FDS-based mountain fire spreading simulation method

Abstract

The invention provides an FDS-based mountain fire spreading simulation method, which is used for solving the problems of insufficient modeling precision of complex terrains, insufficient multi-parameter coupling, contradiction between calculation efficiency and precision and the like in the prior art. The method comprises the following steps: preprocessing topographic data of a simulation area, and extracting topographic features including a gradient and a slope direction; dividing grids for the simulation area, and dynamically adjusting the grid resolution according to the gradient; loading a meteorological condition dynamic coupling module, and carrying out dynamic simulation on fire spreading in combination with data in the forest combustible thermophysical property parameter database; the forest combustible thermophysical property parameter database comprises vegetation types corresponding to each grid in the simulation area, thermophysical property parameters corresponding to each vegetation type, density, height and water content; and the meteorological condition dynamic coupling module is configured to realize bidirectional coupling of a wind field and a combustion process by quantifying fire-to-wind effects and wind-to-fire effects, and dynamically update wind speed, wind direction and temperature boundary conditions.

Description

Technical Field

[0001] This invention belongs to the field of wildfire spread simulation technology. Background Technology

[0002] Wildfire spread simulation is of great significance for forest fire prevention, emergency response, and ecological protection. FireDynamics Simulator (FDS), as a professional fire simulation software, has been widely used in building fire research, but its direct application in wildfire spread simulation in complex terrain presents several challenges. 1. Difficulty in terrain modeling: Traditional FDS struggles to accurately represent the impact of complex mountainous terrain (such as cliffs and gullies) on fire intensity. 2. Complex multi-parameter coupling: Factors such as terrain slope, combustible material distribution, and meteorological conditions (wind speed, wind direction, temperature) interact, making accurate description difficult by traditional models. 3. Low computational efficiency: High-precision simulation of large-scale forest scenarios requires extremely high computational resources and suffers from poor real-time performance. 4. Lack of verification mechanisms: Existing methods for comparing and verifying simulation results with actual wildfire spread processes are limited.

[0003] Existing wildfire spread simulation technologies mainly include empirical models, physical models, and hybrid models. Empirical models, such as the Rothermel model, establish empirical formulas for fire spread rate and environmental parameters based on experimental data, but do not consider complex physical processes such as turbulence. Physical models, such as FDS and WFDS (Wildland-urban interface Fire Dynamics Simulator), solve fire processes based on computational fluid dynamics (CFD), but have limited ability to handle complex terrain and combustible material distribution. Hybrid models combine the advantages of empirical and physical models, but still suffer from problems such as difficulty in parameter calibration and low computational efficiency. Summary of the Invention

[0004] This invention proposes a wildfire spread simulation method based on FDS, aiming to solve the problems of insufficient accuracy in complex terrain modeling, insufficient multi-parameter coupling, and the contradiction between computational efficiency and accuracy in existing technologies.

[0005] A method for simulating wildfire spread based on FDS includes: preprocessing the topographic data of the simulation area to extract topographic features including slope and aspect; dividing the simulation area into grids and dynamically adjusting the grid resolution according to the slope: generating fine-grained grids when the slope exceeds a preset threshold and generating coarse-grained grids when the slope is below the threshold, forming a non-uniform hierarchical grid structure; loading a dynamic coupling module for meteorological conditions and combining data from a forest combustible thermal property parameter database to dynamically simulate fire spread; wherein, the forest combustible thermal property parameter database includes: the vegetation type corresponding to each grid in the simulation area, and the thermal property parameters, density, height, and water content corresponding to each vegetation type; the dynamic coupling module for meteorological conditions is configured to achieve bidirectional coupling between the wind field and the combustion process by quantifying the reaction of fire to wind, the resistance of fire to wind, and the effect of wind on fire, and dynamically updating the boundary conditions of wind speed, wind direction, and temperature.

[0006] In some examples, the dynamic coupling module for meteorological conditions includes: a meteorological data input unit configured to receive real-time data from a meteorological station or a meteorological field output from a WRF model; a terrain wind field correction unit configured to calculate local wind speed and direction changes caused by terrain based on fluid dynamics principles, and to correct the meteorological data acquired by the meteorological data input unit; a dynamic boundary condition adjustment unit configured to adjust boundary meteorological parameters in real time, with the values ​​dynamically updated according to time, fire development, or environmental changes; and a fire-wind interaction model, which quantifies the reaction of fire to wind, the resistance of fire to wind, and the effect of wind on fire to achieve bidirectional coupling between the wind field and the combustion process.

[0007] In some examples, the fire-wind interaction model includes: a fire plume model driven by thermal buoyancy that characterizes the reaction of fire to wind; a flame drag force model that characterizes the drag effect of fire on wind; and a wind-driven model that characterizes the effect of wind on fire.

