A method for simulating large-scale forest fire spread based on the FARSITE model and the WRF-SFIRE model

Abstract

This invention discloses a large-scale forest fire spread simulation method based on the FARSITE and WRF-SFIRE models, relating to the field of forest fire early warning technology. First, this invention uses a non-dominated sorting genetic optimization algorithm, aiming to determine the optimal combustible material model parameters and fire spread simulation results by ensuring consistency between the fire combustion profile and fire radiation intensity simulated by the FARSITE model and satellite observations. Then, the optimal combustible material model parameters are used as the combustible material model parameters in the WRF-SFIRE model, and the fire spread results simulated by the FARSITE model are used as initial conditions to drive the WRF-SFIRE model to simulate the forest fire spread situation over the next 6 hours, thereby obtaining reliable large-scale forest fire spread simulation results. The forest fire spread simulation optimization method described in this invention is simple to operate and comprehensively considers the advantages of both the FARSITE and WRF-FIRE models, enabling effective and rapid application to large-scale forest fire spread simulation, which is of great significance for precise forest fire prevention and control.

Description

Technical Field

[0001] This invention relates to the field of forest fire early warning technology, and in particular to a multi-objective optimization method for large-scale forest fire spread simulation using collaborative satellite observation. Background Technology

[0002] Using fire behavior models to simulate forest fire spread not only allows for timely assessment of forest fire development trends to better support fire rescue decision-making, but also serves as a crucial means of predicting potential forest fire behavior characteristics (such as fire spread rate and intensity) and extreme fire behaviors (such as crown fire and flying fire) at the landscape scale. Accurate simulation of forest fire spread is essential for clarifying the fire occurrence and development process, and also helps build a reliable forest fire behavior case library to improve the accuracy of forest fire risk assessment and support precise forest fire prevention and control.

[0003] Combustible load and moisture content are key factors influencing forest fire development and have been incorporated into combustible model parameters for use in fire behavior models such as FARSITE and WRF-SFIRE to simulate fire spread. Currently, commonly used combustible models fall into two categories: Anderson combustible models (13 types) and Scott and Burgan combustible models (43 types), the latter offering more detailed classifications of combustible types. These standard combustible models correspond to a set of combustible parameters or properties, but their default values ​​may not be applicable to southwestern China. Therefore, combining field combustible surveys can help determine initial combustible model parameters and improve the reliability of forest fire spread simulation results as much as possible. Regarding the accuracy evaluation of fire spread simulation, most studies only consider the consistency between simulated fire spread profiles based on satellite-observed fire combustion profile analysis. This accuracy assessment method based on combustion profiles is easily affected by human factors, meaning that satellite-observed fire spread patterns may not reflect fire spread under natural conditions. In addition to the fire's combustion profile, the thermal radiation released during fire spread is also noteworthy and closely related to the amount of combustible material. Satellite remote sensing technology has the capability to monitor fire radiation intensity over a wide area with high precision, providing a new reference for assessing the accuracy of forest fire spread simulations and can be used to optimize combustible material model parameters.

[0004] Furthermore, while the FARSITE model is well-suited for simulating mesoscale forest fire spread, its application to large-scale fires is hampered by the significant amounts of water vapor and heat released during combustion, which can alter the local atmospheric field and increase the uncertainty in fire spread simulation. The WRF-SFIRE model, which considers atmosphere-fire coupling effects, is suitable for large-scale forest fire spread simulation, but its fire behavior modules are relatively simple. Combining the advantages of both models, the fire spread simulation results from the FARSITE model can be used to drive the WRF-SFIRE model in forest fire spread simulations, yielding more reliable large-scale forest fire spread simulation results. This can provide crucial information and decision support for the precise prevention and control of major and high-intensity forest fires. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a multi-objective optimization method for large-scale forest fire spread simulation based on collaborative satellite observations, which is highly scalable and computationally simple.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model, comprising the following steps:

[0007] Step 1: Determine the fire area and collect the parameters required for the FARSITE model and WRF-SFIRE model to simulate the spread of forest fires. The input parameters include air temperature, relative humidity, wind speed and direction, rainfall, elevation, slope, aspect, forest canopy structure parameters, and combustible material model.

