|
Title |
Monitoring the Algal Blooms of Daecheong Lake Using XGBoost Based on High-resolution Satellite Imagery
|
|
Authors |
황인태(Hwang In-tae) ; 김양완(Kim Yang-wan) ; 김기영(Kim Ki-young) ; 박종민(Park Jong-min) |
|
DOI |
https://doi.org/10.15681/KSWE.2026.42.3.243 |
|
Keywords |
Chl-a; HLS; SHAP; TP; TSI; XGBoost |
|
Abstract |
Recent climate change has led to increased water temperatures and altered precipitation patterns, which have intensified harmful algal blooms in major reservoirs across Korea. This situation underscores the necessity for continuous and extensive spatial monitoring. In response, this study developed a machine learning framework to estimate Total Phosphorus (TP) and Chlorophyll-a (Chl-a) concentrations in Daecheong Lake, utilizing Harmonized Landsat Sentinel-2 (HLS) satellite imagery and the eXtreme Gradient Boosting (XGBoost) model. Following this, trophic conditions were assessed using the Trophic State Index (TSI). HLS surface reflectance data from 2013 to 2024 were combined with in situ measurements from national monitoring networks. SHAP (Shapley Additive Explanations) analysis was conducted to identify the most effective spectral inputs, revealing that the Aerosol (0.43?0.45 μm), Green (0.53?0.59 μm), and NIR (Band 8A) bands were the most significant contributors to predictions of both TP and Chl-a. Model validation showed correlation coefficients (R) of approximately 0.62 for both variables, with RMSE and MAE values of 0.018 mg/L and 0.010 mg/L for TP, and 0.023 mg/L and 0.011 mg/L for Chl-a, respectively. While the model successfully captured overall temporal trends, it tended to underestimate peak concentrations during summer months. The spatial distributions of TSI, derived from predicted TP and Chl-a, closely aligned with periods of official algal bloom warnings, with Chl-a-based TSI exhibiting greater sensitivity to short-term environmental changes. These findings demonstrate the effectiveness of HLS-based XGBoost modeling for monitoring eutrophication in large reservoirs.
|