[0008] In some examples, the terrain data preprocessing includes: acquiring terrain information using LiDAR or DEM; calculating local elevation gradients; and generating a slope matrix. S ( x , y ) and aspect matrix A ( x , y ).

[0009] In some examples, when the terrain slope is <10°, the grid size is 20-50m; when the terrain slope is 10°≤30°, the grid size is 10-20m; and when the terrain slope is ≥30°, the grid size is 5-10m.

[0010] In some examples, the vegetation types in the forest combustible thermal property parameter database are obtained by classifying remote sensing images.

[0011] In some examples, the dynamic boundary condition adjustment unit includes: a thermal convection coefficient correction model. h c ( x , y )= h c0 ×(1+ k × S ( x , y ) / 90°) In the formula: h c0 As the reference convection coefficient, k Slope influencing factors S ( x , y () represents the slope.

[0012] In some examples, the dynamic boundary condition adjustment unit further includes: a wind speed correction factor. f v :

[0013] A ( x , y ( ) represents the slope direction. θ The angle between the slope and the wind direction is used as a correction factor for the dynamic adjustment of the ventilation boundary in the FDS.

[0014] In some examples, distributed parallel computing is implemented based on the MPI framework.

[0015] In some examples, the data output by the FDS is parsed to extract the fire spread rate, thermal radiation intensity, and burned area, and then visualized.

[0016] This invention improves terrain modeling accuracy: through adaptive mesh technology, the representation error of complex terrain is reduced to within 5%, an improvement of approximately 30% compared to traditional methods. This invention enhances the accuracy of multi-parameter coupling: considering the vertical profile of wind speed and airflow changes caused by terrain, the wind speed field simulation error is reduced from 15% to 8%. This invention optimizes computational efficiency: through regional adaptive mesh and parallel computing, the simulation speed for large-scale scenarios (1km × 1km) is increased by 40%-60%. Compared with actual wildfire cases, the prediction error for fire spread rate is reduced from 25% to 12%, and the prediction error for burned area is reduced from 30% to 18%, significantly outperforming existing technologies. Attached Figure Description

[0017] Figure 1 This is a flowchart of a terrain preprocessing method according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of adaptive grid division for complex terrain in one embodiment of the present invention.

[0019] Figure 3 This is a diagram of the combustible material parameter database architecture in one embodiment of the present invention.

[0020] Figure 4 This is an overall architecture diagram of a wildfire spread simulation system based on FDS in one embodiment of the present invention.

[0021] Figure 5 This is a simulation effect diagram of one embodiment of the present invention. Detailed Implementation

[0022] 1. High-precision terrain slope modeling is achieved by using complex terrain preprocessing methods applicable to FDS. Figure 1 The process of terrain preprocessing method is demonstrated.

[0023] 1.1 Collect terrain data using LiDAR or high-precision DEM (Digital Elevation Model).

[0024] 1.2 Process the terrain data to generate slope and aspect distribution data, and identify key terrain features (such as ridges, valleys, and cliffs).

[0025] The slope is obtained by calculating the local elevation gradient using a sliding window (e.g., 3×3 size), as shown in the formula: (1) In the formula, S(x,y) represents the terrain slope matrix; and The elevation gradients in the x and y directions are calculated using the central difference method. (2) The slope aspect is defined as the direction of the projection of the slope normal onto the horizontal plane, and the formula is: (3) In the formula, A(x,y) is the slope aspect matrix, and the result range is [0, 360°], which is used to determine the orientation of the terrain (such as windward slope and leeward slope).

[0026] 1.3. Divide the grid density according to the slope S(x,y).

[0027] The classification criteria in the table below are for illustrative purposes only and can be adjusted according to actual circumstances.

[0028]

[0029] 1.4 Generate a mesh using an octree mesh generation algorithm based on the mesh density.

[0030] The core idea is to recursively divide the three-dimensional space, generating fine meshes in high-density areas and coarse meshes in low-density areas, thus forming a hierarchical mesh structure.

[0031] Algorithm principle: Based on the idea of ​​recursive segmentation in three-dimensional space, a grid density gradient field is established through terrain slope parameters. Fine-grained grids are generated in steep terrain areas and coarse-grained grids are generated in gentle slope areas, forming a hierarchical grid structure.

[0032] Algorithm flow: Initialization: Define the entire simulation region as the root node of an octree, and set the initial mesh size to [value missing]. L 0 (e.g., 100m, corresponding to the maximum segmentation depth of 4 layers), construct a terrain slope data index table.

[0033] Recursive partitioning: Calculate the average slope of the covered area for each node. S 平 ;like S 平 For angles ≥30°, perform an octree partitioning operation: generate 8 child nodes (each dimension is bisected), with a child node mesh size of [value missing]. L 0 / 2; Segmentation stops when the maximum number of segmentation layers is reached (e.g., 4 layers are set). If L If 0 is 100m and the maximum segmentation depth is 4 layers, then the minimum grid size is... L 0 / 16 = 6.25 m .

