Improving Rainfall Prediction Accuracy Using an LSTM-Driven Model Enhanced by M-PSO Optimization

Abstract

Precipitation, particularly rainfall, plays a crucial role in the economic productivity of the agricultural sector. In regions characterized by unpredictable rainfall patterns, accurately predicting future precipitation is vital for designing effective rainwater harvesting systems and formulating strategies to address potential challenges. The contemporary meteorological community faces a significant dilemma when it comes to forecasting heavy rainfall, as it has far-reaching implications for economic stability and human survival. Moreover, heavy rainfall often serves as a primary trigger for recurring natural disasters like floods and droughts, which impact regions worldwide on an annual basis. The constantly changing nature of our climate presents a formidable barrier to achieving highly precise forecasts of precipitation using traditional statistical methods. Current models used for forecasting rainfall demonstrate less than optimal performance when dealing with complex and non-linear datasets. This study presents a novel method that evaluates the effectiveness of combining Long Short-Term Memory (LSTM) with Modified Particle Swarm Optimization (M-PSO) compared to established rainfall forecasting systems. Experiments conducted using this proposed LSTM-M-PSO model have yielded significant improvements in Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) when predicting monthly rainfall. Consequently, the proposed LSTM-M-PSO method showcases its suitability for applications in global climate projection, particularly when working with extensive datasets. Its improved accuracy in forecasting rainfall holds promise for addressing the critical challenges posed by unpredictable precipitation patterns and their significant impacts on agriculture and society.