Uncertainty of satellite-based solar resource data Marcel Suri and Tomas Cebecauer GeoModel Solar, Slovakia 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany 22-23 October 2015
About GeoModel Solar Solar resource, meteorological and photovoltaic simulation data, software and expert services for solar electricity industry SolarGIS online database and PV software Planning and project development Asset management Forecasting Bankable consultancy and project studies Solar resource assessment Photovoltaic performance assessment Regional solar mapping and monitoring http://solargis.info http://geomodelsolar.eu 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 2
Requirements for solar resource data in PV Historical data Prospecting Planning and due diligence Recent data Monitoring Performance evaluation and asset management Forecasting Intraday Day ahead 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 3
Requirements for solar resource data in PV Historical data Prospecting Planning and due diligence Recent data Monitoring Performance evaluation and asset management Forecasting Intraday Day ahead 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 4
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 5
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 6
Historical data: old ground measurements Limited number of high-grade measuring sites Large number of lower-accuracy sites Many sites stopped operation Older data may not represent well the recent climate Typical features (lower accuracy sites) Lower accuracy equipment Less strict procedures: maintenance, calibration, cleaning Less rigorous or missing quality control and gap filling High uncertainty Difficult to evaluate if data not available (at least) at hourly time step 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 7
Historical data: old satellite models NASA the only global database Regional initiatives, e.g. NREL/SWERA Typical features Simple methods, simple inputs Low resolution Low accuracy (limited or no validation) Only monthly averages Inconsistency: spatial, time Static (no updates or sporadic) GHI difference (yearly) between NASA SSE and SolarGIS 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 8
Old practices: Historical data for longterm assessment TMY for selected sites (NSRDB in the US): Mix of measured and modeled data Monthly values of ground-measured data Spatial interpolation Monthly values of modeled data Synthetic hourly data Most common method of evaluation Expert-based weighted average of data from several sources Subjective Cannot be validated Missing continuity Missing interannual variability Deviation in longterm annual assessment ±10% to ±15% or more in GHI 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 9
Old practices: Historical data for longterm assessment TMY2 (NSRDB) Satellite-modelled data (SolarAnywhere) Source: Solar Today 6/2012 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 10
Old practices: Recent data for performance evaluation Typical situation Low accuracy sensors are installed Mistakes in installation Little maintenance Insufficient cleaning No rigorous data quality control Problematic gap filling => High (unknown) uncertainty => Disputable results 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 11
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 12
Requirements for solar resource data Global (continental) coverage Long climate record Validated accuracy (based on at least hourly data) High temporal resolution (at least hourly) High spatial resolution (at least 4-5 km) Continuity Climate history for longterm assessment Recent data for performance assessment Nowcasting and forecasting of solar power Way to go: modelled data supported by high-quality ground measurements 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 13
How to acquire solar resource data On-site measurements Satellite-based solar models Source: SolarGIS Source: GeoSUN Africa Forecasting: + numerical weather models Source: NOAA 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 14
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 15
Ground (on-site) measurements ADVANTAGES High frequency measurements (sec. to min.) Higher accuracy, if properly managed LIMITATIONS Limited geographical representation Limited time availability Costs for acquisition and operation Maintenance and calibration Data quality control Source: GeoSUN Africa 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 16
Ground (on-site) measurements ADVANTAGES High frequency measurements (sec. to min.) Higher accuracy, if properly managed LIMITATIONS Limited geographical representation Limited time availability Costs for acquisition and operation Maintenance and calibration Data quality control Source: GeoSUN Africa 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 17
Ground measurements: Instruments Instruments and their accuracy 1 DNI RSR 2 SPN1 Pyrheliometers First class ±4.5% ±5% ±1.0% GHI RSR 2 SPN1 Pyranometers Second class First class Secondary standard ±3.5% ±5% ±10% ±5% ±2% 1 Theoretical uncertainty for daily summaries, at 95% confidence level 2 Approximately, after post processing Source: Delta-T Devices, K.A.CARE, Pontificia Universidad Católica de Chile 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 18
Ground measurements: Instruments Instruments and their accuracy 1 DNI RSR 2 SPN1 Pyrheliometers First class ±3.5% ±5% ±1% GHI RSR 2 SPN1 Pyranometers Second class First class Secondary standard ±3.5% ±5% ±10% ±5% ±2% 1 Theoretical uncertainty for daily summaries, at 95% confidence level 2 Approximately, after post processing 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 19
Ground measurements: Quality control Identified issues Missing data Unrealistic values Time shifts Shading Artificial trends Possible reasons Problems with data logger Missing power Data transmission Time is not aligned Nearby objects + terrain Insufficient cleaning Misaligned sensors or tracker Calibration Night-time Data passed QC Shading Physical limits, Consistency Other issues 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 20
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 21
Modern satellite-based models ADVANTAGES Continuous geographical coverage Spatial resolution approx. 3+ km Frequency of measurements 15 and 30 minutes Spatial and temporal consistency Calibration stability High availability (gaps are filled) Up to 21+ years history variability of weather LIMITATIONS Lower accuracy of high frequency estimates Data inputs: JMA, ECMWF, NOAA, SRTM 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 22 Source: SolarGIS
Modern satellite-based solar resource data: Interannual variability Yearly GHI: Standard deviation (1999 to 2014) 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 23
