Predicting the Maximum Temperature at Two Meters Above Earth's Surface Using CNN-LSTM Attention Mechanism
Ravi Bhasker, Nishant Kumar, Ravi Kumar Burman, Abhishek Kumar, Shoaib Alam, Kamaldeep Kamaldeep · 2024
Climate monitoring, energy management, and agriculture are just a few of the sectors that depend on accurate temperature forecasts. By integrating a CNN and LSTM with an attention mechanism, this study seeks to develop a more precise temperature forecasting model. Using meteorological data gathered from NASA's Prediction of Worldwide Energy Resources (POWER) platform, the suggested model is intended to forecast the maximum temperature at two meters above Earth's surface. Benchmarks for comparison are established time series forecasting models like ARIMA and FB Prophet. The attention mechanism helps the CNN-LSTM model better focus on significant time steps. CNN layers are used to capture spatial patterns in the data, and LSTM layers are used to characterize temporal dependencies. A number of metrics, including RMSE, MAE, RRSE, and R2, are used to assess the models' performance. As evidenced by the much reduced RMSE (1.61) and higher R2(0.92), the CNN-LSTM with attention mechanism performs better than traditional models in temperature prediction, according to the results. As compared to conventional methods, the study finds that deep learning models—especially CNN-LSTM with attention—offer a more reliable and accurate method for predicting temperature. This suggests new directions for future research in real-time forecasting, model optimization, and the incorporation of other environmental variables.