OPTIMIZING FLOOD EARLY WARNING SYSTEMS: THE ROBUSTNESS OF THE ADDITIVE HOLT-WINTERS MODEL IN FORECASTING SEASONAL RIVER STAGES

Rising flood risks in tropical riparian zones due to climate variability necessitate robust early warning systems. This study evaluates the effectiveness of the Additive Holt-Winters Exponential Smoothing model in forecasting seasonal river water levels to enhance flood mitigation strategies for the Bengawan Solo River. A quantitative approach was employed using secondary data from the Jurug Observation Post (January 2021–January 2025), aggregated into 49 monthly observations. The analysis proceeded through four systematic stages: stationarity testing (ADF), seasonal decomposition, parameter optimization () using RStudio, and forecast verification. Decomposition analysis revealed a distinct additive seasonal pattern with consistent peaks in February and a historical downward trend (), validating the model selection. The model achieved exceptional accuracy with a Mean Absolute Percentage Error (MAPE) of 0.4725%, outperforming traditional intuitive monitoring. Forecasts for 2025 indicate water levels will remain below the alert threshold, though seasonal peaks require vigilance. Unlike complex neural network models requiring extensive datasets, this study offers novelty by demonstrating that the parsimonious Holt-Winters method provides a strategic balance between computational efficiency and high accuracy (error) in data-limited regions. These findings imply that disaster management agencies can shift from reactive emergency response to proactive maintenance of flood control infrastructure based on reliable medium-term projections.