A forest fire risk assessment method considering potential forest fire behavior characteristics in the neighborhood

Abstract

The present invention discloses a method for assessing forest fire risk level taking into account potential forest fire behavior characteristics in the neighborhood, and relates to the technical field of forest fire early warning. The present invention extracts the ignition probability, fire spread speed, fire intensity, and combustible, meteorological, topographical and human activity factors corresponding to the fire point pixel and the non-fire point pixel, constructs a forest fire behavior characteristic case library, and then constructs a potential forest fire behavior characteristic prediction for the ignition probability, fire spread speed and fire intensity based on the random forest algorithm, and uses the entropy weight method to determine the weights of various potential forest fire behavior characteristics to characterize the forest fire risk level, and finally comprehensively considers the potential forest fire behavior characteristics of the Moore neighborhood to assess the forest fire risk level. The forest fire risk level assessment method described in the present invention is simple to operate, and it fully considers a variety of potential forest fire behavior characteristics and their influences, further improves the reliability of forest fire risk level assessment, and can provide a scientific basis for the allocation of forest fire prevention and control resources, the formulation of emergency response plans, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of forest fire early warning, and in particular to a forest fire risk level assessment method taking into account potential forest fire behavior characteristics in a neighborhood. Background Art

[0002] Accurate assessment of forest fire risk is of great significance to combustible management and forest fire prevention and control. Traditional forest fire risk forecasts are mainly based on forest fire probability assessments, and do not include potential forest fire behavior characteristics such as forest fire spread and fire intensity, resulting in weak targeted allocation of forest fire prevention and control resources and inadequate emergency response plans. In contrast, forest fire risk level assessment belongs to the category of forest fire hazard assessment, which usually takes into account the risk of forest fire occurrence and fire intensity characteristics, and can improve the targeted, practical and accurate nature of forest fire prevention and control.

[0003] The wildfire triangle model points out that combustibles, weather and terrain are key factors affecting the occurrence and development of forest fires, and has been widely used in forest fire risk modeling. In terms of fire risk assessment with coordinated forest fire behavior characteristics prediction, most studies use meteorological parameters to drive fire behavior models and process models to achieve forest fire risk forecasting and warning. This method has appeared in foreign forest fire risk forecasting systems that are in commercial operation, such as the Canadian Forest Fire Risk Rating System and the US National Fire Risk Rating System. Although these systems consider the mechanism of forest fire occurrence and development, the fire risk warning modules involved are relatively simple, and the combustible information is not fully considered. Another study attempted to use a fire behavior model with a physical mechanism to simulate potential fire behavior for characterizing forest fire risks, but this method is difficult to apply on a large scale. At present, mainstream machine learning algorithms have been widely used in forest fire risk forecasting and warning research. By constructing a spatiotemporal big data mining model for forest fire behavior characteristics, it can further serve large-scale and high-precision forest fire risk assessment work. It is worth noting that according to the mechanism of forest fire spread, if a forest fire occurs in a neighboring pixel, the central pixel may also be affected by the forest fire. Therefore, considering the potential forest fire behavior of neighboring pixels is necessary to improve the accuracy of forest fire risk level assessment. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a forest fire risk level assessment method which has strong scalability, is simple to calculate and takes into account the potential forest fire behavior characteristics of the neighborhood.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a forest fire risk level assessment method taking into account the potential forest fire behavior characteristics of the neighborhood, comprising the following steps:

[0006] Step 1: Extraction of forest fire risk factors;

[0007] Extract the combustibles, meteorological, topographical and human activities risk factors for forest fire risk level assessment in the target area. The spatial resolution of the risk factors is 500 meters, the temporal resolution of the combustibles and meteorological risk factors is 1 day, and the topographical and human activities factors are static data;

[0008] Step 2: Construction of forest fire behavior characteristics case library;

[0009] Step 2.1: Extraction of forest fire behavior features;

