<![CDATA[The high variability of rainfall in tropical climates presents a major challenge for agricultural management, as weather uncertainty often leads to inefficient fertilization practices due to nutrient loss. This study aims to develop a robust framework for rainfall prediction, which can inform a flexible and precise crop nutrient scheduling system. Utilizing an hourly rainfall dataset (n=6,624) obtained from IoT sensors, the research proposes an approach that integrates Singular Spectrum Analysis (SSA) for signal decomposition and noise reduction with Gradient Boosting algorithms (LightGBM and XGBoost). Spline interpolation was employed to handle missing data, while SSA served to disentangle deterministic trends from random noise, enabling the models to perform more effectively on the refined dataset. Empirical evaluation demonstrates that the SSA-XGBoost hybrid model achieves superior performance, with an RMSE of 0.0057 and an R² of 0.8278, significantly outperforming the SSA-LightGBM model (R² 0.2879), which struggled to capture non-linear patterns within this dataset. The high predictive accuracy of the SSA-XGBoost model facilitates the implementation of responsive nutrient management strategies, wherein fertilizer application can be deferred during forecasted periods of high rainfall to prevent runoff and environmental pollution. This research contributes to the field of hydroinformatics by demonstrating the effectiveness of combining SSA and XGBoost as a cost-efficient yet high-performance solution for mitigating climate-related risks in tropical wetland agriculture.]]>
Copyrights © 2026
