Analysis of soft Computing models for rainfall prediction

Abstract

Soft computing is a novel intelligent way to predicting natural newlineweather characteristics to tackle the world s many challenges. Surface newlineweather parameters are involved in rainfall prediction research around newlinethe world. Using meteorological weather parameters and Sea Surface newlineTemperature (SST) data from 2010 to 2019, the researchers developed newlinerainfall prediction models for five locations along Tamil Nadu s southeastern newlinecoastline. newlineArtificial Neural Networks (ANN), Fuzzy Inference Systems newline(FIS), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Genetic newlineAlgorithms (GA) and Deep Neural Networks (DNN) were used to newlineproduce rainfall prediction models. One of the research s major newlineobjectives is to discover the best combination of inputs and targets for newlineall rainfall prediction models. newlineSurface meteorological weather observations are defined newlinetemperature-related four input parameters (Temperature, Dew Point, newlineHumidity and Atmospheric Pressure) are used to build rainfall newlineprediction models using ANN, FIS, ANFIS, GA and DNN (RNNLSTM). newlineThen, using soft computing approaches, wind-related input newlineix newlinefactors, such as wind variations (Wind Speed, Wind Direction, Wind newlineGust, Cloud Cover) were integrated as eight input parameters with heat newlinevariants to develop the rainfall prediction model. The nineth Oceanic newlineparameter, Sea Surface Temperature (SST), was introduced as an input newlineparameter to the model to improve accuracy. newlineAn ANN model was created using the Feed-Forward Back newlinePropagation (FFBP) approach. With a PURELIN transfer function, newlinetraining is done with the Levenberg-Marquardt (LM) TRAINLM newlinefunction, while performance analysis is done with the Mean Square newlineError (MSE) function. newlineThe Fuzzy Inference System (FIS) with Sugeno rainfall newlineprediction model was created with input variables such as heat variants newlineand wind variants, as well as their membership functions. Sugeno-type newlineFIS is a method for generating crisp outputs from fuzzy inputs using a newlineweighted average.