expand. Moreover, the lack of job opportunities has pushed more than 5 million Nepali laborers to work 15 overseas. Agriculture is the sector most contributing to the GDP, but raising productivity through irrigation is also constrained by the lack of electricity. While the energy sector has the potential to become a major source of income and to bring Nepal to middle-income status, the sector currently relies on government subsidies to survive since the price charged to consumers for electricity and imported fuel3 consumption does not cover the high cost incurred due to inefficiency of the sector. Subsidies to the energy sector have become a major drain on scarce public resources. Government short-, medium-, and long-term strategies and actions. While the GoN has been continuously promoting off-grid renewable energy development to expand access to energy services in rural areas, it is re-shaping and implementing a strategy for grid-side solutions to deal with the energy crisis in urban areas and, eventually, achieve the long-term power sector objectives. The strategy and actions are to (a) reduce load shedding in the short term, through rehabilitation of existing generation plants to increase supply, rehabilitation of distribution network to reduce system losses, adding generation capacity that can be quickly installed (25 MWp grid- connected solar farm), issuing tax policy to support roof-top solar in urban areas, and launching a power sector reform to address key sector issues; (b) expand access to grid electricity services and reach supply-demand balance in the medium term, through grid extension, commissioning of hydropower under construction (about 1,500 MW), including the Upper Tamakoshi (456 MW) developed by the NEA subsidiary for commission in 2017, and the first 400 kV cross-border transmission line for power import from India (up to 1,000 MW); and (c) ensure universal access to sustainable, reliable and affordable electricity supply in Nepal and generate export revenues to sustain economic growth through development of its huge hydropower potentials and integration into the South Asia regional power market. Regional and GoN commitment to power sector development. The signing of the PDAs of export-oriented Upper Karnali and Arun III projects and a paradigm shift in regional cooperation with signing of the PTA between Nepal and India shows strong commitment from the GoN to expedite the development of hydropower sector. This commitment was further reinforced on multilateral basis through the South Asian Association for Regional Cooperation (SAARC) Framework Agreement on Cooperation in energy, which was signed by all SAARC countries. 3Nepal’s fuel imports, at US$1.36 billion in FY 14, constitute the single largest import item, exceeding by a wide margin of its revenues from all exports of goods and services. 16 The Arun River, one of the main tributaries of the Saptakoshi River, originates at a glacier on the northern slope of Mount Xixabanma in Tibet. The river flows approximately 300 km eastward across the Tibetan Plateau at an elevation of about 4000 m before crossing the Himalayas and plunging into Nepal where it flows in a narrow 30 to 60 m canyon at a very steep slope where the proposed project is located. The dam site is located about 15 km downstream of the border with Tibet. The river has the catchment area of nearly 25,700 km2 (only 400 km2 of which is within Nepal) and an average run-off approximately 200 m3/s. Very little development has so far occurred in the catchment area (see Figure 4). Figure 4. FAO landuse in the Koshi basin The Upper Arun Hydropower Project (UAHP) The Upper Arun Hydroelectric Project (UAHP), which is located on the upper reach of the Arun River (see Figure 5), has been identified as one of the most attractive projects in the Eastern Development Region of Nepal. In 1985, the project site for UAHP was identified during the Master Plan Study of the Koshi River (Water Resources Development, JICA). In the summer of 1986, NEA conducted a Reconnaissance Study. The Joint Venture of Morrison Knudsen Corporation, Lahmeyer International, Tokyo Electric Power Services Co., and NEPECON on behalf of Nepal Electricity Authority (NEA) carried out a feasibility study of this project in 1991. NEA has given priority for the development of this project to augment the energy generation capability of the integrated Nepal Power System due to its relatively low cost of generation and availability of abundant firm energy. The feasibility study carried out in 1991 chose an installed capacity of 335 MW for the Peaking Run-of-River type UAHP. The design discharge of the project was 78.8 m3/sec, and it was expected to generate firm energy of 2050 GWh per year (see Figure 6 and Table 1). In 2008, NEA obtained license from GoN to develop the 17 UAHP. The updated estimated cost was US$446 Million (335 MW/2,050 GWh). In 2011, NEA opened a review of the project. Figure 5. Elevation and glacier cover within the Koshi basin Fifty-two hydropower project sites have been identified within the Koshi basin alone, with a total production potential of 10,909 MW. Thirteen of those are priority projects with a total potential of 8,473 MW. The proposed UAHP is amongst the highest priority projects. Fed by mountain glaciers and aquifers, the most attractive feature of this project is the firm energy it produces throughout the year. Some hydropower experts quote that perhaps there is only one plant in the world with a combined head and flow magnitudes greater than the UAHP – Kemano in British Columbia. The Project may be able to provide up to 85-90 percent of the firm installed capacity even in the dry season of November-March, when load shedding becomes severe in Nepal. 18 Figure 6. Location and plan view of the UAHP (NEA 2014) 19 Table 1. Design parameters for UAHP prefeasibility design – 335 MW PROR Parameter Details Year 2011 Review Comments Catchment Area 25,700 sq.km. (98% in China, Tibet) Firm Discharge 58.7 cumecs (Q95) – Comparatively High Flood Discharge 4000 cumecs (PMF) – Comparatively Low Design Discharge 78.80 cumecs (about Q70) Net Head 492 m (Gross Head 509 m) Installed Capacity 335 MW (at Q70) 750 / 2000 MW at Q40/Q25 Total Annual Energy 2050 GWh (Firm Energy) 2597.3 GWh (Average) Storage for Peaking 2 hours Geology Relatively Sound Project Cost 479.6 M US$ (year 1991) Updated cost: 445.54 M US$ Interconnection to Tumlingtar 220 k V S/S; 45 Grid km from Powerhouse Project 2022/23 - Tentative Commissioning Year Average Energy Tariff 2.1 NRs. Per kWh [NEA Scenario (interest rate = 8.5%, ROE = 14%)] Technical features Access. A road exists up to Num Bazaar, close to the dam site of Arun III HEP. From Arun III dam site, a 23.4 Km long access road and a bridge is required to reach the powerhouse site of UAHP. The elevation at the site ranges from approximately 1100 m at the proposed site of the powerhouse to over 2700 m at headwork site along the ridgeline separating the powerhouse and dam site. The design discharge will be diverted through an intake tunnel to three underground de-sanding basins, a headrace tunnel of 7.8 km (5.5 m diameter), 18 m diameter surge tank, two 450 m long steel lined pressure shafts leading ultimately to the underground powerhouse. In the feasibility study, the project was designed to provide 335 MW peak power (at 95% reliability) to the Nepal system with a maximum demand of 1000 MW. Excess energy was to be exported to India (in the wet season). According to NEA’s 2008 annual report, the maximum demand of the Nepal integrated system will be about 2,050 MW in 2019/20, and the corresponding energy demand will be about 9,563 GWh of which about 2,000 GWh is currently being met by existing hydro installations of 700 MW. Based on the recent review by NEA the annual generation may be increased to 2,597 GWh (from previous value of 2,050 GWh) out of which 742 GWh is dry season energy. 20 The application of the Decision Tree to assess and manage climate risks to the UAHP is described in the following chapter. Chapter 5 then provides a basin-wide overview of hydropower development opportunities. 21 Chapter 4. Case Study Application: The Upper Arun Hydropower Project This chapter describes the application of the Decision Tree to the Upper Arun Hydropower Project (UAHP). The objective of the analysis is to assess whether the proposed project faces climate change risks, and if so, what possible adaptations may be needed. Phase 1 of the Decision Tree – Project Screening In Phase 1 of the Decision Tree, a project is classified according to its potential sensitivity to climate change. Though the 335 MW UAHP has been repeatedly identified as a very attractive potential investment, it must also be classified as having “substantial potential sensitivities to future uncertainties”. Prior to any quantitative analysis, it can be seen that the UAHP project is defined by:  Large capital investment  Significant potential effects on other sectors  Significant reliance on market forces (and international trade arrangements) outside of the control of project planners  Long-lifetime project (> 20-30 years) during which climate and other conditions could change substantially  The uncertain climate factors4 o Temperature (as affects glacier melt contribution and evapotranspiration) o Precipitation (intensity, duration, and frequency) o Monsoon patterns o Glaciers – how much to begin with, estimates on melting/movement Conclusion to Phase 1 Project Screening: The UAHP has potential climate (and other) sensitivities that need to be explored further in Phase 2. Phase 2 of the Decision Tree – Initial Analysis In this phase, the Initial Analysis is a rapid project scoping to determine if the project 4This study did not explicitly evaluate the effect of climate change on natural hazards such as glacier lake outburst floods (GLOFs), flash floods, or landslides. The answers to questions regarding these and other phenomena would be of great value, but were outside of the scope of this study. 22 is sensitive to climate change. This is accomplished through desktop analysis using general planning factors and existing reports on expectations of climate change for the region of interest. In addition, to place climate change uncertainties into an appropriate context, other factors that are also uncertain and could influence the success of the project are identified. If sensitivities are found and judged to be important relative to other uncertainties, then the process proceeds to Phase 3, the Climate Stress Test. A hydropower project would generally be expected to be sensitive to possible changes in climate, such as changes in precipitation that lead to increases or decreases in expected streamflow and thus hydropower generating potential. However, in the case of run-of-the-river hydropower, the flows may be large enough such that the project is relatively insensitive to the possible changes. Thus for hydropower, the rapid project scoping process is used to determine whether the project is insensitive to climate change, and thus wouldn’t need further analysis. In addition, in some cases there may be other factors that require equal or greater attention than climate change. In the rapid project scoping, other factors that are uncertain with potential significant future changes are also identified. This may be conducted through a desktop analysis based on the experience of the planner or through a stakeholder consultation process. In the case where climate change uncertainty is small relative to other important uncertainties, the process can be exited and a broader uncertainty analysis considered. In the case of the UAHP, the important contribution of snowmelt and glacier melt to streamflow, and the potential sensitivity of these factors to changes in temperature and precipitation, indicate that climate may require careful consideration. There were also clear additional factors that were highly uncertain and should be considered in the analysis, such as uncertainties in electricity prices and the risk of cost overruns. Therefore, Phase 2 consisted of gathering climate and hydrologic data for desktop analysis of climate sensitivities and also identifying other key uncertainties through stakeholder consultation. The climate sensitivity of the project was assessed using the hydrologic record and simple planning relationships. First, the planned capacity for UAHP was assessed relative to historical streamflow values. Next, the streamflow elasticities to changes in precipitation and temperature were calculated and applied to hydropower potential using a planning equation. This allows a quick assessment of how climate changes could impact the project. Because these relationships are based on planning factors rather than analysis tailored for the location, they are broad indicators only. Nonetheless, they can indicate the need to proceed to Phase 3. 23 Figure 7. Uwagaon hydrology (MCM/month) As shown in Figure 7, the Upper Arun streamflow is highly seasonal, with monsoon- and melt-water augmented summer flows reaching peaks almost ten times as large as average dry season (November-March)5 flows. This is in sharp contrast to the low interannual variability (cv,interannual = 0.18, see Figure 53 in the Appendix). As a result, the low season flow amount is critical to choosing the generating capacity of UAHP. Figure 8 shows the streamflow exceedance probability for the Uwagaon stream gage, just downstream of the proposed site of the UAHP. Average daily streamflow values in m3/s were ordered and assigned exceedance probability using Weibull plotting position. The values presented are in close, though not perfect, agreement with those presented by NEA in reference to the prefeasibility study, in which the Q70 flow was estimated to be 78.8 m3/s. The discrepancy is likely due to small differences in the data. The data used for this analysis were from the Nepal Department of Hydrology and Meteorology, site 600.1, Uwagaon, with coverage (after disqualification of data from 2007 and 2008 during a QA/QC process) from April 1986 through December 2006. 5The dry season in Nepal, as reported by NEA, is from Maingsir to Falgun. Mansir occurs form approximately mid November to mid December, and Falgun from mid February to mid March. Based on this information, as well as inspection of the hydrograph, we took the dry season to be November to March, inclusive. 24 Figure 8. Uwagaon streamflow exceedance probability (Apr 1986-Dec 2006) The streamflow data in Figure 8 were then converted to GWhr/day using equation (1), which is a simple planning relationship: (1) with net head, H = 492 meters, and the efficiency of the conversion of mechanical energy into electrical energy, e = 0.9. Equation (1) is derived from the fundamental physics describing the translation of potential energy into kinetic energy. A cubic meter of water, weighing 103 kg, falling a distance of one meter, acquires 9.81 x 103 joules (Newton-meters) of kinetic energy. A Watt is a unit of power, equal to a Joule of energy expended per second. Equation 1 expands the example of a single cubic meter of water falling a single meter to the case of many millions of cubic meters of water falling many meters each day. The coefficient 0.002725 is an aggregate unit conversion. Detailed explanation of the derivation and utility of (1) is available in Loucks and van Beek [2005]. The potential (un-capped) hydroelectric power exceedance probability is presented in Figure 9. The horizontal red line in Figure 9 shows the GWH70 (daily hydropower with exceedance probability of 70%) for the 335MW facility based on the NEA’s feasibility study (Q70 = 78.8 m3/s; GWH70cap = 8.04). It nearly passes through, but is somewhat less than, the red dot for GWH70 = 9 calculated using data available for this analysis. 25 Figure 9. Theoretical potential (uncapped) hydropower exceedance probability Figure 10 provides perspective on the capacity of the 335 MW facility relative to the seasonal peak flows of hydropower potential at the site of the UAHP. The time series of hydropower production in Figure 10 is calculated by applying (1) to the time series of streamflow at Uwagaon. The horizontal red line in Figure 10 locates the capacity of the 335 MW facility. Hydropower production potential in excess of the red line would not be generated with a 335 MW facility, implying there would be much potential untapped. Thus consideration of larger capacities may be warranted. 26 Figure 10. UAHP, 335 MW PROR @ Q70, max daily GWhr/day generation relative to daily timeseries of GWhr potential (red line) The elasticity of hydropower performance, measured in terms of NPV, to changes in climate is given in (2). By dividing relative changes in system performance by relative changes in streamflow while holding all other factors constant, it is possible to quantify (in a first-order, approximate sense) the influence of changing streamflow on system performance. If system performance is inelastic to changes in streamflow, then it can be inferred that it is also inelastic to changes in climate. This shorthand analysis can be done with a simple water system model (see Equation 1) by applying change factors on historical streamflow data, without hydrologic modeling or climate data analysis. (ℎℎ − )/ = (ℎℎ − )/ (2) What is needed is a sense for the appropriate range of hydrologic change factors. Based on previous experience in the region, and general familiarity with regional GCM projections, we explored the response of the system to an increase by up to 50% (Qhigh) and decrease by up to 30% (Qlow). This is an expansive range that goes much beyond what regional climate change studies suggest is likely in the next 30-40 years. The NPV was calculated using (3): 27 T Ct NPV   C t 1 1  r t 0 (3) where: Ct= net cash inflow during the period. C0= initial investment r= discount rate t= number of time periods (in our case, months) Based on this first-order approximation, we estimated the Q to be approximately 0.3. Table 2 Design 1, 335 MW PROR, baselines Parameter Value Capital cost $450M O&M cost 2*(125000*(kWhr_cap/1000/24)^0.65) Energy gen. cap. 8.04 GWhr/day Discount rate 5% Economic lifetime 30 years (5 yrs construction) Elasticity ( Q) 0.3 Because a decrease in streamflow leads to a substantially large decrease in system performance over a range that is reasonable given first-order approximations of plausible climate change (based on regional experience), a more in-depth Phase 3 climate change stress test may be warranted. However, Figure 11, with a range of precipitation and temperature change that goes well beyond what is plausible in the lifetime of plant, shows that there is little risk of the project suffering losses due to climate change. It is left to the project planner to decide whether or not further analysis in Phase 3 is needed. In the case of this study, the analysis was continued into Phase 3 at the request of the stakeholders, and investors, and in satisfaction of the intention of the project analysts to evaluate the output of the Decision Tree process through each phase. 