Cross-scale mapping of Leaf Nitrogen Content in arid desert shrubs by integrating UAV and satellite data under sparse sample conditions

Accurate regional-scale mapping of Area-based Leaf Nitrogen Content (LNCarea) is critical for understanding plant physiological processes and nutrient status in arid regions. However, such mapping remains challenging in desert shrublands because harsh environmental condition, sparse vegetation cover, and strong spatial heterogeneity limit the availability and representativeness of field samples, while single-source remote sensing provides insufficient information. To address these challenges, this study developed an integrated framework that combines field observations, unmanned aerial vehicle (UAV) and Sentinel satellite imagery to estimate LNCarea for desert shrubs. To alleviate sample scarcity and imbalance, the Synthetic Minority Over-sampling Technique for Regression (SMOTER) method was applied to generate synthetic samples, which significantly enhanced the model’s generalization at the UAV scale (R2 = 0.78, RMSE = 0.166 g m−2). Using the eXtreme Gradient Boosting (XGBoost) algorithm, this study successfully upscaled shrub-level LNCarea estimation derived from UAV imagery to a regional-scale using Sentinel satellite imagery at 10 m resolution, achieving a high predictive accuracy (R2 = 0.74, RMSE = 0.826 g m−2). SHapley Additive exPlanations (SHAP) analysis showed that raw spectral bands as the most influential predictors, contributing over 47% to the model’s decisions. Spatially, LNCarea exhibited generally low values (Mean = 0.74 g m−2, Standard Deviation = 0.9 g m−2), with approximately 76% of pixels falling within the range of 0–1 g m−2, and showed a distinct decreasing gradient from the oasis fringe towards the desert interiors. Overall, this study proposes a framework that integrates multisource remote sensing with SMOTER-based sample augmentation to address the challenge of sparse samples in desert shrub ecosystems. It enables reliable cross-scale estimation of LNCarea across UAV and satellite observations, facilitating the spatially evaluation of vegetation nutrient status in arid regions.