SOLAR RESOURCE AND PV POTENTIAL OF MALAWI SOLAR MODEL VALIDATION REPORT December 2018 This report was prepared by Solargis, under contract to the World Bank. Solar Power Resource Mapping: Malawi [Project ID: P151289]. This activity is funded and supported by the Energy Sector Management Assistance Program (ESMAP), a multi-donor trust fund administered by the World Bank, under a global initiative on Renewable Energy Resource Mapping. Further details on the initiative can be obtained from the ESMAP website. The content of this document is the sole responsibility of the consultant authors. Any improved or validated solar resource data will be incorporated into the Global Solar Atlas. Copyright © 2018 THE WORLD BANK Washington DC 20433 Telephone: +1-202-473-1000 Internet: www.worldbank.org The World Bank does not guarantee the accuracy of the data included in this work and accept no responsibility for any consequence of their use. 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. Rights and Permissions The material in this work is subject to copyright. Because the World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for non-commercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: +1-202-522-2625; e-mail: pubrights@worldbank.org. Furthermore, the ESMAP Program Manager would appreciate receiving a copy of the publication that uses this publication for its source sent in care of the address above, or to esmap@worldbank.org. All images remain the sole property of their source and may not be used for any purpose without written permission from the source. Attribution Please cite the work as follows: World Bank. 2018. Solar resource and PV potential of Malawi: Solar Model Validation Report. Washington, DC: World Bank. Solar Model Validation Report Regional adaptation of Solargis model based on data acquired in 24-months solar measurement campaign Republic of Malawi No. 141-08/2018 Date: 7 December 2018 Customer Consultant World Bank Solargis s.r.o. Energy Sector Management Assistance Program Contact: Mr. Marcel Suri Contact: Mr. Dhruva Sahai Mytna 48, 811 07 Bratislava, Slovakia 1818 H St NW, Washington DC, 20433, USA Phone +421 2 4319 1708 Phone: +1-202-473-3159 E-mail: marcel.suri@solargis.com E-mail: mailto: dsahai@worldbank.org http://solargis.com http://www.esmap.org/RE_Mapping Solar Model Validation Report Regional adaptation of Solargis model based on 24 months solar measurement campaign Solargis reference No. 141-08/2018 Table of contents Table of contents ............................................................................................................................................. 4 Acronyms ........................................................................................................................................................ 5 Glossary .......................................................................................................................................................... 6 Executive summary.......................................................................................................................................... 8 1 Overview of regionally adapted data layers .............................................................................................. 10 2 Solargis database.................................................................................................................................... 11 2.1 Solar resource data calculated by satellite-based solar model ................................................................ 11 2.2 Combined use of satellite-based model and measurements ................................................................... 12 2.2.1 Site adaptation vs. regional adaptation ......................................................................................... 13 2.2.2 Conditions to be met ...................................................................................................................... 13 3 Ground measurements in Malawi ............................................................................................................ 15 3.1 Solar meteorological stations: specifications and data ............................................................................ 15 3.2 Quality control and harmonization of solar measurements...................................................................... 18 4 Regional adaptation of Solargis model .................................................................................................... 20 4.1 Solargis method ........................................................................................................................................... 20 4.1.1 Reduction of systematic deviation between satellite and measured data ................................. 21 4.1.2 Regional aerosol correction ........................................................................................................... 22 4.1.3 Regional DNI and GHI de-biasing ................................................................................................... 22 4.2 Results and validation.................................................................................................................................. 23 4.2.1 Accuracy estimate of DNI and GHI at the solar meteorological stations ................................... 23 4.2.2 Accuracy-enhanced DNI and GHI maps ........................................................................................ 25 5 Solar resource maps of Malawi ............................................................................................................... 27 5.1 Accuracy enhanced maps of DNI and GHI ................................................................................................. 27 5.2 Uncertainty of solar resource maps............................................................................................................ 29 6 Conclusions ............................................................................................................................................ 32 7 List of figures .......................................................................................................................................... 33 8 List of tables ........................................................................................................................................... 34 9 References .............................................................................................................................................. 35 Support information ....................................................................................................................................... 37 © 2018 Solargis page 4 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months solar measurement campaign Solargis reference No. 141-08/2018 Acronyms AOD Aerosol Optical Depth CFSR Climate Forecast System Reanalysis. The meteorological model operated by the US service NOAA (National Oceanic and Atmospheric Administration) CFSv2 Climate Forecast System Version 2 CFSv2 model is the operational extension of the CFSR (NOAA, NCEP) DIF Diffuse Horizontal Irradiation, if integrated solar energy is assumed. Diffuse Horizontal Irradiance, if solar power values are discussed DNI Direct Normal Irradiation, if integrated solar energy is assumed. Direct Normal Irradiance, if solar power values are discussed. GFS Global Forecast System. The meteorological model operated by the US service NOAA (National Oceanic and Atmospheric Administration) GHI Global Horizontal Irradiation, if integrated solar energy is assumed. Global Horizontal Irradiance, if solar power values are discussed. GTI Global Tilted (in-plane) Irradiation, if integrated solar energy is assumed. Global Tilted Irradiance, if solar power values are discussed. MACC Monitoring Atmospheric Composition and Climate – meteorological model operated by the European service ECMWF (European Centre for Medium-Range Weather Forecasts) MERRA-2 Modern-Era Retrospective analysis for Research and Applications, Version 2 Meteosat Meteosat satellite operated by EUMETSAT organization. MSG: Meteosat Second (MFG and MSG) Generation; MFG: Meteosat First Generation © 2018 Solargis page 5 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months solar measurement campaign Solargis reference No. 141-08/2018 Glossary Aerosols Small solid or liquid particles suspended in air, for example desert sand or soil particles, sea salts, burning biomass, pollen, industrial and traffic pollution. All-sky irradiance The amount of solar radiation reaching the Earth's surface is mainly determined by Earth-Sun geometry (the position of a point on the Earth's surface relative to the Sun which is determined by latitude, the time of year and the time of day) and the atmospheric conditions (the level of cloud cover and the optical transparency of atmosphere). All-sky irradiance is computed with all factors taken into account Bias Represents systematic deviation (over- or underestimation) and it is determined by systematic or seasonal issues in cloud identification algorithms, coarse resolution and regional imperfections of atmospheric data (aerosols, water vapour), terrain, sun position, satellite viewing angle, microclimate effects, high mountains, etc. Clear-sky irradiance The clear sky irradiance is calculated similarly to all-sky irradiance, but without considering the impact of cloud cover. Long-term average Average value of selected parameter (GHI, DNI, etc.) based on multiyear historical time series. Long-term averages provide a basic overview of solar resource availability and its seasonal variability. P50 value Best estimate or median value represents 50% probability of exceedance. For annual and monthly solar irradiation summaries it is close to average, since multiyear distribution of solar radiation resembles normal distribution. P90 value Conservative estimate, assuming 90% probability of exceedance (with the 90% probability the value should be exceeded). When assuming normal distribution, the P90 value is also a lower boundary of the 80% probability of occurrence. P90 value can be calculated by subtracting uncertainty from the P50 value. In this report, we apply a simplified assumption of normal distribution of yearly values. Root Mean Square Represents spread of deviations given by random discrepancies between measured Deviation (RMSD) and modelled data and is calculated according to this formula: 6 * ∑7 (* +,-./0,1 − +31,4,1 5 = & *89 On the modelling side, this could be low accuracy of cloud estimate (e.g. intermediate clouds), under/over estimation of atmospheric input data, terrain, microclimate and other effects, which are not captured by the model. Part of this discrepancy is natural - as satellite monitors large area (of approx. 