Research On The Estimation Of Forest Parameters Based On Remote Sensing Image Data

Forests are the largest reservoir of organic carbon in global terrestrial ecosystems,and forest parameters such as forest biomass,growing stock,and canopy closure can objectively reflect the net primary productivity of forest ecosystems,providing a relevant basis for forestry production management and forest management decisions.Traditional forest parameter estimation has large limitations,which is not only timeconsuming and laborious,but also has certain bias when extending from the sample sites to a larger area.Remote sensing technology has become an important tool for forest parameter estimation because of its characteristics of observation over a large area,high timeliness and low cost.Compared with other remote sensing data sources,optical remote sensing images are still the most widely used data source for forest parameter estimation because of their wide spatial and temporal coverage,diverse resolution and low cost.This thesis mainly focuses on the estimation of forest biomass,growing stock and canopy closure,and takes Chun’an County in Hangzhou city,Zhejiang province as the study area,selects Landsat-8 OLI and GF-6 WFV images,with continuous forest inventory data of Zhejiang Province and DEM data,and selection of modeling factors by Pearson correlation coefficient method and random forest-recursive elimination method,and uses multiple linear regression,BP neural network and random forest modeling to construct models for forest biomass,growing stock and canopy closure based on modeling factor set of Landsat-8 and GF-6 respectively,and the accuracy of different models was compared.The main results are as follows:(1)Among the 23 modeling factors of the Landsat-8 modeling factor set,12 modeling factors were significantly related to forest biomass,growing stock and canopy closure.And among the 26 modeling factors in the GF-6 modeling factor set,the number of significantly correlated modeling factors were 17,13 and 19 respectively.Among all modeling factors,the highest correlation with both biomass and canopy closure was the red-edge vegetation index(NDVIre1)in GF-6 with 0.542 and 0.625,respectively,and the highest correlation with growing stock was the green light band(Band2)in GF-6with-0.486.The correlation between the GF-6 and forest parameters was higher than Landsat-8 overall,and the correlations between the red-edge vegetation indices and the forest parameters was generally higher,and the correlation between the spectral bands or vegetation indices of both remote sensing images and the forest parameters was overall higher than those of the texture features.(2)From the inversion results based on the validation dataset,the inversion accuracy based on GF-6 modeling factor set is higher than that of Landsat-8 when the same modeling method is used;when different modeling methods were used,the inversion accuracy of the random forest and BP neural network models was higher than that of the multiple linear regression model,with the random forest model having the highest inversion accuracy.In general,the random forest model based on GF-6 had the highest inversion accuracy for the test dataset of each forest parameter: biomass R2 was0.579 and RMSE was 27.74 t/ha,growing stock R2 was 0.634,RMSE was 30.06 m3/ha,and canopy closure R2 was 0.678,RMSE was 0.101.(3)The data of forest parameters in the study area obtained by inversion of GF-6based random forest model,statistical analysis shows that: 1)The total biomass was31.8356 million tons,and the biomass per unit area was 89.58 t/ha.The biomass per unit area decreases and then increases with the increase of elevation and increases with the increase of slope,and the biomass per unit area in each slope direction does not differ significantly.2)The total growing stock in the study area was 21,506,300 m3,and the average growing stock is 60.51 m3/ha.The growing stock per unit area decreases and then increases with the increase of elevation and slope.3)The range of forest canopy closure was 0 to 0.96,with an average canopy closure of 0.65,mainly concentrated in the range of 0.6-0.8.