Grassland Biomass Inversion And Drought Vulnerability Evaluation Based On Machine Learning Methods

As an important part of natural ecosystem,grassland ecosystem provides important grazing resources for the development of livestock economy and plays a very important role in regulating climate change and maintaining the balance of ecosystem.The grasslands of Xilin Gol league are located in the arid/semi-arid zone,and with the impact of climate change and human activities,coupled with the relatively homogeneous vegetation type of the region,the vulnerability of grassland ecosystems has increased significantly,and it has become important to conduct quantitative vulnerability assessment of them.Above Ground Biomass(AGB)is the material carrier of grassland ecosystem to obtain energy and fix carbon dioxide,represents the basic level of primary productivity,and determines the capacity of grassland to carry grass and livestock.The change of grassland AGB directly reflects the degree of grassland degradation and pasture desertification.Currently there are more drought vulnerability studies using remote sensing indices characterizing vegetation greenness growth,and less drought vulnerability related studies on vegetation productivity.There is a need to gain insight into the drought vulnerability of AGB of grasslands in Xilin Gol league.Therefore,this paper takes Xilin Gol league as the study area and uses multi-year grassland AGB actual measurement data,meteorological station data,remote sensing data and other multi-source data.Based on the machine learning method,a grassland AGB estimation model was established,the inversion study of grassland AGB was carried out,and the spatial and temporal variation characteristics of grassland AGB in Xilin Gol league from 2000 to 2020 were analyzed.Then,the SPEI(Standardized Precipitation Evaporation Index)was calculated for the short-,medium-,medium-,and long-term time scales of grasslands in Xilin Gol League based on the meteorological station data,and the spatial and temporal variation characteristics were analyzed.On this basis,the drought vulnerability of different grassland types in the study area at long time scales was investigated in conjunction with drought vulnerability theory.The results of the study are as follows:(1)Among the three machine learning algorithms built to estimate the grass AGB,the random forest algorithm built the model with the highest accuracy.The results of the analysis of the contribution of various characteristic variables showed that the contribution of vegetation elements were all relatively high,while the contribution of terrain elements was low.The overall trend of fluctuating increase in grassland AGB from 2000 to2020 in Xilin Gol league,with the highest grassland AGB in 2013,while the lowest grassland AGB value was reached in 2009.Spatially,grassland AGB showed a spatial distribution characteristic of gradually decreasing from northeast to southwest.Among the various grassland types,the total AGB of typical grassland was the highest,followed by meadow grassland and desert grassland the least.Among the 12 banners and counties under the jurisdiction of Xilin Gol League,the grassland AGB was higher in East and West Uzhumqin Banner,and lower in Sunit Left Banner,Sunit Right Banner and Erlianhot City.(2)The cumulative droughts that occurred in the grasslands of Xilin Gol league from 2000 to 2020 were mainly light droughts,and there were differences in the occurrence of droughts in various grassland types at different time scales of SPEI.Among the typical grasslands,the highest frequency of mild drought occurred in SPEI-3 and SPEI-24 reached 28.6%and 14.3%,respectively.In meadow grassland,the highest frequency of mild drought occurred in SPEI-24 reached 19.0%,moderate drought occurred in SPEI-1 reached 14.3%,and severe drought occurred in SPEI-3 and SPEI-12 reached 9.5% and 4.8%,respectively.Among the desert grasslands,SPEI-6 had the highest frequency of mild drought of 19.0%,SPEI-48 had the highest frequency of moderate drought of 19.1%,and SPEI-1 had the highest frequency of severe drought of 9.5%.Among other grasslands,the highest frequency of mild drought was 23.8% in SPEI-24,19.0% in SPEI-1,and 4.8% in SPEI-3.(3)The exposure analysis of AGB of grasslands in Xilin Gol league found that most regions showed a drought-free state,with droughts in2011～2015 mainly occurring in Sunit Right Banner and in 2016～2020mainly in the northern part of Abaga Banner.The results of drought sensitivity calculation of grassland in Xilin Gol league showed that the drought sensitivity of meadow grassland was the highest among various grassland types,and the drought sensitivity of desert grassland was lower.The results of drought adaptation calculation showed that the drought adaptation of Xilin Gol league grasslands was the lowest from 2000 to2005,and the drought adaptation was generally higher from 2011 to 2015.The results of the drought vulnerability study showed that Sunit Left Banner and Sunit Right Banner exhibited low drought vulnerability in each study time period,while the drought vulnerability in East and West Urumqi was generally higher.The results of this paper are important for scientific adjustment of grass-livestock relationship,protection of ecological and environmental security,and sustainable development of grassland resources.