Application of Kalman filter for post-processing WRF-Solar forecasts over Metro Manila, Philippines

•WRF-Solar forecasts were improved using Kalman Filter in 1 site in the Philippines.•WRF-Solar performs better for drier, clear-sky months than wetter, cloudier months.•Kalman filter with only 3 training days reduces irradiance forecast bias by 70 %.•Optimal training days have more than 80 % forecast bias reduction for both seasons. Day-ahead forecasts of global horizontal irradiance (GHI) from WRF-Solar were evaluated against GHI observations from a pyranometer deployed at Manila Observatory (MO; 14.64°N, 121.08°E), Metro Manila, Philippines for the January to March (JFM) and June to August (JJA) seasons in 2020. A clear sky detection method using the pyranometer GHI measurements is employed to enable forecast validation not just for overall performance but also for cloudy and clear sky periods separately. To potentially improve GHI forecasts, the WRF-Solar GHI values were postprocessed using a Kalman filter (KF) tested for different training days (i.e., from 3 to 42 days) to determine the optimal number of training days that minimize the RMSE. KF post-processing, with the shortest number of training periods (3 days), already provides an MBE (RMSE) reduction of 70 to 94 % (8 to 12 %). The optimal training period (14 for JJA; 42 for JFM) for filtered WRF-Solar GHI forecasts leads to an MBE (RMSE) reduction of at least 64 % (17 %) during cloudy periods. However, KF underestimates GHI values for clear sky periods because of reducing the bias of the dominantly cloudy periods over the site. Results from the study, the first of its kind to assess performance of WRF-Solar and KF over the Philippines, will serve as a basis for a computationally efficient alternative to more expensive higher resolution and multiple ensemble member solar forecasts. Future work intends to focus on applying this method over different topographies in the Philippines, given the availability of irradiance data.