Validation of Direct Normal Irradiance from Meteosat Second Generation. DNICast

Transcription

1 Validation of Direct Normal Irradiance from Meteosat Second Generation DNICast A. Meyer 1), L. Vuilleumier 1), R. Stöckli 1), S. Wilbert 2), and L. F. Zarzalejo 3) 1) Federal Office of Meteorology and Climatology, MeteoSwiss 2) German Aerospace Centre, Institute of Solar Research 3) Centre for Energy, Environment and Technological Research CIEMAT EMS Annual Meeting: European Conference for Applied Meteorology and Climatology Sept, Dublin, Ireland

2 Validation of Direct Normal Irradiance from Meteosat Second Generation DNICast Introduction Data Clear-sky estimates All-sky estimates Conclusions

3 Solar radiation estimates obtained from satellite data: Motivations introduction 1. widely used for obtaining large-scale solar resources maps 2. important input for solar resources short-term forecasting (nowcasting) Yearly Direct Normalized Beam Solar Radiation (Wm -2 ) from HelioMont (Stöckli, 2013) MeteoSwiss Stöckli, R. (2013), The HelioMont Surface Solar Radiation Processing, Scientific Report MeteoSwiss, 93, Editor: Federal Office of Meteorology and Climatology MeteoSwiss

4 Solar radiation estimates obtained from satellite data: Goal: Motivations introduction 1. widely used for obtaining large-scale solar resources maps 2. important input for solar resources short-term forecasting (nowcasting) - Assess uncertainties in satellite estimates of solar radiation using high accuracy ground-based data as reference - Look into radiation component uncertainty (direct irradiance) instead of global radiation (global irradiance) - Explore problem of area (satellite) to point (ground-based) comparison: spatial and temporal averaging

5 Background introduction General scheme for radiation estimates both sides can be tested in this study Clear-sky estimates Radiative transfer model (RTM) ( look-up table, LUT) Non-trivial input: Aerosol (AOD, SSA, α, ) Surface Albedo (satellite) Water vapor column (ECMWF) O3 (ECMWF) Cloud modification factor Cloud macrophysical information (e.g., satellite VIS & IR imagery) Typical macrophysical information Cloud cover Cloud brightness Cloud depth ~ COD physics understood as good as aerosol input is 0 1 physics limited in model empirical and quite crude

6 Data & data availability data PSA allows using a combination of 4 pyrheliometers for full year (2015) DNI measurements spread on about 1 MSG SEVIRI pixel (visible) 1-min measurements of BSRN-type quality Long and Ackermann, 2000 ~1 km 2 43% of data

7 Satellite information CMF sat compared to DNI and CMF data CMF val = IImmmmmmmm At clear-sky times DNI theoretical prediction can be compared with data. ttt II cccc CMF val CMF sat measured DNI theoretical clear-sky DNI clear-sky times

8 Similar PDF for all ground-based measurements same for their average Estimates using measured AOD: extremely close to measurements Estimates using global AOD products: - Kinne et al. (2006) climatology: strong underestimation Probability density function clear-sky estimates - CAMS AOD product: small underestimation (too many high AOD events) DNI measurements Using measured AOD Using global AOD product Kinne, S., et al. (2006): An AeroCom initial assessment optical properties in aerosol component modules of global models, Atmos. Chem. Phys., 6, , doi: /acp

9 Inverse CMF distributions clear-sky estimates Validating CMF for clear-sky (should be =1) gives information on uncertainty of clear-sky estimates uncertainty in aerosol input ~1 (~1.5%) 2 PSA DNI, measured AOD PSA DNI, CAMS AOD ~±6% ~ ±12%

10 Similar PDF for all ground-based measurements same for their average Satellite estimates: significant underestimation of (almost-)zero-dni events - Strong effect on 1 st quartile - Compensate for Kinne AOD overestimation Probability density function all-sky estimates

11 Similar PDF for all ground-based measurements same for their average Satellite estimates: significant underestimation of (almost-)zero-dni events - Strong effect on 1 st quartile - Compensate for Kinne AOD overestimation DNI measurements DNI estimates Probability density function all-sky estimates

12 - Kinne agrees better on average, but CAMS better event by event - Significant scatter DNI comparison all-sky estimates underestimation overestimation PSA DNI comparison

13 DNI spatial and temporal averaging all-sky estimates Computing r 2 and RMSE of CMF comparisons for different temporal and spatial averaging (single vs. 4-instrument average at PSA) instantaneous comparison 15-min average 60-min average

14 DNI spatial and temporal averaging all-sky estimates - Temporal & spatial averaging increase correlation and decrease RMSE - At 1-hour temporal averaging, very small difference between PSA with and without spatial averaging 1-hour averaging satisfactory for point-to-area comparison in cloud effect temporal averaging instantaneous 15-min. 60-min. Correlation (r 2 ) PSA all instruments single instrument spatial averaging PAY single instrument RMSE PSA all instruments single instrument spatial averaging PAY single instrument

15 DNI spatial and temporal averaging all-sky estimates - Temporal & spatial averaging increase correlation and decrease RMSE - At 1-hour temporal averaging, very small difference between PSA with and without spatial averaging 1-hour averaging satisfactory for point-to-area comparison in cloud effect temporal averaging instantaneous 15-min. 60-min. Correlation (r 2 ) all instruments PSA single instrument PAY single instrument spatial averaging increase RMSE all instruments PSA single instrument PAY single instrument spatial averaging decrease

16 DNI spatial and temporal averaging all-sky estimates - Temporal & spatial averaging increase correlation and decrease RMSE - At 1-hour temporal averaging, very small difference between PSA with and without spatial averaging 1-hour averaging satisfactory for point-to-area comparison in cloud effect temporal averaging instantaneous 15-min. 60-min. Correlation (r 2 ) PSA all instruments single instrument nearly equal spatial averaging RMSE PAY single instrument all instruments PSA single instrument nearly equal spatial averaging PAY single instrument

17 Conclusions Validation of DNI estimates from satellite data (MSG) using data from multiple ground-based measurements within a single MSG pixel Clear-sky situations: - Main uncertainty from AOD input - Using measured AOD allows agreement within about 2 3 times measurement uncertainty - Use of Kinne climatology leads to strong DNI underestimation - CAMS AOD product: DNI ~ 3% bias, uncertainty: ~ ±15% GHI ~ +1% bias, uncertainty: ~ ±5% - Measured AOD: DNI ~ 1% bias, uncertainty: ~ ±6% GHI ~ +1% bias, uncertainty: ~ ±3%

18 Conclusions Validation of DNI estimates from satellite data (MSG) using data from multiple ground-based measurements within a single MSG pixel All-sky situations: - Significant underestimation of number of zero-dni events (too many small DNI satellite estimates that should be zero DNI) - Satellite estimates with Kinne AOD agree better on average, but those with CAMS or measured AOD are better event by event Satellite & ground measurement community need working together for better satellite-based CMF formulation - Both spatial and temporal averaging allow reduction of cloud-induced scatter in comparisons between area satellite estimates and point ground-based measurements - 1-hour averaging satisfactory for averaging over MSG pixel area

19 Thank you? /! Meyer, A., L. Vuilleumier, R. Stöckli, S. Wilbert and L. F. Zarzalejo (2017). Validation of Direct Normal Irradiance from Meteosat Second Generation, manuscript in preparation.