Comparisons of next-day solar forecasting for Singapore using 3DVAR and 4DVAR data assimilation approaches with the WRF model. (March 2020)

• Record Type: Journal Article
• Authors: Huva, Robert; Verbois, Hadrien; Walsh, Wilfred
• Is Part Of: Renewable energy. Volume 147(2020)Part 1
• Page Start: 663
• Page End: 671
• Publication Date: 2020-03
• Subjects: Numerical weather prediction -- Tropical -- Irradiance forecasting -- WRFDA; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2019.09.011
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: For tropical locations forecasting of solar irradiance at time horizons of 12 h, or longer, can only be achieved with the assistance of Numerical Weather Prediction (NWP) models. NWP models simulate the time evolution of atmospheric processes that are important for the prediction of solar irradiance. We use the Weather and Research Forecasting (WRF) model to simulate the atmosphere over Singapore down to 3-km resolution and for the years 2015–2016. However, by their nature the NWP models suffer from incomplete knowledge of atmospheric initial conditions. The process of Data Assimilation (DA) attempts to minimise the initial condition problem by incorporating observations into the model. DA utilises observations to constrain the state of the model either in a static (3DVAR) or time-evolving (4DVAR) manner. We compare hourly next-day forecasts using 3DVAR and 4DVAR intialisations of the WRF model with observations of surface irradiance across Singapore. Raw results show that 4DVAR has the lowest error for all time horizons and for all sky conditions except clear-sky hours. We then post-process the raw results using the random forest algorithm. Following post-processing, the 4DVAR initialised forecasts remain the best performing with relative RMSE of 37%. All models after post-processing out-perform persistence ensemble and climatological references. Highlights: Data assimilation assists solar forecasting for both raw and post-processed results. Before post-processing 4DVAR is preferred over 3DVAR warm-cycle and then no DA. The 4DVAR approach is best in both raw and post-processed. Overcast conditions are the most difficult to forecast for all simulation types. All NWP simulations out-perform climatological and persistence ensemble benchmarks.