[0034] Mesh optimization: Construct an octree topology table with node coordinates (x, y, z), mesh size L, terrain density level (low density, low density, and high density), and parent / child node association index; perform boundary smoothing to eliminate mesh jump effects. Figure 2 A schematic diagram of adaptive grid partitioning for complex terrain is presented.

[0035] This invention breaks through the limitations of uniform grids in traditional FDS software. Through a mapping mechanism of "terrain features (terrain slope) → grid density", it achieves refined grid division and optimal resource allocation.

[0036] 2. Construct a database of thermal property parameters of forest combustibles to support fine-grained input settings.

[0037] Constructing a database of thermal properties of forest combustibles is a key step in achieving refined fire simulation. Figure 3The architecture of the combustible material parameter database was demonstrated. This process requires a complete technology chain, from data classification, parameter measurement, database construction to spatial distribution mapping, to support the input requirements of the FDS.

[0038] 2.1 Data Classification and Parameter Measurement: A combustible material classification system was established based on vegetation type (e.g., trees, shrubs, herbs), humidity level (e.g., dry, moderate, humid), and density level. Key thermophysical parameters matching the combustible material classification system, including heat release rate, ignition temperature, and combustion efficiency, were obtained using experimental equipment such as thermogravimetric analysis (TGA) and cone calorimeter.

[0039] 2.2 Database Construction and Format Adaptation: An XML-based parameter database was developed to structurally store key experimentally measured thermophysical parameters, and interface rules for the FDS input format were designed. The advantage of the XML format lies in its hierarchical structure, which allows for flexible expansion and facilitates integration with the FDS's ASCII text format. For example, parameters such as combustible material type, density, height, and thermophysical parameters can be nested within XML nodes and automatically converted into the input syntax required by the FDS using scripting tools.

[0040] 2.3 Remote sensing data processing and classification: Remote sensing image data (such as Landsat, Sentinel series or UAV multispectral / hyperspectral data) need to undergo radiometric calibration, atmospheric correction and geometric correction to eliminate sensor errors and environmental interference, and ensure that the spectral information truly reflects the surface features.

[0041] Building upon this foundation, supervised and unsupervised classification methods are combined to classify remote sensing images. Supervised classification utilizes the spectral features of known land cover categories (training samples) to build a classifier that classifies the entire image. Common methods include Support Vector Machines (SVM) and Random Forests. For example, typical forest, grassland, and shrub plots are selected within the study area, and their spectral features are obtained as training samples. A classification model is then constructed using a Support Vector Machine.

[0042] Unsupervised classification: This method does not require predefined land cover categories. It clusters pixels based on spectral similarity, and then manually interprets and defines the clustering results. Common methods include K-Means clustering and the ISODATA algorithm. For example, after identifying potential vegetation types through clustering, it can be classified as coniferous forest, broadleaf forest, etc., based on field survey data.

[0043] Based on the requirements of FDS simulation for combustible material types, remote sensing images are divided into basic categories such as coniferous forest, broadleaf forest, grassland, shrubs, and bare land. For complex vegetation types (such as mixed forests and multi-coverage grasslands), the classification accuracy can be further improved by subdividing tree age, cover level, etc., to more accurately reflect the spatial distribution and combustion characteristics of combustible materials.

[0044] 2.4 Parameter Mapping and Spatial Distribution Modeling: Based on the classification results of remote sensing images (such as coniferous forests, broad-leaved forests, grasslands, etc.), a correspondence is established between vegetation type information and the combustible parameters (type, density, height, water content) required for FDS input. Specifically, this includes type mapping, density mapping, height mapping, and water content mapping.

[0045] (1) Type mapping Land cover categories (such as "coniferous forest") in remote sensing classification results directly correspond to combustible material types in FDS. For example, the "coniferous forest" category in remote sensing classification results can be mapped to the coniferous forest combustible material type in FDS.

[0046] (2) Density mapping By correlating remote sensing classification results with field survey data, a regression model between density and spectral characteristics is established. The specific steps are as follows: Extract spectral characteristics of specific categories from the remote sensing classification results (e.g., near-infrared reflectance of grassland); combine this with field survey data of the corresponding area (e.g., combustible material mass per unit area) to establish a regression model between density and spectral characteristics (e.g., linear regression or random forest model); use the model to convert the spectral characteristics from the remote sensing classification results into density values. For example, grassland density can be estimated through the regression relationship between its cover (cover ratio in the classification results) and the measured density.

[0047] (3) Height mapping The spatial information of remote sensing data can be used to obtain vegetation height. Specific methods include: for vegetation with significant height, such as forests, the canopy height can be inverted from LiDAR point cloud data and mapped to the FDS input file; if LiDAR data is unavailable, a digital elevation model (DEM) can be calculated from stereo pairs of satellite images, and the vegetation height can be extracted by combining the classification results; alternatively, empirical values ​​can be referenced based on the vegetation type (such as coniferous forest and shrub) in the remote sensing classification results (e.g., the average height of coniferous forest is 15-25 meters, and that of shrub is 1-3 meters).