[0008] Step 2: Multi-objective optimization of wildfire spread simulation based on the FARSITE model;

[0009] Step 2.1: Parameterization of the FARSITE model;

[0010] Based on the GLC_FCS30 land cover product, the combustible model of the target area is determined; the global sensitivity of each parameter of the combustible model is determined using the Sobol algorithm; based on the forest stand and combustible information survey data, the range of sensitive parameters is set, and parameters that are weakly sensitive or insensitive are set as default values.

[0011] Step 2.2: Extraction of forest fire behavior features based on satellite observations;

[0012] Based on the Global Fire Atlas dataset, the initial fire location was determined and the combustion profile information of the fire was extracted. The SC coefficient was used to evaluate the spatial simulation accuracy of forest fire spread. The SC coefficient was calculated using Formula 1. Based on the VIIRSNRT active fire point product, the fire radiation intensity characteristics of the combustion zone were extracted. The relative root mean square error (rRMSE) was used to evaluate the simulation accuracy of fire behavior characteristics during forest fire spread. The relative root mean square error (rRMSE) was calculated using Formula 2.

[0013]

[0014] Among them, S obs S represents the area of ​​the burned area of ​​a fire extracted from the GFA dataset. s S represents the area of ​​the fire-burning zone simulated by the FARSITE model. o This represents the area of ​​the overlapping portion of the fire-burning areas;

[0015]

[0016] in, and , respectively, represent the fire radiation intensity simulated by the FARSITE model and observed by satellite for the i-th pixel, and n is the number of satellite fire pixels falling within the combustion profile simulated by the FARSITE model;

[0017] Step 2.3: Optimize sensitive parameters;

[0018] The accuracy of forest fire spread simulation under specific parameters was evaluated using SC coefficient and rRMSE. Based on non-dominated sorting genetic algorithm II, the population size was set to m and the maximum number of generations was set to n. The sensitivity parameters of the FARSITE model were optimized to obtain the optimal results of forest fire spread simulation with high SC coefficient and low rRMSE.

[0019] The SC of the optimal forest fire spread simulation results opt coefficients and rRMSE opt It should satisfy formula (3);

[0020]

[0021] Where, d i To consider the SC coefficient and rRMSE as a comprehensive evaluation index during the i-th forest fire spread;

[0022] Step 3: Simulate large-scale forest fire spread based on the optimized WRF-SFIRE model;

[0023] Step 3.1: Parameterization of the WRF-SFIRE model;

[0024] Based on the GLC_FCS30 land cover product, the combustible models of the target area are divided into two categories: grassland and forest. The optimal combustible model parameters calibrated in step 2.3 are used as the combustible model parameters of WRF-SFIRE. The domain of the WRF-SFIRE model is set to a 4-layer nesting, with the spatial resolution of the fourth atmospheric grid being 300 meters and the spatial resolution of the fire grid being 30 meters.

[0025] Step 3.2: Simulate large-scale forest fire spread based on the WRF-SFIRE model;

[0026] The optimal forest fire spread simulation results obtained based on the FARSITE model are used as the initial fire location information for the WRF-SFIRE model. The WRF-SFIRE model is used to carry out large-scale forest fire spread simulation and output forest fire behavior characteristics such as fire spread rate and fire intensity. To ensure the reliability of the forest fire spread simulation, the above process is repeated every hour.

[0027] Further, the sensitive parameters mentioned in step 2.1 are the surface dead combustible moisture content and load, the moisture content and load of live herbaceous and woody combustibles, and the thickness of the combustible bed, while the insensitive parameters are the combustible surface area to volume ratio, the calorific value of the combustible, and the combustible extinguishing moisture content.

[0028] Furthermore, the initial fire scene pixel mentioned in step 2.2 should satisfy the condition that its burning date is the smallest relative to other fire scene pixels, and the start and end times of the fire scene are determined by the earliest and latest times of the VIIRS NRT fire point pixels.

[0029] Furthermore, the confidence level of the VIIRS NRT fire point pixel applied to the rRMSE calculation in step 2.2 should be "h" or "n".

