Vegetation Aboveground Eco-water Stock Inversion Using Remote Sensing In The Upper Reaches Of Minjiang River

Vegetation eco-water?VEW?refers to the water closely connected with land surface vegetation.It includes the water intercepted or stored by the leaf,root,humus layer,surface soil and plant itself.It is a special transitional zone in the water cycle of terrestrial ecosystem and can regulate surface water and groundwater,affect evapotranspiration with important ecological functions.Vegetation aboveground eco-water stock?VAEWS?is the sum of the water stored by the aboveground part of vegetation,which is an important part of VEW.The upper reaches of Minjiang River have unique geographical location,complex natural environment with environmentally fragile and important ecological function status.Remote sensing technology can effectively explore the stock and distribution characteristics of VAEWS for its large amount of information,fast,efficient and large coverage area.Therefore,the study of VAEWS,by remote sensing technology,in the upper reaches of Minjiang River,is helpful to understand the characteristics and distribution of VEW in the upper reaches of Minjiang River,further analyze the problems of ecological environment deterioration,frequent disasters and water resource shortage,and guide the work of environmental protection,ecological restoration and reconstruction and water resource calculation in this region.On the basis of group's long-term VEW research and related researches at home and abroad,aiming at the problem of the lack of VAEWS inversion in previous studies,this study qualified VAEWS with Sentinel 2 data combining machine learning algorithms in Maoxian upper Minjiang River.Firstly,plot level VAEWS were measured in the field.In the lab,support vector machine was performed to classify land use/cover with Sentinel 2 data.According to the band reflectivity,spectral index,texture index,land use/cover classification data and so on,the regression analysis was carried out by using random forest and artificial neural network through variables selection,and the inversion model of VAEWS was established.Finally,random forest model was selected to invert VAEWS duo to its outperformance.The main results are:?1?Green band,red band and near infrared band,four red edge band,normalized difference vegetation index and elevation were chosen as characteristic band,support vector machine was used to classify land use/cover in the study area,and good classification results were obtained with an overall accuracy of 90.45%and a Kappa coefficient of 0.88.?2?The two machine learning algorithms,random forest and artificial neural network,have great difference in the predictors.In 35 initial prediction variable,random forest algorithm chose blue band,green band,normalized difference near infrared index and normalized difference vegetation index?calculated by band 5 and band 4?,vegetation type,elevation,soil bulk density as the final prediction variables to build model based on 10 fold cross-validation and variable importance order.However,artificial neural network algorithm chose normalized differential vegetation index,normalized difference near infrared index,soil-adjusted vegetation index,entropy,vegetation type,and soil bulk density as the final prediction variables through variable importance order and repeated comparison.?3?A 10 fold cross-validation was used to verify the accuracy of the ecological water stock inversion model established by the two machine algorithms.Random forest performed better with a model efficiency?R²?of 0.74 and a root mean square error?RMSE?of 49 Mg ha^-1 compare to artificial neural network with R² of 0.64 and RMSE of 57 Mg ha^-1.?4?According to the vegetation aboveground ecological water stock inverted by the random forest model,it was found that the VAEWS of different land use/cover types were significantly different with coniferous forest(217.81 Mg ha^-1)>broad-leaf forest(158.78 Mg ha^-1)>orchard(76.07 Mg ha^-1)>shrub-grass(36.63 Mg ha^-1).The total stock were 2.61×10⁷ Mg for coniferous forest,5.45×10⁶ Mg for broad-leaf forest,4.74×10⁶ Mg for shrub-grass,3.55×10⁵ Mg for orchard.The total vegetation aboveground ecological water stock of the study area is 3.66×10⁷ Mg.Spatially,places with larger vegetation aboveground ecological water stock are mainly distributed at relatively higher elevations,while smaller stock places are mainly distributed along two sides of Minjiang River.Based on previous studies on vegetation water content and VEW,this study innovatively quantified vegetation aboveground ecological water stock.The results show that remote sensing technology combined with machine learning algorithm can effectively estimate VAEWS,and VAEWS is closely related to the ecological environment.This study makes for understanding of VEW's characteristics and distribution,ecological environmental protection and water resource calculation in the upper reaches of Minjiang River.