[0048] (4) Water content mapping An empirical model for water content is established based on the multispectral features of remote sensing images. The specific steps are as follows: Use the reflectance combination of near-infrared (NIR) and short-wave infrared (SWIR) bands (such as the NDWI index) as the basis for water content estimation; Combine field survey data (such as measured values ​​of vegetation water content in different seasons) to establish a regression model of water content and spectral index; Input the vegetation type and corresponding spectral index from the remote sensing classification results into the model to estimate its water content.

[0049] 2.5 Generate combustible material distribution parameters in the FDS input file.

[0050] Data format conversion: The established correspondences of combustible parameters (such as type, density, height, and moisture content) and calculation results are structured and organized according to the format requirements of the FDS input file. Since FDS input files typically use ASCII text format, combustible parameters at different locations need to be organized using a specific syntax structure. For example, FDS command blocks are used to define combustible types and their attributes (such as heat release rate and ignition temperature), and a unique identifier is assigned to each type; the spatial locations (such as latitude, longitude, or projected coordinates) in the remote sensing classification results are aligned with the coordinate range of the FDS grid, and a script tool is used to convert pixel-level parameters into attribute values ​​for FDS grid cells; the numerical accuracy and unit system (such as meters, kilograms per square meter) are ensured to be consistent with FDS requirements.

[0051] Parameter assignment and filling: Based on the geographic coordinate information of the remote sensing image, the combustible parameters corresponding to each pixel are assigned to the corresponding spatial location in the FDS input file. For example, for a two-dimensional FDS simulation area, the combustible parameters of the remote sensing image pixels are mapped to the corresponding positions in the FDS grid to form a complete combustible distribution parameter matrix.

[0052] 2.6 Verification and Calibration.

[0053] After completing parameter mapping and generating the FDS input file, field verification is required. This involves selecting a certain number of sample plots within the study area, measuring parameters such as the type, density, height, and moisture content of combustibles in the field, and comparing these measurements with parameters obtained from remote sensing image mapping.

[0054] Based on the results of field verification, errors in the remote sensing image classification and parameter mapping processes are analyzed and corrected. For example, if it is found that the estimated values ​​of combustible density in certain areas deviate significantly from the actual values, the classification model or parameter mapping relationship can be adjusted, and the combustible distribution parameters in the FDS input file can be regenerated until the simulation results match the actual situation.

[0055] 3. The dynamic coupling module for meteorological conditions is used to simulate the changes in wind speed, wind direction, and temperature with terrain and time.

[0056] The dynamic coupling module for meteorological conditions includes a meteorological data input unit, a topographic wind field correction unit, a dynamic boundary condition adjustment unit, and a fire-wind interaction model.

[0057] The meteorological data input unit is configured to receive real-time data from meteorological stations or meteorological fields output by WRF (Weather Research and Forecasting) models.

[0058] The terrain wind field correction unit is configured based on the principles of fluid dynamics to calculate the local wind speed and direction changes caused by the terrain and correct the meteorological data obtained by the meteorological data input unit.

[0059] The dynamic boundary condition adjustment unit is used to adjust boundary meteorological parameters (such as wind speed, wind direction, temperature and humidity) in real time, and its values ​​are dynamically updated according to the time, fire development or environmental changes.

[0060] To address the issue of insufficient coupling between boundary conditions and physical fields in fire simulations under complex terrain, this invention proposes a terrain-feature-driven boundary condition mapping mechanism. This mechanism quantifies the impact of terrain features on FDS boundary conditions (heat transfer and wind speed), enabling adaptive setting of the physical field and thus improving the simulation accuracy of FDS in complex environments.

[0061] Mapping between topography and thermal convection coefficient: In complex terrain, slope directly affects airflow and heat exchange efficiency. Thermal convection coefficient h c With slope S ( x , y The relationship can be modeled as follows: h c ( x , y )= h c0 ×(1+ k × S ( x ,y) / 90°)(4) In the formula: h c0 The reference convection coefficient (e.g., 5 W / (m²)) 2 ·K)), k The slope influence factor (experimentally determined to be 0.2). When the slope... S At a slope of 30°, the thermal convection coefficient increases by about 6.7%, indicating that heat transfer modeling needs to be strengthened in steep slope areas.

[0062] Mapping of Topography and Wind Speed: Valley topography significantly amplifies wind speed. This is achieved through slope aspect. A ( x , y ) and wind direction θ The angle relationship between the two sides is used to define the wind speed correction factor. f v : (5) When the angle between the wind direction and the valley orientation is less than 30°, the wind speed can be increased by 1.5-2 times. This correction factor is directly used for the dynamic adjustment of the ventilation boundary in FDS.

