利用地理随机森林绘制地上生物量：克服大尺度区域的空间非平稳性

[Objective]Forest above-ground biomass(AGB)information and its drivers are crucial to the formulation of forest management activities.However,there are still large uncertainties in the estimation results of AGB,and few studies focus on the spatial variation of AGB driving mechanisms at small scales,which will hinder the precise forest management at specific locations.[Method]We choose Inner Mongolia Autonomous Region(IMAR)as the research area,which has a large distance from east to west,obvious changes in vegetation,climate,soil and topography,and strong spatial non-stationarity.We established four AGB estimation models in this region,namely geographical random forest(GRF),random forest,gradient enhanced regression and support vector regression model.We establish the AGB driving model with GRF method,and analyze the change of the influence of driving factors on AGB in space.[Result]Among the 32 AGB predictors,14 features have significant spatial non-stationarity,involving spectral and environmental variables.Among the driving variables,4 factors are spatially different from the driving mechanism of AGB.The GRF method achieved R2 values of 0.622 and 0.481 in random and spatial cross-validation,respectively,outperforming traditional machine learning methods such as Random Forest(RF),Gradient Boosting Regressor(GBR),and Support Vector Regression(SVR).Furthermore,GRF is less uncertain and more stable.Species diversity is the most important driving factor of AGB,and its influence is similar to the variance of monthly precipitation distribution in the growing season(SDP),and their influence is relatively uniform in the whole study area.The driving force of slope and total precipitation showed significant gradient change during the growth period.[Conclusion]In the western region of IMAR,because the change of slope has a great impact on AGB,the area with low slope should be preferred for planting trees.In the northeast region,the influence of slope is relatively small,but special attention should be paid to the impact of extreme precipitation events on AGB,and monitoring of low-lying areas prone to waterlogging should be strengthened to prevent trees from dying due to excessive waterlogging.In addition,it is recommended to avoid monocarpy planting in all regions,and prioritize multi-species mixed planting to improve the stability and adaptability of ecosystems.