Research On Monitoring And Forecasting Method Of Large-scale Forest Surface Dead Fuel Moisture Content

Forest fires not only emit greenhouse gases to aggravate global change,but also directly threaten human life and property safety.The key factor affecting the occurrence and development of forest fire,forest surface Dead Fuel Moisture Content（DFMC）,the ratio of surface dead fuel moisture content to dry matter mass,is an important input for forest fire risk assessment and fire behavior prediction system.The monitoring and forecasting of forest surface DFMC with large scale and high precision is very important for forest fire risk control and scientific fire rescue decisions.The traditional methods based on field surveys and weather station estimation of DFMC are difficult to be applied at large scales.Remote sensing technology can provide large-scale meteorological variables such as air temperature(T_air)and relative humidity（RH）for DFMC estimation model,and is promising for DFMC monitoring and forecasting of forest surface at large scales.Existing studies mainly use satellite remote sensing data to estimate meteorological variables and input empirical models to estimate DFMC,which relies on a large number of field data and cannot easily be applied in new areas.Although the process-based model with the water and energy balance processes is more robust than the empirical model,the initialization is unclear for large scales,which makes it difficult to apply in large scales,and there is currently a lack of research on the estimation of surface DFMC by combining satellite remote sensing data and process-based model.Meanwhile,satellite data is susceptible to cloud,resulting in a lack of estimated meteorological variables,thus affecting the spatial continuity of DFMC estimation.In addition,forest fire risk and fire behavior prediction require DFMC for a period of time in the future.Existing studies mainly combine meteorological forecast data with low spatial resolution and empirical models to make DFMC forecast,resulting in low spatial resolution and low accuracy.Therefore,how to combine satellite data with high spatial resolution,robust process-based model and meteorological forecast data to achieve high-precision large-scale forest surface DFMC forecast is extremely important for forest fire risk and fire behavior prediction.To address these limitations,this dissertation focused on large-scale forest surface DFMC monitoring and forecasting methods,the main work and achievements are summarized as follows:（1）A practical model initialization scheme and model parallelization method were proposed.Due to the lack of analysis on the impact of initial values on process models in existing studies,it is difficult to determine the appropriate initialization scheme for large-scale application,and then the time required for model come to equilibrium cannot be determined,making it difficult for process-based models to be applied at large scales.This dissertation analyzed the iterative characteristics of the Fuel Stick Moisture Model（FSMM）by giving a set of initial values,and determined the time（warm up time）required for the model coming to equilibrium under different initial values.Based on the warm up time,the initial value of the model can be taken in a preset range.Based on the warm up time,the long-time-series iterative process of the FSMM was divided into several parallel short-time tasks,and then the FSMM was optimized into a parallel mode（PaFSMM）,and the PaFSMM was used to estimate the DFMC at large scales.Results show that compared with FSMM,the estimation efficiency of PaFSMM was greatly improved with time is reduced to 0.3%of the original model.（2）A method of coupling PaFSMM and Long Short-Term Memory（LSTM）to estimate long time series DFMC was proposed.Some of the physical processes in the process-based model are very complex,and the numerical simulation can not fully characterize these complex processes,which makes the process-based model insufficient in mining long-term sequence information.In order to further improve the performance of the process-based model,this dissertation coupled PaFSMM and LSTM,and used LSTM’s ability to mine time series information to correct the estimation results of long time series DFMC output by PaFSMM.The results show that LSTM can effectively improve the estimation accuracy of PaFSMM for long time series DFMC.（3）A large-scale forest surface DFMC monitoring method based on the coupling of geostationary meteorological satellite data and PaFSMM model was proposed.The process-based model requires multiple meteorological variables and some of them are not sensitive.Estimating each meteorological variable based on satellite data is time-consuming and unnecessary.To address this limitation,this dissertation firstly analyzed the sensitivity of meteorological factors required by the process-based model PaFSMM,and determined that the most significant factor affecting the change of DFMC is Relative humidity（RH）.Then,the RH estimation method based on geostationary satellite（KOMPSAT-2A,GK2A）data was explored.Finally,remotely sensed RH and PaFSMM were used to estimate forest surface DFMC at large scales.The results show that the method proposed in this dissertation can ensure the estimation accuracy of surface DFMC while reducing the estimation of meteorological variables from satellite data,and can effectively monitor large-scale forest surface DFMC（R²=0.75,RMSE=3.45%）.（4）A forest surface DFMC reconstruction method for cloud cover areas was proposed.Satellite data can be affected by cloud,resulting missing DFMC values under cloud cover.This dissertation firstly determined the relationship between clear sky pixel Air Temperature(T_air)and all-weather variables,including profile temperature,T_AS,and Solar Radiation（SR）,using GK2A products.Then,this relationship was applied to cloud cover areas to estimate the cloud affected T_air,and the reconstruction method of cloud affected RH was explored by combining the specific humidity data from GK2A.Finally,combined with the reconstructed RH and other meteorological variables provided by Global forecast system,GFS,the PaFSMM was used to estimate large-scale spatial continuous forest surface DFMC.The results show that the DFMC under cloud cover can be reconstructed effectively by using the reconstructed RH and process-based model,PaFSMM.（5）A large-scale forest surface DFMC forecast method based on the integration of geostationary satellite data and weather forecast data was developed.Previous studies used empirical models and low-resolution meteorological forecast data to forecast future DFMC.In this dissertation,the Weather Research and Forecasting Model（WRF）was used to dynamically downscale the RH provided by the coarse-resolution meteorological forecasting data,GFS.Then the RH estimated by GK2A was taken as the observation data to assimilate the RH downscaled.The assimilated RH and other factors provided by GFS are finally input into the PaFSMM to forecast DFMC in the next three hours.The results show that the use of WRF downscaling can significantly improve the forecast accuracy of DFMC,and the assimilation of satellite data can further improve the forecast accuracy of DFMC,which can provide support for forest fire risk and fire spread trend prediction.