[0063] In wildfire spread simulation, fire-wind interaction is a crucial physical process that directly affects the speed, extent, and behavior of fire spread. Fire-wind interaction models need to accurately depict the complex dynamic coupling relationship between flames and airflow fields. The following explanation covers the model's principles, mathematical expression, and implementation process.

[0064] 3.1 Core physical mechanisms and model objectives.

[0065] Wind drives fire: By providing oxygen transport and changing heat convection patterns, wind accelerates the spread of the flame front. At the same time, the blowing of wind causes the flame to tilt, changing the spatial distribution of heat radiation and heat conduction, and affecting the preheating and ignition process of combustibles.

[0066] The reaction of fire to wind: The combustion of flames releases a large amount of heat, heating the surrounding air and forming an updraft driven by thermal buoyancy (fire plume). This causes disturbances in the airflow field, including inducing wind-generated flows, changing the distribution of wind speed and direction, and even forming extreme phenomena such as firestorms in large-scale wildfires. In this invention, the fire-wind interaction model considers the influence of flames on local airflow, achieving two-way coupled simulation.

[0067] The construction of the model objective can quantitatively describe the two-way coupling process of "wind field driving fire development - fire changing airflow field", providing accurate fire-wind interaction boundary conditions and physical field calculation basis for FDS wildfire spread simulation, and improving the accuracy of wildfire spread path and intensity prediction.

[0068] 3.2 Mathematical modeling of fire-wind interaction.

[0069] (I) Basic Framework of Governing Equations The Navier-Stokes equations within the framework of computational fluid dynamics (CFD) are used to describe the airflow field, and the fire-wind coupling is characterized by combining energy equations and chemical reaction kinetics equations. The core governing equations are as follows: 1. Mass conservation equation (continuity equation) (6) in ρ air density, t For time, u =( u , v , w The velocity vector () describes the conservation of airflow mass. The change in air density caused by flame heating drives the airflow motion and is the basic mass transfer constraint of fire-wind coupling.

[0070] 2. Momentum conservation equation (Navier-Stokes equation) (7) In the formula p Pressure; τ Let be the viscous stress tensor, which describes the effect of air viscosity on momentum transfer; g f is the acceleration due to gravity. fire-wind The source term for fire-wind interaction includes the drag force and thermal buoyancy of the airflow caused by the flame. The initial velocity of the wind field. u wind,0 As the initial input condition, the density change (Δ) induced by flame combustion ρ )pass ρ = ρ 0+Δ ρ ( ρ The momentum equation is influenced by the ambient air density (0 represents the ambient air density), thus achieving the initial coupling between the wind field and the fire.

[0071] 3. Energy Conservation Equation (8) In the formula h Enthalpy; k Thermal conductivity; T For temperature; q comb Combustion heat release rate is a key energy source term in fire-wind coupling; the wind field affects oxygen supply and thus alters... q comb Φ represents the viscous dissipation term. Flame temperature. T fire Significantly higher than ambient temperature T 0, heat is transferred to the surrounding air through heat conduction, convection and radiation, heating the air and changing the temperature distribution of the wind field, while the wind field affects the flame shape and combustion intensity by transporting heat.

[0072] 4. Combustion chemical reactions and species transport equations (9) In the formula, Y i Components i (such as the mass fraction of oxygen O2, fuel volatiles, carbon dioxide CO2, etc.); D i The diffusion coefficient is denoted as . ω i Components i The generation / consumption rate. The wind field affects the oxygen supply (O2 transport) and diffusion of combustion products in the flame zone by transporting species through convection, thereby changing the combustion intensity. The combustion reaction consumes O2 and generates heat, which in turn affects the wind field.

[0073] (II) Modeling of key source terms in fire-wind coupling 1. Thermal buoyancy-driven fire plume model (fire's reaction to wind) The thermal buoyancy generated by the flame heating the air is the core factor changing the wind field. The Boussinesq approximation simplifies the effect of density change on the momentum equation, and the thermal buoyancy source term... f buoyancy Represented as: (10) In the formula, β The coefficient of thermal expansion (approximately 1 / ) T 0, T 0 represents ambient temperature. High temperature in the flame zone ( T >> T 0) It reduces the air density, generating upward buoyancy that drives the airflow upward, forming a fire plume, which induces the surrounding air to flow towards the flame zone, changing the wind field structure.

[0074] 2. Model of the drag force of flame on airflow (the drag effect of fire on wind). As a porous, medium-like combustion body, the flame generates a drag force on the passing airflow; the drag force source term f drag Using the Ergun equation form (approximating complex structures such as vegetation flames): (11) In the formula: C d This is the drag coefficient, which is related to flame morphology, vegetation density, etc. (calibrated through experiments or empirical formulas, such as vegetated flames). C d (Take a value of 0.2-1.0). A d This represents the contact area between the flame and the airflow per unit volume, reflecting the degree to which the flame obstructs the airflow; u flame The characteristic velocity of the flame zone is approximated by a value related to the flame front spread velocity. The drag force slows down the airflow as it passes through the flame zone, while also changing the airflow direction, thus affecting the spatial distribution of the wind field.

