Predicting Mountain Forest Parameters Using The Geometric-optical Model And Multi-source Data

As important forest parameters,the leaf area index(LAI),canopy closure(CC),forest height(h)and forest above-ground biomass(AGB)are indispensable for ecological process models and carbon cycle models.Therefore,the accurate estimations of regional or global scale forest parameters are of great significance for a deep understanding of inherent laws of environmental change.With the diversification of remote sensing technology,the single-source remote sensing data has been unable to meet the application demand of the region and high precision.Recently,a large amount of effort has been devoted to the joint utilization of multi-source remote sensing data for the estimation of regional forest parameters.The statistical technologies were adopted as the major methods for the multi-source remote sensing data,and the physical models were rarely used.The statistical analysis has a high requirement on the quantity and distribution of observations,and the structure of the models is varied,which has limited universality in the monitoring of regional forest parameters.The most physical models were based on the assumption of a flat surface,which ignored the influence of the sloping terrain on the observed reflectivity of the sensors,and failed to meet the application requirements under the complex topographical conditions.In addition,the mixed pixel is also one of the main problems in the inversion of forest parameters over mountainous regions.The model algorithms for currently existing global-scale surface parameter products,such as GLOBCARBON LAI,CYCLOPES LAI,assume that the surface is composed of a single vegetation type,which does not take into account the mixing characteristics of the pixels and then will bring great uncertainties to the mountain forests with high surface heterogeneity.In the sum,the regional application,topographic influence,and mixed pixel decomposition have become the three major scientific problems in the joint retrieval of the multi-source remote sensing data.In response to these three problems,this study has proposed methods for the prediction of the mountain forest height,the canopy closure,and the effective leaf area index(LAIe).Furthermore,the forest AGB model was constructed based on vegetation indices,topographic indices and these structure parameters with physical significance.The research includes the following three main aspects:(1)Predicting forest height using the GOST model and multisource remote sensing data for sloping terrainsA reliable method was developed for estimating regional mountain forest height,which could be applicable for forests on various sloping terrains.Based on the relationship between the four components of the GOST model and the canopy structural parameters,this method was designed to map high-quality forest height information from airborne LiDAR and multispectral data.Firstly,the Sequential Maximum Angle Convex Cone(SMACC)algorithm was used to generate image endmembers and corresponding abundances in Landsat imagery.Secondly,LiDAR-derived forest metrics,and topographical factors were used to calibrate and validate the GOST,which aimed to accurately decompose the SMACC mixed forest pixels into sunlit crown,sunlit background and shade components.Then,The endmembers extracted from the Landsat7 imagery were respectively combined into different subsets of the components.Finally,the forest height of the study area was retrieved based on a back-propagation neural network and a look-up table.The GOST showed good performance in forest height estimations of coniferous forest on sloping terrains,and the R~2 values were above 0.70.The study demonstrated the tremendous potential of the GOST model for quantitative mapping of forest height on sloping terrains with multispectral data and LiDAR inputs.(2)Predicting canopy closure and effective leaf area index using the Li-Strahler geometric-optical model and multisource remote sensing data.A reliable method was developed for estimating regional CC and LAIe.Based on the Li-Strahler geometric-optical model,this method was developed to solve the mixed pixel problem,and further to realize the prediction of regional CC and LAIe from airborne LiDAR and multispectral data.First,based on the airborne LiDAR-derived canopy height product,the CC and LAIe were estimated over airborne LiDAR coverage.Second,the illumination background component was calculated based on the simplified relationship with canopy gap,CC and LAIe.Then,the reflectance of illumination background was calculated based on the linear decomposition model.Finally,the forest CC and LAIe were estimated by using Li-Strahler geometric-optical model over the study area.Results showed that the retrieval method proposed in this thesis could be used effectively in the inversion of regional CC and LAIe.(3)Multi-parameter synergic retrieval of forest AGB.The regional forest AGB was predicted based on the forest height,CC and LAIe,optical remote sensing information and topographic information.Firstly,the linear model of forest AGB was constructed based on the forest height with physical meaning.Secondly,the multivariate linear regression models of forest AGB were constructed based on multiple parameters.Then,the nonlinear remote sensing retrieval model of forest AGB was constructed based on the machine learning algorithm of support vector regression(SVR).Finally,the abilities of the linear model,the multivariate linear regression model,and the SVR model were compared,and thus the regional forest AGB was retrieved by the selected optimal model.It was found that the models based on the linear model and multivariate linear model had a considerable ability to estimate the forest AGB;while the independent verification showed that these two models had the overestimation;the SVR machine learning algorithm had higher estimation accuracy than the linear model and multivariate linear model,and there was no overestimation,In summary,this research aimed at the large-area expansion and terrain problems in multi-source remote sensing data for estimating forest AGB,and proposed the effective methods for estimating forest parameters respectively.The results showed that the coupling methods could be effectively improved.The problems of estimating the forest height of the over-10-degree slope and decomposition of the mixed pixels were solved.The high-precision mapping of AGB over the study area was finally performed.The results of this thesis not only provided support for the completeness of the theoretical analysis for the GOST model,but also provided innovative inspirations for remote sensing of forest parameters in mountainous areas.