3.3 x 4.0 km for MSG satellite pixel), while sensor sees only micro area of approx. 1 sq. centimetre. On the measurement side, the discrepancy may be determined by accuracy/quality and errors of the instrument, pollution of the detector, misalignment, data loggers, insufficient quality control, etc. Solar irradiance Solar power (instantaneous energy) falling on a unit area per unit time [W/m2]. Solar resource or solar radiation is used when considering both irradiance and irradiation. © 2018 Solargis page 6 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months solar measurement campaign Solargis reference No. 141-08/2018 Solar irradiation Amount of solar energy falling on a unit area over a stated time interval [Wh/m2 or kWh/m2]. Uncertainty Is a parameter characterizing the possible dispersion of the values attributed to an of estimate, Uest estimated irradiance/irradiation values. In this report, uncertainty assessment of the solar resource model estimate is based on a detailed understanding of the achievable accuracy of the solar radiation model and its data inputs (satellite, atmospheric and other data), which is confronted by an extensive data validation experience. The second source of uncertainty is ground measurements. Their quality depends on accuracy of instruments, their maintenance and data quality control. Third contribution to the uncertainty is from the site adaptation method where ground-measured and satellite-based data are correlated. © 2018 Solargis page 7 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Executive summary This report describes accuracy enhancement of Solargis solar resource data for Malawi based on the ground measurements collected at three solar meteorological stations across the country. These solar meteorological stations were installed and operated by GeoSUN Africa (South Africa), under funding from the World Bank, over years 2016 to 2018. The accuracy-enhanced solar model makes it possible to calculate time series for any location with lower uncertainty. This effort results in more accurate regional data, which are needed in solar energy yield calculation and financial evaluation of an solar project to be developed in the region. The major benefit is higher confidence and lower costs of solar projects development. used data layers represent long-term yearly and monthly averages of Direct Normal Irradiation (DNI) and Global Horizontal Irradiation (GHI), and they cover a period of the last 24 years: from 1994 to 2017. The data is calculated by aggregation of sub-hourly map-based time series calculated for the territory of Malawi with 1-km spatial resolution. The aggregated data layers are delivered in the format that is compatible with Geographical Information Systems (GIS). Additionally, printable maps are available in digital format and ready-to-use for large- format printing. The accuracy of the data layers is enhanced by the regional adaptation of the Solargis model with use of ground measurements acquired at three high-standard solar meteorological stations located in Malawi. The measurements helped to reduce systematic deviation of the data inputs to the Solargis model as driving factors of the uncertainty in the region. Individual improvements of the Solargis model at the stations in Malawi are shown in the table below. As a result of the Solargis model adaptation, the calculated GHI and DNI data are available with reduced uncertainty. Table 0.1: Position of solar measuring stations in Malawi Site location Nearest town Latitude Longitude Elevation Measurement station host [°] [°] [metres a.s.l] Chileka airport Blantyre -15.67984 34.97229 767 Malawi Meteorological Services Kasungu airport Kasungu -13.01530 33.46840 1065 Malawi Meteorological Services Uni Mzuzu Mzuzu -11.41990 33.99530 1285 Mzuzu University Table 0.2: Model output changes, due to regional adaptation, at the solar measuring stations Site DNI original DNI adapted Difference GHI original GHI adapted Difference to original to original [kWh/m2] [kWh/m2] [%] [kWh/m2] [kWh/m2] [%] Chileka 1765 1608 -8.9 1993 1859 -6.7 Kasungu 1825 1647 -9.7 2077 1988 -4.3 Mzuzu 1737 1449 -16.6 2022 1861 -8.0 * All values in this table show results of the regional adaptation of Solargis model long-term average yearly values. GHI – Global Horizontal Irradiance, DNI – Direct Normal Irradiance. The data is derived from the GIS layers after terrain disaggregation. © 2018 Solargis page 8 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 0.3: Uncertainty of yearly values for Malawi, for original and regionally-adapted Solargis model Direct Normal Irradiation Global Horizontal Irradiation Global Tilted Irradiation DNI GHI GTI (fixed at optimum tilt) Low Medium Low Medium Low Medium Original data < ±12% < ±21% < ±9% < ±13% < ±9% < ±13% After adaptation ±5% to ±7% < ±10% ±4% to ±5% < ±6% ±4.5% to ±5% < ±7% Best-achievable* ±3.5% - ±2.5% - ±3.0% * Uncertainty only achievable by site-specific model adaptation based on many years of high-quality measurements (values are shown as a model data reference) The accuracy-enhanced solar model makes it possible to calculate more accurate time series and derived data products. This project reduced substantially the uncertainty of primary solar parameters that are key inputs in calculation of solar electricity yield and financial prediction. Thus results of this project increase confidence in the technical design and performance evaluation of any solar power plant in the region, and it also increases reliability of financial estimates. © 2018 Solargis page 9 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 1 Overview of regionally adapted data layers Solargis is a high-resolution global database which includes solar resource and meteorological parameters, which are important for development and operation of solar power plants. The regionally adapted solar model provides more accurate and reliable primary solar parameters for Malawi: Global Horizontal Irradiation (GHI) and Direct Normal Irradiation (DNI). This results in a reduced uncertainty of derived solar parameters, such as Diffuse horizontal Irradiation (DIF), Global Tilted Irradiation (GTI) and subsequently also Photovoltaic power potential (PVOUT), see Table 1.1. Lower uncertainty reduces financial risk and improves engineering quality of the design of solar power systems. For a project-specific site, the uncertainty can be further reduced by the model site adaptation and the use of local measurements. This stage of the project delivers: • Validated Solargis models for the region • Harmonized and accuracy-enhanced solar resource data for Malawi: yearly and monthly long-term averages of GHI and DNI • Historical time series and aggregated data represents the last 24 years (1994 to 2017) and it is available at high spatial and temporal resolution. • The accuracy enhanced model can deliver time series data for any location at lower uncertainty. Table 1.1: Description of GIS data layers that were accuracy enhanced by regional model adaptation Acronym Full name Unit Type of use Type of data layers GHI Global Horizontal kWh/m2/year Reference information for the Long-term annual and monthly Irradiation kWh/m2/day assessment of flat-plate photovoltaic averages (PV) and solar heating technologies (e.g. hot water) DNI Direct Normal kWh/m2/year Assessment of Concentrated PV (CPV) Long-term annual and monthly Irradiation kWh/m2/day and Concentrated Solar Power (CSP) averages technologies. It is also important for simulation of flat-plate PV tracking technologies. DIF Diffuse Horizontal kWh/m2/year Complementary parameter to GHI Long-term yearly and monthly Irradiation kWh/m2/day and DNI average of daily totals GTI Global Irradiation at kWh/m2/year Assessment of solar resource for PV Long-term yearly and monthly optimum tilt kWh/m2/day technologies average of daily totals OPTA Optimum angle ° Optimum tilt to maximize yearly PV - production PVOUT Photovoltaic power kWh/kWp/year Assessment of power production Long-term yearly and monthly potential kWh/kWp/day potential for a PV power plant with free- average of daily totals standing fixed-mounted c-Si modules, mounted at optimum tilt to maximize yearly PV production Note: in italics, we