28 Figure 11. Elasticity of NPV to changes in precipitation (left) and temperature (right) It would be perfectly reasonable to jump out of the Decision Tree based on the findings of Figure 11, labeling the project “climate robust”, and the choice to continue to Phase 3 in this study should not be taken as a mandate to other similar applications. Phase 3 of the Decision Tree – The Climate Stress Test In Phase 3, the Climate Stress Test is applied to the planned project to systematically identify its vulnerabilities so that they can be assessed and managed if necessary. Although a focus of the analysis is climate change, the stress test can also include other uncertain factors, such as those financial, construction and natural hazards identified in Phase 2 of the UAHP analysis. The climate stress test consists of a carefully designed sensitivity analysis combined with a data mining algorithm to identify problematic scenarios for the project. Here, scenario refers to a range of factor (or factors) over which a project does not meet its performance objectives. Once these problematic scenarios are identified, they are investigated to determine whether they represent real concerns that should be addressed (leading to Phase 4) or are so unlikely that they can be safely reviewed and discarded. To collect necessary hydrologic data and models, and to identify key uncertain factors in addition to climate change, an initial project workshop with key stakeholders was held in Nepal in September 2014. Through a consultative process, a set of information relative to the planning of the project was generated during the workshop. Stakeholders were convened representing the Government of Nepal (GoN), local NGO’s, local academics, and hydrologic, climate and energy services of the government. The information was organized according to the Decision Matrix presented in Table 3 which shows the key categories of information that are relevant to the analysis formulation. 29 Table 3. The Decision Matrix developed during the initial consultations with Nepalese experts and NEA Uncertainties Investment and Policy Options Natural Uncertainties Upper Arun HP Precipitation - How much, when, where 335 MW (Q70) – original design Impacts of future climate change 750 MW (Q40) – possible alternative Temperature – Impact on glacial melt and 2000 MW (Q25) – possible alternative evapotranspiration Sedimentation Seismic risk and disasters Nepal Future System and Operations National markets International agreements Electricity prices Project Variables Capital costs Lifetime of the projects Discount rate Metrics of Success Existing Models and Data Hydropower Performance Watershed System Net Present Value WEAP model built by IWMI Power generation Dry season Hydrological model Wet season SWAT model built by IWMI (basin scale) – Annual baseline flows (1971-2000) Key uncertain factors that could potentially affect the project are listed in the first box (upper left). There are some risks generally categorized as “natural uncertainties,” some others related to the future energy system and effects of prices, and typical uncertainties associated with large projects, such as capital costs, lifetime and the choice of discount rate. The second box (proceeding clockwise) identifies the options available to the planner. In this case the options considered related to the generating capacity of the project. The initial design capacity for UAHP was 335 MW but the capacity could be varied and two larger capacities were added as options (later in the analysis, additional capacities between the end points were added in response to stakeholder input). Next, the existing models and datasets were identified. In this case, although there were existing models, it was clear that further model development would be required because none were specifically designed for the UAHP location. Finally, the metrics by which planners and stakeholders would measure success were defined. In this case, decision makers were concerned with two key metrics: a) whether the investments passed a cost-benefit test, i.e., a positive NPV, and b) the production of electricity, including in the dry season. These questions can help decision-makers more fully understand the context of the investment and also the degree to which climate change is an important source of uncertainty. 30 The implementation of the climate stress test for UAHP required the development and application of several models and algorithmic tools. These include:  Stochastic climate/weather generator – generates wide range of weather time series representative of plausible climate changes  Hydrologic model – estimates streamflow for a given climate/weather timeseries  Hydroelectricity works simulation model – estimating hydropower production for a given inflow  Economic model to evaluate the economic performance of the UAHP under multiple climate and economic futures  Patient Rule Induction Method (PRIM) – data mining algorithm for identification of problematic scenarios The climate stress test was conducted in two phases. First, the effect on the project of a wide range of climate changes was determined. Second, the climate effects were incorporated into a climate stress test that also included the non-climate factors previously identified. For each of the non-climate factors, a plausible range was specified. The climate stress test then sampled the climate and non-climate uncertainties to identify the factors or combinations of factors that led to problematic outcomes. The metrics for assessing outcomes included the NPV and hydropower generation, including special attention to low season production. Effects of Plausible Climate Changes The key to the Decision Tree’s approach to assessing the effects of climate change is that the analysis is conducted without the need for climate change projections and all the uncertainties and processing that entails. Rather than assessing the sensitivity of a project to the climate projections that happen to be used, this process assesses sensitivity to climate change itself. It does so using a stochastic climate/weather generator specifically designed for this purpose. No pre-judgments or limiting assumptions are employed. Instead, the climate is varied systematically until the vulnerabilities of the project are revealed. Then, if the vulnerabilities are deemed plausible, further investigation into the likelihood of those specific climate changes can be initiated. The stochastic climate/weather generator provides a comprehensive exploration of climate-related vulnerabilities, including natural variability and climate change trends, with additional climate-related uncertainties such as climate change seasonality shifts, changes to precipitation variability patterns, and glacier-related assumptions (i.e., initial mass and ablation rate) explored as needed. The range and type of changes that are explored is based on a review of the existing literature on plausible changes to climate in the region. The goal of this approach is to create a clear understanding of the impacts of climate changes on a project without the 31 uncertainties associated with climate change projections. The results show the vulnerabilities (or opportunities) of a project to climate change, rather than to the climate change projections that happen to be used. Climate change everywhere is difficult to predict, but in South Asia future climate is deeply uncertain. There is great uncertainty in climate projections due to the complex topography, importance of the South Asian Monsoon, uncertainty associated with glacier volumes. This is a difficult location to make climate predictions. Historical analysis has shown increasing temperatures and no clear signal on precipitation. Glaciers are growing in some areas and receding in other areas, and more are receding than growing. Streamflow generally seems to be increasing. All of these factors are explored carefully and comprehensively in the analysis. The climate analysis consisted of the creation and implementation of the stochastic climate/weather generator, hydrologic model, and hydropower simulation model. These are each described in turn. The process is shown schematically in Figure 12. Figure 12. Schematic representation of the Phase 3 process Stochastic Climate/Weather Generator The stochastic climate/weather generator was designed to create a dataset of weather time series that are used to comprehensively explore the effects of climate 32 change on the project’s performance. These include changes to mean climate and careful sampling of climate variability. This provides a much more systematic and comprehensive assessment of climate change than would be provided in the typical GCM-led analysis, and it also much easier to implement. The stochastic climate/weather generator uses resampling of historical local climate, maintaining critical statistics (e.g., mean, standard deviation, serial correlation, skew) and spatial correlations. The weather generator is entirely forward looking. It applies change factors to historical traces of precipitation and temperature and perfectly preserves correlations between the two. There are few assumptions that cannot be varied; it consists of several “knobs” that can be used to change variability or mean conditions. The biggest assumption is that the historical conditions are clearly represented by the observed historical data, and that the climate change is consistently expressed over the region of interest. The method is described in detail in the publication Steinschneider and Brown (2013): "A semiparametric multivariate, multi-site weather generator with low-frequency variability for use in climate risk assessments", Water Resources Research. In the case of this analysis, the weather generator resampled from 60 years of bias- corrected gridded temperature and precipitation data from the APHRODITE database (see Appendix for in-depth discussion of climate data availability and QA/QC). The temperature and precipitation data were resampled in seasonal (wet season/dry season) random sets – the dry season of 1983 might follow after the wet season of 1995, which follows after the dry season of 1966, for example. Using the weather generator algorithm, thirty of these climate traces were generated. The result is traces of climate that represent alternative possible realizations of historical climate due to natural climate variability. Then, to create weather traces consistent with climate changes, change factors were then applied to create weather time series that were consistent with climate change of 0-8 degrees C of increase in temperature and a 40% decrease to 40% increase in precipitation relative to historic values (each applied uniformly throughout the year). Note that these changes go beyond what is plausible for regional climate change. The range is chosen broadly enough to ensure no vulnerabilities are missed. Ultimately the end points of the range do not influence the analysis (since it is not probabilistic) unless they were deemed too narrow. In this way the problem of prescribing the range of future climate change is avoided. The resulting dataset was then used with the hydrologic model to understand the hydrologic response to these climate changes. Hydrologic Model The glacio-hydrologic distributed model applied for this analysis is the HYMOD_DS. The HYMOD_DS is the modeling system created by the University of Massachusetts Hydrosystems Research Group and applied to mountainous regions with sparse data (see Figure 13). The model is designed for parallel processing on supercomputers, allowing calibration by the Massachusetts Green High-Performance Computing 33 Center (MGHPCC). The HYMOD_DS is well suited for analysis in the Himalaya, such as the UAHP. The prototype of the model was built for a World Bank supported study of the Kabul Basin (Wi et al., 2014) and the Brahmaputra Basin (Yang et al., 2014b). The original HYMOD model (Boyle, 2001) is a lumped parameter, rainfall excess model composed of a soil moisture accounting module. Wi et al. (2014) introduced a routing module, which allows water balance results from each cell to be hydrologically connected to the basin outlet, creating a spatially-distributed version of the model. In addition, a temperature-based snow/glacier module was developed to explicitly model the dynamics of melting snowpack and glaciers, with resulting contributions to streamflow. The model structure of the HYMOD_DS modeling system is described in detail in Wi et al. (2014). Figure 13. Diagram of the UMass hydrologic model with glacial component The model calibration and validation is presented in Figure 14. A measure of model fit, the Nash Sutcliffe Efficiency, is 0.91 for the calibration phase and 0.75 for the validation phase. Generally, values of NSE above 0.7 are considered to be very good, and values over 0.9 are considered to be excellent. The calibration and validation of the model give us confidence to use it to explore climate change effects in this basin. 34 Figure 14. Calibration and validation of the hydrologic model Figure 15 presents output of the hydrologic model showing the component contributions of subsurface groundwater flow, snow/glacier melt, and rainfall runoff to streamflow at Uwagaon station. The historical time series shows little change in glacier/snow contribution. The seasonal hydrograph shows that the greatest contribution of meltwater occurs in April/May/June, and the greatest contribution of rainfall runoff occurs July/August/September. Streamflow from November through February is supported almost exclusively by groundwater baseflow. Figure 15. Hydrologic model calibration results showing percent contribution to 35 streamflow Hydropower Model The hydropower model was created in the R modeling environment. Because this project is a run-of-the-river facility, the model was a simple function of inflows and the generating capacity of the project design. Results: Climate Change Effects The results of the climate change analysis are presented as climate response functions, which show the changes in the impacted variable as contours over the range of climate change evaluated. Figure 16 presents the streamflow output of the hydrologic model for the range of climate change considered. The climate change range used was 0-8 degrees C of increase in temperature and a 40% decrease to 40% increase in precipitation relative to historic values (each applied uniformly throughout the year). Though the response of the system to increases in precipitation is fairly monotonic and predictable (more precipitation leads to more streamflow), Figure 16 shows an interesting phenomenon related to temperature. As temperature increases up to approximately 3 degrees C, streamflow increases. This is due to increased meltwater from glacier melt. However, with average temperatures in excess of 3 degrees warmer than historic, the net contribution from the glacier decreases. This is because the glacier recedes to critically low levels of glacier mass (see Figure 17). In the case of the dry season, the effects of temperature are always negative (reducing streamflow). Nonetheless, the temperature effect is quite small in comparison to precipitation. Figure 17 shows the response of the glacier mass within the basin to changes in climate. Temperature is the dominant effect, with higher temperatures causing reduction of glacier mass. Precipitation has a small positive effect. GCM projections centered on the year 2050 are superimposed on the surface to provide their indications of future climate. According to those projections, the glacial mass in the basin will be at about 50-90% of the current mass by the year 2050. 36 Figure 16. Response of streamflow to changes in climate by annual total, dry season (lower left) and wet season (lower right). Changes in precipitation are shown on the x axis and changes in temperature are shown on the y axis. Contour colors represent increasing streamflow, blue in excess of historic mean, and red less than historic mean. 37 Figure 17. Glacier area response surface - Year 2050 Glacier Area (as % Year 2014 Glacier Area) Figure 18 shows the change in streamflow seasonality due to changes in temperature. As temperature increases, winter (dry season) flow increases (as precipitation that falls in the winter is not stored as snow, but runs off immediately into the stream). Peak flow also increases due to enhanced glacial melt. There is also a shift in peak meltwater contribution from Apr-May to Mar-Apr. Generally, temperature increases lead to very small effects on streamflow seasonality with an increase in low season flow under strong warming only (+4-6 degrees C). 38 Figure 18. Seasonality shift in streamflow (MCM/month) with increasing temperature (no change to precipitation). Of greatest interest is the effect of plausible climate change on the economic performance of the project. Figure 19 presents this important result. It shows the climate response surfaces for the NPV of the 335 MW pre-feasibility UAHP design. Holding all other variables constant, the NPV is positive over the entire range of the climate stress test with values of $300M-$800M. (Assumptions: a design life of 30 years, plant load factor6 of 0.75, baseline capital and operations and maintenance costs, dry season selling price of $0.084/kWh, wet season selling price of $0.045/kWh, and discount rate of 5%). 6Plant load factor is here taken to be the plant availability considering shut down for sediment management. For example, a plant load factor of 0.75 means that the plant is shut down one quarter of every month for management of sediment (e.g., flushing) and other purposes. 39 Figure 19. Climate change response surfaces – NPV (average values from 2020-2050) Figures 16-19 provide a comprehensive picture of the hydrologic response to climate change and the response of the economic value of the project to climate change. Figure 19 shows that for the wide range of climate changes considered, the project always has a positive NPV. This implies the project is robust to climate changes. For additional information related to the plausibility of the specific climate changes explored, mean changes from GCM projections can be added to these plots. Prior to using climate projections it is helpful to assess their skill in reproducing the local climate. A simple method for doing so is comparing the observed mean climate for the location of interest with the GCM simulations of the same. This illustrates the bias in each model. If the models show strong biases the information available from them should be interpreted cautiously. Additional means for evaluating model performance include process-level assessments (e.g., ability to reproduce monsoonal flow) but require additional effort. Figures 20-22 show comparisons between the mean climate of the UAHP contributing area with GCM simulations of the same. In general it can be seen that there are extremely large biases, especially in precipitation. Figure 20 shows simulations of monthly precipitation over the historical period in comparison to the observed precipitation. The average model bias is on the order of 75% and several have biases of greater than 100% (meaning the simulations double (or more) the observed precipitation. Figure 21 shows the same for temperature, although in this case the models tend to be negatively biased (underestimating actual temperatures) and the biases are smaller. 40 Figure 20 Hindcast of 48 GCMs for Annual Mean Precipitation – Upper Arun River Basin Figure 21 Hindcast of 48 GCMs for Annual Mean Temperature – Upper Arun River Basin Climate projection-based analyses typically attempt to infer climate changes from model projections by ignoring biases and calculating the change between the historical simulation and a future projection that incorporates increasing greenhouse gas emissions. However, if the changes that are calculated are smaller than the biases, it is difficult to infer a direction of change since the changes are within the range of the model errors (and thus cannot be separated from noise). Figure 22 shows the comparison between the model biases (red) and the future projections (blue) in comparison to the historical observed climate (black). The figure illustrates that the biases are of equal or greater magnitude than the changes. For example, in the case of precipitation, the average bias is approximately 50% and the average change is approximately 1%. 