Estimation Biomass Model Of The Huanglong Mountain In Northern Shaanxi Based On Remote Sensing

As the main body of the terrestrial ecosystems, forestry is the largest carbon pool in theworld. The biomass of the forest ecosystem not only reflects complex relationship betweenthe material circulation and energy flow in the forest and it’s environment, but also provide aplatform to study the forestry and ecological issues. Therefore, the study of the forest biomasshas a great significance to studying the global carbon cycle and climate change.Using the Huanglong Mountain as the study area, based on forest sublots and TMimages, the biomass estimation model was built and judged reasonably. The various datarequired for the establishment of forest biomass model were introduced, including TM image,forest survey data of Huanglong Mountain in2006. First, biomass was calculated by usingforest survey data and the growing stock biomass conversion model table; eight vegetationindex were extracted from the remote sensing data (SAVI2, DVI, PVI, SAVI, TVI, NDVI,RDVI);6gray values form Bands1-7except the Band6; elevation, slope, aspect wereextracted from DEM; the canopy density was from forest inventory data. Used by the17indicators as a variable factor, The biomass models were built respectively in SPSS andMATLAB. The conclusions of this study were summarized as follows:(1)17variable factors were extracted form TM data and Digital elevation model (DEM)and forest inventory data, in which16factors were significantly associated with biomass atthe0.01level except the aspect. All the gray value of bands from TM had a significantcorrelation with biomass, TM7has the most significant, the correlation coefficient reached0.265. In this study, all vegetation index extracted from TM data were significantly associatedwith biomass. DVI、PVI、NDVI、RDVI had positive correlations, PVI had a better reflectionof the biomass than others. However, elevation, slope, aspect had a bad correlation with thebiomass. the canopy density showed the most positive correlation among17variable factors.(2)Using the correlation analysis,16variable factors which had a significantcorrelation with biomass were selected out, including7vegetation index(SAVI2, DVI, PVI,SAVI, TVI, NDVI,RDVI),2geoscience factors(elevation, slope),6gray values of TM bands1-5and band7, canopy density from forest inventory data.3estimation models were built byusing multiple regression analysis in the SPSS, there was a comparison between the3estimation models by coefficients, coefficient of determination and standard error. Finally theoptimal model of multiple regression analysis was selected out. B=-32.292+104.715×YBD +802.569×B5+0.262×elevation+0.216×slope-751.766×B7-312.832×PVI.(3)Finally, a comparison was taken between the multiple linear regression model and BPartificial neural network model, the result showed that, for the linear regression model, thecoefficient was0.554, the root mean square error was14.98t hm-2, the prediction accuracywas73.08%, for the BP artificial neural network model, the coefficient was0.738, the rootmean square error was10.71t hm-2, the prediction accuracy was87.49%. The model of BPartificial neural network was more suitable for the Huanglong mountain.