indicate data layer that have been accuracy enhanced indirectly from GHI and DNI Table 1.1 describe the primary data layers GHI, DIF and DNI that have been processed by the regionally adapted solar model as part of the country data delivery. The other data layers, such as GTI and PVOUT were re-computed using the accuracy enhanced GHI, DIF and DNI. © 2018 Solargis page 10 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 2 Solargis database 2.1 Solar resource data calculated by satellite-based solar model Solar radiation is calculated by numerical models, which are parameterized by a set of inputs characterizing the cloud transmittance, state of the atmosphere and terrain conditions. A comprehensive overview of the Solargis model is made available in a recent book publication [1]. The methodology is also described in [2, 3]. The related uncertainty and requirements for bankability are discussed in [4, 5]. In the Solargis approach, the clear-sky irradiance is calculated by the simplified SOLIS model [6]. This model allows fast calculation of clear-sky irradiance from the set of input parameters. Sun position is a deterministic parameter and is described by the algorithms with satisfactory accuracy. Stochastic variability of clear-sky atmospheric conditions is determined by changing concentrations of atmospheric constituents, namely aerosols, water vapour and ozone. Global atmospheric data, representing these constituents, are routinely calculated by world atmospheric data centres: • In Solargis, the new generation aerosol data set representing Atmospheric Optical Depth (AOD) is used. The calculation accuracy is strongly determined by quality of aerosols, especially for cloudless conditions. The aerosol data implemented by MACC-II/CAMS and MERRA-2 projects are used [7, 8, 9, 10]. • Water vapour is also highly variable in space and time, but it has lower impact on the values of solar radiation, compared to aerosols. The GFS and CFSR databases (NOAA NCEP) are used in Solargis, and the data represent the daily variability from 1994 to the present time [11, 12, 13, 14]. • Ozone absorbs solar radiation at wavelengths shorter than 0.3 µm, thus having negligible influence on the broadband solar radiation. The clouds are the most influencing factor, modulating clear-sky irradiance. Effect of clouds is calculated from the satellite data in the form of a cloud index (cloud transmittance). The cloud index is derived by relating radiance recorded by the satellite in spectral channels and surface albedo to the cloud optical properties. In Solargis, the modified calculation scheme of Cano has been adopted to retrieve cloud optical properties from the satellite data [15, 16]. To calculate all-sky irradiance in each time step, the clear-sky global horizontal irradiance is coupled with the cloud index. Direct Normal Irradiance (DNI) is calculated from Global Horizontal Irradiance (GHI) using a modified Dirindex model [17]. Diffuse irradiance for tilted surfaces is calculated by the Perez model [18]. The calculation procedure also includes terrain disaggregation, while the spatial resolution is enhanced with use of the digital terrain model to 250 meters [19]. Solargis model version 2.1 has been used. Table 2.1 summarizes technical parameters of the model inputs and of the primary data outputs. This model was enhanced by regional adaptation based on the ground solar measurements (Chapter 4). © 2018 Solargis page 11 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 2.1: Input data used in the Solargis model and related GHI and DNI outputs for Malawi Inputs into the Source Time representation Original Approx. grid Solargis model of input data time step resolution Cloud index Meteosat MFG and MSG 1994 to 2004 30 minutes 2.8 x 3.3 km satellites (EUMETSAT) 2005 to date 15 minutes 3.3 x 4.0 km Atmospheric optical MACC/CAMS* (ECMWF) 2003 to date 3 hours 75 km and 125 km depth (aerosols)* MERRA-2 (NASA) 1994 to 2002 1 hour 50 km Water vapour CFSR/GFS 1994 to date 1 hour 35 and 55 km (NOAA) Elevation and horizon SRTM-3 - - 250 m (SRTM) Solargis primary data - 1994 to date 15 minutes 250 m outputs (GHI and DNI) * Aerosol data for 2003-2012 come from the reanalysis database; the data representing years 2013-present are derived from near- real time (NRT) operational model 2.2 Combined use of satellite-based model and measurements The fundamental difference between a satellite observation and a ground measurement is that a signal received by the satellite radiometer integrates a large area, while a ground station represents a pinpoint measurement. This results in a mismatch when comparing instantaneous values from these two observation instruments, mainly during intermittent cloudy weather and changing aerosol load. Nearly half of the hourly Root Mean Square Deviation (RMSD) for GHI and DNI can be attributed to this mismatch (value at sub-pixel scale), which is also known as the “nugget effect” [20]. The satellite pixel is not capable of describing the inter-pixel variability in complex regions, where within one pixel, diverse geographical conditions vary (e.g. along the coast, near mountains). In addition, the coarse spatial resolution of atmospheric databases such as aerosols or water vapour is not capable of describing local patterns of the state of the atmosphere. These features can be seen in the satellite GHI and DNI higher bias due to an imperfect description of aerosol load as well as identification of local specific cloud properties from satellite data. Satellite data have inherent inaccuracies, which have a certain degree of geographical and time variability. DNI is particularly sensitive to the variability of cloud information, aerosols, water vapour, and terrain shading. The relationship between the uncertainty of global and direct irradiance is nonlinear. Often, a negligible error in global irradiance may have a high impact on the direct irradiance component. The solar energy projects require representative and accurate GHI and DNI time series. The satellite-derived databases are used to describe long-term solar resource for a specific site. However, their problem when compared to the high-quality ground measurements is a slightly higher bias and partial disagreement of frequency distribution functions, which may limit their potential to record the occurrence of extreme situations (e.g. very low atmospheric turbidity resulting in a high DNI and GHI). A solution is to correlate satellite-derived data with ground measurements to understand the source of the discrepancy, and subsequently, to improve the accuracy of the resulting time series. © 2018 Solargis page 12 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 The Solargis satellite-derived data are correlated with ground measurement data with two objectives: • Improvement of the overall bias (removal of systematic deviations) • Improvement of the fit of the frequency distribution of values. Limited spatial and temporal resolution of the input data, and the simplified nature of the models results in the occurrence of systematic and random deviations of the model outputs when compared to the ground observations. The deviations in the satellite-computed data, which have a systematic nature, can be reduced by site adaptation or regional adaptation methods. 2.2.1 Site adaptation vs. regional adaptation The terminology related to the procedure of improving the accuracy of the satellite data is not harmonized, and various terms are used: • Correlation of ground measurements and satellite-based data; • Calibration of the satellite model (its inputs and parameters); • Site adaptation or regional adaptation of satellite-based data. The term site adaptation or regional adaptation is more general and well explains the concept of adapting the satellite-based model (by correlation, calibration, fitting and recalculation) to the ground measured data. • Site adaptation aims to adapt the characteristics of the satellite-based time series to the site-specific conditions described by local measurements. • Regional adaptation aims to identify systematic patterns of deviation at the regional scale and correct them rather than focusing on a specific site. In this study, we apply a regional adaptation of the Solargis model to improve its performance at the regional level. Its advantage is that the database in the given region has reduced uncertainty over the whole territory which has been assumed. To obtain the best accuracy for a specific location, it is preferred to apply the model site adaptation, as it focuses on matching the model outputs to the specific local climate conditions described by the ground measurements. 2.2.2 Conditions to be met Four conditions are important for successful adaptation of the satellite-based model: 1. High quality DNI and GHI ground measurements for at least 12 months must be available; optimally data for 2 or 3 years should be used: for Malawi, the measurements are available for a period of 24+ months. 2. For regional-adaptation, the sites should be distributed over the whole territory, to provide information for the major climatic regions: for Malawi, three sites are selected to represent whole territory 3. High quality satellite data must be used, with consistent quality over the whole period of data. Solargis model fulfils this condition. 