41 Figure 22 Hindcast (1971-2000) and Forecast (2036-2065) of 48 CMIP5 GCMs (1971-2000) - Upper Arun River Basin Given that note of caution, the changes derived from the projections were calculated. Figure 23 and Figure 24 show the ranges of projections of temperature and precipitation changes (and seasonal effects) for the available GCM projections. On average, the multi-model, multi-run ensemble of GCMs projections show no clear signal in terms of precipitation change. Temperature projections generally show moderate increases in temperature, as would be expected. More warming is anticipated in the winter than in the summer. 42 Figure 23. CMIP5 precipitation change projections: 1971-2000 vs 2036-2065 Figure 24. CMIP5 temperature change projections: 1971-2000 vs 2036-2065 Figure 25 brings together the results of the climate stress test with the mean climate changes derived from statistically downscaled climate projections for the basin. The figure shows that for most climate projections, the NPV of the 335 design is little 43 changed from the baseline estimate, and in many cases, has higher NPV. There are few cases where the NPV is lower than the baseline estimate; in all cases the NPV remains positive. Figure 25. NPV of the 335 MW design with CMIP5 climate change projections (centered 2050) superimposed. In general, the results can be interpreted as not providing any strong concerns that there are problematic climate changes expected, given the response to changes in climate shown in Figure 16, and most important, the consistently positive NPV shown in Figure 19 for the widest range of climate changes. The results presented here are averaged across 30 realizations of internal climate variability (30 plausible re-creations of climate timeseries with statistics similar to the historic), the full range of which are presented in Table 4. In keeping with the low interannual variability previously described, the differences between results across the 30 traces are small. 44 Table 4 Internal variability across 30 traces of stochastic climate variability Result Mean 8226 MCM/yr Annual Flow Max 8575 4% Min 7940 -3% Mean 585 MCM/yr Dry Season Flow Max 612 5% Min 540 -8% Mean 7641 MCM/yr Wet Season Flow Max 7965 4% Min 7378 -3% Mean $501M NPV Max $529M 6% Min $476M -5% It should be noted that the climate change factors used in this analysis were uniformly applied throughout each year. An item for deeper research would be an exploration of seasonally-specific change factors to reflect a potential non-uniform shift. Given the seasonal structure of the existing weather generator, the tools are available to support such an exploration. It may therefore be worthwhile, in future research, to evaluate the potential effect of ranges of natural climate variability on system performance, however, such an analysis would take almost no help from GCMs, which simulate seasonal and sub-seasonal climate dynamics in this region with very little credibility. In summary, the 335 MW pre-feasibility design appears to be quite robust to climate changes. However, the effect of climate change in combination with other uncertainties may reveal problematic scenarios. For a full assessment of the robustness of the project design, these non-climate factors are added to the analysis. Full Climate Stress Test including Non-Climate Factors To expand the vulnerability analysis to include non-climate uncertainties, variables that are typically included in a hydropower project’s analysis, and that were identified by stakeholders were added to the analysis. These are the lifetime of the investment, capital costs, electricity prices, the discount rate, the plant load factor, and the electricity generation. Table 5 shows the original value for each of the uncertain factors and the expanded range of plausible values. Figure 26 shows this visually. Note that the ranges are not an attempt to bound the plausible range of each variable. Due to the deeply uncertainty nature of the variables this would be inherently problematic. Human nature causes us to underestimate uncertainty and the resulting ranges could be too small. In addition, trying to select plausible ranges of deeply uncertain 45 variables could lead to endless discussion of what the end points should be. Instead, the ranges are chosen to be so broad that no plausible risks will be missed. Thus the ranges are designed to extend into the implausible. The analysis will show which values of specific variables are relevant to the design. At that point judgments can be made about whether those key values of specific variables are plausible or not. The values’ ranges were informed by the literature and historical data, and from consultations with energy experts. For example, delays in the implementation of hydropower investments can lead to extreme examples in costs. For one, the Marshyangdi Dam’s final capital costs tripled so experts of Nepal hydropower sectors suggested we considered a range of capital costs from a lower bound of the options’ stated costs, to an upper bound of 300% of the stated costs. Similarly, the electricity price may vary if the GoN were to begin exporting electricity to India – in which case, it could increase significantly, up to 0.15 USD/kWh, according to experts. The Plant Load Factor and lifetime ranges include considerations of different options of sediment management and sediment damage effects. With optimal sediment management, the plant shuts off a few hours per months in the wet season (Plant Load Factor <1). As a result of optimal management, a plant’s lifetime may increase and conversely, bad sediment management can decrease it significantly. Table 5. The values utilized in the original analysis and in this study’s uncertainty analysis. Original This Analysis Analysis A. Projected B. Range A. Uncertainties Value min max Wholesale price of electricity (US$/kWh)a x3 times Wet Season (Apr-Oct) 0.045 0.045 0.135 Dry Season (Nov-Mar) 0.084 0.084 0.252 Discount Rate 0.12 0.03 0.12 Estimated Lifetime of the Plant (years) 30 15 36 Plant Load Factor 0.90 0.60 0.90 Capital Costs (2013 US$) x3 times 335 MW 446M 446M 1338M Climate Change:7 Average temperature change 0°C 0°C +6 °C 7The ranges of change in precipitation and temperature are somewhat less at this stage of the analysis, as a subset of all 81 precipitation-temperature shift combinations were selected for inclusion with all other uncertainties. Temperature increases in excess of 8 degrees C, for example, were deemed “unlikely” (based largely on inspection of the GC M projections), and excluded at this stage. 46 Change in precipitation 0% -40% +40% The discount rate is also uncertain. It is a political choice and often highly- contested (Arrow et al., 2013). It shapes how we allocate resources between the present and the future (Gollier, 2011). A higher discount rate signifies an urgency to satisfy present needs, whereas a lower discount rate expresses concerns for the long- term effects of an investment. Although the World Bank typically uses discount rates of 10% to 12%, no single discount rate is appropriate for all projects and it may be difficult for stakeholders to come to consensus (Hoekstra, 1985; Oxera, 2011). In this study, a high discount rate lessens the importance of optimal plant management – indeed, after 25 years, the plant’s additional value generated becomes nearly null after discounting. After consultations with NEA and World Bank experts, we chose a range from 3 to 12% to explore both long term considerations of the sustainability of the investment and the short term objectives of avoiding load shedding in the immediate future. Hydropower production is a function of inflows. The previous section described the development of the climate change and hydrologic modeling analysis. These results are incorporated into this process. The results provide the potential electricity production (GWh/month) for the range of climate changes considered. Because it was impractical to run all 81 climate scenarios through the PRIM analysis, 13 representative hydrologic sequences were selected. The 13 climates are representative of a range of climate change slightly greater than the range projected by the GCMs. By exploring these 13 scenarios with the other factors, the combined effects of climate and non-climate factors that may pose risks to the UAHP in the next 30 years are assessed. 47 Figure 26. Uncertainty space for the analysis of building a new hydropower plant on the Upper Arun. The black dots represent the initial assumptions. Again, the NPV is used as the primary performance metric, in addition to considerations of total and low flow season energy production. The NPV was now calculated not only across multiple climates, but also across multiple futures, to identify the future’s specific characteristics that may lead to a negative project’s performance. Probabilities are not assigned to values in this range. The sampling ranges are used to explore and identify vulnerabilities. The ranges are used to answer the question, “What could the future bring and how would it affect our investment?” rather than “What will the future bring?” If vulnerabilities are found, their likelihood can be assessed at that point. Next, 500 futures were statistically generated, each a combination of a value for each uncertainty. Latin Hypercube Sampling was used to sample the combinations efficiently. It covers the uncertainty space with a reduced number of samples (Saltelli et al, 20048). We chose the 500 cases using Latin Hypercube Sampling for the five non climate parameters and a full factorial design for the 13 streamflows changes. In total, 6,500 futures were used to stress test the 335 MW dam design. The modeling was performed in monthly time steps to capture the flow variability across seasons. The NPV of the 335 MW was estimated for each of these 6,500 futures. 8Saltelli, A., Tarantola, S., Campolongo, F., and Ratto, M. 2004. Sensitivity in practice: A guide to assessing scientific models. John Wiley & Sons Ltd, West Sussex, England. 48 Table 6 shows two rows of the many in the results table. In addition to the NPV, for each scenario, the 60th percentile of the cumulative power generation during the dry season was also estimated, due to the importance of dry season production. Table 6. Sample from the database with the 6,500 futures Inputs Outputs Other four Uncertainties* Precipitation Change (%) P60 Electricity Gen, Dry Price of Electricity Dry Net Present Value (B Capital Cost (M usd) Season (GWh) (usd/kWh) Future ID Option usd) 1 335 MW 0.126 1,115 1 […] 0.06 1,214 2 335 MW 0.081 943 0.6 […] 1.39 584 * Other uncertainties are the discount rates, Plant Load Factor, lifetime of the plant, and average temperature change. The results of the analysis are summarized in the answers to the following questions. 1. How does the 335 MW UAHP perform across a wide range of plausible future conditions? Figure 27 shows the sensitivity of the system performance, measured in NPV, to changes in each factor. Within the selected range, the system is most sensitive to changes in capital cost and electricity price, holding all other factors at baseline values. Increases in capital cost cause the NPV to drop, and in the extreme of the considered range, to go negative. The same is true of discount rate. Increases in the selling price of electricity have a strong positive effect on NPV. By comparison, the effect of changes in climate (wide though the considered changes may be) is not large. 49 3500 3000 2500 Net Present Value ($M) 2000 1500 1000 baseline 500 - 0 -500 -1000 PLF Cap cost Climate Elec price Project life Discount rate Figure 27 Sensitivity of NPV to changes in each factor holding all others at baseline values The performance of the investment options was investigated in 6,500 plausible futures, defined by combinations of the above variable inputs, including changes to climate and non-climatic factors together. The 335 MW projects results in positive NPV in the vast majority of these futures. Considering all uncertainties, there are some futures in which it has a negative NPV. These specific conditions that lead to negative NPV’s can be identified and investigated further to better understand whether they are likely and whether they might be mitigated. 2. Under what conditions does the 335 MW design for the UAHP fail to meet our target of NPV>0? The combination and range of factors that lead to problematic results (NPV < 0) were identified using the Patient Rule Induction Method (PRIM). Scenario Discovery identifies the conditions common to those futures in which the project’s NPV is negative. As described more in details in Appendix B 9 , we apply the scenario 9Appendix B provides a more complete summary of the scenario discovery results. In particular, three scenarios, each with low coverage and medium density, are required to achieve adequate total coverage and density to describe 50 discovery algorithm PRIM to the database of 6,500 cases to identify the combinations of uncertain conditions that most reliably distinguish the cases in which it does not satisfy the objective from the cases in which it does. The analysis revealed that three conditions together describe a scenario in which the 335MW AUHP’s NPV is negative: a. actual capital costs increase by 100% or more, exceeding 900M USD, AND b. precipitation (mean annual) decreases by more than 30% AND c. the electricity price in the wet season remains less than 0.08 USD Assumptions about lifetime of the investment, the discount rate, the actual plant load factor, or changes in average temperature are less important predictors for determining whether the 335 MW UAHP is economically sound. The co-occurrence of these conditions causes the project to have a negative NPV. Although probabilities are difficult to assign, these conditions would generally be considered not likely based on the available evidence. In particular, there is no indication that mean annual precipitation would decrease by more than 30% either according to the climate model projections, trends in historical data or fundamental climate theory. There are other sets of conditions that would threaten the performance of the 335 MW design (see black dots in Figure 64) and they should not be ignored. However, the three conditions identified above are the most illustrative of potentially problematic conditions for guiding the decision makers’ choices. 3. Are those conditions sufficiently likely or unacceptable that other options should be considered? The analysis of the 335 MW design of UAHP has revealed that the project is robust to changes in climate. It has also shown that in some conditions primarily related to increases in capital costs and low electricity prices, a major reduction in precipitation could cause the project’s NPV can be negative. However, this is considered unlikely. The climate stress test also showed that streamflow may shift to earlier in the year. This is an important finding, based mostly in glacier and snow-melt dynamics, which indicates that more streamflow may be available at the very time that it is the future conditions where the 335MW AUHP does not perform well. For simplicity, we only describe here the one scenario from this group of three with highest coverage and density. 51 most needed – to produce high-value dry-season hydropower. Winter precipitation in a warmer climate is more likely to fall as rainfall (not snow), and contribute immediately as runoff to streamflow; and water currently stored in glaciers will melt earlier in the season, and at increasing rates (until melt-down to critical levels at which net streamflow contribution from glaciers diminishes). At this stage it would be perfectly appropriate to jump out of the Decision Tree process, labeling the project “climate robust”. However, at the request of the stakeholders and investors, the process was continued to Phase 4. The motivation was to evaluate the performance of alternative designs with a larger installed capacities that might better capitalize on opportunities for additional hydropower generation at the UAHP site. While typically Phase 4 (and adaptation generally) is considered a process by which climate risks are managed, instead here the phase was used to investigate opportunities. As stated with respect to the decision to continue on from phase 2 to phase 3, the choice to continue to phase 4 in this study is specific to the needs of the stakeholders, and should not be taken as a mandate for all other similar studies. Phase 4 of the Decision Tree – Risk Management This section evaluates risk management options, with discussion of robustness concepts and explanation of the estimation of the likelihood of risks/opportunities. As described above, in the case of UAHP, the uncertainty of the future presents opportunities in addition to risks. For this purpose, design capacities of greater than 335 MW were evaluated for the UAHP. In doing so, the analysis now investigates the ability to take advantage of opportunities in addition to assessing robustness across risks. A new metric, regret, is introduced to quantify the ability to take advantage of opportunity. Regret is the difference between the performance of a design and the performance of the best design for a given future. The goal is to minimize regret. The Phase 4 repeats the analysis conducted in Phase 3 but now considers additional design capacities and the regret metric. The results are described below. Additional Design Options The GoN had chosen the original 335 MW design based on Q70. In 2011, a review of the initial project suggested evaluating also Q25 and Q40. Finally, after consultation with NEA, additional options were identified for evaluation: the Q40, Q35, Q30, and Q25. Using these values and a choice of designs to span a representative range the following hydropower design capacities were evaluated, in addition to 335 MW: 750 MW, 1000 MW, 1355 MW, and 2000 MW (Figure 28). 52 Figure 28. UAHP max daily GWhr/day generation relative to daily time series of GWhr potential: red = 335 MW; blue = 750 MW; green = 1000 MW; purple = 1355 MW; orange = 2000 MW. Calculations of the capital costs and the electricity generation were conducted in the same way as for the 335 MW design described in Phase 3. The other uncertainties, i.e., lifetime of the plant, discount rates, plant load factor, electricity prices, temperature change, and precipitation change, remain constant across all options, as in Table 5. For all uncertainties, the same ranges were maintained as used in Phase 3 (Table 7). 