CNN-LSTM Approach for Forecasting Daily Maximum and Minimum Temperatures: A Case Study of Southeast Morocco

Accurate prediction of maximum and minimum temperatures is crucial in many domains, such as energy planning, agriculture, and water resource management. In this work, applying Deep Learning (DL) techniques for predicting these climate variables has been investigated. Thus, an approach based on Long-Short-Term-Memory (LSTM) networks and Convolutional Neural Networks (CNNs) was proposed. CNNs are used to extract spatial features from meteorological data, while LSTMs capture temporal dependencies in the corresponding time series. The combination of CNNs and LSTM makes it possible to fully exploit the information contained in weather data. Historical data including variables such as Maximum Daily Temperature (MaxDT) and Minimum Daily Temperature (MinDT), Maximum and Minimum Daily Relative Humidity (MaxDRH) and (MinDRH), Mean Daily Wind Speed (MeanDWS) and Maximum Daily Wind Speed (MaxDWS), Mean Daily Wind Direction (MeanDWD), and Daytime Insolation Duration (DINSOL) are used to train the implemented model. To measure the performance of the hybrid model, we used evaluation criteria such as the Root Mean Square Error (RMSE), the Mean Absolute Error (MAE), and the coefficient of determination (R). The results indicate that our DL methodology excels at providing accurate forecasts for the maximum and minimum temperatures of the day. Indeed, the hybrid model test provided an R-value of up to 96.9% for extreme daytime temperatures in Errachidia City, situated in Southeast Morocco.