4. There has to be a systematic difference identified between both data sources. © 2018 Solargis page 13 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Systematic difference can be measured by two characteristics: • Bias (offset) • Systematic deviation in the distribution of hourly or daily values (in the histogram) Systematic difference can be stable over the year or it can slightly change seasonally for certain meteorological conditions (e.g. typical cloud formation during a day, seasonal air pollution). The data analysis should distinguish systematic differences from those arising at occasional events, such as extreme storms. The episodically occurring differences may mislead the results of adaptation, especially if a short period of ground measurements is only available. If one of the four above-mentioned conditions is not fulfilled, the model adaptation will not provide the expected results. In fact, such an attempt may provide even worse results. For the quantitative assessment of the accuracy enhancement procedures, the following metrics are used: • Metrics based on the comparison of all pairs of the hourly daytime data values: Mean Bias, Root Mean Square Deviation (RMSD) and histogram in an absolute and relative form (divided by the daytime mean DNI values); • Metrics based on the difference of the cumulative distribution functions: KSI (Kolmogorov-Smirnov test Integral) [21] The normalized KSI is defined as an integral of absolute differences of two cumulative distribution functions D normalized by the integral of critical value acritical: , where critical value depends on the number of the data pairs N. As the KSI value is dependent on the size of the sample, the KSI measure may be used only for the relative comparison of fit of cumulative distribution of irradiance values. More about the Solargis site adaptation can be found in [22] and more general description is in [23]. © 2018 Solargis page 14 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 3 Ground measurements in Malawi 3.1 Solar meteorological stations: specifications and data Data from the measuring stations in Malawi was collected and harmonized with the objective of acquiring reference solar radiation data for reducing the uncertainty of the solar models. The quality data from three meteorological stations were available for this assessment (Tables 3.1 and 3.2, Figure 3.1). Positions and detailed information about measurement sites is also available on the Global Solar Atlas website: https://globalsolaratlas.info/?c=-13.239945,35.189095,7&s=-13.453737,34.046592&e=1. The instruments are summarized in Tables 3.3 to 3.5. Table 3.1: Summary of information for installed solar meteorological stations in Malawi Project name Solar Resource Mapping in Malawi Project ID 7173190 Project framework Energy Sector Management Assistance Program (ESMAP) Project leader Solargis s.r.o. Data measurement points 3 stations (1x TIER 1 and 2x TIER 2) Measurement service provider GeoSUN Africa, assisted by SGS Malawi Maintenance service provider Malawi Meteorological Services (trained by GeoSUN Africa) Table 3.2: Overview information on solar meteorological stations operated in Malawi No. Site name Latitude Longitude Altitude Measurement station host [º] [º] [m a.s.l.] 1 Chileka -15.67984 34.97229 767 Malawi Meteorological Services 2 Kasungu -13.01530 33.46840 1065 Malawi Meteorological Services 3 Mzuzu -11.41990 33.99530 1285 Mzuzu University © 2018 Solargis page 15 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Figure 3.1: Position of the solar meteorological stations Table 3.3: Instruments used for measuring solar radiation No. Site name Station type DNI GHI DIF 1 Chileka TIER 1 CHP 1, Kipp & Zonen CMP 10, Kipp & Zonen CMP 10, Kipp & Zonen 2 Kasungu TIER 2 RSR 2, Irradiance Inc. CMP 10, Kipp & Zonen RSR 2, Irradiance Inc. RSR 2, Irradiance Inc. 3 Mzuzu TIER 2 RSR 2, Irradiance Inc. CMP 10, Kipp & Zonen RSR 2, Irradiance Inc. RSR 2, Irradiance Inc. © 2018 Solargis page 16 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 3.4: Instruments installed at Tier 1 and Tier 2 solar meteorological stations Parameter Instrument Type Manufacturer Uncertainty GHI Secondary standard pyranometer CMP 10 Kipp & Zonen < ±2 % (daily) DIF Secondary standard pyranometer CMP 10 Kipp & Zonen < ±2 % (daily) DNI First class pyrheliometer CHP 1 Kipp & Zonen < 1 % (daily) GHI 2 Rotating shadowband radiometer with LI200 RSR 2 Irradiance Inc. Indicatively ±5 % DIF 2 Rotating shadowband radiometer with LI200 RSR 2 Irradiance Inc. Indicatively ±8 % DNI 2 Rotating shadowband radiometer with LI200 RSR 2 Irradiance Inc. Indicatively ±5 % WS TIER 1 station wind speed sensor (at 10 m) 05103 R.M. Young ±0.3 m/s TIER 2 station wind speed sensor (at 3 m) 014A Met One ±1.5 % WD TIER 1 station wind direction sensor (at 10 m) 05103 R.M. Young ±3 ° TIER 2 station wind direction sensor (at 3 m) 024A Met One ±5° TEMP Temperature probe (at 2 m) HMP 155 Vaisala ±0.45°C RH Relative humidity probe HMP 155 Vaisala ±1.7% RH AP Barometric pressure sensor PTB110 Vaisala ±1.5 hPa Hydrological PWAT Tipping-bucket rain gauge TB4 ±3% services Campbell ± (0.06% of reading + - Data logger CR1000 Scientific offset) At the TIER 2 stations, solar radiation is measured by secondary standard pyranometers of high quality and accuracy (GHI), and by RSR 2 instruments (GHI, DNI and DIF). At TIER 1 station we used secondary standard pyranometers (GHI and DIF) and a first class pyrheliometer (DNI). Overview of the data availability, time step and measured parameters is shown in Tables 3.4, 3.5 and Figure 3.2. Table 3.5: Overview information on solar meteorological stations operating in the region No. Site name Parameters Time step Period of data used in this study 1 Chileka GHI, DNI, DIF 1 min 19 March 2016 – 31 March 2018 2 Kasungu GHI, GHI2, DNI2, DIF2 1 min 18 March 2016 – 31 March 2018 3 Mzuzu GHI, GHI2, DNI2, DIF2 1 min 18 March 2016 – 31 March 2018 Year, month 2016 2017 2018 Station 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Chileka Kasungu Mzuzu Figure 3.2: Availability of solar resource measurements (GHI, DNI and DIF). © 2018 Solargis page 17 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 In this report, a complete set of data from the measurement campaign is used for regional adaptation. As the measurement stations have been installed in March 2016, the regular envisaged period for the data analysis starts in April 2016 and ends in March 2018. 3.2 Quality control and harmonization of solar measurements Prior to the comparison with satellite-based solar resource data, the ground-measured irradiance was quality- controlled by Solargis. Quality Control (QC) is based on methods defined in SERI QC procedures, Younes et al. and Long et al. [24, 25, 26, 27] and implemented in-house by the company Solargis. The tests are applied in two runs: (i) the automatic tests are run to identify the obvious issues; next (ii) by visual inspection we identify and flag inconsistencies, which are of a more complex nature. Visual inspection is an iterative and time-consuming process. The quality control methods and results are in detail described in the report “24 Months Solar Resource Report, Republic of Malawi, Report number: 141-07/2018” [28], here we present only a brief summary. Based on the quality control results we conclude that the solar radiation measurements come from the high accuracy (CMP10, CHP1) and medium accuracy (RSR2) equipment that is professionally operated and maintained. Some issues were identified during the data quality control of the whole period of ground measurement campaign: • Several periods of inconsistency between independent GHI, DNI and DIF measurements is seen in the data (Chileka station), mainly in the first year of measurements. This might be a result of insufficient cleaning. The Malawi Meteorological Services took measures to improve the cleaning schedule in the second year. • Effect of dew on CPM10 instruments in the morning hours (mainly Kasungu and Mzuzu station). • The measurements are partially affected by morning or late afternoon shading from surrounding objects or trees (Mzuzu station). The measurements affected by these operation conditions were excluded from further analyses. The issues above have implication on the uncertainty of ground measurements. Therefore the affected data values had to be excluded so that they do not impact the results of the Solargis model adaptation (Chapter 4). In evaluation of the measurement uncertainty several factors are considered: 1. Thermopile pyranometer CMP10 has lower nominal uncertainty than the RSR2 instrument. Therefore, use of CMP10 data has preference over RSR2, in the regional adaptation. 2. The thermopile pyranometers are more susceptible to soiling. Some issues of this type were identified in the first year of measurements in the Chileka station, and affected data was excluded from further analysis. 3. Instruments are used in challenging environmental conditions (higher temperature, high humidity, dew, etc.), and their possible impact was evaluated and the measured data values excluded. Table 3.6 summarises the finding of the quality control. © 2018 Solargis page 18 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 3.6: Quality control summary Description Chileka Kasungu Mzuzu Station description, metadata Instrument accuracy Instrument calibration Data structure Cleaning and maintenance information Time reference Quality control complexity Quality control results Time period Other issues Legend: Quality flag Very good Good Sufficient Problematic Insufficient Not specified © 2018 Solargis page 19 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 4 Regional adaptation of Solargis model Ground measurements from three solar meteorological stations in Malawi are used for the regional adaptation of the Solargis model (Figure 3.1). In addition, measurements from Misamfu station in Zambia was used to improve regional adaptation of the model in the Northwest. After adapting the model for the region, it was run to produce accuracy enhanced time series. By aggregating the time series, for every grid cell, the model has been used to generate a new version of GHI and DNI data layers and maps of Malawi. 