[0030] Furthermore, m and n mentioned in step 2.3 are 30 and 50, respectively;

[0031] Furthermore, h in step 3.2 is 6.

[0032] The beneficial effects of this invention are as follows: The multi-objective optimization method for simulating large-scale forest fire spread described in this invention is simple to operate. First, it improves the accuracy of spatial simulation of forest fire spread and prediction of fire behavior characteristics simultaneously by utilizing multi-objective optimization of model parameters. Then, based on the coupling of two fire spread models, it further improves the reliability of large-scale forest fire spread simulation. This method can provide key information support such as fire situation for precise prevention and control of forest fires. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall method flow of the present invention.

[0034] Figure 2 This is a location map of a forest fire case in a specific embodiment of the present invention.

[0035] Figure 3 The results are based on the sensitivity analysis of the combustible material model parameters in a specific embodiment of the present invention.

[0036] Figure 4 This is the optimal simulation result of forest fire spread using the FARSITE model in a specific implementation scheme of the present invention.

[0037] Figure 5 The results of large-scale forest fire spread simulation using the WRF-SFIRE model are shown in a specific implementation scheme of this invention. Detailed Implementation

[0038] The following description, in conjunction with specific embodiments and accompanying drawings, further illustrates the multi-objective optimization method for large-scale forest fire spread simulation based on collaborative satellite observations provided by this invention:

[0039] A multi-objective optimization method for simulating large-scale forest fire spread using collaborative satellite observations, such as Figure 1 As shown, it includes the following steps:

[0040] Step 1: Prepare input data for the fire behavior model.

[0041] Select the forest fire in Qiaodi Township, Xichang City, Sichuan Province (April 20, 2014) (see...) Figure 2 Conduct case analysis.

[0042] Specifically, it includes two parts:

[0043] Step 1.1: Preparation of input data for the FARSITE model (see...) Figure 1 Hourly air temperature, relative humidity, wind speed and direction, and precipitation parameters were extracted from ERA5-Land meteorological reanalysis data; elevation, slope, and aspect parameters were extracted from GDEMV3 products; forest canopy height was extracted from global forest canopy height products published by Peter Potapov; canopy cover was extracted from GFCC products; canopy height under branches was obtained by fitting canopy height and leaf area index; canopy bulk density was calculated by combining canopy combustible load and canopy height under branches; the spatial resolution of all parameters was resampled to 30 meters.

[0044] Step 1.2: Preparation of input data for the WRF-SFIRE model. The raw meteorological data used were ERA5 reanalysis data, with a temporal resolution of 1 hour and a spatial resolution of 0.25°. The main meteorological elements used were those from each pressure layer, as well as surface meteorological elements and soil moisture content parameters (see...). Figure 1 The required elevation (DEM) data is extracted from the GDEMV3 product (see...). Figure 1The required geographical data, such as soil type, can be downloaded from https: / / www2.mmm.ucar.edu / wrf / src / wps_files.

[0045] Step 2: Multi-objective optimization of wildfire spread simulation based on the FARSITE model;

[0046] Specifically, it includes three parts:

[0047] Step 2.1: FARSITE Model Parameterization. Based on the GLC_FCS30 land cover product, the combustible models of the study area were divided into grassland and forest categories. Using the Sobol algorithm, the sensitive parameters were determined as follows: 1-h / 10-h / 100-h surface dead combustible moisture content and load, live herbaceous and woody combustible moisture content and load, and combustible bed thickness. The insensitive parameters were combustible surface area to volume ratio, combustible calorific value, and combustible extinguishing moisture content (see...). Figure 3 );

[0048] Step 2.2: Extraction of Forest Fire Behavior Features Based on Satellite Observations. Fire combustion profile information is extracted from the Global Fire Atlas (GFA) dataset, and the initial fire location is determined. Based on the satellite-observed fire combustion profile information, the Sorensen Coefficient (SC, Equation 1) is calculated to assess the accuracy of the spatial simulation of forest fire spread. Based on the medium-to-high confidence VIIRS NRT active fire point information, the fire radiation intensity (RI, kW / km²) from satellite observations and model simulations is calculated. 2 The relative root mean square error (rRMSE, Equation 2) between the two is used to assess the accuracy of the simulation of fire behavior characteristics during the spread of forest fires.

