An Improved Statistical Method for Rainfall Forecasting in Sri Lanka using the WRF Model

摘要

Weather Research and Forecasting (WRF) model is one of the forecast models which is used in Sri Lanka for weather forecasting specifically considering rainfall forecasting. This model offers multiple parameters which can be combined in many ways and to be used for simulating the model according to the regional aspects. According to the parameters of the WRF model which is applicable to the region including Sri Lanka, there are twelve model simulations of the WRF model in order to generate twelve rainfall forecasts for a given location at a specific time. The common practice is to select a random forecast value from these generated twelve forecasts. In this research, the daily rainfall patterns over Sri Lanka during the year 2019 were statistically analyzed with the consideration of the main climatic zones as Wet, Intermediate, and Dry. Best regressions were fitted, based on principal component analysis for the climate zones using the data from the first six months of 2019. Moreover, fitted regressions were tested using the data from the next three months of 2019. According to the calculated Mean Square Error (MSE) of the fitted regressions, the results demonstrated a better accuracy compared to the individual model simulations of the WRF model for all three zones. For the Wet zone, MSE of the fitted regression was decreased by 20.5% compared to the minimum MSE value of the twelve model simulations, for the Intermediate zone, the MSE decrease was by 41.3% and for the Dry zone, the MSE decrease was by 5.5%. Thus, the proposed method can be considered as an improved method based on principal component analysis, for rainfall forecasting in Sri Lanka using the WRF model.