4.1 Solargis method Solargis regional model adaptation aims at reducing the bias (systematic deviation) at the level of the region. Bias reduction improves also RMSD (Root Mean Square Deviation, i.e. random deviation) and KSI (Kolmogorov Smirnov Index, i.e. difference between frequency distribution of the measured and satellite-based data). The original Solargis data show a regional pattern of overestimation, compared to the ground measurements, for both GHI and DNI. The highest difference is seen at Mzuzu station, where the systematic deviation between the ground measurements and the satellite data exceeds 12% and 20%, for yearly values GHI and DNI respectively. Such discrepancy is far beyond uncertainty usually seen in Solargis satellite data in this region, as typical expectation of the satellite data systematic error is within ±6 to ±8% for GHI and ±12 to ±15% for DNI). The detailed inspection of the ground measurements and satellite data indicates three possible sources (and their combination) of this deviation: • Specific geographical conditions of the Mzuzu site, where local climate is influenced by short distance to the Lake Malawi and orography, both determining strongly changing microclimate patterns. The change of yearly solar radiation values − within a distance of 20 km in East-West direction − is 9% and 21% for GHI and DNI, respectively. Considering size of the satellite pixel (approx. 3.3 x 4.0 km), the satellite sensors are capable capturing these strong microclimate gradients only partially. The limitations given by the spatial resolution of the satellite data is one source of its mismatch with pinpoint measurements at the meteorological station. • Ground measurements at the Mzuzu site are affected to some extent by local shading (trees and buildings nearby the station). The high frequency variability of solar irradiance, driven by small scattered and fast changing clouds, makes it difficult to distinguish the local shading from drop of irradiance due to clouds. Moreover, the permanent shading of the sky dome by surrounding objects (in certain sky segments up to 17 degrees above horizon) partly reduces the diffuse irradiance, thus influencing the measurements of GHI and DIF, subsequent influencing calculation of DNI (GHI and DIF measured by RSR instrument). • Performance of current satellite models is less accurate in conditions of high occurrence of scattered clouds and persistent clouds in the tropical regions. Similar features (mainly microclimatic gradient) can be found also in Chileka site, where higher mismatch between the model data and measurements is seen. The performance of the satellite model shows higher discrepancies in conditions of high intermittency of solar irradiance. The DNI is overestimated mostly for lower irradiance (Figure 4.1, left) while GHI is overestimated mainly in medium to high irradiance values (Figure 4.1, right). This indicates insufficiencies (i) in the clear-sky model (quantifying the cloudless conditions), which is mainly controlled by aerosol data input as well as (ii) limitations of the cloud model. In regional adaptation, we have © 2018 Solargis page 20 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 focused on (i) accuracy improvement of the model inputs, namely Aerosol Optical Depth (AOD), and (ii) on reducing the overestimation by de-biasing the GHI and DNI values. Figure 4.1: Comparison of hourly original satellite-model DNI and GHI with ground measurements. Model systematically overestimates the measurements – Chileka Left: DNI, Right: GHI More about the Solargis site adaptation method in [22]. 4.1.1 Reduction of systematic deviation between satellite and measured data Deviation between the original satellite model output and ground-measurements was analysed individually for each site. We focussed on understanding of the following differences: 1. Deviation for the entire period of measurements 2. Seasonal patterns of deviation 3. Deviation patterns in various weather situations 4. Differences in the cumulative distribution of values. Understanding of discrepancies and their sources was determining factor for the selection of appropriate methods for reduction of systematic deviation between satellite-based calculation and measured data. The reduction was conducted in two steps: 1. Regional aerosol correction. Coefficients for adaptation of Aerosol Optical Depth (ADO) were derived for individual ESMAP meteorological stations (Chapter 4.1.1). Aerosol adaptation coefficients were interpolated between the solar meteorological sites to extend over the territory of Malawi (Chapter 4.1.2). Adapted (accuracy enhanced) aerosol values were used for the recalculation of the satellite model outputs. 2. Regional GHI and DNI de-biasing. Larger residual discrepancies were reduced between the ground measurements and the AOD corrected model, in the second step, by regional de-biasing of the GHI and DNI layers. GHI and DNI de-biasing correction factors derived for the ESMAP stations were spatially interpolated and applied to calculate the final GHI and DNI data layers. © 2018 Solargis page 21 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 4.1.2 Regional aerosol correction Based on the comparison of satellite model data and ground measurements, correction factors for AOD (Atmospheric Optical Depth) values were calculated. They were developed by comparing the cloudless situations with theoretical clear-sky profiles. The input aerosols were corrected separately for low and medium to high aerosol load concentrations. For each group a separate set of monthly correction factors was identified. Next, these factors were harmonized to avoid abrupt month-by-month changes that might be rather a representation of specific weather situations, rather than a systematic problem in the model. In this phase, correction factors from the neighbouring sites were also compared to avoid issues in a spatial context, especially if sites are located in similar geographic conditions. In a case of nearby sites with contradicting deviations, the corrections for individual sites have to be balanced to avoid high spatial mismatch. The bias is removed only partially to maintain the spatial consistency of corrections. This approach helps to avoid steep changes of correction coefficients in the area. In case of Malawi, the deviations in individual stations were spatially consistent and only small changes of correction coefficients were introduced by harmonization. For each site, the adaptation procedure results in a set of monthly correction values for low and medium AOD load. The correction was developed to respect the seasonal and spatial context of the data. The main objective of spatial interpolation of aerosol correction coefficients is to extend the correction coefficients identified at the sites to the territory of Malawi to a wider region. To achieve this goal, an interpolation technique was used. The selection of interpolator is based on the assumption that the spatial distribution of aerosols is controlled by air mass movement. Applying the spatial interpolation, we extended the validity of the correction factors identified at the solar meteorological stations to the entire territory of Malawi and neighbouring areas. To maintain the stability of model in a wider regional context, also effect of corrections from neighbouring regions (Zambia) was introduced. The interpolation was applied separately for each month and two aerosol load conditions. The output of the interpolation is a set of 24 aerosol correction layers (2 layers per month). In the last step, the satellite-based model was recalculated using aerosol correction layers for the full period of 24 years for the whole territory of Malawi. Thus, the consistency of accuracy-enhanced GHI, DNI and DIF components is maintained. The aerosol adaptation method removes one important source of discrepancies between satellite-model data and the ground measurements. Because of fundamental difference between the modelling using satellite data and ground measurements, even if bias was to a large extent reduced, some mismatch between the data is still present in the data after the AOD correction. The discrepancies are present in the regions with a high bias of original data (mainly at the Mzuzu station). The residual bias was reduced by regional de-biasing of GHI and DNI. 4.1.3 Regional DNI and GHI de-biasing The residual differences of AOD corrected model data were analysed at the measurement stations and the yearly GHI and DNI correction (de-biasing) factors were derived. The proposed reduction of the bias is applied within the regional context, where the representativeness of the individual stations within the given region (Chapter 4.1), and quality of measurements (Chapter 3.3) are considered. By applying the correction factors, we aim to reduce regional bias patterns, rather than the specific microclimatic features. Therefore, for individual stations a small residual bias at the level of solar meteorological stations can be expected. Similarly