India plans to add 100 GW of solar electric power generation by 2022 to an existing system (with an installed capacity of close to 330 GW, as on January 2018, from all sources). Selection of sites for such a large infrastructure investment is definitely an important decision.
Here, we examine the performance of four models of annual solar photovoltaic production that are appropriate for site selection against actual generation data from utility-scale plants in the Indian state of Gujarat. We find that a simple Bird clear sky index predicts the annual PV plant production with an error of 14Â±5%, while the satellite data (at 10 km Ã 10 km resolution) in National Renewable Energy Laboratoryâs National Solar Radiation Database (NSRDB) has a prediction error of 9Â±6%. Data from solar monitors at Indiaâs Solar Radiation Resource Assessment (SRRA) stations, which are 46â95km away from the power plants, has a prediction error of 7Â±3%. However, a power plant model using SRRA weather data to predict output correction for solar cell temperature performs best with an error of 6Â±6%. The inter-annual variability of the annual mean irradiation for 2000â2014 in Gujarat is Â±3.6% for direct normal irradiance (DNI). Thus, we find that for site selection based on annual PV production, the inter-annual variability of irradiance is larger than the differences between models; in the absence of coincident data, satellite data (NSRDB) could be a preferred choice.