Research On Weak Sensitive Parameters Retrieval Using Vegetation Canopy Reflectance Model

From the 1970s, the researches on vegetation canopy reflectance model (CRM),such as SAIL (scattering by arbitrarily inclined leaves) and GEOSAIL CRMs were booming. Numerous classical and amazing CRMs were developed. The estimation of vegetation key variables based on these CRMs were then risen over the past 20 years,and severed in the fields of crop yield estimation, ecological environment protection,disaster assessment and early-warning, water resources management, etc. However, the progress for the CRMs development and their applications seems to be slow in recent years since no significant CRM was proposed over past 10 years. More specifically, the estimations of vegetation key variables were generally focused on these model sensitive parameters over and over again, ignoring the fact that there are many important but model weak sensitive parameters that can also be retrieved based on these classical CRMs, however seldom people care about them. The model sensitive parameters are indeed important but they cannot fully meet the requirements for the society, especially for a rapidly evolving society like today.In this thesis, the methodologies for the estimation of model weak sensitive parameters were explored based on previous studies on these model sensitive parameters retrievals. Specifically, two important variables, grassland aboveground biomass (AGB) and vegetation canopy fuel moisture content (or fuel moisture content,FMC) were estimated base the classical CRMs. These two variables are derived from the model weak sensitive parameters, dry matter content (DMC). AGB is key to our understanding of the terrestrial carbon balance and is also a key variable for crop yield estimation. One of the primary variables affecting ignition and spread of wildfire is FMC. The purpose of estimating these two variables is to set up the methodologies for model weak sensitive parameters estimation and to extend the application range of CRMs, and further serves the country in the fields of crop yield estimation, ecological security of large scale, wildfire risk assessment and early-warning, global climate change, etc. This thesis includes following five aspects.(1) The strategies for CRM weak sensitive parameters retrieval were presented based on the previous studies on model sensitive parameters retrieval. We pointed out that the only way to retrieve model weak sensitive parameters was to enhance their sensitivity. For that purpose, we proposed five strategies, estimation of model weak sensitive parameters using hyperspectral remotely sensed data, parameterize the model sensitive parameters and then enhance the sensitivity of model weak sensitive parameters, the objective-based method for model weak sensitive parameters retrieval,retrieval of model weak sensitive parameters in a long time-series, retrieval of model weak sensitive parameters based on field measurements.(2) The estimation of key vegetation variables using a CRM is generally hampered by the ill-posed inverse problem which will largely decrease the accuracy level of retrieved variables of interest. In this study, we focused on alleviating the ill-posed inverse problem based on the Bayesian network algorithm to allow the improvement of the LAI and CWC retrieval. Previous studies generally treated the free pareamters while igonored the fact that these parameters were naturally correlated. The correlations that naturally existed between the model parameters were introduced into their prior joint probability distribution (PJPD) of the free parameters that was needed to build the Bayesian network. This treatment allowed the reduction of the probabilities of unrealistic combinations that may confuse the retrieval process, and then increase the accuracy level of retrieved LAI and CWC. The methodology used in this study for alleviating the ill-posed inverse problem will also be used in the next sections.(3) This study presents a novel method to derive grassland AGB based on the PROSAILH radiative transfer model. Two variables, leaf area index (LAI, m2m-2,defined as a one-side leaf area per unit of horizontal ground area) and dry matter content(DMC, gcm-2, defined as the dry matter per leaf area), were retrieved using PROSAILH.The result of LAI × DMC was regarded as the estimated grassland AGB according to their definitions. The well-known ill-posed inversion problem when inverting PROSAILH was alleviated using ecological criteria to constrain the simulation scenario and therefore the number of simulated spectra. A case study of the presented method was applied to a plateau grassland in China to estimate its AGB. The results were compared to those obtained using an exponential regression, a partial least square regression (PLSR) and an artificial neural networks (ANN). The RTM-based method offered higher accuracy than the exponential regression and the ANN. However, the proposed method offered similar performance than PLSR as presented better determination coefficient than PLSR but higher RMSE. Although it is still necessary to test these methodologies in other areas, the RTM-based method offers greater robustness and reproducibility to estimate grassland AGB at large scale without the need to collect field measurements and therefore is considered the most promising methodology.(4) The methods for grassland and forest FMC retrievals were developed. This contains two parts: (1) Estimation of grassland FMC by parameterizing CRM with interval estimated LAI. FMC is an important factor in wildfire risk assessment, but the estimation of LFMC is affected by the LAI if it is retrieved using a CRM. To reduce the influence of LAI in the FMC retrieval process, we used the interval estimated LAI which was estimated using a simple downscaling technique from the MODIS LAI product. The interval estimated LAI was then used to parameterize the PRO SAIL CRM,with the purpose to decrease the influence from the LAI and thereby increase the accuracy of estimated FMC. (2) Estimation of forest FMC using coupled CRMs. This study focused on the retrieval of FMC of forest with two-layered canopy (lower grass canopy and upper tree canopy) by coupling two CRMs: PROSAIL and PROGEOSAIL.The spectra of lower grass canopy were firstly simulated by the PROSAIL model, and then the soil spectra required in the PROGEOSAIL model was replaced by the simulated spectra of lower grass canopy layer. These coupled models allowed to better resemble the two-layered canopy configuration in the study area. The results showed that the accuracy level of retrieved FMC using the proposed methodology was better than that when the PROGEOSAIL model used alone.(5) This thesis developed a preliminary continental scale flammability monitoring system. The methodologies for canopy FMC estimation were applied in Australia and Lu mountain in China for mapping its FMC from the years 2001 to 2015, and the FMC products were therefore generated. To assess the performance of these products, the Logistic model was used to set the relations between FMC and historical wildfires from MODIS burned area product, and then calculated the flammability index (FI) products.Three historical big wildfires in Australia (Canberra fire in 2003, Victoria fire in 2009 and New South Wales fire in 2013) were used to test the FI products. The results showed a good performance of the FMC and FI products in detecting the wildfire risk in early time. Thus, a preliminary continental scale flammability monitoring system was built.