to AOD correction factors, the GHI and DNI de-biasing factors derived for measurement stations were interpolated for the entire territory. Also, for this correction factors the information from neighbouring regions (Zambia) was included in the interpolation scheme to maintain stability and continuity in a wider context. As a result, two correction layers were derived, one for GHI and one for DNI. In the final step the correction factors were used to de-bias GHI and DNI data for the whole territory. Final corrected GHI and DNI data layers were used for spatial analysis of solar resource in Malawi, and for calculation of © 2018 Solargis page 22 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 secondary data layers: diffuse horizontal irradiation (DIF), global radiation on optimally tilted surface (GTI) and potential photovoltaic production (PVOUT). The main focus of the regional adaptation is to determinate correction coefficients for individual sites that are used to remove seasonal and annual systematic deviations in the regional context. The residual discrepancies present in the regionally-adapted data can be removed only in the local context as their source are locally-specific features such as pollution in cities. Such residual discrepancies cannot be extrapolated, and they can be only addressed in the model site adaptation. The coefficients of the regional adaptation were derived for the period with the overlapping ground measurements and model data. These coefficients were implemented in the model to recalculate the full-time series of solar radiation. 4.2 Results and validation 4.2.1 Accuracy estimate of DNI and GHI at the solar meteorological stations Comparing original Solargis data to the ground measurements shows a regional model overestimation pattern for both GHI and DNI,. The model adaptation, for the region, allowed reducing a large proportion of the systematic mismatch between satellite-based data and ground measurements. Tables 4.1 to 4.2 summarize validation of the regional adaptation for all three solar meteorological stations. The original Solargis data represent output of the model, which is based on a standard calculation scheme without considering any corrections derived from the measurements. The regionally-adapted model includes the effect of correction factors calculated from the ESMAP project measurements in Malawi (Chapter 4.1). The GHI validation statistics (Table 4.2) show a comparison of the accuracy-enhanced GHI to the measurements from secondary standard CMP10 thermopile pyranometers. The DNI validation statistics (Table 4.1) shows a comparison of the accuracy enhanced DNI to the measurements from CHP1 (Chileka) and RSR2 (Kasungu and Mzuzu) instruments. The measurements from RSR2 instrument have higher nominal uncertainty. Table 4.5 shows the difference between yearly GHI and DNI values − before and after the model regional adaptation. Terms are explained in Glossary. Absolute values of bias are calculated for daytime hours only. Table 4.1: Direct Normal Irradiance: bias before and after regional model adaptation DNI after regional Meteo station Original DNI data adaptation Bias Bias Bias Bias 2 2 [kWh/m ] [%] [kWh/m ] [%] Chileka 42 10.5 3 0.7 Kasungu 41 9.6 -2 -0.4 Mzuzu 78 20.5 7 1.8 Mean 54 13.5 3 0.7 Standard deviation 21 6.1 4 1.1 © 2018 Solargis page 23 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 4.2: Global Horizontal Irradiance: bias before and after regional model adaptation GHI after regional Meteo station Original GHI data adaptation Bias Bias Bias Bias [kWh/m2] [%] [kWh/m2] [%] Chileka 38 8.3 4 0.9 Kasungu 27 5.4 5 0.9 Mzuzu 59 12.7 16 3.5 Mean 41 8.8 8 1.8 Standard deviation 16 3.7 7 1.5 As a result, at the level of individual sites in Malawi, the mean bias of the adapted GHI and DNI values stays below 1.0% for Chileka and Kasungu stations. For Mzuzu station it is slightly higher: 1.8% and 3.5% for DNI and GHI, respectively. The standard deviation of bias values, considering all three stations, is 1.1% and 1.5% for the DNI and GHI respectively, which is comparable to the inherent uncertainty of ground sensors. A significant improvement was achieved for both Direct Normal Irradiance (DNI) and Global Horizontal Irradiance (GHI). The average bias of DNI for all stations dropped from the range of 9.6% to 20.5% to the range of -0.4% to 1.8%, and the standard deviation considerably reduced from 6.1% to 1.1% by regional adaptation. This confirms removal of region-specific model issues where bias of original DNI data for one station exceeded 20% (Mzuzu). The average Global Horizontal Irradiance (GHI) bias for all stations after adaptation is 1.8% with standard deviation of 1.5% (Table 4.2). High bias of original data, in the range of 5.4% to 12.7% was reduced to the range 0.9% to 3.5% by regional adaptation. The higher residual bias at Mzuzu station is a result of lower representativeness of the measurements from this station (Chapter 4.1) in the regional context. Therefore, in this station, the adaptation correction coefficients were calculated with higher freedom than for the other stations. The regionally-adapted model values better represent the geographical variability of DNI and GHI solar resource. Some residual discrepancies still remain in the output data, but their removal is beyond the possibilities of regional adaptation. The residuals can only be removed for the locations of the meteorological stations in the context of the site-adaptation. Moreover, the residual discrepancies should be evaluated within the context of the quality and accuracy of ground measurements (Chapter 3.3). Table 4.3 shows the comparison of long-term annual averages of GHI and DNI derived from original and regionally adapted GIS data layers. Both the GHI and DNI values from the accuracy-enhanced model are lower, which indicates that the original model has trend of systematic overestimation in the region. Table 4.3: Comparison of long-term average of yearly summaries of original and regionally-adapted values Meteo station DNI annual values* GHI annual values* Original Adapted Difference Original Adapted Difference [kWh/m2] [kWh/m2] [%] [kWh/m2] [kWh/m2] [%] Chileka 1765 1608 -8.9 1993 1859 -6.7 Kasungu 1825 1647 -9.7 2077 1988 -4.3 Mzuzu 1737 1449 -16.6 2022 1861 -8.0 * Values represent GIS data layers, and they may slightly deviate from the version of the point model that calculates site-specific time series. This difference is due to spatial resolution and method of terrain disaggregation in the grid model. © 2018 Solargis page 24 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 4.2.2 Accuracy-enhanced DNI and GHI maps Regionally adapted DNI and GHI long-term averages are lower in the entire region, compared to the original data (Figure 4.2 and 4.3). The absolute change of DNI is higher, as DNI is more sensitive to changes of aerosol load introduced in the first step of the regional adaptation. The change of GHI and DNI values due to regional adaptation are most visible in the region around the Mzuzu station. In case of GHI, slightly higher correction is seen also in the region around Chileka station (Figure 4.3). Figure 4.2: Map of differences in yearly DNI between the original and regionally adapted model © 2018 Solargis page 25 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Figure 4.3: Map of differences in yearly GHI between the original and regionally-adapted model © 2018 Solargis page 26 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 5 Solar resource maps of Malawi 5.1 Accuracy enhanced maps of DNI and GHI Figure 5.1: Accuracy-enhanced DNI yearly long-term average © 2018 Solargis page 27 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 After the regional adaptation of the Solargis satellite model, full time series representing a period of 24 years (1994 to 2017) are aggregated into long-term yearly averages of DNI and GHI (Figure 5.1 and 5.2). Important outcomes of this exercise are two maps with reduced uncertainty (Chapter 5.2). The spatial (grid) resolution of the output maps is 1 km. Figure 5.2: Accuracy-enhanced GHI yearly long-term average © 2018 Solargis page 28 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 5.2 Uncertainty of solar resource maps Solargis model is based on the use of the best available algorithms and input data, and it has been calibrated and validated for all geographies. Therefore, the model has robust and uniform behaviour in all conditions. Validation sites in Malawi show consistent bias within the expected range, except for the Mzuzu station. The results reflect specific microclimatic conditions and lower representativeness at some sites of the regional context (Mzuzu and partially Chileka, Chapter 4.1) as well as limitations of the Solargis model. The regional adaptation reduced these discrepancies (Chapter 4.1.3). The adaptation was applied to remove systematic model deviations, not the discrepancies coming from specific microclimate. After the model adaptation the bias values are significantly lower (Tables 4.1 and 4.2). For practical use, the statistical measures of accuracy have to be converted into uncertainty, which better characterizes probabilistic nature of a possible error of the model estimate. Uncertainty is based on the assumption of normal distribution of errors of solar radiation model, which is a simplification given by availability of validation data and limited geographical knowledge. Typically, in the industry, a long-term yearly average estimate is required, often denoted as