Remote Sensing Estimation And Change Analysis Of Grassland Aboveground Biomass In Northern China

Long-term and comprehensive monitoring of the spatiotemporal dynamics of grassland aboveground biomass（AGB）is key to assessing grassland productivity,maintaining grassland sustainable utilization,and analyzing the impacts of climate and human interventions on grassland ecosystems.Remote sensing is the powerful way to rapidly monitor grassland AGB at a regional scale.Currently,the research on remote sensing monitoring of grassland AGB focuses on multi-factor machine learning model construction and spatiotemporal dynamics analysis of grassland AGB at the regional scale.However,there are some limitations and shortcomings in the existing studies in terms of multivariate database construction,variable selection,model comparative analysis,sample size,observation years,study area,and driving factor analysis.It is still a difficult challenge to construct an accurate estimation model of grassland AGB in a vast region（covering a variety of grassland types）to support the analysis of AGB dynamics and its driving factors over a long time series.To address the problems of existing studies on grassland AGB monitoring and clarify the spatiotemporal dynamics of grassland AGB and its driving factors in northern China in recent decades,extensive grassland AGB measurements（collected in northern China during the grassland growing season of 2000–2019）,MODIS data（reflectance and vegetation indices）,and environmental factors（climate,topography and soil）were employed to construct the grassland AGB models in northern China and five zones,respectively,using four machine learning algorithms i.e.,random forest（RF）,support vector machine（SVM）,artificial neural network（ANN）and extreme learning machine（ELM）combined with four variable selections i.e.,successive projections algorithm（SPA）,genetic algorithms（GA）,least absolute shrinkage and selection operator（LASSO）and stepwise regression（STEP）.The spatial distributions of annual grassland AGB from 2000 to 2019 were simulated based on the optimal grassland AGB estimation model.The temporal change from 2000 to 2019 and possible future trend after 2019 of grassland AGB were comprehensively analyzed by the Slope model and Hurst exponent.The influences of natural and anthropogenic factors on the dynamics of grassland AGB were explored quantitatively using the Geodetector model and LMG method.The results showed that:（1）The RF model constructed from the variables selected by the SPA was more suitable for remote sensing estimation of grassland AGB in the study area.Four variable selection methods greatly simplified the models,reduced the multicollinearity among variables,and improved the prediction accuracy of machine learning models of grassland AGB.In contrast,SPA showed stable selection results and high operational efficiency,and SPA had the best ability to eliminate multicollinearity.Besides,the models constructed with the variables selected by SPA performed better.For the four machine learning algorithms,considering the prediction accuracy,model stability and computational efficiency,RF performed better than SVM,ANN and ELM.（2）When the sample size was sufficient,the sample plots were evenly distributed within the study area,and the modeling variables covered various vegetation indices and environmental factors,the machine learning models of grassland AGB constructed in the whole study area had higher stability and estimation accuracy than that of the models for each zone.The optimal estimation model of grassland AGB for the study area was the SPA-RF model constructed using all AGB samples within the study area,with R~2 and RMSE in the test dataset of 0.64 and 596.88 kg/ha,respectively.（3）The 20-year（2000-2019）average annual maximum grassland AGB in northern China ranged from 161.83-3663.09 kg/ha,with an overall spatial distribution pattern of being low in the central and western parts and high in the southeastern part.This was similar to the spatial gradient of annual precipitation.The annual maximum grassland AGB in most（85.87%）regions showed an increasing trend during 2000–2019.The future trend of grassland AGB after 2019 may be optimistic,as reflected by more grassland AGB was predicted to increase rather than decrease（71.12%vs.28.88%）.（4）The main driving factors of grassland AGB dynamics in the study area were precipitation,solar radiation,soil organic matter and livestock density.The influence of natural factors on AGB dynamics was greater than that of anthropogenic factors.In addition,the influences of the driving factors on grassland AGB dynamics in the study area were not independent but showed mutual enhancement.There was more bivariate enhancement between natural factors,while nonlinear enhancement was more likely to occur between anthropogenic factors,and between anthropogenic factors and natural factors.（5）There were spatial differences in the influence of climate drivers on the spatiotemporal dynamics of grassland AGB in the study area.The spatiotemporal dynamics of grassland AGB in 52.95%of grassland areas were dominated by precipitation;the AGB dynamics in 17.67%of the grassland areas were dominated by temperature,and the dominant influence of temperature on AGB dynamics was mainly in the semi-humid areas with better water and heat conditions;while solar radiation dominated 29.38%of the AGB spatiotemporal dynamics in the grassland areas,and solar radiation mainly played a stronger role in the grassland areas with relatively moderate water and heat conditions and sufficient sunshine.