| Forest ecosystems are one of the most important parts of terrestrial biosphere,which cover approximately 30% of Earth's total land area,not only provide valuable ecosystem goods,but also play a critical role in the global carbon cycle.Above ground biomass(AGB)is an important ecological indicator for the assessment of forest productivity and carbon sequestration rates.Traditional forest biomass estimation is mainly based on field measurements in relatively small sampling areas,which is time-consuming,expensive and spatially constrained.Remote sensing technique has been increasingly used for monitoring forest AGB at multiple spatial and temporal scales.Combing airborne LiDAR and satellite optical imagery could take advantage of both high precision three-dimensional structure information and large area coverage to achieve more accurate forest AGB at regional scale.The Three Gorges region of China is widely known due to the construction of the Three Gorges Dam,which brings tremendous impact on the eco-environment around it.Accurate forest AGB estimation of this area is of great significance for both natural resource protection and environment sustainable development.In this study,we combined field measurements,airborne LiDAR,PHI-3 hyperspectral imagery,ZY-3,Landsat 8 OLI and MODIS data to estimate forest AGB by scaling models in the Three Gorges region of China.Firstly,sixty variables,including 26 LiDAR metrics and 34 hyperspectral vegetation indices(VIs)extracted from the airborne imaging spectroscopy were selected and a stepwise regression model was used to estimate AGB in the flight area(11 km×14 km)located in the Shennongjia Forest Nature Reserve.Secondly,a typical study area(40 km×60 km)covered the flight area was used for up-scaling,where we built AGB estimation models using classic stepwise regression and machine learning algorithms for needle-leaved forest(NLF),broad-leaved forest(BLF)and needle-broad-mixed forest(NBMF),respectively,based on 13 Landsat VIs,4 topographic factors,8 texture indices from ZY-3 high spatial resolution data and AGB estimation in the flight area.Finally,a scaling model combining the 30 m AGB in the typical study area and 250 m MODIS BRDF,LAI,VIs time series data and forest classification was built up using SVM algorism for mapping the AGB over the Three Gorges region,and we validated the results by filed measurements at county level.The main conclusions are as follows:(1)The AGB results in the flight area showed that LiDAR structural metrics combined with stepwise regression method could estimate AGB with a relatively high accuracy(R2=0.80,p<0.0001),and the introduction of hyperspectral data could slightly improve the regression accuracy(R2=0.83,p<0.0001).We validated the AGB results by 10-fold cross validation and the mean value of R2 was 0.798.(2)In the typical area,the SVR(support vector regression)always outperformed the stepwise regression and RF(random forest)algorithm in each forest type.The ten-fold cross validation results showed that SVR had a best accuracy(R2=0.637,RMSE=12.976 t/ha)for NLF,followed by BLF(R2=0.611 RMSE=13.522 t/ha)and NBMF(R2=0.504,RMSE=16.046 t/ha).We validated the results by 20 filed plots and found that SVR estimated AGB was accurately(R2=0.689),while the RF estimation results tended to underestimation for high value and overestimation for low value.(3)In the Three Gorges region,MODIS BRDF parameters showed significant correlation with AGB,especially for the weight of geometrical-optical scattering in red band(r =0.438).The accuracies of SVR model for AGB upscaling tended to be stead with the increasing training sample size.For NLF and BLF,the SVR accuracies R2 showed significant variation(0.53-0.65)when the sample size was less than 2000 and stayed pretty stable around 0.63 and 0.65,respectively.We validated the results by filed measurements at county level and found that the estimated results in this paper were consistent with the traditional forest inventory data(R2=0.6184). |
PDF Full Text Request