[0010] Based on the GFA fire dataset, GFA stands for Global Fire Atlas, the location information of the ignition point of each forest fire case in the study area is determined, and the corresponding fire spread speed is extracted; the maximum fire radiation power corresponding to the location of each forest fire ignition point is extracted to characterize the fire intensity characteristics;

[0011] Furthermore, the number of pixels in the single forest fire case should be ≥2;

[0012] Step 2.2: Extract feature information of non-fire point pixels;

[0013] Based on the GFA fire dataset, for each forest fire, the location and occurrence time information of other fire burning pixels except the fire point pixel are extracted;

[0014] Step 2.3: Construct a case library of forest fire behavior characteristics for fire point pixels and non-fire point pixels;

[0015] For the fire starting point pixels and non-starting point pixels, the characteristics of combustibles, meteorology, terrain and human activities at the corresponding locations are extracted to build a case library of forest fire behavior characteristics (FBC) consisting of forest fire risk factors, ignition probability (FGP), fire spread speed (ROS) and fire reaction intensity (FRI);

[0016] Furthermore, the ignition probability of the fire pixel in the forest fire behavior characteristic case library is set to 1, and the ignition probability of the non-fire point pixel is set to 0;

[0017] Furthermore, the fire spread speed and fire intensity characteristics in the forest fire behavior characteristic case library are divided into five levels by the natural break method, namely low, lower, medium, higher, and high, and are assigned values ​​of 1, 2, 3, 4, and 5 respectively;

[0018] Step 3: Construction and evaluation of prediction model of potential forest fire behavior characteristics;

[0019] Step 3.1: Balance the samples of forest fire behavior characteristic case library;

[0020] The SMOTEENN algorithm is used to balance the samples in the case library of ignition probability, fire spread speed, and fire intensity characteristics to ensure that the number of samples in each category is equal.

[0021] Furthermore, the ratio of the two samples in the ignition probability feature case library is 1:1, and the ratio of each category in the fire spread speed and fire intensity feature case library is 1:1:1:1:1;

[0022] Step 3.2: Model construction and evaluation;

[0023] Based on the forest fire behavior characteristic case library, the training set and test set were randomly divided; the nonlinear relationship between combustibles, meteorology, terrain and human activity factors and ignition probability, fire spread speed and fire intensity characteristics was explored using the random forest (RF) algorithm, and the prediction models of latent ignition probability, fire spread speed and fire intensity were constructed according to their respective nonlinear relationships; a confusion matrix was constructed, and the overall accuracy and Kappa coefficient were used to evaluate the performance of the prediction model;

[0024] Furthermore, the ratio of the training set to the test set is 7:3;

[0025] Furthermore, the number of trees (n-tree) of the potential forest fire behavior feature prediction model based on the random forest (RF) algorithm is 300, and the maximum depth (max-depth) is 10;

[0026] Further, the probability of the ignition probability feature prediction model predicting a value of "1" is defined as the ignition probability of the forest fire;

[0027] Further, the category with the maximum probability output by the fire spread rate characteristic prediction model is defined as the potential fire spread rate characteristic level ROS;

[0028] Further, the category with the maximum probability output by the fire intensity characteristic prediction model is defined as the potential fire intensity characteristic level FRI;

[0029]

[0030] Among them, OA represents the overall accuracy, KC represents the Kappa coefficient, N is the total number of samples, and x ij is the value of the i-th row and j-th column in the confusion matrix, x i+ represents the total number of rows in the confusion matrix, x +i represents the total number of columns in the confusion matrix, * represents a multiplication operation, and r represents the number of rows in the confusion matrix;

[0031] Step 4: Forest fire danger level assessment integrating potential forest fire behavior characteristics;

[0032] Step 4.1: Determine the weights of potential fire behavior characteristics;

[0033] Using the prediction model of ignition probability, fire spread speed, and fire intensity constructed in step 3, the ignition probability, fire spread speed, and fire intensity of all samples in the forest fire behavior characteristic case library are calculated respectively; using formulas 3, 4, and 5, the weights w of various potential forest fire behavior characteristics (FBC) to characterize the forest fire risk level are determined j ;