[0075] 3. Wind-driven model of flame spread (the effect of wind on fire) Wind fields accelerate the spread of the flame front by altering thermal convection and oxygen supply, thus increasing the flame spread rate. v f With wind speed u wind The coupling relationship is modeled using an empirical model: (12) In the formula: v f0 The speed of flame spread in the absence of wind (determined by the type of combustible material); k w The wind speed influence factor (related to the properties of combustibles and terrain, such as grassland fires)k w Take 0.3-0.5). θ The angle between the direction of wind and the direction of fire spread, cos θ This model reflects the effective component of wind speed in the direction of flame spread. The oxygen provided by the wind field enables more complete combustion, and thermal convection accelerates the preheating of combustibles, thus affecting the speed of flame spread. This model quantifies the driving effect of the wind field on the fire and feeds it back into the fire development.

[0076] 4. FDS Simulation Optimization Process Figure 4 This is the overall architecture diagram of a wildfire spread simulation system based on FDS. In the input file generation stage, an automated script integrates terrain data, combustible material distribution parameters (such as type, density, height, and moisture content), and meteorological conditions (such as wind speed, wind direction, and temperature) to generate an input file conforming to the FDS ASCII format, ensuring structured mapping and parameter consistency across multiple data sources. Based on this, an adaptive mesh generation strategy is adopted. By dynamically identifying flame front regions (such as high-temperature gradients and areas of abrupt changes in combustion rate), high-density meshes (e.g., 0.1-meter resolution) are deployed in key areas, while low-density meshes (e.g., 1-meter resolution) are used at the fire edge or in low-dynamic areas, thereby reducing computational resource consumption while maintaining simulation accuracy. For the computational needs of large-scale forest scenarios, distributed parallel computing is implemented based on the MPI (Message Passing Interface) framework. Task partitioning and inter-node communication optimization improve the computational efficiency of the multi-processor cluster. Finally, a dedicated post-processing tool is developed, integrating an FDS output data parsing module to support real-time visualization of key indicators such as fire spread rate, thermal radiation intensity, and burned area. Combined with time series analysis functions, it enables quantitative assessment and risk prediction of dynamic characteristics of fire behavior.

[0077] Application Examples: I. Preliminary Preparation Stage (Manual Operation) 1. Simulation Area and Target Setting The wildfire area to be simulated is manually selected (e.g., a forest farm in a mountainous area, with an area of ​​approximately 1 km × 1 km), and the simulation objectives are clearly defined (e.g., predicting the fire's spread path, burned area, and heat radiation intensity within 24 hours). Historical wildfire data for this area (if available) is collected simultaneously as an important basis for verifying subsequent simulation results.

[0078] 2. Multi-source data acquisition and preprocessing Terrain data: Manually acquire LiDAR point cloud data (accuracy ≥ 0.5m) or high-precision DEM data (resolution ≤ 1m) for the area to ensure key terrain features such as ridges, valleys, and steep slopes within the coverage area.

[0079] Combustible material data: The distribution map of vegetation types in the area was collected manually (such as the proportion of coniferous forests, shrubs, and grasslands), and the thermophysical parameters of typical combustible materials (such as heat release rate and ignition temperature) were measured through field sampling for database calibration.

[0080] Meteorological data: Meteorological data for the simulation period are manually obtained from weather stations or WRF model outputs, including initial wind speed (e.g., 2-5 m / s), wind direction (e.g., southeast wind), temperature (e.g., 25-30℃), and humidity (e.g., 30%-50%).

[0081] II. Data Processing and Parameter Configuration Stage (Automatic System + Manual Assistance) 1. Terrain Feature Extraction and Adaptive Mesh Generation The system automatically performs the following steps: It calls the terrain preprocessing program to filter and interpolate the LiDAR / DEM data, generating a regularized terrain grid. It automatically calculates the slope matrix S(x,y) and aspect matrix A(x,y) (based on the central difference method using a 3×3 sliding window, see formulas (1), (2), and (3)). It automatically identifies key terrain features (ridges, valleys, and steep slopes ≥30°).

[0082] Manual assistance steps: Check whether the slope / aspect distribution output by the system is reasonable (e.g., whether steep slope areas match the actual terrain). If errors exist, manually adjust the filtering parameters and recalculate. Confirm the grid density division standard based on the slope S (default is the standard in step 1.3: low-density grid for S<10°, medium-density grid for 10°≤S<30°, and high-density grid for S≥30°). The threshold can be fine-tuned according to the specific characteristics of the area.

