Improving the outputs of regional heavy rainfall forecasting models using an adaptive real-time approach

In this work, an adaptive real-time approach is presented to improve the rainfall forecasts of the weather research and forecasting (WRF) model. Heavy rainfall events are considered to make 6-, 12-, 18-, and 24-hour ahead predictions, and to define various initial conditions of the forecasting model. Comparing various forecasts and observations specifies precise forecasting time line and physical settings of the WRF model that would lead to an improved forecasting model. An adaptive real-time approach is defined using a combination of observed precipitation and WRF forecasts. Forecasted precipitation values for each event are updated by the observed precipitation at each three-hour interval. A sigmoidal curve is fitted to the observations and remaining forecasts to provide modified values. The process continues step by step, and thus an adaptive real-time forecasting model is developed to improve predictions. The methodology is applied to Tehran and Golestan watersheds in Iran, and the results indicate that the developed methodology improves WRF forecasts in terms of categorical and statistical metrics.