Performance of aerosol optical depth forecasts over the Middle East: Multi-model analysis and validation

Abstract

A primary source of error for predictions of solar irradiance in clear-sky conditions is the total aerosol optical depth (AOD). Dust aerosol loading can also be significant in arid regions such as the Middle East, thus considerably decreasing the solar resource while increasing the detrimental effects of soiling on collectors at solar power plants, particularly during dust storms. Many photovoltaic (PV) and concentrated solar power (CSP) plants have been or will be constructed in the Middle East, making AOD forecasting a pressing issue for plant and grid operators. In this study we present a climatological analysis of 1–3-day AOD forecasts from a two-year period (2018–2019) from three operational models: the NASA Goddard Earth Observing System Model, Version 5 (GEOS-5), the NEMS GFS Aerosol Component (NGAC) model, and the Copernicus Atmosphere Monitoring Service (CAMS) Near-Real-Time (NRT) model. AOD predictions from these models are validated against daily-average observations from 20 Aerosol Robotic Network (AERONET) stations across the Middle East. It is found that GEOS-5 is the best model on average, with the smallest fractional gross error and near-zero modified normalized mean bias. CAMS NRT is the next-best model, while NGAC, which has the coarsest grid spacing of the three models examined here, generally performs poorly. In addition to standard error metrics to characterize the overall performance of the models, a multi-site time series analysis is performed to assess how well these models represent significant dust storm events in the UAE in July 2018 and in Kuwait in April 2018.