[0083] 2. Combustible material parameter mapping and database access The system's automated steps include: 1. Calling the remote sensing image processing program to perform radiometric calibration, atmospheric correction, and geometric correction on the input remote sensing image (e.g., Sentinel-2 multispectral data). 2. Automatically using the Support Vector Machine (SVM) algorithm to classify the remote sensing image and output the distribution of combustible material types (e.g., coniferous forest 60%, shrubs 20%, grassland 20%). 3. Based on the classification results, automatically retrieving corresponding parameters from the "Forest Combustible Material Thermal Property Parameter Database" in XML format (e.g., heat release rate of coniferous forest). 500kW / m 2 The ignition temperature is 350℃, and the density (e.g., grassland density 1.2 kg / m³) is calculated using a regression model. 2 ), height (e.g., 20m on average for coniferous forests), and water content (e.g., 15%).

[0084] Manual assistance steps: Compare the remote sensing classification results with the field vegetation survey data. If the classification error in a certain area is >10% (e.g., shrubs are mistakenly classified as grassland), manually correct the classification labels and remap the parameters. Check whether the parameters called from the database are reasonable (e.g., under extreme drought conditions, the moisture content of combustibles can be manually reduced).

[0085] 3. Meteorological conditions and fire-wind model parameter configuration The system automatically performs the following steps: It invokes the "Meteorological Condition Dynamic Coupling Module" to convert the meteorological data (wind speed, wind direction, temperature) output by WRF into an initial boundary condition format recognizable by FDS. It then automatically loads the governing equations (mass conservation, momentum conservation, energy conservation equations) of the fire-wind interaction model and initializes parameters such as the drag coefficient Cd (e.g., 0.3) and the wind speed influence coefficient kw (e.g., 0.4).

[0086] Manual operation steps: Manually set the simulation time step (e.g., 10s) and total duration (e.g., 24 hours), and specify the output indicators (e.g., fire spread rate, thermal radiation intensity, burned area). Confirm the parallel computing parameters (e.g., using a 4-node MPI cluster, with 4 CPU cores allocated to each node).

[0087] III. FDS Simulation Operation Phase (Automatic System) 1. Input file generation The system automatically runs a script to integrate terrain grid data, combustible material distribution parameters, and meteorological boundary conditions, generating an input file conforming to the FDSASCII format. This file contains the following key information: grid definition (adaptive grid based on octree algorithm, with high-density areas measuring 5-10m and low-density areas measuring 20-50m); combustible material properties (spatial distribution of type, density, and thermal properties); initial wind field (wind speed 2m / s, southeast wind) and dynamic update rules (based on a fire-wind interaction model); and output control (recording temperature and velocity field data every 30 minutes).

[0088] 2. Parallel computing execution The system automatically invokes the MPI framework, assigns the simulation task to 4 computing nodes, and starts the FDS simulation: Nodes 1-2 are responsible for the calculation of the combustion process in high-density grid areas (steep slopes, ridgelines); Nodes 3-4 are responsible for the coupled calculation of wind field and fire spread in low-density grid areas (gentle slopes, plains); the nodes synchronize boundary data (such as the position of the flame front and wind speed correction factor) in real time through 10 Gigabit Ethernet.

[0089] 3. Simulation process monitoring The system automatically runs a monitoring program, displaying the CPU utilization (maintained at 70%-90%), memory usage (not exceeding 64GB), and simulation progress (e.g., "10 hours completed, 14 hours remaining") of each node in real time. If a calculation anomaly occurs (e.g., memory overflow), the system will automatically pause and issue an alarm.

[0090] IV. Results Analysis and Verification Phase (Automatic System Operation + Manual Operation) 1. Output data analysis and visualization The system automatically performs the following steps: After the simulation, it automatically parses the binary result file output by FDS and extracts the fire spread velocity (e.g., average 0.8 m / s) and thermal radiation intensity (e.g., peak 15 kW / m²). 2 ), burned area (e.g., 120 hm²) 2 Indicators such as fire spread path animation (one frame per hour), temperature field cloud map, and wind speed vector map are generated using post-processing tools.

[0091] Manual operation steps: Manually check the reasonableness of the visualization results (e.g., whether the flames are spreading rapidly along the ridgeline, whether the wind speed is increasing in the valley). Compare the simulation results with historical wildfire records and calculate the error (e.g., the predicted burned area error is 15%, within the range of ≤18% as stated in the document).

[0092] 2. Model calibration (if the error exceeds the limit) If the predicted fire spread rate error is greater than 12%, the simulation is repeated after manually adjusting the parameters: correct the moisture content of the combustible material (e.g., from 15% to 12% to simulate a drier environment); fine-tune the wind speed influence coefficient kW (e.g., from 0.4 to 0.5 to enhance the driving effect of wind on the fire).