53 Table 7. Updated uncertainty table, with all options Original This Analysis Analysis A. Projected B. Range A. Deep Uncertainties Value min max Wholesale price of electricity (US$/kWh)a x3 times Wet Season (Apr-Oct) 0.045 0.045 0.135 Dry Season (Nov-Mar) 0.084 0.084 0.252 Discount Rate 0.12 0.03 0.15 Estimated Lifetime of the Plant (years) 30 15 36 Plant Load Factor 0.90 0.60 0.90 Capital Costs (2013 US$) x3 times 335 MW 446M 446M 1338M 750 MW 1010M 1010M 3030M 1000 MW 1345M 1345M 4035M 1350 MW 1822M 1822M 5466M 2000 MW 2690M 2690M 8070M Potential GWh generation depends on: Average temperature change 0°C 0°C +6 °C Change in precipitation 0 -40% +40% Opportunity Metric: Regret In Phase 3, NPV was the main metric to evaluate a project’s performance with a threshold set at zero to identify problematic scenarios. Here NPV is still used but additional performance metrics are added. These are the hydropower production during the dry season. In addition, the regret metric is added to evaluate the degree to which each design is able to take advantage of potential opportunities. Regret can also be viewed as a way of calculating the robustness of a project across multiple futures. The regret for a design measures the difference between that design and the optimal design for a given future (Savage (1950)). The regret metric can be used to identify designs that do well across different futures relative to the best design. The design with the maximum regret is the project that performs less robustly, or least takes advantage of opportunity. Conversely, the project that minimized this regret, i.e., that performs better than the other across the different futures, is the one that best takes advantage of opportunities. This is a common rule from decision analysis known as minimax regret. In this study, regret is measured in 54 terms of NPV, e.g., the lower the NPV compared to the optimal project’s, the higher the regret. The regret metric is used to compare the five different design options, by first identifying for each future the option with zero regret, i.e., the option that performs better than the others. The regret of an option i in the future state of the world s is defined by: regret(, ) = max(( ′ , ) − (, )) (eq. 5) ′ where i’ is the best design for a given future and the performance criteria is the net present value of the project (NPV). The best option is the one that has the minimum maximum regret, which is: min (max((, ))) (eq. 6) The metric above is dependent on extreme scenarios. One way to avoid that is be to compare the regret of each project to the regret of not implementing the project. In that case, in order to find the most robust option one would solve: max(regret(,)) min (regret( ℎ) ) (eq. 7) This methodology allows choosing strategies that will minimize the losses whatever the future brings. However, using the minimization of the maximum regret to make decision assumes a very risk-averse behavior. Less risk averse decision makers, more interested in maximizing potential gains would rather explore the maximization of the maximum profit. In this case, for each scenario we would identify the dam with the highest NPV, regardless of the adequacy of that system configuration in other less fortuitous scenarios. However, this metric is more sensitive to the scenarios we are exploring, i.e., the ranges of the uncertainty variables. Results: Sensitivity and vulnerability to climate of different design options First, we tested these new design options for their sensitivity to climate changes alone, as in Phase 3. We maintained all economic variables constant and changed only temperature and precipitation. Figure 29 illustrates the changes in electricity production of the two extreme dam designs, 335 MW and 2000 MW. In all cases, the production seems to be more sensitive to precipitation changes, than to temperature. It is interesting to notice the different behavior in the wet and dry season, Figure 30 and Figure 31 respectively. In the wet season, the 2000 MW dam is able to produce up to 5 times more electricity than the 335 MW. Instead, in the dry season, the two produce nearly the same. However, if we look at the minimum monthly production, although slight, there is a difference between the two dams – which could potentially help avoid load shedding. 55 Figure 29. Changes in annual electricity production, 335 MW and 2000 MW options. Dots locate average values of temperature and precipitation projected by the CMIP5 generation of IPCC GCMs over the approximately lifetime of the project, 2020-2050 (Green: RCP 2.6; Blue: RCP 4.5; Yellow: RCP 6.0; Purple: RCP 8.5). Figure 30. Wet season electricity production in the wet season, 335 MW and 2000 MW options Figure 31. Dry season electricity production, 335 MW and 2000 MW options 56 Figure 32. Minimum monthly electricity production, 335 MW and 2000 MW options Figure 33. NPV response to changes in precipitation and temperature Figure 33 shows the NPV response of the five designs to changes in precipitation and temperature. Again, the NPV is more sensitive to changes in precipitation. Figure 34 shows that overall, the NPV varies due to changes in climate, but it remains positive. In both Figures, it appears that only the larger designs, 1355 MW and 2000 MW, may have a negative NPV as result to climate changes. If we were to select options by looking at their median performance, the two options with the highest median NPV are the 750 MW or 1000 MW. 57 Figure 34. Economic performances across different climate scenarios, maintaining all other variables constant. The shaded area indicates a negative NPV. Figure 35 and Figure 36 show the same information in a different fashion, for the 335 MW and 2000 MW respectively. The two plots show that maintaining all economic variables constant, and instead looking at the 81 different climate series (two columns further on the left), the NPV (right column) of the small design ranges from approximately 500 to 800 M USD, whereas that of the larger design loiters around zero, with half cases below zero (black lines). The grey areas below show the range of possible NPVs, if other conditions were to change. The plots, called parallel coordinates plots, are in fact interactive, and best demonstrated explored using the slider bars shown. If we look at all the range shaded in grey, for the small design, the NPV may range from 5B USD to -1B USD, whereas for the large design, the range is approximately from 15B USD to -7B USD. This shows that other uncertainties may drive the range of the projects’ performances. We are interested in identifying what determines the different performances of the dam. Therefore, we continue the comparative evaluation of the dams’ performances by stress testing them under a full combination of deep uncertainties. Specifically, we identify what are the conditions that make the different investment options fail to meet the policy makers’ goals. 58 Figure 35. 335 MW performance under 81 different scenarios of climate change 59 Figure 36. 2000 MW performance under 81 different scenarios of climate change 60 Results: Comparing all options’ performances under multiple uncertainties The analysis so far suggests that other larger designs may better capture the opportunity of higher flows than the 335MW design. This section examines the performance of the different options that the GoN may pursue. The analysis aims to help decision makers answer key questions: What option is sufficiently robust relative to the others? What are the tradeoffs between robustness and electricity production? What are the remaining vulnerabilities of the preferred option? To answer these questions, we run the same economic model as in Phase 2 for each of the four additional options over the same 6,500 futures shown in Figure 26 and perform scenario discovery to assess the conditions in which they do not meet decision makers’ objectives. Scenario discovery reveals that increases in capital costs and low electricity prices remain the two main uncertainties most relevant for determining the AUHP’s performance. How do the different options perform across a wide range of plausible future conditions? What is the most robust of these investment options? The climate change analysis showed that each of the designs were robust to the broad range of plausible climate changes, although the largest designs of 1355 MW and 2000 MW did show vulnerability to drier conditions. The other designs were clearly robust to climate change (Figure 34). However, when we incorporate other uncertainties, all options perform badly in some futures. As we described above, an additional metric for comparing different projects’ performances, especially in terms of realizing opportunities, is the regret metric. In this study, we investigate which project performs better across futures, i.e., which option minimizes the maximum regret (the relative lowest NPV) across the 6,500 futures. The option that minimizes the maximum regret across these futures is the 1000 MW design dam. The results imply that the 1000 MW design is best able to take advantage of opportunities without suffering too much in other futures. In addition, if the priority were the electricity generation during the dry season and its tradeoffs with costs, the best option would be again the 1000 MW design. A comparison of the regret with the production in the dry season shows again the best option is the 1000 MW (Figure 37). The figure denotes each strategy with a different colored mark. The 1355 MW seems to perform similarly to the 1000MW, but as we described above, it is more vulnerable to changes in climate – so it seems wiser to pick the 1000 MW design. 61 Figure 37. Tradeoffs between maximum regret and production in the dry season (note that decision makers were not interested in 1355 MW and 2000MW, as they are sensitive to changes in climate). Despite the fact that even when looking at different robustness metrics, the 1000 MW design performs better than the other options, All projects perform badly in some futures, when we stress test them under different combinations of climate and economic uncertainties. Considering non-climate factors as well as climate changes, the 335 MW design has the least number of future scenarios where it has a negative NPV while the 2000 MW design has the largest number of negative scenarios. Thus there is a potential trade-off between the design that better captures the opportunity presented by the higher flows (the 1000MW design) and shows a better balance between regret and production in the wet season (Figure 37) and that which has a negative NPV in the least number of futures (the 335MW design). Even the 1000 MW dam remains vulnerable in 1,693 out of 6,500 futures. This does not mean that the plant has a 26% chance of failing (1693/6500). Indeed, we are not able to associate sure probabilities to any of the inputs used to calculate the NPV of the dam. Therefore, a further step in the analysis is to carefully evaluate the scenarios under which the 1000 MW design performs poorly and consider whether those scenarios are likely and whether then could be mitigated. This scenario is identified below. Under what conditions does the 1000 MW UAHP fail to meet our target of NPV>0? Rather than looking at the probability distribution function (since we cannot be sure about the probabilities of the different deep uncertainties) of the 1000 MW UAHP performance, the data mining algorithm is used to identify sets of conditions that describe the futures where the design does not perform well. The analysis reveals 62 that the 1000 MW design is vulnerable to the following conditions:  The electricity price in the wet season is less than 0.10 USD, AND  Actual capital costs increase by more than 80%, AND  Precipitation does not increase significantly (by greater than 10%) These three conditions need to occur at the same time. Again, the lifetime of the investment, the discount rate, the actual plant load factor, and changes in average temperature are less important in determining whether the projects are economically sound. Electricity prices and capital costs emerge as key considerations for the design of UAHP. These are not the only conditions in which the plant would have a negative NPV, but are the most prominent set of conditions that represent vulnerability of the 1000 MW design because they do appear plausible. The analysis helps focus on the these factors, which will make the project fail and thus policy makers can reduce the risk significantly if they address these conditions. But it should not make us ignore other risk factors. Are those conditions sufficiently likely that this design should be reconsidered? The three conditions described above need to occur simultaneously for the 1000 MW dam to result in a negative NPV. While these factors represent uncertainties, they are also factors over which project managers exhibit some control. The final step in the analysis is to evaluate whether the vulnerabilities of the 1000 MW project are likely to occur and if necessary, what can be done to mitigate the risks. Before deciding to invest in the 1000 MW, decision-makers should first carefully assess whether the set of vulnerable conditions are likely to happen together. What are the odds that electricity price will be lower than 0.10 USD/kwH and at the same time, that the project’s capital costs will increase by more than 80%? We know, based on GCM projections, that the odds of precipitation increasing by 10% are difficult to assess but also cannot be ruled out. In general, further consideration of these conditions is warranted to assess whether they are of concern. Such analysis goes beyond the scope of this study. The objective of the study is to identify the key uncertain factors that influence the success of the design, and the level at which they do so, such that further targeted analysis can be conducted if warranted. Some means of managing these uncertainties can also be considered. For example, for electricity price, Government and investors may negotiate to make sure the price does not fall below the threshold that influences the vulnerability of the investment. Power purchasing agreements then are important for the success of the larger capacity (1000 MW). Currently, the electricity price is set at 0.045 USD/kWH. A price of 0.10 USD/kwh may seem difficult to achieve if the market remains mainly national. However, if the GoN set a deal for electricity exports, for instance to India, 63 prices may quickly increase. Ensuring that electricity prices increase to at least 0.10 USD/kwh is one way to ensure the project meets success. Alternatively, project managers could carefully manage capital costs to avoid overruns greater than 80%. For example, they could minimize the conditions for possible delays in implementation, one of the reasons that cause major increases in capital costs. In Nepal, experts claim that about 90% of hydropower projects overshoot their initial cost estimates, but the exact cost overruns are unclear. NEA and other experts of the Nepal region should support the Government in a discussion on the plausibility of incurring into a 80% higher costs than initially estimated. If the GoN deemed this highly unlikely, they could proceed with the project even with a lower than 0.10 USD/kwh electricity price – as the likelihood of both conditions to materialize would be low and acceptable. However, the information on the vulnerability thresholds would remain useful throughout the project implementation phase. If costs happened to start rising, the GoN would possess enough information for monitoring the risk and devising solutions to avoid finding itself in scenario in which both conditions were true. These risks and their plausibility or acceptability need to be considered by decision makers and they themselves need to make the final choice. The policy-makers themselves will in the end decide whether the presented risks are sufficiently acceptable to justify the investment. This is precisely the advantage of DMU approaches, like those implemented in this analysis via the decision tree’s phases: the decision is back in the policy-makers hands, and risks and opportunities are presented in as a transparent way as possible. These types of analysis can help them make informed choices, even when they cannot have confidence about what the future will bring. Conclusions The Decision Tree was implemented to assess the climate risks to the UAHP and to provide context for those risks by considering the effects of other uncertainties. The analysis proceeded through the four phases of the Decision Tree, reflecting the vulnerabilities to climate change and non-climate factors that were revealed. Interestingly, in this case the potential for climate opportunities, in the case of increased potential hydropower generation were also revealed. The analysis of the original 335 MW design found that this design capacity is quite robust to climate change and non-climate factors as well. Climate change itself poses very little risk to this design. Under certain scenarios where capital costs nearly double, the project does become sensitive to electricity prices and major reductions in precipitation. Control of capital costs is an important consideration for this project but the likelihood of this scenario is very low (mainly because there are no 64 indications of such massive declines in precipitation). The analysis of the 335 MW design revealed that there might be opportunities for increased hydropower production of the UAHP. Thus, a Phase 4 analysis was conducted to assess alternative designs. In this case, the adaptation or risk management analysis was in fact focused on realizing a potential climate opportunity. The analysis revealed that the 1000 MW design was a potentially better choice than the original 335 MW UAHP, both in terms of minimizing the highest regret and of tradeoffs between power production in the dry season and cost. The better performance during the dry season would help reduce the chances of load shedding, which a particular concern of NEA. Looking at the specific vulnerabilities of the option, the 1000 MW design is robust to climate change, although when considering non-climate factors it does pose higher risk than the original 335 MW design. That is, the scenarios where the project would have a negative NPV are more plausible than the 335 MW design (although probability of those scenarios has not been assessed). Increases in capital costs, lower electricity prices, and precipitation are key factors. If the 1000 MW design were to be chosen, it would be important to protect against those risks, to make sure the GoN can fully exploit the opportunities that this larger design may bring. Given that a reliable assessment of the probabilities of occurrence of these conditions is not possible, the GoN will have to evaluate whether these risks are acceptable or not – and make their choice accordingly. It is understood that the analysis presented in this section, which points to the preferability of the 1000 MW design, would best be used only as a screening tool to help identify opportunities for efficient use of available water resources under a range of plausible future climate and other conditions. The recommendations contained in this section need to be thoroughly tested and verified by engineering designers able to explore issues related to such matters as tunneling, geology, seismology, electricity transmission, firm energy requirements, and electricity market effects. We submit these systems analysis findings for their consideration during the appropriate stage of the feasibility study. 