generalized and finely tunable rainfall prediction based on optimized forecasting model

Abstract

Agriculture is the most important factor in India for survival. Rainfall is the most critical factor for agriculture. A new approach for designing the network structure in a deep neural network (DNN) based forecasting model for rainfall prediction is presented. The number of atmospheric parameters determines whether the rain will fall. We suggest a systematic approach for analysing various parameters affecting rainfall in the atmosphere in addition to identifying seasonal and diurnal variables, the ground-based weather features Temperature, Relative Humidity, Dew Point, Solar Radiation, Precipitable Water Vapor (PWV) are also identified, and a detailed feature correlation study is presented. Forecasting rainfall in non-monsoon sessions required input values of temperature, wind velocity, humidity, and cloud cover. Among the parameters used to evaluate results were MSE (mean square error), CCI (correlation coefficient), COE (coefficient of efficiency), and MAE (mean absolute error). By using the proposed approach, the proposed scheme establishes that deep learning (DL) can be effective in improving rainfall prediction quality, exploiting the potential of neural network (NN) in using a weather simulator. In terms of daily precipitation characteristics, this various model is applied to 12 different sites under various climatic reinforcements.