Vegetation Cover Estimation And Biomass Simulation In Horqin Sandy Land Using Ground-based Hyperspectral Remote Sensing

The spectral parameters and quantitative models are the key components of real-time monitoring and accurate diagnosis of vegetation growth. With people’s sustained concern over ecological environment, its development on monitoring, evaluation, management and other aspects urgently requires for an information accessable technology with lower cost and consumption, higher precision, density and stability. The application on vegetation ecology by ground-based hyperspectral remote sensing technology makes rapid monitoring of vegetation growth, and accurate estimation of vegetation physicochemical properties possible.In this paper, the study area was in Agula ecohydrological experiental area, which is in southeast of Kezuohouqi, Tongliao city, Inner Mongolia. Using the FieldSpec HandHeld portable field spectroradiometer of Analytical Spectral Device (ASD, Inc., Boulder, CO, USA), a series of observations on typical vegetation were carried out, and on this basis, four groups of controlled in situ experiments were conducted to evaluate the anisotropic reflectance effect (ARE) on spectral mixture analysis (SMA) for vegetation cover estimation. Finally, simulation of aboveground dry and green biomass (ADB and AGB) with hypersepctral vegetation index on two natural grassland sites in study area was perfomed. The main results were as follows:1. Due to the variation of pigments, liquid water, and internal structure in vegetation leaves, the reflectance spectrum of different vegetation type is different. A phenomenon called "different material with same spectrum", however, still exists. The influence of external environment on vegetation growth is very significant, and the different vegetation growth of vegetation may reflect on the different vegetation canopy reflectance spectral curves, which refered to as "same material with different spectrum".2. The effective of the LSMM for Carex duriuscula cover estimation in thress experiments using both visible and near-infrared bands without considering ARE was poor, the corresponding RMSEs were 0.160 and 0.164, respectively. In order to reduce the influence of spectral similarity between the water and Carex duriuscula endmembers, the Carex duriuscula cover estimation accounting for ARE only applied near-infrared bands (750nm～900nm). Results showed that improved accuracy was found, with the RMSE dropped dramatically from 0.164 to 0.079, for more than 50%. Future research should emphasize the consideration of reflectance anisotropy as another source of intraclass variability.3. Under natural conditions, even if a region with the same vegetation type, the vegetation parameters within its scope will have large spatial heterogeneity, especially in the situation, where big differences in local environmental factor, such as soil, water and terrain condition can be found. So the assumption made in common researchs about the vegetation parameters, such as leaf area index and aboveground biomass in small area with small spatial heterogeneity is incorrent.4. There was no significant correlation between the aboveground dry and green biomass and the single waveband original hyperspectral reflectance; the best correlation coefficient R< 0.45, in C4 site, and R>-0.25 in C3 site, which was affected by the different vegetation communites within itself.5. The best correlation coefficient of four vegetaiton index (SRVI, NDVI, SAVI and EVI) based on original expression wavebands was lower than that based on all available wavebands. By using the multivariate linear stepwise regression analysis, NDVI and EVI were selected as model variables from four vegetation indexes in C3 site for ADB, with R2 and RMSE were 0.732 and 57.5 g/m2. As for the ADB and AGB in C4 site, as well as AGB in C3 site, SAVI and EVI were selected as model variables from four vegetation indexes, and the R2 were 0.676,0.693 and 0.769, and the RMSE were 65.8 g/m2,134 g/m2 and 184 g/m2.