65 Chapter 5. Basin-level assessment of hydropower investment portfolios under uncertainty Introduction This chapter demonstrates an approach for aiding in the selection of efficient combinations (portfolios) of hydropower investments in Nepal at river basin scale. The current Decision Tree Framework was designed for individual project scale analysis and so may to be applied differently at system scale. In this chapter we propose an approach aimed at regional or national assessments of water infrastructure under uncertainty. The proposed approach is ultimately intended to aid Nepal’s national hydropower planning; proving the concept at the basin-level is an important step towards this goal. Nepal has an integrated national electricity grid, meaning supply-demand deficits manifest at the national level, and so must be addressed at that scale. Achieving the most efficient use of capital and natural resources to achieve electricity generation goals is a government priority. The ability of hydropower systems to function and meet their intended goals depends on a multitude of factors related to the supply and demand of both energy and water. Uncertainty surrounds future changes in many of these factors, which could affect new assets put in place, so a decision-making under uncertainty approach is designed and applied. The goal is to help identify portfolios of hydropower assets that could provide an acceptable mix of benefits under a wide range of plausible futures. This proof-of-concept study was completed in a few months; further work would be completed if applied to national energy planning; this is addressed in the discussion section of this chapter. Proposed hydropower system investment appraisal approach General problem In complex water resources systems water users are interdependent and competition for limited resources may occur in some periods. In Nepal, this type of stress occurs mainly in the dry season, owing to the abundance of water resources during the wet season when monsoon rains coincide with peak snow and glacier melt runoff. To make decisions about the most appropriate hydropower investments given available water resources, or how to allocate water to different sectors of the economy, it is valuable to understand the trade-offs, which will need to be navigated to achieve competing objectives. For example, storage-type hydropower schemes may impound water which would otherwise support downstream abstractions for irrigation, urban water supply, or ecosystem services. In complex systems it is difficult to understand 66 the multitude of interactions; how increasing attainment of one objective impacts on the attainment of others, and how this varies in space and time. Water resources system simulation models can help decision makers test ‘what-if’ scenarios for operation or development of their system. Information can be provided about the likely impacts and benefits of, for example, building a proposed dam with a particular storage capacity and operating rules. Where there are a large number of possible combinations of proposed dam investments and ways of operating those dams (many billions in the case of the Koshi Basin) to provide various benefits, it can quickly become unfeasible to test all combinations to evaluate which is likely to perform best. This issue is only compounded by the need to test portfolios against multiple future states of the world where water supplies and demand are unknown. Our approach therefore requires some way to efficiently evaluate the vast number of options under different futures to identify the best performing portfolios (the most efficient, i.e. Pareto-optimal, and robust given a range of future conditions). Portfolios are considered to be robust if they are efficient across a broad range of futures, i.e. they perform acceptably under less favorable conditions yet are able to capitalize on favorable conditions. Next we describe how the proposed approach meets these requirements: the ability to filter through a large number of options considering several measures of performance given a range of plausible futures. Proposed basin-level hydropower investment approach Because the performance of hydropower assets is highly dependent on environmental factors (river flows, water management rules, upstream and downstream water use, etc.), our basin-level approach applies an integrated water resource management approach. We use a river basin simulation model to predict the performance of the system over time (our analysis simulates a 30-year future period at a monthly time-step) given various conditions and decisions. The simulation model tracks flows and storages throughout the water resource network over time and performance metrics ensure that the most relevant aspects of a system’s performance are related to the user at the end of a simulation (trial). Examples of performance metrics are: hydropower generation and ability to meet irrigation and public water supply demands. To address the multi-criteria (metrics that quantify success) and uncertainty-related aspects of the hydropower investment problem, we propose a four-phased approach. This first phase involves building and/or using a system simulation model with stakeholders, which is capable of representing the impacts of interventions (in our case hydropower investments and their operation) under different scenarios of the future. The system simulator acts as an impact assessment tool – predicting impacts of any combination of interventions and scenarios. Building such a model is a significant investment as it must be trusted by stakeholders and include measures of performance they can identify with. 67 The second phase is a preliminary uncertainty assessment which explores with stakeholders what are the relevant sources of uncertainty in their system and how these uncertainties could impact proposed interventions (defined as the combination of a portfolio of assets and their operating rules or ‘policies’). In the third phase the impact model is linked to an automated search process. This filters through the many possible combinations of investments to find, given many plausible futures, which ones achieve the highest performance. This search repeatedly uses the impact model to test different combinations of assets under a set of future scenarios. The output of this phase is a set of the best performing (approximately Pareto-optimal) portfolios and their operations which can be visualized within trade-off plots. From these a small group of seemingly promising alternatives are singled out for further consideration by stakeholders and/or decision makers. Although not considered here, the storage capacity behind storage dams could also be identified as part of the best performing portfolios. Because in the third phase, for computational reasons, only a restricted set of water- related scenarios can be included in the search, the fourth phase is a rigorous stress test where preferred alternatives are exposed to a set of plausible future socio- economic conditions. The final output is one or more preferred investment portfolios that work acceptably over a wide range of future scenarios, and have broad stakeholder support. Thus, the proposed hydropower investment assessment method can be summarized into 4 phases: I. System characterization – identification of Preliminary, key performance metrics and system function II. Vulnerability assessment – identification of key factors of uncertainty affecting system performance III. Intervention screening - automated search for robust alternatives considering all metrics; trade-off assessment leading to a small set of preferred alternatives. IV. Stress testing of preferred alternatives to help select an acceptable portfolio of future investments. The following text provides further technical and practical details on each phase. Phase I: System characterization The approach begins with a stakeholder engagement phase where over an extended period, planners from relevant organizations are consulted to agree on what metrics of system performance are most relevant to evaluate success or failure of proposed interventions. The metrics can be iteratively defined and refined using a water resource system simulator, which the stakeholders must agree provides a sufficiently accurate assessment of impacts, and constitutes an agreed and trusted evaluation tool. A model of appropriate sophistication should be used in order to capture the system’s most salient features (this may require representing non-linearities in 68 system function which river basin simulators do well). Phase II: Vulnerability assessment This step can be quantitative (sensitivity analysis using system model) or qualitative stakeholder consultation aimed at identifying, describing and quantifying the relevant sources of uncertainty. If quantitative, it amounts to a multifactor sensitivity analysis of the existing system and/or proposed plan under hundreds of combinations of future conditions, aimed at evaluating the system’s sensitivity to future stresses. The output is a description of how current or proposed assets are vulnerable to certain future conditions or combinations of realizations of uncertain factors such that appropriate scenarios can be chosen for phases 3 and 4. Phase III: Automated search Where multiple hydropower schemes interact (as they could in the Koshi, particularly if storage schemes are considered) and objectives conflict (e.g. hydropower generation, irrigation provision, environmental flow support during Nepal’s dry season), an automated multi-criteria approach for sifting through the billions of possible configurations can help. When multiple definitions of success co-exist in real-world engineered systems, there is no single best solution to a portfolio investment problem. Rather, there are multiple trade-offs available whereby the degree to which each objective is achieved impacts on the achievement of all the objectives with which it conflicts. In this case decision makers need to select a balance between the benefits perceived by different stakeholders. Multi-criteria search algorithms can be coupled to stakeholder approved system simulation models to automate the search for promising solutions (those which work well over an agreed set of initial future scenarios). The solution of the multi-criteria search process is not a single optimal solution, but an approximately Pareto-optimal (‘efficient’, ‘non-dominated’, or simply ‘best’) set of solutions, i.e. those for which any further increase in one objective (benefit), would require a decrease in performance of one or more other objectives. If the system simulator and metrics of performance have been agreed upon, the set of best alternatives (investment and policy portfolios) will be of interest to stakeholders. After deliberations, a select few alternatives will be considered in the final phase. Phase IV: Stress test The search process identifies promising investments, which scored well over an initial set of water-related future scenarios. But do they perform well under a much wider and rigorous analysis? That question is answered in this fourth phase. Here we assess what combinations of future socio-economic stresses would cause our efficient alternatives (from Phase 3) to fail, and ask whether these failures are acceptable or probable given the required combination of stresses. If not acceptable or highly probable, further improvements to the proposed future system must be made or other alternatives selected. Otherwise, the hydropower investment portfolio 69 assessment is complete. Various analytical methods can be applied in this phase; as with the UAHP project-level analysis, we use a scenario discovery method called PRIM to identify potential failure scenarios. This step is inspired by ‘Robust Decision Making’ (Lempert et al., 2003) which first described such an approach. Box 2: Application of the basin-scale method Q: Should every river basin be subjected to the analysis applied to the Koshi Basin? A: Basin scale analyses that consider climate change are not required by World Bank policy, and not all basins need this type of analysis, but they can be useful. River basins are themselves complex systems that are governed by natural physical laws and water inputs, and by human uses of water and the operation of man-made infrastructure; project-scale infrastructure itself is a nested complex system within a water basin. In recognition of this fact, most river basin organizations already conduct some level of basin system analysis to aid in operations and capital project planning (e.g., water master plans). The river basin planning applied to the Koshi Basin will be particularly useful under the following conditions: 1. A substantial new potential investment is planned, involving multiple sites and infrastructure options; 2. Multiple uses of water and water infrastructure exist and are planned (e.g., hydroelectric generation, municipal and agriculture supply, and flood control), and complex trade-offs between these water uses already exist or could result which are expected to be challenging to manage; 3. Decisions about what and how to make new investments are sensitive to one or more uncertain factors, such as climate change, future demand for electric power, and the price of electricity or alternative forms of energy. Because almost all parts of a system can affect the performance or many or all of the other parts, basin-scale analysis can reveal insights that are surprising and would not be available from simpler, independent analyses with a smaller scope. Application to the Koshi basin This section describes the basin-level analysis of the Koshi Basin to investigate which portfolios of additional dam development could be both efficient and robust for addressing Nepal’s electricity needs while maintaining other water-related benefits. Phase I: System characterization The Koshi Basin currently supplies 119MW of Nepal’s 729MW grid connected hydropower capacity from its 5 existing run-of-river schemes (Table 8). Table 8. Installed capacity of existing grid connected hydropower dams in the Koshi Basin Hydropower dam Existing capacity (MW) Sunkoshi HEP 2.5 70 Baramchi HEP 4.2 Indrawati III 7.5 Khimti 60 Bhote Koshi 45 We include 5 further proposed dams in the Koshi Basin as investment options in our analysis (Figure 38, Table 9). Although the approach can search for the ideal storage capacity behind a dam, in our application we investigate the specific designs identified by the JICA (1985) report on Koshi Hydropower potential (storage volume and/or installed capacity) for each asset, except for the proposed Upper Arun Hydroelectric Project (UAHP) where 5 alternative designs are considered (as in the project-level analysis). In this basin-scale analysis we also search for the best operating rules for storage- type hydropower schemes, when they are included in a portfolio. These rules can impact on performance at the monthly time-step of the modeling, owing to these schemes’ capacity to store water between months. Run-of-river type schemes are associated with only limited water storage facilities (if any) in order to maximize generation during peak electricity load hours of the day. Their operating rules are assumed to not affect performance at the model’s monthly time-step so they are not subjected to the search process. Changing the operation of storage-type dams can lead to substantial variations in performance (in terms of the balance of benefits achieved) from the same physical infrastructure. As a result, storage dams have more capacity to adapt to future conditions than run-of-river dams do. Figure 38. Schematic of the IRAS-2010 Koshi Basin model showing the 5 new hydropower 71 reservoirs being considered in this hydropower investment assessment. Existing reservoirs are also displayed Figure 39. Shaded area shows extent of IWMI Koshi basin model used as a data source for the system simulation model built for this study. A river basin system simulation model (Figure 38) was built using information and data from a pre-existing International Water Management Institute (IWMI) model (Figure 39). The model was built using the generalized Interactive River-Aquifer Simulation-2010 (IRAS-2010) model (Matrosov et al, 2011) which is adaptable and computationally efficient and has an appropriate level of sophistication for this type of study, including both storage and run-of-river dams. Table 9 Proposed hydropower projects included as options in the basin-scale analysis Project name Type of scheme Generating capacity Capital cost (US$m) (MW) Sun Koshi 3 Storage 536 1690.5 Dudh Koshi Storage 300 1144 Upper Tamakoshi Run-of-river 456 441 Upper Arun Run-of-river 335 – 2000 depending on 446 – 2600 depending decision on gen. capacity Arun-3 Run-of-river 900 423.2 The system characterization identified 8 Koshi basin performance metrics relevant to investment appraisal through the September 2014 project group visit to Nepal and subsequent contacts with NEA and other stakeholders: 1) capital investment 72 2) dry season electricity generation 3) total annual electricity generation 4) firm electricity generation (99.5% reliability) 5) urban water supply deficit 6) agricultural water supply deficit 7) flood peak at the basin outlet 8) number of environmental flow failures downstream of dams The output of this phase was the calibrated system simulation model and its performance metrics (also reported on under Phase III). Phase II: Vulnerability assessment The initial uncertainty analysis aimed to identify with stakeholders the most important uncertain factors affecting system performance was done qualitatively; the chosen factors are reported below. The preliminary assessment of uncertainties was also completed through workshop exercises at the September 2014 project group visit to Nepal and subsequent contacts with NEA and other stakeholders. Sources of uncertainties considered significant included climate change impacted flows, other water demands and potential future requirements for environmental flow releases. Further sources of uncertainty (included in the phase 4 stress test) are in line with then non-hydrological variables described in Table 7 for the project-level analysis. In some cases a quantitative analysis would help more rigorously identify sources of uncertainty; in this application, the system identification was done qualitatively through workshop facilitation and consultation. Three uncertain factors were identified in relation to the flows available for hydropower generation. River flows Hydropower generation is closely linked to availability of river water, which varies both seasonally and inter-annually. Climate change presents an additional layer of uncertainty on river flows as shown by the analysis in Chapter 4 of this report. Using the flow response surfaces from the project-level analysis (Figure 16), we selected five bands of plausible flow changes around the bands spanned by GCM outputs, to test portfolios against a broad range of plausible future flow conditions. In practice this meant a maximum 10% reduction in flows and a maximum 60% increase in flows, with 17.5% change steps between them. These change factors are applied to the baseline (1971-2000) monthly flow series from IWMI’s hydrological model of the Koshi basin. Abstraction demand increases 73 Abstractions for agricultural and urban demands can both affect and be affected by hydropower dams depending on whether they occur upstream or downstream respectively. We use present day demands from the IWMI Koshi Basin model (Figure 39) in our water resources simulation model, and also increase these by 50% in line with future population projections for Nepal by 2050 (IDS-Nepal et al., 2014). These two demand scenarios can be considered extremes of abstraction demand uncertainty. Environmental flow releases Review of the relevant literature and modeling of the current system showed environmental flow releases to be a source of significant uncertainty for hydropower producers. A scheme designed for a certain level of flow without any requirement for an environmental flow release which is later subjected to such a requirement by legislation, will suffer from a reduction in generation and therefore revenue and profitability. We applied two scenarios for environmental flow releases, where either a Q95 flow release is required, or it isn’t. The Q95 flow is that which is exceeded 95% of the time, i.e. 19 out of 20 months in the unregulated flow record. Phase III: Automated search for efficient hydropower interventions We used a multi-criteria search process to identify promising hydropower investment portfolios for the Koshi Basin, and the operating rules of any storage-type hydropower dams selected. This combinatorial problem has a large number of possible combinations of assets built and operations (>1020) but the efficient search algorithm requires