Comparison of artificially intelligent methods in short term rainfall forecast

Rainfall forecasting has been one of the most scientifically and technologically challenging task in the climate dynamics and climate prediction theory around the world in the last century. This is due to the great effect of forecasting on human activities and also for the significant computational advances that are utilized in this research field. In this paper our main objective is to forecast over a very short-term and specified local area weather using local data which is not always available by forecast center but will be available in the future by social network or some other methods. For this purpose in this paper we have applied three different algorithms belonging to the paradigm of artificial intelligence in short-term forecast of rainfalls (24 hours) using a regional rainfall data of Bihar (India) as a case study. This forecast is about predicting the categorical rainfall (some pre-defined category based on the amount of total daily rainfall) amount for the next day. We have used two classifier ensemble methods and a single classifier model for this purpose. The ensemble methods used in this paper are LogitBoosting (LB), and Random Forest (RF). The single classifier model is a Least Square Support Vector Machine (LS-SVM). We have optimized each of the models on validation sets and then forecast with the optimum model on the out of sample (or test) dataset. We have also verified our forecast results with some of the latest verification tools available. The experimental and verification results suggest that these methods are capable of efficiently forecasting the categorical rainfall amount in short term.