[0034]

[0035]

[0036] In formula 3, 4, and 5, FBC ij is the jth potential forest fire behavior characteristic of the i-th sample, namely, the fire probability (FGP), fire spread rate (ROS) and fire intensity (FRI) characteristics; P ij represents the element in the i-th row and j-th column of the probability matrix (or deviation matrix), e j represents the standard information entropy of the jth potential forest fire behavior feature; N is the total number of samples; M is the number of potential forest fire behavior feature types (M = 3); w j is the weight of each potential forest fire behavior characteristic;

[0037] Step 4.2: Comprehensively assess the forest fire risk level FDG;

[0038] The predicted ignition probability characteristics are divided into 5 levels using the natural break method, namely low, lower, medium, higher and high, and the values ​​are assigned as 1, 2, 3, 4 and 5 respectively; the forest fire risk level is calculated by weight using formula 6;

[0039] FDG=[w1*FGP+w2*ROS+w3*FRI+A] (6)

[0040] Wherein, [] is the Gaussian rounding function; FRP, ROS, FRP are the ignition probability, fire spread speed, and fire intensity characteristics, respectively, and the values ​​are 1, 2, 3, 4, and 5, corresponding to the forest fire risk level of low, relatively low, medium, relatively high, and high, respectively; w1, w2, and w3 are the corresponding weights, and A is the correction coefficient;

[0041] Step 5: Forest fire danger level assessment integrating potential forest fire behavior characteristics of the neighborhood;

[0042] Specifically, it includes 2 parts:

[0043] Step 5.1: Determine the weight W of the influence of the forest fire danger level (FDG) of the neighboring pixel on the forest fire danger level of the central pixel;

[0044] Drawing on the idea of ​​simulating forest fire spread based on the cellular automaton model, if a forest fire occurs in the neighboring pixels, the central pixel may also be affected by the forest fire. Therefore, considering the Moore neighborhood, formulas (7) and (8) are used to calculate the degree of influence of the forest fire risk level of the neighboring pixels on the forest fire risk level of the central pixel: ij ;

[0045]

[0046]

[0047] Where i0 and j0 are the row index and column index of the center pixel respectively, ROS ij is the potential fire spread rate level of the neighborhood pixel ij, a is assumed to be the size of the pixel; when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; the pixel (i, j) should be a forest pixel;

[0048] Step 5.2: Update the forest fire risk level of the central pixel;

[0049] Using formula (9), the original FDG of the central pixel and the FDG of the neighboring pixels are combined to update the forest fire risk level FDI' of the central pixel;

[0050]

[0051] Where [] is the Gaussian rounding function, is the forest fire risk level of the central pixel, is the original forest fire risk level of the central pixel, calculated by formula (6); when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; pixel (i, j) should be a forest pixel.

[0052] Beneficial effects of the invention: The forest fire risk level assessment method of the invention is simple to operate. First, the random forest algorithm is used to predict the potential forest fire behavior characteristics (ignition probability, fire spread speed, and fire intensity). Then, the three potential forest fire behavior characteristics of the central pixel and the neighboring pixel are fused based on the entropy weight method to carry out forest fire risk level assessment. Compared with the forest fire risk forecasting method that only focuses on the probability of occurrence assessment, this method fully considers a variety of potential forest fire behavior characteristics and their impacts, and further improves the accuracy of forest fire prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the overall method flow of the present invention.

[0054] Figure 2 It is a location map of the study area in a specific implementation scheme of the present invention.

[0055] Figure 3 It is the accuracy evaluation result of the potential forest fire behavior characteristic prediction model in the specific implementation scheme of the present invention.