only a relatively small number of trials using a fast simulator (e.g. 2 million trials in 24 hours) to converge on the best options. Results are presented as trade-off scatter plots which show the set of most efficient (‘Pareto-optimal’, ‘best’) intervention options. With trade-off scatter plots, the preferences of decision makers need not be specified in advance of the search process, but instead can be expressed at the results analysis stage by considering the levels of performance that are acceptable for each metric, given the trade-offs available. This approach has the potential to identify win-wins where gains in one or more objectives can be achieved with little or no impact on others. The search or ‘optimization’ problem is formulated as follows: what combination of assets and operating rules: 1) Minimizes urban water supply deficits, 2) Minimizes capital costs, 3) Minimizes agricultural water supply deficit, 4) Minimizes maximum flood peak at the basin outlet, 5) Maximizes dry season electricity generation, 6) Maximizes total annual electricity generation, 74 7) Maximizes firm electricity generation (99.5% reliability), 8) Minimizes environmental flow failures (flow is less than Q95) downstream of dams We recognize in relation to minimizing environmental flow failures that our metric is a crude measure of environmental performance, but note sophistication could be increased on the basis of more detailed analysis of the environmental flow requirements. For simplicity a fixed environmental flow (Q95) was used but more detailed future work could consider a seasonally varying environmental flow. Uncertainty cases To facilitate the assessment and interpretation of investment bundle (portfolio) robustness, three ‘cases’ (i.e., groupings of scenarios) are searched (Table 10). The ‘best case’ looks at future scenarios which are broadly favorable to hydropower generation owing to greater availability of water resources (we assume environmental flow releases won’t be required in future and that abstraction by consumptive uses will not increase). Our goal in the best case search is to find those portfolios of new hydropower dams which are best able to capitalize on such futures. The ‘worst case’ search identifies portfolios that do best under less favorable conditions for hydropower generation (environmental releases are required and abstraction increases). The average case search is for portfolios that perform best on average, across all scenarios. Table 10. Three uncertainty cases used in the search process and the circumstances under which performance for all metrics is evaluated Uncertainties Best case Average case Worst case 5 Flows (-10% to +60% scaling) X X X No Environmental flow release X X Environmental flow release X X No abstraction demand increase X X Abstraction demand increase X X For each uncertainty case we solve a different optimization problem, so that we can identify promising portfolios of assets under expected, favorable or worst case conditions. Results Below we provide and discuss the results of the multi-criteria search (optimization) exercise. We focus first on the most efficient investment portfolios for addressing the dry season electricity supply deficit for simplicity of presentation, although three different metrics of generation performance were used. Figure 40 shows the best 75 ‘average case’ portfolios identified for balancing 8 objectives (i.e. an 8 -dimensional set in 2 dimensions). Each point shows the performance of a unique portfolio of infrastructure investments and their operation (where storage dams are included). Figure 40. Performance of all portfolios identified by the search under average case conditions, in terms of investment and dry season generation Figure 41 shows the same plot as Figure 40 but with each point colored to indicate which of the UAHP capacity options is part of each efficient portfolio. The larger capacity Upper UAHP options are implemented in higher investment portfolios, partly because the cost of each option increases in proportion to its capacity. Figure 41. The same plot from Figure 40 with coloring used to show which UAHP capacity 76 option is used in each portfolio The members of the best (most efficient) set of interventions for generating increased dry season energy for minimal investment are isolated in Figure 42. Portfolios are numbered and detailed to show how they vary. Figure 42 indicates there is no additional advantage in terms of dry season generation from investments above US$ 4700m. This is likely because larger capacity (and more expensive) schemes remain limited by low flows for generating electricity in the dry season. This does not mean that there are no other benefits of higher levels of investment. Figure 42. The most efficient set of interventions for increasing Koshi Basin dry season electricity generation alone with portfolios numbered and detailed for reference. As stated in the section describing Phase II, we identified two additional ‘extreme’ cases (best and worst) for hydropower generation to illustrate the range of possible conditions under which hydropower portfolios may be operating by the 2050s (Table 10). Figure 43 shows how the interventions identified by the search under each of the three uncertainty cases and three generation metrics relate to each other. It shows, from the variations in UAHP options alone, that the best average case performance does not ensure the best performance under more extreme conditions. It also shows that spreads of performance between the three cases is not consistent between generation metrics – total annual generation demonstrating more consistency than firm or dry season energy. This is because the best and worst cases will have a greater impact during periods of low flows, which disproportionately affect both 77 firm and dry season generation. Figure 43. Comparison of the best performing interventions for three types of energy generation and for each of three uncertainty cases. Variations between the uncertainty case portfolios (indicated by the UAHP option used) show that robustness to extreme conditions does not necessarily result from good average performance. Robustness analysis We analyzed the portfolios identified by the three searches in order to identify portfolios that are robust (i.e. efficient across the three uncertainty cases and three generation metrics). In order to differentiate the relative robustness of these portfolios, we calculated the total maximum (normalized) regret of each. Regret in this basin-scale analysis is very similar in concept to the description given in the project-level analysis. Minimizing the maximum regret is a way of improving overall performance. Maximum regret represents the worst performing metric among the eight associated with a portfolio, i.e., all other metric scores are better. As described in the project-level analysis, this represents a conservative approach, minimizing risk rather than attempting to maximize possible benefits. In our case regret is calculated for each of the eight performance metrics and normalized for each to give a value between 0 for the best performance and 1 for the worst performance. Regret is calculated separately for each uncertainty case to help 78 understand the relative strengths of infrastructure portfolios in taking advantage of opportunities afforded by favorable conditions in contrast to limiting risks from unfavorable conditions. We total the three uncertainty case maximum regrets and judge lower total maximum regret scores to indicate greater robustness. Regret is only one factor which decision makers may wish to consider, alongside the balance of benefits provided. Table 11 shows the eight infrastructure portfolios efficient for investment and energy generation across the three uncertainty cases (i.e. robust). Portfolios which were efficient under only one or two of the uncertainty cases did not meet the criteria defined above for robustness and were therefore not considered for further analysis. The UAHP is included in robust portfolios at three of its five available capacities – 335, 750 and 1000MW. Five out of the eight robust portfolios are four or five dam combinations. The performance metric values shown in the table are averages across 20 scenarios. Generation levels for UAHP shown in Table 11 are not directly comparable with those from the project-level analysis which did not use this averaging process. Generation performance of the Arun-3 & UAHP-335 portfolio (Rank 4) is less than the sum of performance for portfolios comprising these dams individually (Rank 6 and 8) because the generation from existing dams is included in each result. The relationship between maximum regrets for the three uncertainty cases for the eight infrastructure portfolios in Table 11 is shown in Figure 44. It also shows that the operation of the storage dams in portfolios ranked 1, 5 and 7 can be changed to decrease maximum regrets (i.e. increase robustness). Maximum regrets are reduced by operating these dams for efficiency in terms of all the performance metrics used, rather than just those for investment and energy generation. This demonstrates the benefit of accounting for all stakeholder interests. The reduction in regret is mainly for worst case performance for portfolios 5 and 7, whereas for portfolio 1 only best case regret is reduced by the change in operation. 79 Figure 44 Relationship between maximum regrets for the three uncertainty cases for infrastructure portfolios from Table 11 when they are operated to maximize efficiency across either investment and energy generation objectives alone, or across all objectives. All-objective efficiency results in lower total maximum regrets for portfolios 1, 5 and 7, showing the increased robustness available by changing the objectives of dam operation, even with the same infrastructure. 80 Table 11. Portfolios of infrastructure which are efficient and robust for investment and energy generation objectives under all 3 uncertainty cases. Ranking is performed on the basis of minimizing the total maximum regret for the 3 uncertainty cases. Performance met ric values are each portfolio’s best investment and energy performance result under the average uncertainty case (i.e. averages across 20 scenarios). Rank 1 2 3 4 5 6 7 8 Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 UAHP-750 UAHP-1000 UAHP-750 UAHP-750 UAHP-1000 UAHP_335 Arun-3 Infrastructure portfolio UAHP-335 U. Tamakoshi U. Tamakoshi U. Tamakoshi Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi Sun Koshi 3 Sun Koshi 3 Sun Koshi 3 Sun Koshi 3 Investment 3008 4258 4603 869 4699 446 5044 423 (US$m) Dry season generation 2172 2322 2264 1600 2466 1008 2424 805 (GWh/yr) Performance metric value Annual generation 7375 6834 8049 4772 7839 3067 8467 2335 (GWh/yr) Firm generation 241 280 270 94 302 64 287 48 (GWh/month) Urban water 0.02 0.16 0.02 0.03 0.03 0.03 0.04 0.03 deficit (Mm3/yr) Irrigation deficit 1.88 6.89 1.55 1.48 2.40 1.48 1.98 1.48 (Mm3/yr) Max flood (m3/s) 3699 3774 3774 3775 3663 3775 3665 3775 Environmental flow failures 826 793 855 602 951 543 1038 522 (occurrences) Worst case 0.74 0.68 0.73 1.00 0.97 1.00 0.96 1.00 Maxim regret um Average case 0.58 0.68 0.73 0.72 0.76 0.82 0.89 0.88 Best case 0.68 0.68 0.73 0.65 0.79 0.79 0.82 0.86 81 Total 2.00 2.03 2.19 2.36 2.52 2.61 2.66 2.74 82 Considering efficiency in terms of all the performance metrics used adds 17 other portfolios and their operating rules that are robust to the three uncertainty cases (Figure 45). This is a significant expansion on the eight alternatives previously identified. Furthermore, 11 of these portfolios and their operations outperform the highest ranked portfolio and operation from Table 11 in terms of total maximum regret. These 11 portfolios, their average case performance and maximum regrets are shown in Table 12. The portfolio ranked 5th in Table 11 includes the same infrastructure as the highest ranked portfolio in Table 12, but with the Dudh Koshi storage dam operated in favor of a greater range of objectives. Storage dams are more prevalent in the portfolios which produce robust performance for all metrics than they were in the portfolios providing robust energy generation only. All 11 portfolios in Table 12 include one or both of the storage dam options. Where only one storage dam is included, this is always Dudh Koshi. We note that the 5 of 11 robust portfolios for all objectives comprise 3 dam combinations, which were absent from the energy-focused portfolios in Table 11. Energy focused portfolios included one and five dam combinations which are absent from Table 12. Figure 45 Relationship between maximum regrets as in Figure 44 with 17 other additional infrastructure portfolios added which are efficient across all objectives. Eleven of these result in lower total maximum regrets, showing the further benefits of maximizing efficiency in relation to all objectives, rather than a restricted set of investment and energy objectives. From Figure 40 through to Figure 42 we refined the set of efficient portfolios down to only those which are best for capital investment and dry season electricity generation, before analyzing across the uncertainty cases and other generation metrics (Figure 11 43) for maximum regret to define and rank eight robust portfolios (Table 11). Figure 46 shows the extent to which dry season energy generation must be limited (traded- off) from that achieved by green colored portfolios from Table 11 in order to increase other benefits – in this case environmental flow performance. As illustrated in Figure 45, these trade-offs have the positive effect of also reducing total maximum regret (increasing robustness). Figure 46. Average case performances of efficient and robust portfolios of hydropower infrastructure investments considering different sets of objectives. The extent of the red points shows that maximum dry season generation is reduced by around 300GWh/year to balance other objectives efficiently and robustly. Three portfolios from Table 12 are labelled according to their robustness rank and composition. Two are implementations of the same 4th ranked portfolio with operating rules which maximise generation but differ in their environmental performance. This difference depends on how environmental performance is balance with other objectives not shown. The 6 th ranked portfolio is shown here operated to trade-off dry season generation to maximise environmental performance. 11 Table 12. Portfolios of infrastructure which are efficient for all objectives under all 3 uncertainty cases and lead to lower total maximum regret than the highest ranking infrastructure combination and its operation from Table 11. Ranking is performed on the basis of minimizing the total of maximum regret for all 3 uncertainty cases. Performance metric values are each portfolio’s best performance result under the average uncertainty case. Rank 1 2 3 4 5 6 7 8 9 10 11 Dudh Koshi Dudh Koshi UAHP-1000 Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi Dudh Koshi UAHP-1000 Dudh Koshi UAHP-1355 Dudh Koshi Sun Koshi 3 Koshi 3 Koshi 3 Tamakoshi Tamakoshi UAHP-750 UAHP-335 UAHP-335 UAHP-750 UAHP-335 UAHP-750 750MW Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 Arun-3 UAHP Infrastructure portfolio U. U. Sun Sun Investment (US$m) 2567 2013 2912 3704 3008 1590 2144 3258 3835 3353 3389 Dry season generation 2131 1994 1821 2179 1986 1403 1424 1159 1225 2135 2060 (GWh/yr) Performance metric value Annual generation 5960 6141 5727 7035 7189 4436 5048 3255 3918 6651 7036 (GWh/yr) Firm generation 233 184 173 200 198 150 171 120 126 233 199 (GWh/month) Irrigation deficit 2.09 2.05 1.52 2.76 1.33 2.05 1.86 0.92 8.81 1.97 2.10 (Mm3/yr) Urban water deficit 0.03 0.03 0.02 0.04 0.02 0.03 0.03 0.02 0.28 0.03 0.03 (Mm3/yr) Max flood (m3/s) 3749 3672 3678 3676 3669 3673 3752 3673 3670 3690 3775 Environmental flow 700 673 599 824 775 614 609 572 526 819 715 failures (occurrences) Worst case 0.43 0.36 0.54 0.59 0.74 0.58 0.87 0.52 0.67 0.75 0.87 Maximum Average case 0.41 0.41 0.46 0.59 0.51 0.56 0.49 0.69 0.62 0.57 0.54 regret Best case 0.41 0.55 0.46 0.59 0.56 0.67 0.50 0.68 0.61 0.59 0.54 Total 1.25 1.33 1.47 1.76 1.80 1.81 1.85 1.89 1.89 1.91 1.94 11 Phase IV: Basin-scale stress test The automated search identified portfolios of investments that perform acceptably over up to 20 plausible water availability futures. In reality, the uncertainty is such that any future package of investments could be vulnerable to other, non-water related factors. To illustrate this concept we now test a promising portfolio from Phase III for a number of additional plausible future socio-economic scenarios. Similarly to analysis done in the Decision Tree for the project-level analysis, in this step we perform a detailed stress test to assess what set of future conditions could threaten the economic viability of preferred alternatives. Specifically, a preferred alternative is simulated over a sampling of all uncertainties used at the basin (flows, demands, environmental flows) and project (also including capital costs and duration of construction) scale. As in Chapter 4, the PRIM scenario discovery algorithm is used to identify failure scenarios within a large number of future system simulations. As an illustration, we stress tested a high performing portfolio’s economic performance under the same 500 economic futures used in the stress test of the UAHP. This time however, we used 20 future scenarios of flows and power generation, for a total of 10,000 runs. The resulting NPV of the portfolio ranges from approximately US$ -5 billion to US$ 14 billion. Figure 47. Performance of a high performing portfolio across 10,000 futures As Figure 47 shows, there are some futures in which the portfolio has a negative 86 NPV. Specifically, this happens in 2,527 of 10,000 simulated futures. Running a PRIM analysis on the results database shows the portfolio is vulnerable to the following conditions:  Actual capital costs are 120% higher (or more) than estimated capital costs, and  The electricity price in the wet season is less than 0.062 USD. These two conditions need to occur at the same time. As with the project-level analysis, the lifetime of the investment, the discount rate, the actual plant load factor, and changes in average temperature and precipitation are found to be less important for this case in determining whether the portfolio is economically sound. And again, policymakers should evaluate the risk and decide whether these conditions are sufficiently acceptable for the investment to move ahead. This is a demonstration of the type of analysis we could run on the most promising (efficient and robust) portfolios, to gain a deeper understanding of their specific vulnerabilities. This time, we ran the portfolio looking at its cumulative costs and power generation. A next step would be to stress test each individual asset within a portfolio individually and explore the cumulative performance of the portfolio. Moreover, as shown at the project level, this analysis allows decision makers to compare the performance of different portfolios and choose the most economically robust given an uncertain future. 