Establishing hybrid deep learning models for regional daily rainfall time series forecasting in the United Kingdom
Geethu Thottungal Harilal,Aniket Dixit,Giovanni Quattrone
DOI: https://doi.org/10.1016/j.engappai.2024.108581
IF: 8
2024-05-23
Engineering Applications of Artificial Intelligence
Abstract:Accurate daily rainfall predictions are becoming increasingly important, particularly in the era of changing climate conditions. These predictions are essential for various sectors, including agriculture, water resource management, flood preparedness, and pollution monitoring. This study delves into the complex relationship between meteorological data, with a focus on the accurate forecasting of rainfall by identifying the impact of temperature variations on rainfall patterns in different regions of the United Kingdom (UK). The meteorological data was collected from the National Aeronautics and Space Administration (NASA) and covers daily observations from January 1, 1981, to July 31, 2023, in four distinct regions of the UK: England, Wales, Scotland, and Northern Ireland. The main objective of this research is to introduce hybrid deep learning models, namely Convolutional Neural Networks (CNN) with Long Short Term Memory (LSTM) and Recurrent Neural Networks (RNN) with Long Short Term Memory (LSTM), for predicting daily rainfall using time-series data from the four UK countries, specifically designed for daily rainfall forecasting of four regions in the UK. The models are fine-tuned using the hyperparameter optimisation method. Comprehensive performance evaluations, including Loss Function, Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), are employed to compare the effectiveness of our proposed hybrid models with established baseline models, including LSTM, stacked LSTM, and Bidirectional LSTM. Additionally, a visual analysis of actual and predicted rainfall data is conducted to identify the most proficient forecasting model for each region. Results reveal that the proposed hybrid models consistently outperform other models in terms of both quantitative performance metrics and visual assessments across all four regions in the UK. This research contributes to improved rainfall forecasting methodologies, which are critical for sustainable agricultural practices and resource management.
automation & control systems,computer science, artificial intelligence,engineering, electrical & electronic, multidisciplinary