[0056] Figure 4 This is the mapping result of the average forest fire risk level assessment in the study area from February to April 2024 in the specific implementation plan of the present invention. DETAILED DESCRIPTION

[0057] The forest fire risk level assessment method taking into account the potential forest fire behavior characteristics of the neighborhood provided by the present invention is further described below in conjunction with specific embodiments and drawings of the specification:

[0058] A method for assessing forest fire risk level that takes into account the potential forest fire behavior characteristics of the neighborhood, such as Figure 1 As shown, the following steps are included:

[0059] Step 1: Extraction of forest fire risk factors. Yunnan Province, China was selected as the study area ( Figure 2 ) to extract risk factors such as combustibles (moisture content of canopy combustibles, moisture content of dead combustibles on the surface, leaf area index, forest type), meteorology (wind speed, air temperature, relative humidity, rainfall), terrain (elevation, slope and slope aspect), and human activities (distance from roads, distance from residential areas) for forest fire risk assessment in the target area;

[0060] For the moisture content of canopy fuel, it was extracted from the global canopy fuel moisture content (LFMC) dataset published by the inventor team (URL: https: / / zenodo.org / record / 7690340#.ZAAt0T1Bybg); for the moisture content of dead surface fuel, it was extracted from the fire risk index product released by the European Center for Medium-Range Weather Forecasts (ECMWF) (URL: https: / / cds.climate.copernicus.eu / cdsapp#! / dataset / cems-fire-historical?tab= overview); for leaf area index, it was extracted from the MCD15A2H product set; for forest type, it was extracted from the MCD12Q1 product set; for meteorological factors, they were all extracted from the ERA5-Land dataset, and the daily average air temperature, daily average air relative humidity, daily average wind speed and daily cumulative rainfall were synthesized; for terrain factors, elevation information was first extracted from the GMTED 2010 dataset, and then the slope and aspect characteristics were extracted using the SAGA tool; for human activity factors, road and residential area information was first extracted from the National Geographic Information Resource Directory Service System (https: / / www.webmap.cn / main.do?method=index), and then the Euclidean distance from the road and residential area was calculated as the human activity factor; the spatial resolution of all risk factors was 500 meters, the temporal resolution of combustible and meteorological risk factors was 1 day, and the terrain and human activity factors were static data;

[0061] Step 2: Construction of forest fire behavior characteristics case library;

[0062] Specifically, it includes 3 parts:

[0063] Step 2.1: Extraction of forest fire behavior characteristics. Based on the Global Fire Atlas (GFA) fire dataset, determine the location information of the ignition point of each forest fire case (the number of pixels should be ≥ 2) in the study area from 2003 to 2016, and extract the corresponding fire spread speed; based on the MOD14A1 / MYD14A1 temperature anomaly / fire L3 level products, extract the maximum fire radiation power corresponding to the location of each forest fire ignition point to characterize the fire intensity characteristics;

[0064] Step 2.2: Extract feature information of non-fire point pixels. Based on the GFA fire dataset, for each forest fire, extract the location and occurrence time information of other fire burning pixels except the fire point pixels;

[0065] Step 2.3: Construction of a case library of forest fire behavior characteristics for fire point pixels and non-starting fire pixels. For fire point pixels and non-starting fire pixels, extract the characteristics of combustibles, meteorology, terrain and human activities at the corresponding locations, and construct a case library of forest fire behavior characteristics (FBC) consisting of forest fire risk factors, ignition probability (FGP), fire spread speed (ROS) and fire reaction intensity (FRI) (see Figure 2); the ignition probability of the fire pixel in the forest fire behavior characteristic case library is set to 1, and the ignition probability of the non-fire point pixel is set to 0; the fire spread speed and fire intensity characteristics in the case library are divided into 5 levels by the natural interruption method, namely low, lower, medium, higher, and high, and are assigned values ​​of 1, 2, 3, 4, and 5 respectively;

[0066] Step 3: Construction and evaluation of prediction model of potential forest fire behavior characteristics;

[0067] Specifically, it includes 2 parts:

[0068] Step 3.1: Balance samples in the forest fire behavior feature case library. Use the SMOTEENN algorithm to balance samples in the ignition probability, fire spread speed, and fire intensity feature case libraries to ensure that the number of samples in each category is equal;

[0069] Step 3.2: Model construction and evaluation. For the forest fire behavior characteristic case library, the training set and test set were randomly divided in a ratio of 7:3; using the random forest (RF) algorithm, the number of trees (n-tree) was set to 300 and the maximum depth (max-depth) was set to 10, and the nonlinear relationship between combustibles, meteorology, terrain and human activity factors and ignition probability, fire spread speed and fire intensity characteristics was explored respectively, and potential forest fire behavior characteristics (ignition probability, fire spread speed, fire intensity) prediction models were constructed respectively; among them, the probability predicted as "1" by the ignition probability feature prediction model was defined as the ignition probability of the forest fire; a confusion matrix was constructed, and the overall accuracy (OA, formula 1) and Kappa coefficient (KC, formula 2) were used to evaluate the model performance; Figure 3 The statistical distribution of OA and KC of the potential forest fire behavior feature prediction model when the training set and test set were randomly divided 50 times is given. It can be seen that the ignition probability feature prediction model has the best performance (OA≥0.9, KC≥0.9), followed by the potential fire spread speed feature prediction model (OA≥0.8, KC≥0.8), while the performance of the potential fire intensity feature prediction model is slightly worse (OA≥0.79, KC≥0.73);

[0070]

[0071] Where N is the total number of samples, x ii is the value of the i-th row and j-th column in the confusion matrix, x +i and *x +i are the sum of the i-th row and i-th column respectively;

[0072] Step 4: Forest fire danger level assessment integrating potential forest fire behavior characteristics;

[0073] Specifically, it includes 2 parts:

[0074] Step 4.1: Determine the weight of potential forest fire behavior characteristics. Using the prediction model of ignition probability, fire spread speed, and fire intensity characteristics constructed in step 3, calculate the ignition probability, fire spread speed, and fire intensity characteristics of all samples in the forest fire behavior characteristic case library; based on the entropy weight method (Formulas 3, 4, and 5), determine the weight w of various potential forest fire behavior characteristics (FBC) to represent the forest fire risk level. j , that is, the weights of the ignition probability, fire spread speed, and fire intensity characteristics are 0.33, 0.35, and 0.32, respectively;

[0075]

[0076] In formulas (3)-(5), FBC ij is the jth potential forest fire behavior characteristic of the i-th sample, namely, the fire probability (FGP), fire spread rate (ROS) and fire intensity (FRI) characteristics; N is the total number of samples; M is the number of types of potential forest fire behavior characteristics (M = 3); w j is the weight of each potential forest fire behavior characteristic;

[0077] Step 4.2: Comprehensive evaluation of forest fire danger level (FDG). The predicted ignition probability characteristics are divided into 5 levels using the natural break method, namely low, lower, medium, higher and high, and the values ​​are assigned as 1, 2, 3, 4 and 5 respectively; the forest fire danger level is calculated by weight using formula (6);

[0078] FDG=[0.33*FGP+0.35*ROS+0.32*FRI+0.5] (6)

[0079] Where, [] is the Gaussian rounding function; FRP, ROS, and FRP are the ignition probability, fire spread speed, and fire intensity characteristics, respectively, and the values ​​are 1, 2, 3, 4, and 5, corresponding to low, relatively low, medium, relatively high, and high forest fire risk levels, respectively;

[0080] Step 5: Forest fire danger level assessment integrating potential forest fire behavior characteristics of the neighborhood;

[0081] Specifically, it includes 3 parts:

[0082] Step 5.1: Determine the weight (W) of the influence of the forest fire danger level (FDG) of the neighboring pixel on the forest fire danger level of the central pixel. Drawing on the idea of ​​simulating forest fire spread based on the cellular automaton model, if a forest fire occurs in the neighboring pixel, then the central pixel may also be affected by the forest fire. Therefore, considering the Moore neighborhood, use formulas (7) and (8) to calculate the influence of the forest fire danger level of the neighboring pixel on the forest fire danger level of the central pixel t ij ;

