Study On Estimation Of Aboveground Biomass Of Grasslands In The Three-River Headwaters Region Based On Deep Learning Models

Grassland is an important part of the terrestrial ecosystem and plays an important role in the terrestrial carbon cycle.The grassland ecosystem in the source area of the Three Rivers is the most sensitive and fragile area in China.Carrying out research on the dynamic change of above-ground biomass in the source areas of the Three Rivers can provide a basis for the sustainable use and scientific management of grassland resources in the area.The purpose of this study was to explore the spatial and temporal distribution and change trend of aboe biomass in the grassland of sanjiangyuan from 2001 to 2015,combines the above ground biomass(AGB)data of 148 field-measured grasslands in the source area of the Three Rivers from 2005 to 2007 and the enhanced vegetation index,meteorology,and altitude data over the same period to construct a grassland above-ground biomass estimate based on a deep confidence network(DBN)The model is compared with the multiple linear regression model(MLR)and support vector machine.The aim is to propose an effective and operable grassland biomass estimation model.The grassland AGB in the study area is estimated with high accuracy.Provide a basis for sustainable use of resources and scientific management.The main results are as follows:(1)Meteorological data(photosynthetic effective radiation,temperature,precipitation and relative humidity),remote sensing data(EVI,NDVI),elevation data and soil data(soil PH,soil bulk density,soil water content and soil organic carbon content)were used as alternative driving factors for the model,and correlation analysis was conducted with the measured biomass data.The results showed that photosynthetic effective radiation,elevation data,average annual photosynthetic effective radiation,average annual moderate annual precipitation and measured biomass in 2005-2007 had significant correlations,so these five variables were used as the driving factors of the model.(2)The R2 and RMSE of the deep belief network model training and testing were 0.83,50.17 g/m2 and 0.81,43.75 g/m2,respectively.The simulation results were better than the machine learning model,indicating that the model could be used to estimate the above-ground biomass of grassland in the source area of the three rivers.(3)The average aboveground biomass in the Three-River Headwaters Region from 2001 to 2015 was 172.55 g/m2,showing a slow rising trend,with an average increase of 0.29 g/m2 y-.The mean aboveground biomass in 2006 was the highest(185.56g/m2),and the lowest(161.03 g/m2)in 2010.Among the three types of grassland,alpine grassland,alpine meadow and sparse grassland,the mean AGB of alpine meadow is the highest(221.36 g/m2),while that of sparse grassland is the lowest(125.21 g/m2).There was obvious heterogeneity in spatial distribution,which decreased from southeast to northwest.Among them,the above-ground biomass of tramalai grassland in the north was the lowest(106.6 g/m2),and that of henan county in the east was the highest(345.71 g/m2).(4)From 2001 to 2015,the above-ground biomass of grassland in the Three-River Headwaters Region was mainly stable and restored.Among them,61.57%of the grasslands remained stable,31.44%of the grasslands showed a recovery trend,and 6.99%showed a deterioration trend.