Modern satellite solar resource data: Models Models used in operational calculations Typically semi-empirical models Scientifically validated Tuned for different geographies Fast and stable results Differences between approaches Satellite and atmospheric data preprocessing (radiometry and geometry) Multispectral and multiparametric cloud detection Management of various phenomena (high albedo, low angles ) Integration of atmospheric data into clear-sky model DNI and transposition models Correct management of terrain effects 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 24
Modern satellite solar resource data: Data inputs Input data Cloud index: satellite data Aerosols, water vapour, ozone Correct representation of spatial and time variability Differences between approaches Preprocessing Adapted for the specific models Geographical and temporal stability: Meteorological models are constantly changing Satellite sensors are degrading and upgrading 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 25
MFG 4-7 GMS 5 MFG 5,7 GOES 8,12,13,14 GOES 10, 11, 15 MSG 1,2,3 MTSAT 1,2 Satellite data: Availability (SolarGIS) 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995 1994 PRIME IODC GOES East Pacific GOES West 0 57.5-75 145-135 GOES 9 Source: NOAA, EUMETSAT, JMA 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 26
Satellite data: spatial and time resolution Cloud index Time resolution 15 and 30 minutes Spatial resolution 3 to ~7 km Further from the image center pixel geometry is distorted (for better visualization 100-km blocks are shown) GHI and DNI is affected primarily by cloud transmissivity Source: EUMETSAT 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 27
Aerosol data: Daily time resolution Solar Village (Riyadh), Saudi Arabia Ilorin, Nigeria Source: ECMWF, AERONET, SolarGIS MACC-II AOD (aerosols) vs. AERONET ground measurements 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 28
Terrain Terrain altitude and shading is modelled with high accuracy NASA SSE MSG native resolution Disaggregated with DEM 1 4 x 5 km 250 x 250 m 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 29
Why satellite data do not match perfectly the ground measurements? Ground measurements may deviate from satellite data, because of: Size of the satellite pixel and sampling rate Resolution and limitations of the input atmospheric data Imperfections of the models Site specific microclimate Issues in ground measurements Example: SolarGIS (Peru) Source: SolarGIS 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 30
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 31
Model uncertainty: Validation metrics Bias: systematic model deviation Root Mean Square Deviation (RMSD) and Mean Average Deviation (MAD): spread of deviation of values Correlation coefficient (R) Kolmogorov-Smirnoff index (KSI): representativeness of distribution of values High-accuracy ground measurements are to be used 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 32
Model uncertainty: Validation metrics Bias: systematic model deviation Root Mean Square Deviation (RMSD) and Mean Average Deviation (MAD): spread of deviation of values Correlation coefficient (R) Kolmogorov-Smirnoff index (KSI): representativeness of distribution of values High-accuracy ground measurements are to be used 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 33
Bias: SolarGIS uncertainty of yearly estimate GHI ±3.9%** DNI ±7.6%** * 68.27% occurrence: standard deviation (STDEV) assuming simplified assumption of normal distribution ** 80% occurrence: calculated as 1.28155 STDEV can be used for an estimate of P90 values 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 34 Source: SolarGIS
Root-Mean Square Deviation: GHI Uncertainty of hourly, daily and monthly values Global Horizontal Irradiation: DLR-PSA Almeria, Spain RMSD Hourly RMSD Daily RMSD Monthly RMSD Values Bias RMSD Hourly Daily Monthly [W/m 2 ] [%] [%] [%] [%] GHI 23005 2.8 0.6 12.0 5.4 1.5 DNI 21645-14.5-2.6 22.3 13.1 3.7 Source: DLR-PSA, SolarGIS 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 35
Root-Mean Square Deviation: DNI Uncertainty of hourly, daily and monthly values Direct Normal Irradiation: DLR-PSA Almeria, Spain RMSD Hourly RMSD Daily RMSD Monthly RMSD Values Bias RMSD Hourly Daily Monthly [W/m 2 ] [%] [%] [%] [%] GHI 23005 2.8 0.6 12.0 5.4 1.5 DNI 21645-14.5-2.6 22.3 13.1 3.7 Source: DLR-PSA, SolarGIS 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 36
Model uncertainty for Global Horizontal Irradiation SolarGIS high uncertainty High latitudes High mountains Variable aerosols Reflecting surfaces Snow and ice Rain tropical region SolarGIS low uncertainty Arid and semiarid regions Low aerosols Hourly values Daily Monthly Yearly ±4 to ±8% Values are indicative, based on the analysis of 200+ sites Uncertainty for ground sensors considers that they are well maintained, calibrated and data are quality controlled 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 37
Model uncertainty for Direct Normal Irradiation SolarGIS high uncertainty High latitudes High mountains Variable aerosols Reflecting surfaces Snow and ice Rain tropical region SolarGIS low uncertainty Arid and semiarid regions Low aerosols Hourly values Daily Monthly Yearly ±8 to ±15% Values are indicative, based on the analysis of 130+ sites Uncertainty for ground sensors considers that they are well maintained, calibrated and data are quality controlled 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 38
Contents Historical approaches Solar resource data needs in PV Ground measurements Satellite-based solar resource modelling Uncertainty of satellite-based models Conclusions 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 39
Conclusions 1/2 How SolarGIS data compare to ground measurements? Limits Uncertainty of instantaneous values lower than solar sensors Inherent discrepancy, mainly high frequency measurements (e.g. 15-minute) Advantages Uncertainty of aggregated values Comparable to lower accuracy sensors Better than data from insufficiently managed ground monitoring Radiometric stability and continuity Historical data available (from 1994 onwards) + recent data + forecasting Model can be adapted by ground measurements 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 40
Conclusions 2/2 SolarGIS data uncertainty Without Site adaptation GHI: ±4 to ±8% DNI: ±8 to ±15% After site adaptation (best achievable): GHI: ±2.5 DNI: ±3.5 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 41
Thank you for attention! http://solargis.info http://geomodelsolar.eu Source: SolarGIS 4th PV Performance Modelling and Monitoring Workshop, Köln, Germany, 22-23 October 2015 42