Remote Sensing Estimation Of Forest Biomass In Jiulian Mountain Nature Reserve Of Jiangxi Province

Taking Jiulian Mountain Nature Reserve of Jiangxi Province as the study area,using Landsat5 TM image of November 3rd as data source,combined with plot data of Continuous Forest Inventory in 2009,this study established remote sensing estimation model based on "3S" technique in order to provide a basis for forest management of study area.Firstly,the biomass of each plot was calculated according to the relative growth model combined with tally data.Then,after a series of processing for remote sensing image like radiometric calibration,atmospheric correction,geometric correction,SCS+C terrain correction,six gray values from Band1 to Band 5 and Band 7,nine kinds of vegetation indices(RVI,NDVI,DVI,?VI,etc.),seven kinds of band ratio and derived data(TM4/TM2,TM7/TM3,TM3/TM1,etc.),the principal components(PCA1,PCA2,PCA3),tasseled cap factors(BI,GVI,WI,)were extracted from TM data.Slope and aspect were extracted from DEM.Finally,the remote sensing estimation model of biomass were established by using the unitary regression model,multivariate regression model and BP neural network model.The main conclusions are as follows:(1)A total of 30 variable factors were extracted from TM image and DEM,in which 16 factors were significantly associated with biomass at the level 0.01.Five gray value of bands had a significant negative correlation with biomass.RVI,NDVI,IIVI,MSI,these 4 vegetable indices were significantly correlated with biomass.The correlation of vegetation indices were significantly higher than that of single band.The correlation of principal components,tasseled cap factors,terrain factors(slope,aspect)with biomass were not high.(2)BP neural network model is better than unitary regression model and multiple regression model.The determination coefficient of BP neural network was 0.867 and its root mean square error was 22.94.The determination coefficient of the multiple linear regression model was 0.725,and its root mean square error was 39.09.At last,BP neural network model was chosen to estimate the forest biomass.The total forest biomass of study area was 2.89 million tons,the range of biomass density was 11.47 t/ha-376.47t/ha,the average forest biomass density was 159.07t/ha.(3)The forest biomass in the study area was mainly distributed at the altitude of 330-1000m.As the altitude increases,biomass density showed a trend of upward downward upward downward;Forest biomass was mainly distributed at slope of 16-35°,the biomass density with slope increased firstly and then decreased;The biomass distributed evenly at the different aspects.