Accurate rainfall forecasting addresses many issues, including energy production, agriculture, and most importantly disaster planning and water resource management. Many regions, including Jordan, suffer from water scarcity, which is a major concern for governments. Relying on traditional solutions is no longer possible with the ongoing climate change and weather fluctuations, so it has become necessary to rely on data analysis using artificial intelligence to increase the accuracy of rainfall forecasting. The variability in data from one place to another and over time periods leads to inconsistencies in the accuracy of rainfall prediction in models. This research presents a deep learning hybrid model for rainfall prediction. The model combines convolutional neural networks (CNN) and long short-term memory (LSTM) networks. It utilizes CNN's capabilities in analysing short-term spatial patterns and the ability to model long-term temporal dependencies in LSTM. The proposed model outperforms traditional machine learning and other deep learning approaches. The performance was examined on a dataset from the Ministry of Water and Irrigation, which includes daily rainfall amounts and some climate variables for the period 1987–1918. The study compared the performance of CNN-LSTM with other machine learning and deep learning models It obtained good values in the common key measures such as 9.98 Root Mean Squared Error, 6.8 Mean Absolute Error, and 6.24% Mean Absolute Percentage Error. Despite the complexity and variability of the data, the model outperformed other models in rainfall prediction. The performance of the model can be increased by incorporating bidirectional LSTM layers and additional weather features. The result proves the robustness of the new model in processing complex rainfall data, and it helps in water resources management, disaster awareness and agriculture decisions.