Prediction of Rainfall in Karnataka Region using optimised MVC-LSTM Model

Abstract

Forecasting precipitation is a prominent topic of study in meteorology. Predictions of precipitation using statistical analysis, learning methods are only a few of the methods that have been offered in the past. Organizations tasked with preventing natural catastrophes might benefit from using meteorological time series data prediction in their decision-making processes. The volume, dimension, and frequency of updates to Time Series data are quite high. The study of period sequence data for forecasting is a crucial part of the practical application. Time series data analysis allows for more precise rainfall forecasts, which is useful for assessing the severity of potential droughts and floods. Precipitation forecasting publications have employed a wide range of approaches. This improved forecast accuracy may be attributed to these methods. In this research, we apply a deep learning technique to examine the precipitation records from the Karnataka Division. This article presents a network consisting of a generator and a predictor for predicting spatial-temporal rainfall data. To imprisonment spatial correlations and build high-resolution data from sparse comments, it uses a multi-layer perceptron (MLP) in its generative module. A Multivariate Convolutional (MVC-LSTM) network is used to create the prediction unit; this network is able to capture the interplay between many variables and their temporal and spatial correlations. The Honey Badger approach finds the appropriate weight for the LSTM, which improves the model's ability to classify data. Additionally, this report suggests several avenues for further study in the fields of rainfall prediction and time series data analysis.