Improving Short-Term Drought Prediction Using Lstm with Lagged Spei Features: a Comparative Study with Arima

Keywords:
Drought Prediction, Lstm, Arima, Spei, Time Series Forecasting

Abstract:
Accurate short-term drought forecasting is essential for the implementation of early warning systems and managing water resources. This study investigates the applicability of data-driven models for one-month-ahead drought forecasting by comparing a traditional ARIMA method with two configurations of Long Short-Term Memory (LSTM) neural networks. The Standardized Precipitation Evapotranspiration Index (SPEI) at a one-month scale is computed using precipitation and temperature data. Two LSTM models were developed. The first LSTM model uses only meteorological inputs, while the second uses lagged SPEI values in order to capture temporal persistence in drought conditions. The performance of both models is evaluated using MAE, RMSE, and MAPE on an independent test dataset. The results show that ARIMA achieves the lowest MAE and RMSE, confirming that short-term SPEI dynamics have strong autoregressive characteristics. The LSTM model enhanced with lagged SPEI information outperforms the standard LSTM, showing increased stability and predictive accuracy. The results highlight the additional role of deep learning in drought prediction and propose that integrating domain-specific lagged indicators can enhance model effectiveness, enabling the development of data-driven decision support systems.

CITATION:

IEEE format

Copy to clipboard

N. Tihi, S. Popov, S. Stankov, I. Vecštejn, “Improving Short-Term Drought Prediction Using Lstm with Lagged Spei Features: a Comparative Study with Arima,” in Sinteza 2026 - International Scientific Conference on Information Technology, Computer Science, and Data Science, Belgrade, Singidunum University, Serbia, 2026, pp. 213-219. doi:10.15308/Sinteza-2026-213-219

APA format

Copy to clipboard

Tihi, N., Popov, S., Stankov, S., Vecštejn, I. (2026). Improving Short-Term Drought Prediction Using Lstm with Lagged Spei Features: a Comparative Study with Arima. Paper presented at Sinteza 2026 - International Scientific Conference on Information Technology, Computer Science, and Data Science. doi:10.15308/Sinteza-2026-213-219

BibTeX format
Download

RefWorks Tagged format
Download