The Estimation And Dynamic Modeling On The Aboveground Biomass Of Pinus Densata In Shangri-la Based On Landsat

Based on 58 plots of field investigationin 2015 and 96 plots of continuous forest inventory in 1987,1992,1997,2002,2007,three kinds of aboveground biomass estimation models of pinus densatacombined with Landsat image data and DEM datain Shangri-la city were built.They are single-period pinus densata biomass estimation model,multi-period pinus densata biomass estimation model and pinus densatadynamicbiomass estimation model.Each model used four kinds modeling methods,they were multiple linear regression(MLR),partial least squares regression(PLSR),random forest(RF)and gradient boosted regression trees(GBRT).At the same time,theses models were evaluated,including fitting effect and prediction effect.By comparing with each other,the best biomass estimation model was selected.Finally,based on the multi-period pinus densata biomass estimation model and pinus densata biomass dynamic estimation model,pinus densata biomass had been calculated.And the predictive quality of the two methods had been estimated and contrasted.So,we can draw the following conclusion.(1)In single-period pinus densata biomass estimation model,R5T4 EN,R9T4CC,R5T4 SM,R5T5CC,TM73 and DVI have important influence on biomass.Through comparing four methods,GBRT is selected to be the best model.Its coefficient of determination(R2)is 0.958,and its precision estimation(P)is 73.878%(2)In multi-period pinus densata biomass estimation model,R5T4 CC,R5T1HO,ND32,R5T2 CC,R5T7CC and R9T4 CC have important influence on biomass.Through comparing four methods,GBRT is selected to be the best model.Its coefficient of determination(R2)is 0.949,and its precision estimation(P)is 63.524%(3)In pinus densata dynamic biomassestimation model,R9T7 HO and R9T7 CO have important influence on biomass.Through comparing four methods,GBRT is selected to be the best one.Its coefficient of determination(R2)is 0.945,and its precision estimation(P)is 60.422%(4)Comprehensive comparing three models,we can find that thenon-parametric method's fitting effect and prediction effect was generally superior to the parametric method's.It shows that non-parametric method had the better effect on the study of pinus densataaboveground biomass estimation model.Particularly,GBRT is the best one.(5)Comparing the modeling factors of three kinds model,we can find that the quantity of correlation(CC)was the most,which was important to biomass.It shows that correlation(CC)was the most important kind factor to biomass.(6)In the factors which was important to biomass,the quantity of the factor that was derived from near infrared band(TM4)is the best.It shows that near infrared bandwas the most important band in the bands of Landsat image.(7)The method based on pinus densata biomass dynamic estimation model is the better way to calculateaboveground biomass.Also,GBRT is the best method.