Ước tính độ đục từ ảnh LANDSAT - 8: So sánh mô hình Ridge và Partial Least Squares Regression tại phá Tam Giang, Thành phố Huế

Abstract

Turbidity strongly regulates ecosystem dynamics in coastal lagoons by controlling light availability, nutrient cycling, and aquatic productivity. This study investigated turbidity in the northern Tam Giang Lagoon (Hue, Vietnam) by integrating field measurements with Landsat - 8 imagery. Two regression approaches - Ridge and Partial Least Squares Regression (PLSR) - validated their performance for turbidity estimation, revealing an outperformance of PLSR, achieving R² = 0.79, RMSE = 0.41, and MAE = 0.31 using a subset of Pearson based selection features, compared to Ridge with R² = 0.65. Variable importance analysis highlighted the dominant role of red–NIR bands (R865_655) and shortwave infrared bands (R2201, R1609) in explaining turbidity dynamics. Spatial maps generated by PLSR identified three hotspots close to river mouths, shallow aquaculture zones, and agricultural margins, where turbidity exceeded 6 Formazin Nephelometric Unit (FNU), while central and southern lagoon waters remained clearer. These findings demonstrate that satellite-based turbidity modeling provides reliable, cost-effective monitoring across complex lagoon environments. By capturing both natural variability and anthropogenic impacts, this approach offers crucial insights for sustainable water quality management and adaptive responses to climate and human pressures in coastal ecosystems.