Optimized Intelligent Systems for Predicting Rainfall

Water is essential for all human activities, and rainfall is one of the critical sources of this precious commodity. Prediction of how much rainfall and when it is most likely to occur will assist concerned officials in planning for its storage and subsequent distribution. Meteorological agencies predict rainfall using statistical or dynamic models. Because of the complexity involved in rainfall prediction and limitations of existing techniques, prediction skill improvement is necessary. Recently, intelligent systems such as Artificial Neural Networks, Fuzzy Inference Systems, Support Vector Machines, and Genetic Algorithms are being used by researchers for prediction. Many network parameters are required to be selected for the use of these systems, and the choice of the parameters affects the accuracy of the model. Experimental discovery of the parameters is one way, and the other way is to use optimization algorithms. In this paper, various optimization techniques used in computationally intelligent systems are surveyed for rainfall prediction. The optimization techniques mainly used for this purpose are Particle Swarm Optimization, Genetic Algorithm, and Ant Colony Optimization. In all the research articles, case study of a certain geographical area for rainfall prediction with a different set of inputs and different forecasting lead times is presented, and hence comparison between the models is difficult. For rainfall prediction, model input selection is equally important to the selection of model parameters as a set of predictors change with increasing or decreasing geographical area and forecast lead time. This paper attempts to identify optimization techniques suitable for medium-range rainfall forecasting over a small geographical area.