Improvement of Aerosol Optical Depth Data for Localized Insolation Forecasting

Solar generation is variable in nature; the amount of incoming solar radiation available for harvesting by solar panels is influenced by factors including the position of the sun, clouds, aerosol properties, and water vapor in the atmosphere. It is well-known that atmospheric aerosols can attenuate solar insolation by scattering and absorption and therefore affect the solar radiation budget. The National Solar Radiation Database (NSRDB) developed by the National Renewable Energy Laboratory (NREL), United States, is simulated by the Physical Solar Model (PSM) with the input information of surface reflectance, cloud and aerosol properties from satellite measurements and model reanalysis (Sengupta et al., 2018). The current aerosol optical depth (AOD) data used in PSM is from National Aeronautics and Space Administration's Modern Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) and is regridded from the resolution of 0.5^o × 0.5^o to 4 km × 4 km using an elevation-based scaling. However, the elevation-based downscaled AOD data may not appropriately represent the spatial distribution of aerosol, especially over highly polluted area (e.g., due to wildfires, industrial and transportation emissions) where the gradient of AOD could be large and localized AOD profiles may vary considerably within the specified spatial resolution. Based on the assumption that the surface characteristics change relatively slowly in time, Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle Implementation of Atmospheric Correction (MAIAC) focuses on characterizing surface reflectance by using a sliding window technique and improving pixel and image processing algorithm to retrieve AOD at 1 km resolution with high accuracy. Therefore, the goal of this analysis is to investigate whether using MODIS MAIAC AOD product with spatial resolution of 1 km may improve localized insolation forecasting as compared to using the current 4-km elevation-based MERRA-2- AOD.

We take the following steps to achieve this goal:

1. In order to verify if the 1-km MAIAC AOD can better represent the detailed spatial distribution of localized aerosols, we first explore the MAIAC AOD and MERRA-2 AOD in areas known to have relatively high levels of air pollution in the US and in other countries within NSRDB domain, including the San Joaquin Valley in California, U.S., and Sao Paulo, Brazil. The MAIAC AOD data from 2018 to 2019 are used in this study. Then, the ground-based measurements of AOD data from Aerosol Robotic Network (AERONET) are further used as true values of AOD to evaluate the accuracy of the MODIS MAIAC AOD and MERRA-2 AOD. Therefore, two AERONET sites, Bakersfield and SP-EACH, are chosen to represent the aerosol loading for San Joaquin Valley and Sao Paulo, respectively.
2. To investigate whether the high resolution AOD data can improved the NSRDB, the 1-km gridded MAIAC AOD data will be used as input for the PSM to simulate global horizontal irradiance (GHI) and direct normal irradiance (DNI). The accuracy and bias of simulation results will be evaluated by the GHI and DNI from Surface Radiation Budget Network (SURFRAD) and the magnitude of uncertainty of GHI and DNI will be compared with that from the current NSRDB.

The preliminary results for the first step suggest that the MAIAC product can provide detailed spatial information of aerosol loading, while the MERRA-2 AOD cannot resolve the spatial distribution of aerosol, especially within urban areas where gradients of AOD exist within 4 km grids such as the Sao Paulo City and surrounding coastal region, or with high levels of AOD such as the San Joaquin Valley in July, August, and November of 2018.

Moreover, tThe monthly mean MAIAC AOD is averaged over 14 and 16 data points closest to the location of SP-EACH and Bakersfield sites, respectively. The nearest data points of the monthly mean MERRA-2 AOD to these sites are chosen to compare with MAIAC AOD. For the SP-EACH site in Sao Paulo, the root-mean-square error (RMSE) and mean bias error (MBE) for MAIAC AOD is 0.0606 and -0.0457, respectively, while the RMSE and MBE for MERRA-2 AOD is 0.0731 and -0.0717. For the Bakersfield site in San Joaquin Valley, the RMSE and MBE for MAIAC AOD is 0.0613 and -0.003, respectively, while the RMSE and MBE for MERRA-2 AOD is 0.0808 and -0.051. This result indicates there is higher degree of accuracy and less bias when using MAIAC AOD and the result is consistent for both locations.

In the next step, we will proceed to evaluate the impacts of the above-demonstrated improvements on surface-level solar radiation as compared with projections using the existing NSRDB, and implications of siting solar PV based on the more highly resolved radiation data on estimates of power output will be analyzed. Overall, this research can help mitigate the uncertainty of simulated GHI and DNI, informing decision making and investment in PV deployment.