P50 value (in case of normal distribution this is equivalent to median). Besides P50, project developers, technical consultants and the finance industry request about uncertainty of long-term yearly GHI or DNI and P90 estimate, that is more conservative and it is calculated by subtracting the uncertainty from P50 value. The uncertainty in this report is calculated for 80% probability of occurrence, thus P90 value shows an estimate at 90% probability of exceedance. The uncertainty of regionally adapted satellite-based DNI and GHI is determined by uncertainty of the model, ground measurements, and the model adaptation method. More specifically it depends on [29]: 1. Parameterization and adaptation of numerical models integrated in Solargis for the given data inputs and their ability to generate accurate results for various geographical and time-variable conditions: • Data inputs into Solargis model: accuracy of Meteosat satellite data, MACC-II/CAMS and MERRA-2 aerosols and GFS/CFSR/GFS water vapour • Solis clear-sky model and its capability to properly characterize various states of the atmosphere • Simulation accuracy of the Solargis cloud transmittance algorithms, being able to properly distinguish different states of various surface types, albedo, clouds and fog • Diffuse and direct decomposition by Perez model • Transposition from global horizontal to in-plane irradiance (for GTI) by Perez model • Terrain shading and disaggregation by Ruiz-Arias model 2. Uncertainty of the ground-measurements, which is determined by: • Accuracy of the instruments • Maintenance practices, including sensor cleaning, service and calibration • Data post-processing and quality control procedures. 3. Uncertainty of the model adaptation at regional scale and residual uncertainty after adaptation The uncertainty of the estimate Uncertest in this study is estimated from the model uncertainty of the Solargis model Uncertmodel, the uncertainty of the measurements Uncertmeas, and the uncertainty of the model adaptation method Uncertadapt: ,.@ = B+31,4 6 + +,-. 6 + -1-D@ 6 Combined uncertainty of the yearly estimate Uncertest is estimated empirically, based on the experience and accuracy evaluation of the model and measurements (Chapter 3 and Chapter 4.2.1). We consider it to have © 2018 Solargis page 29 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 probabilistic nature and it is derived primarily from the statistical measures calculated at three solar meteorological stations. The expert estimate of the combined user uncertainty in this report assumes 80% probability of occurrence of values, i.e. 90% probability of exceedance. Table 5.1 summarizes the estimated uncertainty. The uncertainty from the interannual variability of solar resource is not considered in this study. Based on today’s knowledge and experience, we assume that the lowest achievable uncertainty (assuming uncertainty of the model and of the measurements at P90) of satellite-based long-term estimates is indicatively ±2.5% for GHI and ±3.5% for DNI. The uncertainty, at best possible limits, can only be achieved if the following conditions are met: • Best available models and approaches are applied • Input data (satellite, atmospheric, etc.) are quality controlled and homogenized • Satellite model is adapted for local geography by high quality ground measurements, available for a period of at least 4 to 5 years • Ground measurements are available for GHI, DNI and DIF, measured by high-standard meteorological instruments and equipment, applying best operation and maintenance practices. The lowest uncertainty levels can only be achieved by site-adaptation for a very local region around meteorological stations with site-specific microclimatic conditions recorded in ground measurements. In the case of the regional adaptation used in this study, the uncertainty is usually higher because it describes uncertainty of any selected location in the broader region. Moreover, a residual discrepancy between ground measurements, and the model data can be found after regional adaptation (Tables 4.1 and 4.2). The model adaptation approach, described in this study, is designed to remove only regional discrepancy patterns, not to resolve site-specific issues. The uncertainty levels of regionally adapted data (Table 5.1) are higher than the best achievable results by site- specific adaptation. It is estimated that for the majority of Malawi territory the regionally-adapted model has uncertainty of yearly values at the level of ±4% to ±5% for GHI, and ±5% to ±7% for DNI. We expect higher uncertainty in regions with more complex geography, which is partly a result of uncertainty of ground measurements, suboptimal distribution of measurement stations throughout country and inherently higher model uncertainty in regions with specific micro-climate (e.g. occurrence of convective clouds close to steep slopes of the Lake Malawi). These uncertainties can be reduced by the use of longer period of high quality ground measurements and by measurements from more meteorological stations located in the region. Table 5.1: Uncertainty of the model estimate for original and regionally-adapted annual GHI and DNI and how does it compare to the best-achievable uncertainty case. Direct Normal Irradiation Global Horizontal Irradiation Global Tilted Irradiation Low Medium Low Medium Low Medium Original data < ±12% < ±21% < ±9% < ±13% < ±9% < ±13% After adaptation ±5% to ±7% < ±10% ±4% to ±5% < ±6% ±4.5% to ±5% < ±7% Best-achievable* ±3.5% - ±2.5% - ±3.0% The expected model uncertainty in the regions of Malawi is presented in Table 5.2 and Figure 5.3. The map is derived by expert evaluation of the distribution and the quality of ground measurements within the context of regional geography and capability of regional adaptation method. © 2018 Solargis page 30 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Table 5.2: Geographic distribution of the model uncertainty Model uncertainty Region DNI GHI GTI for yearly estimates Low Flat and monotonous terrain ±5% to ±7% ±4% to ±5% < ±4.5% to 6% Medium Complex terrain < 10% < ±6% < ±7 % Figure 5.3: Geographic distribution of the model uncertainty in Malawi L: low; M: medium © 2018 Solargis page 31 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 6 Conclusions This project reduced the uncertainty of DNI and GHI solar resource database and the resulting yearly and monthly maps representing the territory of Malawi. It benefits form the systematic and diligent work on (i) setting up a network of solar meteorological stations with high-standard equipment and on (ii) implementation of rigorous practices in operation and maintenance of solar equipment. Well-linked to this infrastructure is satellite-based solar radiation model Solargis, which has proven quality and reliability of time series, and derived site-specific data products and map-based outputs. Reduced uncertainty The typical uncertainty of the Solargis model estimate has been reduced from the original range of ±12.0% (exceptionally for specific regions above 20%) for DNI yearly values to the range of ±5.0% to ±7.0% (exceptionally ±10.0%) for accuracy enhanced values. For yearly GHI the uncertainty reduction is seen from the original range of ±9.0% to the range of ±4.0% to ±5% (exceptionally ±6.0%) for the accuracy enhanced values. The uncertainty in Malawi is split into low and medium uncertainty regions. Besides reducing systematic deviation (bias), the regional model adaptation also results in the indirect improvement of other data quality indicators such as reducing random deviation (quantified by Root Mean Square Deviation) and by improving the probability distribution of hourly values (quantified by Kolmogorov-Smirnoff Index). There is direct benefit in using higher-quality DNI and GHI data in the solar energy yield estimation, which in turn is used for optimising a technical design and for calculation of financial parameters of a project. Role of solar measuring stations in maintaining sustainable solar data infrastructure Receiving data from a number of high-quality measuring stations enables an improved understanding of the geographical and temporal variability of solar resource in regions of Malawi. Even though regional adaptation reduced uncertainty, it is important to maintain the operation of the solar meteorological stations, with special focus on the following cases: • For new sites, relevant to any larger solar power project, it is important to install and operate a solar meteorological station with objective of reducing uncertainty of long-term estimates to achievable minimum (see Table 5.1). • For existing sites, the meteorological stations together with satellite data make it possible to maintain high quality and bankability of solar resource and meteorological data for sustainable performance evaluation of solar power plants in the region. • Keeping solar measuring stations in operation is of strategic importance to maintain quality of satellite models and of solar power forecasts. © 2018 Solargis page 32 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 7 List of figures Figure 3.1: Position of the solar meteorological stations ......................................................................................... 16 Figure 3.2: Availability of solar resource measurements (GHI, DNI and DIF). ......................................................... 17 Figure 4.1: Comparison of hourly original satellite-model DNI and GHI with ground measurements.................... 