Adaptability Analysis Of Models For Vegetation Canopy Fuel Moisture Content Retrieval From Remote Sensing

In recent years,wildfires have occurred more frequently throughout the world with global warming,which not only pose great damage to the regional ecosystem,release much greenhouse and contaminating gases,aggravate the greenhouse effect,but also threaten people's life and properties.Consequently,high-precision,near-real-time and large-scale wildfire risk assessment and early warning are significantly essential.Vegetation canopy fuel moisture content(CFMC)is considered as the critical factor for wildfire risk assessment and early warning.Due to the optical satellites providing large-scale,near real-time and multi-band remotely sensed imagery,CFMC retrieval based on optical remotely sensed data has gradually been favored by related researchers in the field of wildfire science.Among the methodologies,the inversion method based on radiative transfer models(RTMs)performes more universal than the conventional empirical method.However,due to the differences in the structural complexity of different vegetation types,the applicability of different physical models is limited.Moreover,there are very limited researches investigating the model applicability for CFMC retrieval.To this end,this study aims to explore the applicability of different physical models to CFMC inversion for different vegetation types.Firstly,as for shrubland CFMC retrieval,previous researches consistently applied PROSAILH model based on the assumption that shrubland is uniformly and continuously distributed.No study,however,took the different structure and distribution between closed and open shrubland into consideration,which induces many uncertainties into CFMC retrieval.Secondly,in terms of coniferous forest CFMC retrieval,PROSPECT model was used,despite which is designed to model the spectra of broad leaf.Therefore,the suitability of PROSPECT for needle leave should be investigated.Moreover,as the critical factor of wildfire risk assessment,the are still confusing.This thesis uses CFMC field measurements and the products of Moderate Resolution Imaging Spectroradiometer(MODIS)to analyze the models'suitability for shrubland and coniferous forest CFMC retrieval.Besides,we also analyze the spatiotemporal pattern of CFMC in the regions with frequent wildfires and its indicative effect on the wildfire occurrence based on global CFMC product.This dissertation provides suggestions of the models selection for different vegetation types and canopies with complex and diverse structure,as well clarifies the importance of developing a fire risk assessment and early warning system which integrates the remote sensing-based CFMC product.The main work are summarized as follows:(1)We firstly evaluated the suitability of multiple models for closed and open shrubland CFMC retrieval.The 14 shrubland sites extracted from the National Fuel Moisture Database(NFMD)were used for the validation of models'performance,and these sites were separated into closed and open shrubland based on the International Geosphere-Biosphere Programme(IGBP)classification scheme.MODIS reflectance product MCD43A4.006 was used as the observational source.The multiple models we evaluated included three radiative transfer models(RTMs),i.e.PROSAILH,PROGeo Sail and PROACRM,and two empirical models,i.e.vegetation index-based(VI-based)and relative vegetation index-based(VI_max-min-based).Meanwhile,we analysed the structure and distribution of closed and open shrubland based on MODI vegetation continuous fields product,which was combined with the principle of models construction to analyse the suitability of multiple models.The results recommended the PROACRM RTM for closed shrubland and the VI_max-min-based model for open shrubland.(2)This thesis also evaluated the performance of PROSPECT RTM to model the spectra of needle leave and retrieve coniferous forest CFMC at both leave and canopy scale.At the leave level,we applied the unbiased data created by LIBERTY model,which was considered as the true values of needle leaf,to assess the performance of PROSPECT RTM.The results showed that PROSPECT performed well in modeling the spectra of needle leaf and retrieving the chlorophyll content and water content with R² of 0.98 and0.99,respectively.At the canopy level,PROSPECT was combined with Geo Sail and successfully retrieved coniferous forest CFMC and captured the CFMC time series variability(R²=0.57,RMSE=13.81%).However,LIBERTY model should be improved to further describe the effect of dry matter content on the leaf absorption properties.(3)Finally,we took the 2018 California fires,the 2019 Australian bushfires and the Amazon fires as examples and analyzed the spatiotemporal pattern of forest,shrubland and grassland CFMC for each region,including seasonal variability,inter-annual variability and the spatiotemporal pattern in the fire year,and further the indicative effect of these variation properties on the wildfire occurrence,based on the 2001?2019 global CFMC product.The results showed that CFMC of each region and each vegetation type in the fire year almost presented lower than the average level of 19 years.CFMC presented significant indicative effect on the fire occurrence particularly for the grassland and shrubland in California and Australia.However,such a significant indicative effect cannot be found in Amazon rainforest.This can be explained by that rainforest CFMC presents weaker seasonal and inter-annual variability duo to the stronger physiological regulation ability to regulate the water status,and most of the fires in Amazon were caused by the unreasonable mankind activities.In addition,by comparing the burn areas with CFMC variability,it can be concluded that wildfires tend to break out when CFMC continues to drop to a certain threshold.Moreover,the results also showed that CFMC can distinguish between fire and non-fire seasons.