[0049]

[0050] Among them, S obs The area of ​​the fire-burning zone (km²) extracted from the GFA dataset. 2 ), S s The fire burning area (km²) simulated by the FARSITE model 2 ), S o The area of ​​the overlapping part of the fire burning area (km²) 2 ).

[0051]

[0052] in, and , respectively, represent the fire radiation intensity simulated by the FARSITE model and observed by satellite for the i-th pixel, and n is the number of satellite fire pixels falling within the combustion profile simulated by the FARSITE model;

[0053] Step 2.3: Sensitivity parameter optimization. Based on the Non-Dominated Sorting Genetic Algorithm II (NSGA-II), the population size was set to 30 individuals and the maximum number of generations to 50. The SC coefficient and rRMSE were optimized simultaneously. The optimal sensitivity parameters for the FARSITE model and the corresponding forest fire spread simulation results were selected using formula (3). The results show that using the optimized combustible model parameters results in higher accuracy of forest fire spread simulation (see...). Figure 4 The accuracy of forest fire spread simulation using the standard Scott and Burgan combustible models (GR2 and TU1) is lower, with SC = 0.807 and rRMSE = 4.80%. Compared to these models, the accuracy is lower, with SC = 0.756 and rRMSE = 19.9%, when compared to the standard models for grassland (GR2) and forest (TU1). Although the standard combustible model can obtain fire combustion profiles consistent with satellite observations, it still presents significant uncertainties in predicting fire radiation intensity. This demonstrates that the present invention, by comprehensively considering both the SC coefficient and rRMSE index, can obtain more reliable forest fire spread simulation results.

[0054]

[0055] Where, d i To consider the SC coefficient and rRMSE as a comprehensive evaluation index during the i-th forest fire spread;

[0056] Step 3: Optimization of large-scale forest fire spread simulation based on the WRF-SFIRE model;

[0057] Specifically, it includes two parts:

[0058] Step 3.1: WRF-SFIRE Model Parameterization. Based on the GLC_FCS30 land cover product, the combustible models of the study area are divided into grassland and forest categories, and the optimal combustible model parameters calibrated in Step 2.3 are used as the combustible model parameters for WRF-SFIRE. The domain of the WRF-SFIRE model is set to a 4-layer nesting, with spatial resolutions of 27 km, 9 km, 3 km, and 300 m for the four atmospheric grids, and a spatial resolution of 30 m for the fire grid. The parameterization schemes for the main physical processes of the WRF-SFIRE model are as follows: microphysical processes – WSM3 scheme, longwave radiation – RRTM scheme, shortwave radiation – Dudhia scheme, near-surface layer – Monon-Obukhov scheme, land surface processes – Noah-MP land surface process scheme, planetary boundary layer – YSU scheme (layer 4 closed), cumulus parameterization – Kain Fritsch scheme (layers 3 and 4 closed).

[0059] Step 3.2: Forest Fire Spread Simulation Based on the WRF-SFIRE Model. The results of the FARSITE model simulation are used as the initial fire line and directly input into the WRF-SFIRE model to conduct a large-scale forest fire spread simulation over the next 6 hours. Using the fire spread results simulated by the FARSITE model at 18:00 on April 20th as the initial fire line, the WRF-SFIRE model is used to simulate the fire spread characteristics from 18:00 on April 20th to 00:00 on April 21st, such as... Figure 5 As shown, the results indicate that the simulation results of large-scale forest fire spread based on the WRF-SFIRE model in a short period of time are relatively reliable.

