Estimating vegetation aboveground biomass in Yellow River Delta coastal wetlands using Sentinel-1, Sentinel-2 and Landsat-8 imagery

Abstract

Accurate analyzing the spatial pattern and spatial uncertainty of vegetation aboveground biomass (AGB) in coastal wetland is critical for addressing sustainable blue carbon management goals. Eight models based on Extreme Gradient Boosting (XGBoost) method were established to analyze the capability of Sentinel-1 (S1), Sentine-2 (S2) and Landsat-8 (L8) data for predicting AGB in coastal wetlands of the Yellow River Delta (YRD), China. Spatial uncertainty of AGB was quantified by Quantile Regression Forest (QRF) method. The results showed that AGB model based on S2 achieved higher model performance (R²: 0.74, RMSE: 171.23 g/m²) compared with those based on L8 (R²: 0.59, RMSE: 198.84 g/m²) and S1 (R²: 0.43, RMSE: 219.60 g/m²). The AGB model based on S1, S2, L8 and other predictive variables including the terrain and biophysical factors (S1S2L8plus) achieved the highest model performance (R²: 0.80, RMSE: 154.98 g/m²) among all the models. Red-edge related-spectral indices derived from S2 were proved to be important predictors in AGB modelling. The spatial uncertainty quantified by QRF showed the spatial prediction uncertainties of AGB models based on S2S2L8plus and S2 were lower than AGB model based on S1L8. The results of this study demonstrate the suitability of optical remote sensing data especially S2 and the weak capability of S1 in modelling AGB in coastal wetlands of the YRD. The regularly modelling, mapping and uncertainty estimations of AGB could help guide the sustainable blue carbon management in coastal wetlands.