TransformBioRegressor: A Transformer-based Grassland Biomass Estimator using Multi-source Data

Accurately monitoring aboveground biomass (AGB) in grassland ecosystems is critical for sustainable grassland management and ecological protection. In this paper, TransformBioRegressor is proposed for estimating AGB by utilizing multi-source data, including remote sensing, geographic, and meteorological data. It is a transformer-based model that integrates self-attention mechanisms with multi-layer perceptron (MLP) structures. Extensive experiments have shown that TransformBioRegressor consistently outperforms other models, including XGBoost, Random Forest, and MLP, in terms of Root Mean Square Error (RMSE), Rsquared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), and Mean Absolute Percentage Error (MAPE). TransformBioRegressor achieves an RMSE of 59.47 g/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, a MAPE of 27.85%, and an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.80 when using full spectral bands. Additionally, this study analyzes the spatial and temporal distribution of AGB in West Ujimqin Banner, a representative grassland region in northern China, and identifies precipitation as the most significant environmental factor influencing biomass, further enhancing the understanding of grassland dynamics.