Real-time rainfall projection in hilly areas using ARLSTM model: a case of Dharamshala, India

Abstract Rainfall forecasting is essential in many industries, including agriculture, hydrology, urban planning, disaster management, and renewable energy. However, typical forecasting approaches frequently fail to reflect the complex spatiotemporal dynamics inherent in rainfall patterns, particularly in hilly areas. Since, the mountainous terrain influences wind patterns, moisture flow, and temperature inversions which affect precipitation patterns, adding challenge to meteorologists. So, designing a predictive forecasting model is essential for people and governments to make preventive measures in time. In recent years, the introduction of Internet of Things (IoT) technology and advances in machine learning (ML) techniques have provided a promising route for improving rainfall prediction accuracy. Real-time data on environmental elements such as temperature, wind speed, air pressure, and humidity can be collected using IoT sensor networks. This research paper investigates the efficacy of the Auto-Regressive Integrated Moving Average - Long Short-Term Memory (ARIMA-LSTM) model for real-time rainfall prediction in hilly towns like Dharamshala. Utilizing real-time data collected from sensors and using edge to preprocess it before sending it to the cloud, this study explores the integration of traditional time series analysis techniques with deep learning architectures to enhance predictive accuracy. The propounded model is paralleled against alternative forecasting approaches and shows a significant reduction in Root Mean Square Error (RMSE), Relative Root Mean Square Error (RRMSE), and Mean Square Error (MSE). Furthermore, it had a 0% Absolute Percentage Bias (APB), a high Nash-Sutcliffe Efficiency (NSE) of 0.99, and a low Theil’s coefficient (U2) of 0.01. These results demonstrate that the ARLSTM model outperforms other forecasting models of rainfall prediction.