5.5 Recommendations and discussion The results of the basin-level analysis allow a range of different questions to be answered depending on decision makers interests. At their simplest, results allow an assessment of which UAHP designs are part of the most robust investment portfolios for the basin. Being a component of a robust portfolio is not the same as being the best design for an individual project considered in isolation however, as other factors come into play at the basin level. For example, the performance of a UAHP design must compete not only with other UAHP designs, but with other proposed dams. Basin-level trade-off analysis can provide decision relevant information about the best way to invest limited funds towards achieving stated goals or the amount of investment needed to fully achieve those goals. As decision makers ask increasingly sophisticated questions, the answers become more complex and nuanced. As the trade-off plots show, no portfolios achieve a perfect solution, but valuable information is provided about the trade-offs available and the robustness to a range of future uncertainties. This can assist decision makers in balancing societal goals and achieving robustness in complex real world water resource and energy systems. In this basin-scale analysis we have not discussed whether current investment decisions should be modified on account of climate change as we do not know existing basin-level plans. The proposed approach is recommended when there is a 87 need to consider a range of uncertainties, including climate change in relation to hydropower and broader water resources infrastructure development. It could be used to check planning decisions already made, applied in a process of basin-wide development planning, used to update existing designs (pre-construction) or help modify operation of existing dams in conjunction with proposed investments as part of a basin-level optimization exercise. Specifically, results indicate:  Investment portfolios that do well under ‘average’ future conditions don’t necessarily work well over more extreme conditions (best or worst case) (see Figure 43)  Storage dams feature in all robust portfolios for addressing the full range of objectives, showing the value storage provides. (Although not implemented here, the approach allows optimization of dam storage capacities as well)  UAHP at 335, 750 or 1000MW is included in robust portfolios when considering investment and energy generation metrics only (see Table 11). This is an indication of the lack of increased dry season and firm energy generation returns on investment in higher installed capacity, as these types of electricity are limited primarily by river flows. These same three UAHP capacities feature in robust portfolios for addressing all objectives.  The lowest regret robust portfolio for addressing all objectives includes Arun-3, UAHP 750MW and Dudh Koshi. Four of the top 11 portfolios in the total regret ranking contain the 750 MW design, three the 335MW, two the 1000MW and one the 1355MW design. In the end the specific UAHP design chosen will depend on the particular balance of benefits decision-makers find appealing, the robustness they require, and how much they would like to spend on developing Koshi hydropower resources.  The stress test analysis example suggests that the key vulnerability factors for the NPV of a high performing basin portfolio are capital expenditure and electricity price, in line with the project-level analysis. Although climate change and other flow related uncertainties have a significant impact on NPV, these economic factors outweigh such impacts. The basin-scale analysis identifies the 750MW UAHP option as part of the most robust portfolios, whereas the project level analysis showed the 335MW and 1000MW options to be more robust. This is because each study applies different criteria to define robustness. The basin-scale analysis includes objectives which are more likely to be affected (positively or negatively) by the storage-type dams it includes, but which are less relevant to the project level analysis of run-of-river UAHP. Although the basin-scale portfolio stress test focuses on NPV, portfolio robustness is defined by the lowest total maximum regret across all 8 water-related metrics of performance. Decision maker assessment of these metrics and their trade- offs and robustness should help decide which portfolios are preferred. 88 Limitations and future work This basin-scale study is intended to demonstrate an approach for identifying promising portfolios of hydropower investment studies under uncertainty. Hydrological inputs to the basin modeling were generated by combining previous Koshi Basin model outputs from IWMI with detailed hydrological modeling done for the UAHP project level analysis. Applications of the basin-scale approach to support real decision-making would benefit from extending the ‘bottom-up’ UAHP hydrological modeling to the whole basin. One limitation of this demonstration has been the regional scale of the study. Energy planning is typically done at national scale such that supply and demand can be considered concurrently. At the national scale it will be possible to consider the national supply-demand balance – this concept is of little relevance at the basin scale as much of the energy demand may be outside the basin and external supplies may be unrepresented, even though they may prove preferable. If applied nationally, the basin scale approach covered here could consider energy demand and its uncertainty. Demand is currently spread relatively evenly throughout the year, although slightly higher during the dry season, which is winter (with more demands for heat and light). It is possible with climate change that this balance could change, with increased cooling demands during the warm summer season. The basin–scale stress test conducted here was limited by its treatment of the portfolio as a consistent whole (i.e. same across all new hydropower schemes), with consistent changes in capital expenditure, project life etc., whereas in reality these factors would vary between individual members of the portfolio. This is an area of active development by the authors; in the near future we will be able to test the impacts of the uncertainties on each individual component of the portfolio. The disaggregation will facilitate the investigation of the impacts of different scheduling options for selected portfolios and this may reveal different vulnerabilities. The use of NPV as the sole indicator of performance in the stress test is a further limit which the authors are working to expand to also consider performance metrics from the search (Phase 3). A key consideration for decision makers is likely to be the returns on capital investment; for this the assumed linear interpolation of cost with UAHP capacity here may not give a realistic picture of the returns on investing in higher capacity options. Traditional least-cost energy planning may be simple but it generates the least cost schedule of asset investment/implementation. The approach demonstrated here recommends efficient groupings of interventions, but does not explicitly solve for their sequencing in time. Recent research (Haasnoot et al., 2013) highlights that considering investment decisions as dynamic (evolving over time) adaptive pathways (i.e. trajectories of actions over time) is a useful concept for planning under uncertainty. Searching for robust and cost-effective schedules of asset investments is possible but was not attempted in this proof of concept work, but it has been carried 89 out successfully in other projects by the authors. Ideally a regional infrastructure investment analysis would be undertaken well in advance of a project level analysis, but in this case that has not been possible, so they were completed concurrently as a proof of concept. Conclusions Basin-scale and project-level hydropower investment analyses are complementary. Basin-scale analysis provides a range of bigger picture information around the efficiency of investment portfolios considering many system goals. The investment portfolio trade-off analysis allows decision makers to improve decisions based on a broad definition of performance. The Phase IV stress test provides an additional level of robustness testing, including in our case socio-economic uncertainties which are difficult to incorporate in a water resources simulation model. The answer to whether UAHP 335MW is adaptable to climate change from a basin perspective is positive; it is a component of 3 of the lowest regret portfolios considering all objectives. Whether seeking to address solely electricity generation objectives, or a wider range of objectives for the basin, the same three capacity options for UAHP (335, 750 and 1000MW) figure in the most robust portfolios. The key difference is in the combination of UAHP and other run-of-river dams with storage-type dams which have a greater range of operation. Storage-type dams feature in all robust portfolios for addressing the broad range of objectives considered here. In addition to robustness, decision-makers need to assess what balance of benefits they would like to see in the basin and the level of funding available in order to select an investment plan for it. Ideally a national level assessment would follow a similar approach to show how the Koshi Basin portfolios perform in the context of all options available across the country. Different basin characteristics may present different vulnerabilities to climate change and variability in particular. From a government perspective, it is important to identify the most promising infrastructure portfolios nationally for addressing projected energy demands over the coming decades. Our recommendation is to build capacity to be able to undertake system studies such as this one, but at national scale incorporating supply and demand of water and energy. 90 Chapter 6. Conclusions Hydropower development in South Asia is complicated by the complexity of the natural environment and concerns related to climate change. Project such as these natural face a number of risks, including climate change, high sediment loads, and environmental, social and financial risks. These risks should be addressed in order to create promising investments that perform well throughout their design life and provide the expected benefits. However, the existence of risks does not mean a hydropower project is risky. It does mean that a systematic approach to risk assessment is needed. IDA 17 requires rigorous assessment of hazards such as climate change in all projects. This is clearly consequential in World Bank South Asia hydropower and reservoir projects. Previous efforts to assess climate change risks have relied on attempts to forecast future climate through extensive scientific efforts with climate models. However, the future climate is inherently uncertain and these efforts have provided vague results of questionable reliability. IEG (2012) concluded that climate-science led efforts have been unhelpful for decisions and new approaches are needed. The Decision Tree Framework was developed specifically for this purpose. It is designed to guide a planner through a systematic assessment of climate risks to infrastructure investment. It tailors the level of analysis to the risks identified in a tiered process. The Decision Tree is an implementation of general Decision Making Under Uncertainty approaches and this application represents the culmination of learning in these areas, bringing together advanced physical and economic modeling to assess climate and non-climatic risks. These techniques can be applied to plan hydropower investments subject to a range of future climate scenarios, including glacier melting and sediment load uncertainties, while also taking into account other uncertain factors, such as electricity price, non-hydropower fuel price, the magnitude of co-benefits (such as flood control), construction cost, and projected energy demand. The analysis employed the Decision Tree Framework to the climate change risks to the pre-feasibility design of 335 MW hydropower capacity for the Upper Arun Hydropower Project (UAHP). In addition to climate change risks, other non-climate factors were identified through stakeholder discussions and included in the analysis. In addition, performance metrics for project evaluation were identified as the economic value of the project (Net Present Value) and the total and low season hydropower production. The Decision Tree consists of four Phases that are employed in series if the assessment requires it, based on the risks that are uncovered. This document demonstrated the application of the four Phases to the UAHP. It should be noted that all four phases were employed as a demonstration, and would not necessarily have been necessary in this case otherwise. The analysis was responsive to the desires of stakeholders. Phase 1 and Phase 2 are initial screening steps designed to clear 91 projects with low climate sensitivity without needing further assessment. Phase 3 began with the thorough quantitative analysis of the UAHP project using the climate stress test. In this case, a hydrologic model, hydropower simulation model and stochastic weather generated were needed. These were used to identify the climate conditions that are problematic for the project. In addition, a range of non-climate factors that could also affect project performance were identified and used as part of the stress test. In the case of the 335 MW UAHP, the climate stress test revealed a problematic scenario associated with very high capital cost increases (almost double estimated costs). Interestingly, it also revealed the potential for higher capacity hydropower production. Effectively, climate change adaptation in this case addresses potential opportunities rather than risks. For these reasons, the analysis proceeded to Phase 4. In this phase, the risks identified in Phase 3 were carefully assessed and alternative designs were evaluated. Overall, the 335 MW design was quite robust to the range of climate changes considered which were not limited by GCM projections. The climate stress test sampled a wide range of changes in climate and variability and revealed the vulnerability of the design to plausible climate changes. In this case, climate change itself posed little risk to the 335 MW design. In fact, the analysis revealed the opportunity to consider larger capacities. The analysis also provided insight into the general hydrologic response of the Upper Arun River in Nepal’s Koshi river basin to changes in climate. It showed that streamflow increases moderately with warmer temperatures up to about +3C, after which streamflow decreases moderately due to declining contributions from glacial melt. The streamflow during the low flow season was found to decline slightly with warmer temperatures but the effect was small. Precipitation effects are as expected, increases leading to increased streamflow, and the effects are much larger than the temperature effects. Projections for the region, of unknown credibility, indicate warmer future temperatures and no clear signal for precipitation. The assessment next considered of possible alternative (larger capacities) designs for UAHP, based on the earlier finding of potential opportunity. The results indicate that each of the capacities was robust to plausible climate changes. Only the largest capacity exhibited vulnerabilities to climate changes that were fairly extreme (e.g., 20% reduction in precipitation). Among non-climatic factors, the price of electricity and construction costs emerged as key risk factors, specifically if the former remains as low as it is today (around 0.04 USD/kwh) and the latter increases by more than 50% due to delays in implementation or other implementation problems. The analysis concluded that the original pre-feasibility design of 335 MW was robust to all uncertainties with few problematic scenarios. However, the design of 1000MW emerged as an attractive alternative, providing the best combination of robustness and opportunities, including dry season production but is more sensitive to increases in capital costs and electricity prices. These risks should be carefully addressed if this design is selected. Specifically, given the limitations of the supply-side approach taken in this study (the large spill energy in the wet season and the assumption that 92 all energy produced is purchased), a supply-demand balance study would be an appropriate follow-up to this one. Models such as VALORAGUA or other production simulation models could be used. The application of the Decision Tree to the UAHP development resulted in unambiguous assessment of climate and other risks and the identification of two project designs that are robust to a wide range of possible climate changes. This document concentrates mostly on an exploration of the risks and potential benefits of investment in the UAHP, with the understanding that it is a “preferred” investment alternative in Nepal’s hydropower sector. The project-level analysis is supplemented by a basin-scale analysis that confirms UAHP as a promising investment. The basin-scale analysis of hydropower infrastructure investment considered five proposed hydropower dams, including UAHP, alongside the Koshi’s five existing dams. The assessment identified robust portfolios of investments (those that perform acceptably well over a range of plausible futures) considering firm, dry season, and total annual electricity generation as well as environmental flow impacts, urban and agricultural water supply and flood alleviation (for storage schemes). The approach identifies how different portfolios of hydropower assets (scheme, capacity, operating policies) lead to different trade-offs between these performance metrics. The analysis provides decision relevant information to a range of investment questions, of varying sophistication. UAHP schemes (of various capacities) were members of the best (i.e. most efficient & robust) hydropower investment portfolios. As shown by the project-level analysis, larger UAHP schemes are difficult to justify for dry season electricity generation alone, as flows are the limiting factor. Smaller UAHP capacities remain members of efficient portfolios despite unfavorable conditions for hydropower generation while larger capacity options are better able to take advantage of increasing flows, which may result from climate change. When optimizing over a range of futures for all eight objectives, four of the 11 most robust portfolios contained the 750 MW UAHP design, three the 335MW, two the 1000MW and one the 1355MW design. All 11 included one or both storage dam options, underlining the potential benefits of storage dams. Ultimately decision-maker preferences and the Koshi hydropower development budget will direct which combination of schemes, and which specific UAHP design, are most appropriate. A stress test similar to the one applied in the project-level analysis was applied to one of the promising Koshi basin portfolios of hydropower investments. This showed non-climate uncertainties of construction cost and electricity price to be the key risk factors for the achievement of expected gains from this hydropower investment. The portfolio of investments maintained acceptable economic performance throughout the range of flow scenarios considered in the project-level study. Both project- and basin-level studies acknowledge the limited capacity of a run-of- river hydropower scheme such as UAHP to address dry season electricity shortage, 93 even in a basin with relatively reliable flows. Larger schemes can only make better use of available flows up to a point. The studies also confirm that larger investments are only efficient where the increased wet season generation which becomes possible, can be used or sold. Although this work explored hydropower investment options at the project and basin-scale, there could be benefits to undertaking such work at national scale. Such a study could borrow elements of the analyses presented here. The Decision Tree, as a framework for the staged application of approaches for Decision Making Under Uncertainty to (climate change) risk management, helps decision makers o Assess climate change threats without first needing to predict the future climate o Understand the strengths and limitations of their projects in wider range of future climate and non-climate conditions o Identify adaptation strategies, projects, and choices that are critical to long-term success Moreover, DMU approaches facilitate conflict resolution by providing o A transparent and accessible analysis that invites testing of multiple project designs o A rigorous treatment of a range of stakeholder views on how the future will unfold (provided that there is a major stakeholder presence in the development of the analysis, and a continuing opportunity for inputs as new information arises) o An opportunity to incorporate multiple metrics that facilitate discussion and agreement. References Friedman, J.H., Fisher, N.I., 1999. Bump hunting in high-dimensional data. Stat. Comput. 