Claims

1. A method for simulating large-scale forest fire spread based on the FARSITE model and the WRF-SFIRE model, comprising the following steps: Step 1: Determine the fire area and collect the parameters required for the FARSITE model and WRF-SFIRE model to simulate the spread of forest fires. The input parameters include air temperature, relative humidity, wind speed and direction, rainfall, elevation, slope, aspect, forest canopy structure parameters, and combustible material model. Step 2: Multi-objective optimization of wildfire spread simulation based on the FARSITE model; Step 2.1: Parameterization of the FARSITE model; Based on the GLC_FCS30 land cover product, the combustible model of the target area is determined; the global sensitivity of each parameter of the combustible model is determined using the Sobol algorithm; based on the forest stand and combustible information survey data, the range of sensitive parameters is set, and parameters that are weakly sensitive or insensitive are set as default values. Step 2.2: Extraction of forest fire behavior features based on satellite observations; Based on the Global Fire Atlas dataset, the initial fire location was determined, and the combustion profile information of the fire was extracted. The SC coefficient was used to evaluate the spatial simulation accuracy of forest fire spread. The SC coefficient was calculated using Formula 1. Based on the VIIRSNRT active fire point products, the fire radiation intensity characteristics of the combustion zone were extracted. The relative root mean square error (rRMSE) was used to evaluate the simulation accuracy of fire behavior characteristics during forest fire spread. The relative root mean square error (rRMSE) was calculated using Formula 2. Among them, S obs S represents the area of ​​the burned area of ​​a fire extracted from the GFA dataset. s S represents the area of ​​the fire-burning zone simulated by the FARSITE model. o This represents the area of ​​the overlapping portion of the fire-burning areas; in, and , respectively, represent the fire radiation intensity simulated by the FARSITE model and observed by satellite for the i-th pixel, and n is the number of satellite fire pixels falling within the combustion profile simulated by the FARSITE model; Step 2.3: Optimize sensitive parameters; The accuracy of forest fire spread simulation under specific parameters was evaluated using SC coefficient and rRMSE. Based on non-dominated sorting genetic algorithm II, the population size was set to m and the maximum number of generations was set to n. The sensitivity parameters of the FARSITE model were optimized to obtain the optimal results of forest fire spread simulation with high SC coefficient and low rRMSE. The SC of the optimal forest fire spread simulation results opt coefficients and rRMSE opt It should satisfy formula (3); Where, d i To consider the SC coefficient and rRMSE as a comprehensive evaluation index during the i-th forest fire spread; Step 3: Simulate large-scale forest fire spread based on the optimized WRF-SFIRE model; Step 3.1: Parameterization of the WRF-SFIRE model; Based on the GLC_FCS30 land cover product, the combustible models of the target area are divided into two categories: grassland and forest. The optimal combustible model parameters calibrated in step 2.3 are used as the combustible model parameters of WRF-SFIRE. The domain of the WRF-SFIRE model is set to a 4-layer nesting, with the spatial resolution of the fourth atmospheric grid being 300 meters and the spatial resolution of the fire grid being 30 meters. Step 3.2: Simulate large-scale forest fire spread based on the WRF-SFIRE model; The optimal forest fire spread simulation results obtained based on the FARSITE model are used as the initial fire location information for the WRF-SFIRE model. The WRF-SFIRE model is used to carry out large-scale forest fire spread simulation and output forest fire behavior characteristics such as fire spread rate and fire intensity. To ensure the reliability of the forest fire spread simulation, the above process is repeated every hour.

2. The large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model as described in claim 1, characterized in that, The sensitive parameters mentioned in step 2.1 are the surface dead combustible moisture content and load, the moisture content and load of live herbaceous and woody combustibles, and the thickness of the combustible bed. The insensitive parameters are the combustible surface area to volume ratio, the calorific value of the combustible, and the combustible extinguishing moisture content.

3. The large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model as described in claim 1, characterized in that, The initial fire cell mentioned in step 2.2 should satisfy the condition that its burning date is the smallest relative to other fire cells. The start and end times of the fire are determined by the earliest and latest times of the VIIRS NRT fire point cell.

4. The large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model as described in claim 1, characterized in that, The confidence level of the VIIRS NRT fire point pixels applied to the rRMSE calculation in step 2.2 should be "h" or "n".

5. The large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model as described in claim 1, characterized in that, In step 2.3, m and n are 30 and 50, respectively.

6. The large-scale forest fire spread simulation method based on the FARSITE model and the WRF-SFIRE model as described in claim 1, characterized in that, The h mentioned in step 3.2 is 6.

Images