H41A-0765:
WRF Performance Skills in Predicting Rainfall Over the Philippines

Thursday, 18 December 2014
Gay Jane P Perez and Jay Samuel Combinido, University of the Philippines, Quezon City, Philippines
Abstract:
The Weather Research and Forecasting (WRF) model has been used for predicting rainfall over the Philippines. The period of October 2013 to May 2014 is chosen for the evaluation because of the unprecedented number of new ground instruments (300 to 500 automated rain gauges). It also gives us a good statistical representation of wet and dry seasons in the country. The WRF model configuration makes use of NCEP FNL for the initial boundary condition. Hindcasts are produced at 12-km resolution with 12 hours up to 144 hours lead-time. To assess the predictability of rainfall, we look at the dichotomous case, wherein we evaluate if the model is able to predict correctly the number of rainfall events. The left column in Figure 1 shows the monthly Percent Correct and Critical Success Index (CSI) for different lead-time. Percent Correct represents how well the model performs, 1 being the highest score, with equal bearing on correct positives and correct negatives. On the other hand, CSI is a balanced score that accounts for false alarm and missed events – it has a range of 0 to 1, where 1 means perfect forecast. Results show that during the wet season (October, November and December), PC is approximately 0.7 while in dry season (January, February and March), PC reaches values of around 0.9, which suggests improvement in the performance from wet to dry season. The increase in performance is attributed to the increase in number of correct negatives during the dry season. The CSI score, which excludes the correct negatives, shows that the ability of WRF to predict rainfall events drastically decline in December or during the transition from wet to dry season. This is due to the inability of WRF to pinpoint exact locations of small convective rainfall events. The predictability of actual rainfall values is indicated by the Mean Absolute Errors (MAE) and Root Mean Square Errors (RMSE) in Figure 1. The MAE for 3-hour accumulated rainfall is smallest during the dry season.