Monitoring the Spatiotemporal Dynamics of Wetland Vegetation Aboveground Biomass in the Yellow River Delta

In the Yellow River Delta, the aboveground biomass (AGB) of Tamarix chinensis, Suaeda salsa, and Phragmites australis reflects wetland productivity and carbon sequestration capacity. However, due to redundancy among multi-source remote sensing features, large-scale AGB estimation remains a challenge. Field measurements from sample plots were integrated with multi-source remote sensing features from the Diaokou River and Qingshuigou regions. A machine learning framework combining feature selection, VIF-based multicollinearity assessment was developed. The optimal feature selection strategy varied among machine learning algorithms. The random forest model combined with stepwise regression (RF-STEP) achieved the highest estimation accuracy (R2 = 0.75, RMSE = 116.91 g/m2, RPD = 2.24). SHAP analysis revealed that vegetation indices including TVI and NDVI dominated the prediction. Wetland AGB generally increased during the growing seasons from 2017 to 2024, with mean values of 225.13 g/m2 in the Diaokou River region and 170.15 g/m2 in the Qingshuigou region. A larger proportion of the Qingshuigou region exhibited moderate AGB increase according to pixel-wise fitted slopes, at 50.46%, compared with 38.50% in the Diaokou River region. Areas with negative fitted slopes occupied 23.65% of the Diaokou River region, implying potential localized degradation. These research findings support wetland ecological restoration and sustainable management in the Yellow River Delta.