21 Figure 4.2: Map of differences in yearly DNI between the original and regionally adapted model ........................ 25 Figure 4.3: Map of differences in yearly GHI between the original and regionally-adapted model ........................ 26 Figure 5.1: Accuracy-enhanced DNI yearly long-term average ................................................................................. 27 Figure 5.2: Accuracy-enhanced GHI yearly long-term average ................................................................................. 28 Figure 5.3: Geographic distribution of the model uncertainty in Malawi .................................................................. 31 © 2018 Solargis page 33 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 8 List of tables Table 0.1: Position of solar measuring stations in Malawi ...................................................................................... 8 Table 0.2: Model output changes, due to regional adaptation, at the solar measuring stations .......................... 8 Table 0.3: Uncertainty of yearly values for Malawi, for original and regionally-adapted Solargis model ............. 9 Table 1.1: Description of GIS data layers that were accuracy enhanced by regional model adaptation............ 10 Table 2.1: Input data used in the Solargis model and related GHI and DNI outputs for Malawi ......................... 12 Table 3.1: Summary of information for installed solar meteorological stations in Malawi................................. 15 Table 3.2: Overview information on solar meteorological stations operated in Malawi...................................... 15 Table 3.3: Instruments used for measuring solar radiation ................................................................................... 16 Table 3.4: Instruments installed at Tier 1 and Tier 2 solar meteorological stations ............................................ 17 Table 3.5: Overview information on solar meteorological stations operating in the region ................................ 17 Table 3.6: Quality control summary ......................................................................................................................... 19 Table 4.1: Direct Normal Irradiance: bias before and after regional model adaptation....................................... 23 Table 4.2: Global Horizontal Irradiance: bias before and after regional model adaptation ................................. 24 Table 4.3: Comparison of long-term average of yearly summaries of original and regionally-adapted values . 24 Table 5.1: Uncertainty of the model estimate for original and regionally-adapted annual GHI and DNI ............ 30 Table 5.2: Geographic distribution of the model uncertainty ................................................................................ 31 © 2018 Solargis page 34 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 9 References [1] Perez R., Cebecauer T., Šúri M., 2013. Semi-Empirical Satellite Models. In Kleissl J. (ed.) Solar Energy Forecasting and Resource Assessment. Academic press. [2] Cebecauer T., Šúri M., Perez R., High performance MSG satellite model for operational solar energy applications. ASES National Solar Conference, Phoenix, USA, 2010. [3] Šúri M., Cebecauer T., Perez P., Quality procedures of Solargis for provision site-specific solar resource information. Conference SolarPACES 2010, September 2010, Perpignan, France. [4] Cebecauer T., Suri M., Gueymard C., Uncertainty sources in satellite-derived Direct Normal Irradiance: How can prediction accuracy be improved globally? Proceedings of the SolarPACES Conference, Granada, Spain, 20-23 Sept 2011. [5] Šúri M., Cebecauer T., 2014. Satellite-based solar resource data: Model validation statistics versus user’s uncertainty. ASES SOLAR 2014 Conference, San Francisco, 7-9 July 2014. [6] Ineichen P., A broadband simplified version of the Solis clear sky model, 2008. Solar Energy, 82, 8, 758-762. [7] Morcrette J., Boucher O., Jones L., Salmond D., Bechtold P., Beljaars A., Benedetti A., Bonet A., Kaiser J.W., Razinger M., Schulz M., Serrar S., Simmons A.J., Sofiev M., Suttie M., Tompkins A., Uncht A., GEMS-AER team, 2009. Aerosol analysis and forecast in the ECMWF Integrated Forecast System. Part I: Forward modelling. Journal of Geophysical Research, 114. [8] Benedictow A. et al. 2012. Validation report of the MACC reanalysis of global atmospheric composition: Period 2003-2010, MACC-II Deliverable D83.1. [9] Cebecauer T., Šúri M., Accuracy improvements of satellite-derived solar resource based on GEMS re-analysis aerosols. Conference SolarPACES 2010, September 2010, Perpignan, France. [10] Cebecauer T., Perez R., Šúri M., Comparing performance of Solargis and SUNY satellite models using monthly and daily aerosol data. Proceedings of the ISES Solar World Congress 2011, September 2011, Kassel, Germany. [11] GFS model. http://www.nco.ncep.noaa.gov/pmb/products/gfs/ [12] CFSR model. https://climatedataguide.ucar.edu/climate-data/climate-forecast-system-reanalysis-cfsr/ [13] CFSv2 model http://www.cpc.ncep.noaa.gov/products/CFSv2/CFSv2seasonal.shtml [14] Molod, A., Takacs, L., Suarez, M., and Bacmeister, J., 2015: Development of the GEOS-5 atmospheric general circulation model: evolution from MERRA to MERRA2, Geosci. Model Dev., 8, 1339-1356, doi:10.5194/gmd-8- 1339-2015 [15] Meteosat satellites MFG and MSG, EUMETSAT. http://www.eumetsat.int/website/home/Satellites/CurrentSatellites/Meteosat/index.html [16] Cano D., Monget J.M., Albuisson M., Guillard H., Regas N., Wald L., 1986. A method for the determination of the global solar radiation from meteorological satellite data, Solar Energy, 37, 31-39. [17] Perez R., Ineichen P., Maxwell E., Seals R. and Zelenka A., 1992. Dynamic global-to-direct irradiance conversion models. ASHRAE Transactions-Research Series, pp. 354-369. [18] Perez, R., Seals R., Ineichen P., Stewart R., Menicucci D.,1987. A new simplified version of the Perez diffuse irradiance model for tilted surfaces. Solar Energy, 39, 221-232. [19] Ruiz-Arias J.A., Cebecauer T., Tovar-Pescador J., Šúri M., 2010. Spatial disaggregation of satellite-derived irradiance using a high-resolution digital elevation model. Solar Energy, 84,1644-57. © 2018 Solargis page 35 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 [20] Zelenka A., Perez R., Seals R., Renne D., 1997. Effective Accuracy of Satellite-Derived Hourly Irradiances. Theoretical Appl. Climatology, 62, 199-207. [21] Espinar B., Ramírez L., Drews A., Beyer H. G., Zarzalejo L. F., Polo J., Martín L., 2009. Analysis of different comparison parameters applied to solar radiation data from satellite and German radiometric stations. Solar Energy, 83, 1, 118-125. [22] Cebecauer T., Suri M., 2016. Site-adaptation of satellite-based DNI and GHI time series: overview and Solargis approach. SolarPACES conference, September 2016, Cape Town, South Africa. [23] Polo J., Wilbert S., Ruiz-Arias J.A., Meyer R., Gueymard C., Šúri M., Martín L., Mieslinger T., Blanc P., Grant I., Bolan J., Ineichen P., Remund J., Escobar R., Troccoli A., Sengupta M., Nielsen K.P., Renné D., Geuder N., Cebecauer T., 2016. Preliminary survey on site-adaptation techniques for satellite-derived and reanalysis solar radiation datasets. Solar Energy, 132, 25–37. [24] NREL, 1993. User’s Manual for SERI_QC Software-Assessing the Quality of Solar Radiation Data. NREL/TP- 463-5608. Golden, CO: National Renewable Energy Laboratory. [25] Younes S., Claywell R. and Munner T, 2005. Quality control of solar radiation data: Present status and proposed new approaches. Solar Energy 30, 1533-1549. [26] Long C.N., Dutton E.G., 2002. BSRN Global Network recommended QC tests, V2.0. Rapport technique BSRN. http://epic.awi.de/30083/1/BSRN_recommended_QC_tests_V2.pdf [27] Long C.N., Shi Y., 2008. An automated quality assessment and control algorithm for surface radiation measurements. Open Atmos. Sci. J., 2, 23-37. [28] Suri at el, 2018. Annual Solar Resource Report for solar meteorological stations after completion of 24 months of measurements, Republic of Malawi. Solargis report No.129-07/2018, delivered for the World Bank. [29] Šúri M., Cebecauer T., 2014. Satellite-based solar resource data: Model validation statistics versus user’s uncertainty. ASES SOLAR 2014 Conference, San Francisco, 7-9 July 2014 © 2018 Solargis page 36 of 38 Solar Model Validation Report Regional adaptation of Solargis model based on 24 months of solar measurement campaign Solargis reference No. 141-08/2018 Support information Background on Solargis Solargis is a technology company offering energy-related meteorological data, software and consultancy services to solar energy. We support industry in the site qualification, planning, financing and operation of solar energy systems for more than 18 years. We develop and operate a new generation high-resolution global database and applications integrated within Solargis® information system. Accurate, standardised and validated data help to reduce the weather-related risks and costs in system planning, performance assessment, forecasting and management of distributed solar power. Solargis is ISO 9001:2015 certified company for quality management. This report has been prepared by Tomas Cebecauer, Marcel Suri, Branislav Schnierer, Nada Suriova, Juraj Betak, Daniel Chrkavy and Artur Skoczek from Solargis All maps in this report are prepared by Solargis Solargis s.r.o., Mytna 48, 811 07 Bratislava, Slovakia Reference No. (Solargis): 141-08/2018 http://solargis.com © 2018 Solargis page 37 of 38