Numerical simulation of weather over Kenya using the weather research and forecasting -environmental modeling system

Abstract

This study investigated the accuracy and skill of the Weather Research and Forecasting- Environmental Modelling System (WRF-EMS) to simulate weather over Kenya. The study period was March to May 2011, a long rain season. The data used in the study included the observed daily rainfall and temperature for 27 stations over Kenya, obtained from Kenya Meteorological Department and initialization data for WRF-EMS obtained from the Global Forecasting Model (GFS). Model simulation was done for the period of study and the model output for the 27 stations was used to determine the performance of the WRF-EMS model over Kenya in terms of skill and accuracy. The methods of analysis included spatial distribution comparison, correlation analysis, absolute mean error, root mean square error analysis and categorical statistics. Analysis of the simulated and observed spatial distribution of rainfall over the study area indicated that the WRF model was capable of reproducing the observed general pattern, although in some cases the model displaced the location of occurence of maximum rainfall. The spatial pattern of the observed temperature was well captured by the WRF model. Correlation between the forecasted and observed rainfall indicated significant values over most stations when the entire season was considered. On monthly basis the high correlation were observed in March while they were relatively low in April and May. This was attributed to more localized systems that is associated with Rainfall for April and May which may not have been captured by the model. Similar correlation results were noted for temperature. Root Mean Square Error (RMSE) and Absolute Mean Error (AME) values were generally small for most of the stations for both the rainfall and temperature, which is an indication of small deviation of the forecasted values compered to observed. Categorical statistics that included Frequency Bias Index (FBI), Equitable Threat Score (ETS) and True Skill Statistic (TSS) indicated higher skill of the model for the low threshold of less than lmm/day of rainfall, implying that the model was able to detect the occurrence of rainfall but failed to determine the exact amounts. This was also in line with the fact that the hit rate for most of the stations was higher than 50% for low threshold of less than lrnm/day. Overall, the model has skill in forecasting both rainfall and temperature but may fail to give the exact location of occurence of storms, therefore, during the months of enhanced rainfall in the months of April and May the model forecast needs to be complemeted by other models or forecast methods. There is therefore, need to improve its performance over the domain through reviewing the parameterization of small scale physical processes and its ability in simulating weather on the medium and long range scale.