Research On Remote Sensing Retrieval Method Of Forest Fuel Load

Forest fire is a natural disturbance phenomenon that exists in ecosystems and is very important for its formation,development and succession.However,with global climate change and urban expansion,the frequency,area and intensity of forest fires have shown a significant upward trend.It has brought more negative impacts and even disasters to mankind.The forest fuel load(FL)represents the total dry matter weight of all vegetation within a unit area that is susceptible to combustion.It is the main material basis and energy source for forest fire burning.Forest live fuel load(Live FL)and dead fuel load(DFL)affect the occurrence and development of canopy fires and surface fires respectively.Notably,fine fuels such as foliage fuel load(FFL),canopy active fuel load(CAFL),and surface litter fuel load(LFL)are often the initial contributors to combustion,potentially leading to extreme fire behaviors like spot fires.Their spatial distribution affects the spread and intensity of forest fires.They are key parameters for modeling potential forest fire behavior and carbon emission estimates.The implementation of largescale,spatiotemporal dynamic monitoring offers comprehensive data support and decision-making assistance for forest fire managers,greatly enhancing the practical significance of forest fire prevention and control efforts.However,the traditional fuel load monitoring method based on ground surveys has problems such as small monitoring range,high cost,and low efficiency.Remote sensing technology has the potential for wide-ranging dynamic monitoring of fuel loads,but currently relies mainly on optical remote sensing data to construct empirical statistical methods.There is a lack of research into the collaborative modeling of multi-source remote sensing data,such as Synthetic Aperture Radar(SAR),and its joint characterization of the spatial distribution characteristics of fuel loads,which makes it difficult to accurately depict the complex nonlinear relationship between remote sensing signals and fuel loads.Additionally,in the scattering signals generated within complex forest structures,there is a lack of research focused on extracting characteristic information for key categories of fuel load.This deficiency makes it challenging to construct an effective fuel load inversion model that can accurately simulate the scattering mechanisms within forests.Moreover,because of the forest canopy cover,deriving surface dead fuel loads from remote sensing data continues to be a challenging bottleneck.To address the aforementioned limitations,this dissertation focuses on research on remote sensing retrieval method of forest fuel load,and the main works are summarized as follows:(1)The applicability of multi-source remote sensing data in multi-category fuel load inversion was analyzed.This dissertation explores the ability of multi-band SAR and multi-spectral optical data to characterize FFL,CAFL,branch fuel load,stem fuel load and DFL.Univariate and multivariate collaborative analyzes were conducted based on correlation analysis and partial least squares regression respectively.Results show that Lband SAR data,especially HV polarization,has a strong positive correlation with the Live FL.Single type data has limited ability to represent fuel load,and the collaboration of active and passive remote sensing data can achieve better complementarity,but all remote sensing data are difficult to characterize DFL.These conclusions provide important reference guidance information for subsequent targeted research on the inversion of key categories of fuel loads.(2)A live fuel load(Live FL)inversion model considering the spatial distribution characteristics of forests was developed.The widely used water cloud model has limitations in retrieving Live FL as it does not fully incorporate vegetation coverage information.This dissertation coupled optical remote sensing data and water cloud model based on the strong ability of optical remote sensing data to characterize coverage.Results demonstrate that the coupled model outperforms standalone models,particularly when incorporating the green band,shortwave infrared band,and vegetation index.These additions significantly enhance the accuracy of Live FL inversion.This method mitigates the saturation issue in dense forest areas by supplementing the coverage information,providing a more accurate depiction of the complex nonlinear relationship between remote sensing signals and fuel load.(3)A foliage fuel load(FFL)inversion method based on spatiotemporal feature mining of multi-source data is proposed.In view of the application limitations of current FFL inversion research based on optical remote sensing in cloudy,rainy and foggy areas,this dissertation constructed an FFL inversion model that collaborates with active and passive remote sensing data and their spatiotemporal features.First,the time series data of Sentinel-1 and Landsat 8 OLI from 2014 to 2021 are reconstructed based on the harmonic model,and the time series features are extracted.Secondly,the spatial texture features are extracted based on the gray level co-occurrence matrix.The potential of spatiotemporal features in FFL inversion was explored by setting up multiple random forest independent comparison experiments.Results show that the addition of spatiotemporal features effectively improves the retrieval effect of active and passive remote sensing on FFL.In addition,this method can be used as an effective alternative for fuel load estimation in cloudy,rainy and foggy areas.(4)A canopy active fuel load(CAFL)inversion model based on scattering signal decomposition and vegetation scattering model was constructed.Canopy active fuel include leaves and twigs with a diameter less than 0.6 cm,which are key factors affecting the occurrence and development of canopy fires.Starting from the scattering mechanism,this dissertation first uses the H/alpha polarization decomposition method to analyze the scattering mechanism of the forest in the study area.The classic three-component polarization decomposition method was used to extract the volume scattering information in the Radarsat-2 C-band full polarization data to characterize the distribution of the main scatterers,namely leaves and twigs.A semi-empirical model based on vegetation scattering model that directly simulates forest canopy scattering was constructed to eliminate the interference of surface scattering and secondary scattering.Results show that this method can achieve effective inversion of CAFL by decomposing the microwave scattering signal of the complex structure inside the forest and modeling the canopy scattering mechanism.(5)A litter fuel load(LFL)inversion method taking into account ecological processes was proposed.In view of the bottleneck problem of remote sensing data being difficult to directly represent surface fuel information,this dissertation starts from the ecological process of the accumulation of dead fuels and breaks down it into two processes: accumulation and decomposition.First,the change detection algorithm was used to identify undisturbed forests and invert the FFL from 2000 to 2021 to explore its temporal phenological change characteristics.The cross difference between adjacent years was used to represent the amount of leaf litter in that year,that is,the accumulation of dead fuels.Secondly,the annual decomposition status of dead fuels is characterized through a litter decomposition model and a decomposition rate calculated based on meteorological data.Finally,the above two processes are iterated on a yearly basis and organically combined.Results show that this method can effectively invert the large-scale spatial distribution of LFL by disassembling the ecological formation process of LFL.It opens up a new idea for inverting the surface dead fuel load based on remote sensing technology,which can provide data support for forest fire risk assessment and early prevention and control.