Verification of WRF forecasts with TRMM rainfall over India during monsoon 2010: CRA method

Abstract

WRF model forecast during monsoon season 2010 have been verified with daily observed TRMM rainfall at 0.25° spatial resolution. Conventional grid-point verification technique has been deployed to calculate common scores like mean errors and root mean square errors. Widely used categorical skill scores have also been computed for seven different rainfall thresholds. The scores only could find the general nature of the model performance depicting the degradation of forecast accuracy exceeding moderate rainfall category with 7.5 mm threshold. Subsequently the object oriented contiguous rain area (CRA) method has been considered for rainfall forecast verification. The method also confirmed that the performance of the model deteriorates along with the increase in rainfall amount. At the same time, within CRA method, the decomposition of mean square error has clearly found out that the major error has been occurred due to displacement of rain object or event in the model forecast compared to observation. In general, volume error contributes less as compared to pattern error in day-1 forecasts of the model whereas they are comparable in day-2 forecasts. Applying statistically significant best-fit criteria, the forecast CRAs have been successfully matched to observed events more than 80% of the time for 35.5 mm threshold and 50% for 64.5 mm thresholds respectively during whole monsoon season. The percentage of match further increases when 10 different active spells of the season has been considered separately.