Enhancing Missing Data of Water Level with Artificial Intelligence: A Case Study on the Kretek, Opak Watershed, Yogyakarta

Accurate hydrological data is essential for effective water resource management and flood risk mitigation. However, missing water level data poses significant challenges for predictive modeling. This study investigates the performance of four models-Random Forest, XGBoost, ConvLSTM, and NARX-LSTM in imputing missing water level data at Automatic Water Level Recorder (AWLR) Kretek along the Opak Watershed. The models leverage upstream water level observations from AWLR Pulo and rainfall data collected from five stations across the watershed. The missing data, spanning July to November 2018, represents a critical period for downstream flood risk assessment. Using historical data from 2013-2023, the models were trained and validated to predict missing data during periods where no actual observations are available. Hyperparameter tuning significantly improved model performance, further establishing Random Forest's highest accuracy in addressing hydrological data gaps. Model performance was assessed using metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>). Among the tested approaches, Random Forest demonstrated superior performance, yielding the lowest errors and highest R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, making it reliable choice for handling hydrological time-series gaps. By comparing performance metrics, this research contributes the critical role of AI-based models in addressing missing hydrological data, providing actionable insight for flood management and emergency planning in regions prone to extreme weather variability.