Enhancing Drought Prediction Accuracy with Integrated GSTARIMA-DNN Models Utilizing Penman-Monteith Method for Data Acquisition Devi Munandar , Budi Nurani Ruchjana , Atje Setiawan Abdullah , Hilman Ferdinandus Pardede
1) Doctoral Program of Mathematics, Faculty of Mathematics and Natural Sciences, Universitas Padjadjaran, Sumedang 45363, Indonesia
2) Department of Mathematics, Faculty of Mathematics and Natural Sciences, Universitas Padjadjaran, Sumedang 45363, Indonesia
3) Department of Computer Science, Faculty of Mathematics and Natural Sciences, Universitas Padjadjaran, Sumedang 45363, Indonesia
4) Research Center for Artificial Intelligence and Cybersecurity, National Research and Innovation Agency (BRIN), Jakarta Pusat 10340, Indonesia
Abstract
Accurate drought prediction is essential for effective water resource management and agricultural planning. Traditional methods often fail to capture the complex spatiotemporal dynamics of drought patterns. This study introduces an integrated model combining Generalized Spatiotemporal Autoregressive Integrated Moving Average (GSTARIMA) with Deep Neural Networks (DNN) to enhance drought prediction accuracy. We collected monthly meteorological data, including temperature, solar radiation, humidity, wind speed, and atmospheric pressure, from NASA POWER for West Java. Evapotranspiration data were calculated using the Penman-Monteith method to provide a comprehensive input dataset. The GSTARIMA component captures spatial dependencies, while the DNN component models nonlinear temporal patterns. Data were partitioned into training and testing sets, and the model was trained and validated using these subsets. Performance evaluation was conducted using Mean Absolute Percentage Error (MAPE) as the primary metric. Results indicate that the integrated GSTARIMA-DNN model significantly outperforms traditional methods, demonstrating superior accuracy in predicting drought events. The use of Penman-Monteith evapotranspiration data played a crucial role in enhancing the model^s predictive capability. This study underscores the effectiveness of integrating GSTARIMA and DNN models for drought prediction, offering a robust tool for researchers and practitioners in climate science. Future research should explore further refinements and applications of this integrated modeling approach in various climatic and geographical contexts.