[0093] V. Final Output (Automatic System Execution) The system automatically generates a simulation report, which includes: an overview of the terrain and combustible material distribution in the simulation area; time series curves of key fire spread indicators (speed, area, thermal radiation) within 24 hours; comparative analysis and error assessment with historical data; and risk warning suggestions (such as high-risk areas being within 200m on both sides of the ridgeline). Figure 5 It is a simulation rendering.

[0094] Hardware Platform: CPU: Supports parallel computing; at least 8 cores (≥3.0GHz) are required for small to medium-scale scenarios, and 16 cores or more are recommended for large-scale scenarios (1km×1km and above). Supports Hyper-Threading technology to optimize parallel task scheduling under the MPI framework. GPU (Optional): Professional graphics card (such as NVIDIA Quadro, AMD Radeon Pro), ≥8GB of VRAM, supporting high-performance graphics interfaces for real-time visualization of simulation results; if GPU-accelerated computing is involved, it must support CUDA / OpenCL and have ≥16GB of VRAM. Memory: ≥32GB DDR4 / DDR5 for small to medium-scale scenarios, ≥64GB for large-scale scenarios, with memory bandwidth ≥25GB / s to ensure fast read and write of massive amounts of data (terrain, combustible material parameters, etc.). Storage: 1TB or more NVMe SSD (sequential read / write ≥3000MB / s) for storing the operating system, programs, and frequently accessed data; 2TB or more HDD / large-capacity SSD for archiving raw data and historical simulation results.

Claims

1. A wildfire spread simulation method based on FDS, characterized in that, include: The terrain data of the simulation area is preprocessed to extract terrain features including slope and aspect; The simulation area is divided into grids, and the grid resolution is dynamically adjusted according to the slope: when the slope exceeds a preset threshold, a fine-grained grid is generated, and when the slope is below the threshold, a coarse-grained grid is generated, forming a non-uniform hierarchical grid structure. By loading a dynamic coupling module for meteorological conditions and combining it with data from a database of thermal properties of forest combustibles, dynamic simulation of fire spread is performed. The forest combustible thermal property parameter database includes: the vegetation type corresponding to each grid in the simulation area, and the thermal property parameters, density, height and water content corresponding to each vegetation type; The meteorological condition dynamic coupling module is configured to achieve bidirectional coupling between the wind field and the combustion process by quantifying the reaction of fire to wind, the resistance of fire to wind, and the effect of wind on fire, and to dynamically update the boundary conditions of wind speed, wind direction, and temperature.

2. The method according to claim 1, characterized in that, The dynamic coupling module for meteorological conditions includes: The meteorological data input unit is configured to receive real-time data from the meteorological station or the meteorological field output by the WRF model; The terrain wind field correction unit is configured to calculate the local wind speed and wind direction changes caused by the terrain based on the principles of fluid dynamics, and to correct the meteorological data obtained by the meteorological data input unit. A dynamic boundary condition adjustment unit is configured to adjust boundary meteorological parameters in real time, with values ​​dynamically updated based on time, fire development, or environmental changes; and The fire-wind interaction model achieves bidirectional coupling between the wind field and the combustion process by quantifying the reaction of fire to wind, the resistance of fire to wind, and the effect of wind on fire.

3. The method according to claim 2, characterized in that, The fire-wind interaction model includes: a fire plume model driven by thermal buoyancy, which characterizes the reaction of fire to wind; a flame drag force model on airflow, which characterizes the drag effect of fire on wind; and a wind-driven model for the spread of flames, which characterizes the effect of wind on fire.

4. The method according to claim 1, characterized in that, The terrain data preprocessing includes: acquiring terrain information using LiDAR or DEM; calculating local elevation gradients; and generating a slope matrix. S ( x , y ) and aspect matrix A ( x , y ).

5. The method according to claim 1, characterized in that, When the terrain slope is less than 10°, the grid size is 20-50m; when the terrain slope is less than 30° and the 10° is less than or equal to 30°, the grid size is 10-20m; when the terrain slope is greater than or equal to 30°, the grid size is 5-10m.

6. The method according to claim 1, characterized in that, The vegetation types in the forest combustible thermal property parameter database are obtained by classifying remote sensing images.

7. The method according to claim 2, characterized in that, The dynamic boundary condition adjustment unit includes: Thermal convection coefficient correction model: h c ( x , y )= h c0 ×(1+ k × S ( x , y ) / 90°) In the formula: h c0 As the reference convection coefficient, k Slope influencing factors S ( x , y () represents the slope.

8. The method according to claim 7, characterized in that, The dynamic boundary condition adjustment unit further includes: Wind speed correction factor f v : A ( x , y ( ) represents the slope direction. θ The angle between the slope and the wind direction is used as a correction factor for the dynamic adjustment of the ventilation boundary in the FDS.

9. The method according to claim 1, characterized in that, Distributed parallel computing based on the MPI framework.

10. The method according to claim 1, characterized in that, The data output by FDS is analyzed to extract the fire spread rate, thermal radiation intensity, and burned area, and then visualized.

Images