9, 123–143. doi:10.1023/A:1008894516817 Groves, D.G., Lempert, R.J., 2007. A new analytic method for finding policy-relevant scenarios. Glob. Environ. Change, Uncertainty and Climate Change Adaptation and Mitigation 17, 73–85. doi:10.1016/j.gloenvcha.2006.11.006 Haasnoot, M., J. H. Kwakkel, W. E. Walker, and J. ter Maat, 2013, Dynamic adaptive policy pathways: A method for crafting robust decisions for a deeply uncertain world, Global Environmental Change-Human and Policy Dimensions, 23, 485-498. IEG, 2012, Adapting to Climate Change: Assessing the World Bank Group Experience. 94 Lempert, R.J., Popper, S.W., Groves, D.G., Kalra, N., Fischbach, J.R., Bankes, S.C., Bryant, B.P., Collins, M.T., Keller, K., Hackbarth, A., Dixon, L., LaTourrette, T., Reville, R.T., Hall, J.W., Mijere, C., McInerney, D.J., 2013. Making Good Decisions Without Predictions (No. RB-9701), Research Brief. RAND Corporation. Lempert, R., Kalra, N., Peyraud, S., Mao, Z., Sinh Bach, T., Cira, D., Lotsch, A., 2013. Ensuring Robust Flood Risk Management in Ho Chi Minh City. World Bank, Policy Research Working Paper. Matrosov, E. S., Harou, J. J., and Loucks, D. P., 2011. A computationally efficient open-source water resource system simulator - Application to London and the Thames Basin, Environmental Modelling & Software, 26, 1599-1610, 10.1016/j.envsoft.2011.07.013, 2011. Ray, P., and C. Brown., 2015. Including Climate Uncertainty in Water Resources Planning and Project Design - Decision Tree Framework. Washington, DC: World Bank Group. Water Partnership Program (2015). Water Security for All: The Next Wave of Tools. 2013/14 Annual Report. Water Global Practice, The World Bank Group, Washington, DC. 95 Appendix A: Climate Data Analysis and Supplementary Model Results Climate Data Analysis Figure 48 Gridded Climate Data: 0.5 degree grids of APHRODITE daily temperature and precipitation dataset. Blue basin is for Uwagaon streamgage (600.1); green is for Turkighat (604.5); pink is for Simle (606) Circle points are climate stations in the Koshi basin. X ’s are climate stations without usable data. Figure 49 APHRODITE and GPCC are very precise (coeff of corr), but not particularly accurate (Nash Sutcliffe). However, gages tend to be at lower elevation, and topography of most cells includes very high elevation. Gridded data algorithms include topography in spatial averaging. It is therefore difficult to draw conclusions regarding the quality of gridded data. 96 Only cell with data in Upper Arun basin APHRODITE precip (mm) Observed precip (mm) Cell with many observations Figure 50 APHRODITE is the only dataset with daily precip and temperature. It is fairly precise, but not very accurate. It consistently underestimates precipitation in some cells. Figure 51 GPCC typically performs well in the Himalayas (see Brahmaputra analysis), and its record is longest (monthly data from 1900-2010), but contains only monthly precipitation data, and at 2.5 o grids (as opposed to Aphrodite’s 0.5o grids). 97 Figure 52 Cell with the most observations (though also fairly low elevation) 98 Figure 53 APHRODITE, when converted to streamflow, is much lower than other gridded data. GPCC most closely follows TRMM, which uses satellite observations, and tends to be most trusted (with record only from 1998). The choice is then to use GPCC data, “downscale” it using ratios from APHRODITE data (and distribute it into daily values). Climate data summary:  This study will use GPCC monthly precipitation data and APHRODITE daily temperature data.  The GPCC monthly precipitation data will be “downscaled” (from 2.5o to 0.5o) and distributed according to ratios of APHRODITE daily precipitation.  Climate Data: o Spatial resolution: 0.5o gridded data o 1951-2007 99 Figure 54 Wavelet Generator Diagnosis: Wavelet based on GPCC area-averaged precipitation data for the Upper Arun basin from (water year) 1901-2010 100 Figure 55 Autocorrelation at between 10 and 12 years 101 Figure 56 The area-averaged precipitation data are not normal, and are not made normal by transformations using log or Box-Cox 102 Figure 57 The WARM model (with 11-year signal) therefore is not a good fit for the data 103 Figure 58 Neither is the simple ARIMA model Weather generator recommendation:  We have decided to resample seasonally, bootstrapping without an imposed period.  We may resample in 10-year periods (10-year “chunks”) after the initial analysis. 104 Supplementary Model Results Figure 59 NPV 105 Figure 60 Annual hydropower production 106 Figure 61 Dry season hydropower production 107 Figure 62 Wet season hydropower production 108 Figure 63 Minimum month hydropower production 109 Appendix B: Scenario Discovery Methods and Results In this appendix, we provide greater detail on the results presented in Phase 3 and Phase 4 of Chapter 4. We describe the scenario discovery method and then provide full scenario results for the economic performance of the 335MW and the 1000 MW AUHP. Scenario Discovery Method For this analysis, we apply a scenario discovery exercise for identifying scenarios that characterize the vulnerabilities of the proposed design options. Scenario Discovery is an integral part of the Robust Decision Making methodology (Lempert et al., 2013). It uses statistical cluster-finding algorithms to provide concise descriptions of the combination of future conditions that lead a strategy to fail to meet its objectives (Lempert et al., 2013)). The description of these conditions helps focus decision makers’ attention on the most important uncertain future conditions to the problem at stake. These future conditions help decision makers discuss the acceptability of the risks involved with the various options available (Groves and Lempert, 2007). Scenario discovery begins with the creation of the database described in Table 5, which contains the model’s results. Each row of results reports a future (or case), which is a combination of particular levels of each of the uncertainties considered (i.e., a certain price of electricity, capital cost increase, precipitation and temperature changes, discount rates, plant load factor, and lifetime of the plant) and the resulting performance of the project according to the chosen metrics, in our case the NPV. Scenario discovery works well when stakeholders agree on a threshold for the performance metric. This allows the analyst to proceed with differentiating the futures in which the project meets its objectives from the ones in which it does not. In this analysis, the project(s) failed to meet its (their) objective in those futures where NPV was negative (i.e., threshold = 0). As we described in the main report, we chose the 500 futures, or cases, for five of the six parameters and a full factorial design for the 13 streamflows (i.e, the climate dimension). In total, we created 6,500 cases. We then used the Patient Rule Induction Method (PRIM) (Friedman and Fisher, 1999) to analyze the database of futures and identify the set(s) of conditions (which we will call later a “scenario”) that differentiates the vulnerable futures from the successful ones. These sets of condition describe some combination of constraints on one or more of the uncertainties. For instance, a set may indicate that the NPV is negative if the discount rate is higher than a certain level and the streamflows decrease by a certain percentage. PRIM helps identify the main uncertainties that may affect the project’s performance and focus the attention on these relevant parameters. PRIM uses three measures to evaluate the different sets of conditions it identifies:  Coverage: the fraction of vulnerable futures out of all futures captured by 110 the scenario. Ideally, we would want a coverage of 100% - but this is rarely obtainable.  Density: the fraction of all the vulnerable futures captured by the scenario, out of all futures captured. Again, ideally all futures captured should be vulnerable and density should be 100%.  Interpretability: the easy with which users can understand the information conveyed by the scenario. The number of uncertain conditions used to define the scenario serves as a proxy for interpretability. The smaller the number of conditions, the higher the interpretability (Lempert et al., 2013). PRIM generates a set of scenarios and presents tradeoff curves that help the users choose the one scenario with the best combination of density, coverage, and interpretability. Figure 64 Example of Scenario Discovery. Figure 64 illustrates an example of scenario discovery analysis for the 335 MW design. It represents all 6,500 futures as red circles; the filled red circles represent the futures in which the project has a negative NPV (and thus fails to comply with the decision-maker’s objective). The black box is the best scenario describing the futures in which the project has a negative NPV: electricity price increase of less than 70% than the initial price and capital costs more than 100% higher than the initial costs. This figure shows only the constraints on capital cost increase and electricity price that define each scenario, because they are the two most important variables for defining scenarios, i.e. they are the parameters that matter most for the sign of the NPV. However, they do not fully explain all future conditions under which the project may fail to meet the decision makers’ objectives. First, the other parameters still play a small role in explaining the future conditions under which the project may fail. Second, given that this scenario does not have a coverage and density of 100 %, other scenarios are needed to explain the vulnerable futures. In some cases, no single scenario can provide adequate coverage and density. 111 In such cases, as for the AUHP’s analysis, the PRIM algorithm allows the user to iterate this process on the database. The user identifies a scenario and the algorithm removes the cases within that scenario from the database. The user then reruns PRIM and identifies another scenario from the remaining data. This process can be repeated as many times as needed. The resulting set of multiple scenarios may reduce interpretability, but can increase coverage and density (Lempert et al., 2013). Scenario Discovery on the 335MW and the 1000MW designs for the UAHP Table 13 Conditions defining scenarios in which each project has a negative NPV, and the coverage and density for each scenario. Project Conditions Coverage/Density Design 335 MW Scenario 1 Total=30% / 78%  Electricity price wet season (USD/kWh) < 7.9  Capital cost increase > 100% Scenario 1= 13% / 78%  Precipitation < 90% current precipitation Scenario 2= 9% / 78%  Lifetime < 31 years Scenario 3= 8% / 77%  Temperature change < 3.5°C warmer than currently Scenario 2  Electricity price wet season (cents USD/kWh) < 7.9  Precipitation < 30% current precipitation  Capital cost increase > 100%  Lifetime < 34 years  Discount rate > 5.2%  Temperature change > 3.0°C warmer than currently Scenario 3  Electricity price wet season (cents USD/kWh) < 7.8  Capital cost increase > 120%  Lifetime < 32 years  Precipitation < 10% more than current  Discount rate > 5.6%  Temperature change < 2.5°C warmer than currently 1000 MW Scenario 1 Total=54% / 80%  Electricity price wet season (cents USD/kWh) < 10.2  Capital cost increase > 80% Scenario 1= 30% / 82%  Precipitation < 10% more than current Scenario 2= 24% / 77%  Lifetime < 34 years  Temperature change < 3.5°C warmer than currently Scenario 2  Electricity price wet season (cents USD/kWh) < 10.5 112  Capital cost increase > 81%  Precipitation < 10% more than current  Lifetime < 34 years  Temperature change > 2.5°C warmer than currently Table 13 below describes more in detail the full results of the scenario discovery analysis, reported in Chapter 4. For the two preferred project designs, the table shows the constraints on the most relevant uncertain parametres for the performance of the project by the scenario discovery analysis. Moreover, it shows the coverage and density for each scenario. Neither design has simple scenarios with high coverage and density. For instance, the 335MW design option needs at least three scenarios, which have constraints on five/six out of seven uncertain parameters. As a group, these three scenarios have a coverage and density of 30% and 78%. The coverage is still very low, so other conditions which determine vulnerable futures exists, beyond the ones described below. But each individual scenario has even lower coverage. We reported the first scenarios in Chapter 4 for both designs, because they have the highest coverage and density. References Friedman, J.H., Fisher, N.I., 1999. Bump hunting in high-dimensional data. Stat. Comput. 9, 123–143. doi:10.1023/A:1008894516817 Groves, D.G., Lempert, R.J., 2007. A new analytic method for finding policy- relevant scenarios. Glob. Environ. Change, Uncertainty and Climate Change Adaptation and Mitigation 17, 73–85. doi:10.1016/j.gloenvcha.2006.11.006 Lempert, R.J., Popper, S.W., Groves, D.G., Kalra, N., Fischbach, J.R., Bankes, S.C., Bryant, B.P., Collins, M.T., Keller, K., Hackbarth, A., Dixon, L., LaTourrette, T., Reville, R.T., Hall, J.W., Mijere, C., McInerney, D.J., 2013. Making Good Decisions Without Predictions (No. RB-9701), Research Brief. RAND Corporation. Lempert, R., Kalra, N., Peyraud, S., Mao, Z., Sinh Bach, T., Cira, D., Lotsch, A., 2013. Ensuring Robust Flood Risk Management in Ho Chi Minh City. World Bank, Policy Research Working Paper. 113 Appendix C. Stakeholders’ involvement during the project The Project Team in Kathmandu, 8-13 September 2014 The Project Team, consisting of the UMass Amherst Hydrosystems Research Group (as represented by Casey Brown and Patrick Ray), the University of Manchester (as represented by Julien Harou), and the World Bank (as represented by Pravin Karki, Laura Bonzanigo, and Anjali Lohani Basnet), gathered in Kathmandu during the week of 8-13 September 2014 to conduct a series of seminars and workshops on the subject of the incorporation of climate uncertainty in hydropower planning and project design, and to launch a study of the climate vulnerabilities of the proposed UAHP in the Koshi River basin, Nepal. The climate vulnerability study was to be done according to the steps described in the Decision Tree for Decision Making Under Climate Change Uncertainty, as summarized in Figure 65. Once potential climate-related vulnerabilities to the proposed design had been identified, if present, risk management was to be accomplished through iterative application of search, trade- off analysis, and a climate change stress test, as described in Figure 66. Project description and specifications REVISED FIGURE 2 Scheme of Proposed Framework Project Choices, Decision Tree Phase 1: Connections, Desktop Screening Consequences, unCertainties Downscaled IPCC Decision Tree Phase 2: Stake-holder defined performance metrics 5th CC projections Initial Analysis and thresholds, and preliminary adaptation (to inform range of options stress test only), Demographic/Econ Hydrologic and Water Systems models omic Projections (e.g., WEAP and SWAT from IWMI) Decision Tree Phase 3: Risk Assessment (System Stress Test) Ex Post Scenarios – First-cut Climate and Sediment management: Risks to Demographic/Economic Vulnerabilities Identified Hydropower Production, and other Dam Functions Decision Tree Phase 4: Sediment management: Risk Management Adaptation Strategies Benefits and Costs RDM and Many-Objective Trade-off Analysis advanced adaptation tools coupled with Decision Scaling Vulnerability Response Surface Figure 65. Koshi Basin Modeling Project Workflow Monday 114 Patrick met with IWMI at the IWMI offices to plan the workshop for Tuesday the 9th. Patrick and Laura met to discuss the methodologies for the study. Tuesday Patrick and IWMI conducted the Training Workshop on Understanding the Link Between Hydrological and Climate Change Modeling at the Yak and Yeti Hotel. The rest of the Project Team participated as observers. Attendees (a complete list of whom is available in the Stakeholder Meetings report) included: the Department of Soil Conservation and Watershed Management; the Department of Hydrology and Meteorology; the Department of Environment; the Department of Irrigation; the Ministry of Energy; the Department of Electricity Development; the Department of Water Induced Disaster Prevention; the Nepal Electricity Authority; the Nepal Ground Water Resources Development Board; the Nepal Investment Board; Tribhuvan University; Kathmandu University; Pokhara University; a number of Civil Society and Research Institutes. The Project Team met in the evening to plan for Wednesday the 10th, to rehearse the planned games and interactive activities, and to develop work flowcharts, as shown in Figure 66. Wednesday The Project Team and representatives of the Nepal hydropower sector conducted the Seminar on Impacts of Climate Change on Investments in Water Resources and Hydropower Projects: Investment Decision Making Under Deep Uncertainty at the Yak and Yeti Hotel (a complete list of attendees is available in the Stakeholder Meetings report). The Project Team met in the evening, over dinner and drinks, with seminar participants. Thursday The Project Team met with representatives of the Nepal Electricity Authority (NEA) at the World Bank offices in Kathmandu. The NEA team presented A Brief Introduction on Nepal’s Hydro Potential and Power Development. The NEA team joined the World Bank team and the Project Team for lunch. The Project Team attended the Basin Planning Workshop at the Park Village Hotel during the afternoon. Casey presented the Decision Tree to the workshop’s audience. The Project Team met during a social hour with workshop participants. Friday The Project Team met for a business breakfast at the Dwarika’s Hotel. The Team was joined by Raghuveer Sharma of the IFC to discuss the relevance of climate change risk assessments of the type described in the Decision Tree to IFC projects 115 throughout Nepal and the broader south Asia region. The Project Team then toured the HydroLab facilities and discussed partnership with HydroLab as a hosting institution for climate change risk assessments in Nepal. A business lunch concluded the week’s planned activities. Next Steps The week in Nepal helped considerably to structure the next phases of the project. The project was to be comprised of two analyses: 1) a climate stress test of the UAHP; and 2) a broader analysis of possible hydropower investments in the Koshi Basin. An NEA engineer, Mr. Surya Narayan Shrestha, joined the Project Team to add local expertise, and to be trained in the Project Team’s advanced procedures for climate risk assessment and trade-off analysis in order to add capacity to the capabilities of the NEA. Figure 66. Iterative Application of Decision Tree Phase 4 for Climate Change Risk Management 116