[0083]

[0084]

[0085] Where i0 and j0 are the row index and column index of the center pixel respectively, ROS ij is the potential fire spread rate level of the neighborhood pixel ij, a is assumed to be the size of the pixel; when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; the pixel (i, j) should be a forest pixel;

[0086] Step 5.2: Update the forest fire risk level of the central pixel. Use formula (9) to combine the original FDG of the central pixel and the FDG of the neighboring pixels to update the forest fire risk level FDG' of the central pixel;

[0087]

[0088] Where [] is the Gaussian rounding function, is the updated forest fire risk level of the central pixel, is the original forest fire risk level of the central pixel, calculated by formula (5); when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; the pixel (i, j) should be a forest pixel;

[0089] Step 5.3: Map the forest fire danger level in the study area. Figure 4 The spatial distribution of the average forest fire risk level in Yunnan Province from February to April 2024 is given. The monthly average forest fire risk level is calculated by averaging the daily forest fire risk levels of the month according to their assigned values. Six typical forest fire cases that occurred in Yunnan Province during the 2024 fire season were used to verify the forest fire risk level assessment method proposed in the present invention. The results showed that the forest fire risk levels in these fire-occurring areas were all "medium" to "high", which shows that the present invention can better indicate the forest fire risk.

Claims

1. A method for assessing forest fire risk level taking into account the potential forest fire behavior characteristics of the neighborhood, comprising the following steps: Step 1: Extraction of forest fire risk factors; Extract the combustibles, meteorological, topographical and human activities risk factors for forest fire risk assessment in the target area; Step 2: Construction of forest fire behavior characteristics case library; Step 2.1: Extraction of forest fire behavior features; Based on the GFA fire dataset, GFA stands for Global Fire Atlas, the location information of the ignition point of each forest fire case in the study area is determined, and the corresponding fire spread speed is extracted; the maximum fire radiation power corresponding to the location of each forest fire ignition point is extracted to characterize the fire intensity characteristics; Step 2.2: Extract feature information of non-fire point pixels; Based on the GFA fire dataset, for each forest fire, the location and occurrence time information of other fire burning pixels except the fire point pixel are extracted; Step 2.3: Construct a case library of forest fire behavior characteristics for fire point pixels and non-fire point pixels; For the fire starting point pixels and non-starting point fire pixels, the characteristics of combustible materials, meteorology, terrain and human activities at the corresponding locations are extracted to build a forest fire behavior characteristic case library consisting of forest fire risk factors, ignition probability, fire spread speed and fire intensity; Step 3: Construction and evaluation of prediction model of potential forest fire behavior characteristics; Step 3.1: Balance the samples of forest fire behavior characteristic case library; The SMOTEENN algorithm is used to balance the samples in the case library of ignition probability, fire spread speed, and fire intensity characteristics to ensure that the number of samples in each category is equal. Step 3.2: Model construction and evaluation; Based on the forest fire behavior characteristic case library, the training set and test set are randomly divided; the nonlinear relationship between combustibles, meteorology, terrain and human activity factors and ignition probability, fire spread speed and fire intensity characteristics is explored using the random forest algorithm, and prediction models of latent ignition probability, fire spread speed and fire intensity are constructed based on their respective nonlinear relationships; a confusion matrix is ​​constructed, and the overall accuracy and Kappa coefficient are used to evaluate the performance of the prediction model; Step 4: Forest fire danger level assessment integrating potential forest fire behavior characteristics; Step 4.1: Determine the weights of potential fire behavior characteristics; Using the prediction model of ignition probability, fire spread speed, and fire intensity constructed in step 3, the ignition probability, fire spread speed, and fire intensity of all samples in the forest fire behavior characteristic case library are calculated respectively; using formulas 3, 4, and 5, the weights w of various potential forest fire behavior characteristics to represent the forest fire risk level are determined j ; In formula 3, 4, and 5, FBC ij is the jth potential forest fire behavior characteristic of the i-th sample, namely, the ignition probability, fire spread speed and fire intensity characteristics; P ij represents the element in the i-th row and j-th column of the probability matrix or the deviation matrix, e j represents the standard information entropy of the jth potential forest fire behavior feature; N is the total number of samples; M is the number of types of potential forest fire behavior features; w j is the weight of each potential forest fire behavior characteristic; Step 4.2: Comprehensively assess the forest fire risk level FDG; The predicted ignition probability characteristics are divided into 5 levels using the natural break method, namely low, lower, medium, higher and high, and the values ​​are assigned as 1, 2, 3, 4 and 5 respectively; the forest fire risk level is calculated by weight using formula 6; FDG=[w1*FGP+w2*ROS+w3*FRI+A] (6) Where, [] is the Gaussian rounding function; FRP, ROS, FRP are the ignition probability, fire spread speed, and fire intensity characteristics, respectively, corresponding to the low, relatively low, medium, relatively high, and high forest fire risk levels, and the corresponding values ​​are 1, 2, 3, 4, and 5, respectively; w1, w2, and w3 are the corresponding weights, and A is the correction coefficient; Step 5: Forest fire danger level assessment integrating potential forest fire behavior characteristics of the neighborhood; Specifically, it includes 2 parts: Step 5.1: Determine the weight W of the influence of the forest fire danger level of the neighboring pixel on the forest fire danger level of the central pixel; Formulas (7) and (8) are used to calculate the degree of influence of the forest fire danger level of the neighboring pixel on the forest fire danger level of the central pixel. ij ; Where i0 and j0 are the row index and column index of the center pixel respectively, ROS ij is the potential fire spread rate level of the neighborhood pixel ij, a is assumed to be the size of the pixel; when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; the pixel (i, j) should be a forest pixel; Step 5.2: Update the forest fire risk level of the central pixel; Using formula (9), the original FDG of the central pixel and the FDG of the neighboring pixels are combined to update the forest fire danger level FDI' of the central pixel; Where [] is the Gaussian rounding function, is the forest fire risk level of the central pixel, is the original forest fire risk level of the central pixel, calculated by formula (6); when the central pixel is located on the image boundary, the neighboring pixels beyond the image range are not considered; pixel (i, j) should be a forest pixel.

2. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The number of pixels in a single forest fire case in step 2.1 is ≥ 2, and a pixel is a set unit area.

3. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: In the step 2.3, the ignition probability of the fire pixel in the forest fire behavior characteristic case library is set to 1, and the ignition probability of the non-fire point pixel is set to 0.

4. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The fire spread speed and fire intensity characteristics in the forest fire behavior characteristic case library in step 2.3 are divided into five levels by the natural break method, namely low, lower, medium, higher, and high, and are assigned values ​​of 1, 2, 3, 4, and 5 respectively.

5. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The ratio of the training set to the test set in step 3.2 is 7:

3.

6. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The number of trees in the potential forest fire behavior characteristic prediction model based on the random forest algorithm in step 3.2 is 300, and the maximum depth is 10.

7. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The calculation method of the overall accuracy and Kappa coefficient in step 3.2 is: Among them, OA represents the overall accuracy, KC represents the Kappa coefficient, N is the total number of samples, and x ij is the value of the i-th row and j-th column in the confusion matrix, x i+ represents the total number of rows in the confusion matrix, x +i represents the total number of columns in the confusion matrix, * represents a multiplication operation, and r represents the number of rows in the confusion matrix.

8. A method for evaluating forest fire risk level taking into account the behavior characteristics of potential forest fires in the neighborhood as claimed in claim 1, characterized in that: The weights of the ignition probability, fire spread speed and fire intensity characteristics in step 4.1 are directly set to 0.33, 0.35 and 0.32.

Images