COMPARATIVE ANALYSIS OF SARIMA-GARCH AND LSTM FOR SOIL MOISTURE FORECASTING IN INDRAMAYU
Soil moisture is a critical determinant of agricultural productivity and hydrological stability, particularly in Indramayu Regency, a major rice-producing region in West Java vulnerable to seasonal drought. Accurate forecasting is essential for optimizing irrigation management and mitigating climate risks. This study evaluates and compares the performance of a hybrid statistical model, SARIMA-GARCH, against a Deep Learning approach using Long Short-Term Memory (LSTM) networks to forecast soil moisture. The study utilizes weekly data from 2014 to 2023, employing SARIMA-GARCH to address seasonality and heteroskedasticity, while LSTM is optimized to capture non-linear temporal dependencies. The evaluation results on 2023 test data demonstrate that LSTM outperforms the statistical hybrid model. The LSTM model achieved a Root Mean Square Error (RMSE) of 0.0306 and a Mean Absolute Percentage Error (MAPE) of 7.08%, compared to 0.0349 and 8.47% for SARIMA-GARCH. Visually, LSTM produced a smoother trajectory aligned with physical soil inertia, whereas SARIMA-GARCH reproduced the jagged short-term fluctuations of the previous season. Consequently, this study establishes LSTM as the superior architecture for precise soil moisture forecasting to support regional food security policy.
Authors
- Yenni Angraini (ORCID: https://orcid.org/0000-0003-3186-2378)
- Windi Pangesti
- Aurisa Putri Hawin Yusima
- Qaulan Sadiida Putri Sumarna
- Ibrahim Zaidan
- Rafi Adabhi Sunarya (ORCID: https://orcid.org/0009-0008-1398-7250)
Institutions
- IPB University (ID)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24358729
- Primary Topic
- Hydrological Forecasting Using AI
- Type
- preprint