Comparison of deep learning models for weather forecasting in different climatic zones

Weather forecasting has become an integral part of our day-to-day life. Weather holds significant importance in our everyday lives, impacting areas such as how we travel, produce food, and maintain public well-being. Mostly, weather prediction is done with machines learning models, but the use of deep learning techniques in this field in growing. Still, the existing studies are not sufficient to get a clear concept of weather prediction in different climatic zones. Therefore, in this study, selected four deep learning models, RNN, CNN and LSTM, to predict temperature in four climatic zones. We selected four cities, Dhaka, Moscow, Dubai and Brasilia from four different climatic zones. It is seen that the overall accuracy (OA) of LSTM ranged between 85% to 95%, followed by CNN 78% to 91%, and RNN 64% to 94%. Though the OA values of these three models in four climatic zones differs significantly, high AUC values were seen in all scenario. The highest AUC value (0.999) was seen in continental climatic zone for LSTM model and lowest (0.963) in mil climatic zone for RNN.