Estimation techniques of aboveground biomass in grassland ecosystems based on multi-source data integration

Abstract

Above Ground Biomass (AGB) is a core indicator for evaluating the health and sustainability of grassland ecosystems, but traditional ground measurements and single remote sensing estimates have insufficient accuracy in complex scenarios. To this end, this study proposes an artificial intelligence driven framework that integrates deep learning and ensemble learning. Based on multi-source data such as high-resolution remote sensing, time series, ground measurements, and meteorology, a multi-source data integration algorithm is constructed by combining Gaussian pyramid, wavelet transform, and principal component analysis. Then, a U-shaped network (UNet) and extreme gradient boosting (XGBoost) are integrated to construct a prediction model for AGB estimation. Compared with existing methods that rely solely on data concatenation or traditional dimensionality reduction, this study achieves deep integration of multi-scale spatial features and time–frequency dynamics, which improves feature expression and estimation accuracy in complex scenes. Experimental results show that the Pearson correlation coefficient between the features extracted by the proposed multi-source data integration algorithm and the measured AGB reaches 0.95. The proposed model has an estimated accuracy rate of up to 93.04% in three types of grasslands, with an R2 of 0.88 and the minimum root mean square error (RMSE) of 10.69 g/m2. The results indicate that this model can effectively address the shortcomings of traditional methods and existing remote sensing estimations, significantly improve the estimation performance of AGB in complex scenarios.