Attention-Enhanced LSTM Modeling for Improved Temperature and Rainfall Forecasting in Bangladesh

TL;DR

This paper tackles accurate monthly forecasting of Bangladesh's temperature and rainfall by modeling long-range temporal dependencies with an attention-enhanced LSTM. The method combines an LSTM with Bahdanau-style attention and a rich feature set including lag features, rolling statistics, and cyclical seasonality, trained on data from NASA POWER and HDX spanning 1901–2023. They report superior performance over baselines (XGBoost, Simple LSTM, GRU) with $R^2$ values exceeding $0.96$ for rainfall and temperature and low NRMSE, along with enhanced robustness to climate trends and regional variations. The approach supports climate-sensitive decision-making, including flood risk reduction and agricultural planning, and outlines directions for extending to extremes, multi-region generalization, and computational efficiency.

Abstract

Accurate climate forecasting is vital for Bangladesh, a region highly susceptible to climate change impacts on temperature and rainfall. Existing models often struggle to capture long-range dependencies and complex temporal patterns in climate data. This study introduces an advanced Long Short-Term Memory (LSTM) model integrated with an attention mechanism to enhance the prediction of temperature and rainfall dynamics. Utilizing comprehensive datasets from 1901-2023, sourced from NASA's POWER Project for temperature and the Humanitarian Data Exchange for rainfall, the model effectively captures seasonal and long-term trends. It outperforms baseline models, including XGBoost, Simple LSTM, and GRU, achieving a test MSE of 0.2411 (normalized units), MAE of 0.3860 degrees C, R^2 of 0.9834, and NRMSE of 0.0370 for temperature, and MSE of 1283.67 mm^2, MAE of 22.91 mm, R^2 of 0.9639, and NRMSE of 0.0354 for rainfall on monthly forecasts. The model demonstrates improved robustness with only a 20 percent increase in MSE under simulated climate trends (compared to an approximately 2.2-fold increase in baseline models without trend features) and a 50 percent degradation under regional variations (compared to an approximately 4.8-fold increase in baseline models without enhancements). These results highlight the model's ability to improve forecasting precision and offer potential insights into the physical processes governing climate variability in Bangladesh, supporting applications in climate-sensitive sectors.

Figures (11)

• Figure 1: Methodological illustration: Autocorrelation Function (ACF) plots for temperature and rainfall showing significant correlations at selected lag intervals (1, 3, 6, 12 months). The blue shaded region indicates the 95% confidence interval.
• Figure 2: Methodological illustration: Cyclical month feature visualization (2010–2015). The sine and cosine transformations show continuous, smooth transitions between months.
• Figure 3: Methodological illustration: Temperature and rainfall patterns with cyclical month features (2010–2015). The plot reveals seasonal patterns in climate variables.
• Figure 4: Methodological illustration: 12-month rolling statistics for rainfall (1900–2023). The rolling mean, minimum, and maximum values illustrate long-term patterns in precipitation.
• Figure 5: Methodological illustration: 12-month rolling statistics for temperature (1900–2023). The plot reveals long-term trends and seasonal fluctuations in temperature.