An efficient hybrid weather prediction model based on deep learning

Abstract

Weather events directly affect human activities. In particular, extreme weather events with global warming, forest fires, and high air temperatures that cause drought make human life difficult. Effective and accurate weather prediction models are needed to take precautions against such climatic events. Therefore, it is essential to develop models that make precise weather predictions. Technological developments contributed significantly to developing successful deep learning-based weather prediction models. With a high success rate, this study proposed a hybrid weather prediction model based on Convolutional Neural Networks and Recurrent Neural Networks models. The proposed hybrid model was applied to the Jena dataset, which contains 14-parameter, large-scale meteorological data that were never utilized for weather prediction. The experimental results were compared with popular deep learning, machine learning, and statistical methods such as Auto-Regressive Integrated Moving Average, Convolution Neural Networks, Long-Short Term Memory, Multilayer Perceptron, Random Forest, Recurrent Neural Networks, and Support Vector Machine. As a result of these comparisons, the proposed hybrid model obtained the best prediction result for all metrics. For example, according to the weather prediction results for Jena, Germany, the proposed hybrid model got the results of Mean Squared Error: 0.035, Root-Mean-Squared Error: 0.189, Mean Absolute Error: 0.126, and R-Squared: 0.987.