A mountain fire identification method and system fusing remote sensing and ecological process mechanism

Abstract

The application discloses a mountainous area fire identification method and system fusing remote sensing and ecological process mechanism, selects a forest fire disturbance sample point, then coarsely extracts a forest fire disturbance trace, and then finely extracts the forest fire disturbance trace; a plurality of optimal pairing groups of disturbance pixels and reference pixels are set according to the forest fire trace sample, a forest fire disturbance difference proportion index FDDR of a pairing point reference period is calculated, an FDDR threshold range setting meeting normal distribution is obtained through normality test, a regional forest fire disturbance spatial distribution map is generated, and finally the extracted forest fire trace is obtained; and a long-time-series forest fire disturbance spatial distribution map is output. The method can monitor long-term dynamics of forest fires in mountainous areas, and provides necessary support for forest fire management, forest carbon assessment and forest ecosystem recovery; the influence of factors such as complex terrain and vegetation response difference on fire identification is eliminated, and the identification precision of forest fires in mountainous areas is improved.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing detection technology for biomass combustion, and in particular to a method and system for identifying mountain fires that integrates remote sensing and ecological process mechanisms. Background Technology

[0002] Global climate change and human activities have led to a significant increase in the frequency and intensity of forest fires in recent years, with an increasingly wider impact. Catastrophic fires can negatively affect the function of forest ecosystems, such as enhancing the greenhouse effect, altering soil physicochemical properties, disrupting carbon and water cycles, and even degrading ecosystem services.

[0003] Effective identification of fire occurrences and estimation of forest burn area provide crucial scientific support for wildfire prevention, forest ecosystem protection, fire model validation, and land cover change assessment. However, traditional forest fire survey methods, such as field measurements and visual interpretation of aerial photographs, are only suitable for small-scale forest fire identification and location. Rapidly developing remote sensing technology provides strong support for long-term, large-scale forest fire identification. High-resolution remote sensing products can achieve accurate forest fire identification, but conducting long-term, high-resolution fire identification faces challenges such as high costs or data scarcity. Spectral index methods based on long-term, medium-resolution remote sensing products are widely used for identifying long-term dynamics of forest fires, with the normalized burn ratio (NBR) and differential normalized burn ratio (dNBR) increasingly becoming important indices for large-scale forest fire detection. However, this method struggles to eliminate the influence of topographical shadows and vegetation recovery capacity, resulting in less than ideal performance in complex mountainous landscapes.

[0004] To address the technical bottlenecks in identifying large-scale, long-term forest fires in mountainous areas, such as complex terrain and the difficulty in eliminating the impact of post-disaster climate on vegetation recovery, there is an urgent need to develop new methods for identifying large-scale, long-term forest fires in mountainous areas. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method and system for identifying mountain forest fires that integrates remote sensing and ecological process mechanisms. This method can eliminate the influence of factors such as terrain shadows and differences in vegetation recovery capacity in the identification of mountain forest fires, thereby improving the accuracy and speed of identifying mountain forest fires and further supporting the long-term dynamic assessment of carbon sinks in mountain forests.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The mountain fire identification method and system integrating remote sensing and ecological process mechanisms provided by this invention includes the following steps: Selecting forest fire disturbance sampling points: Defining the scope of forest fire disturbance based on forest fire trace samples; Coarse extraction of forest fire disturbance traces: Randomly select dNBR sequence samples that conform to a normal distribution, calculate the differential normalized burning ratio dNBR of the sample points, and determine the optimal threshold range of dNBR. Refined extraction of forest fire disturbance traces: Based on the forest fire trace sample, set up the best pairing group of multiple disturbance pixels and reference pixels, calculate the forest fire disturbance difference ratio index FDDR of the reference period of the paired points, and obtain the FDDR threshold range that meets the normal distribution through normality test. Determine the optimal FDDR threshold range, thereby confirming the optimal FDDR threshold range, and finally confirming it as a forest fire disturbance trace pixel. Generation of regional forest fire disturbance spatial distribution map: Using the sliding window algorithm, the FDDR of the coarsely extracted forest fire disturbance pixels is calculated based on the optimal threshold range of dNBR. The finely extracted forest fire disturbance trace is determined by combining the optimal threshold range of FDDR. Finally, the extracted forest fire trace is obtained. Based on the accuracy evaluation, output a long-term forest fire disturbance spatial distribution map that meets the preset requirements.

[0007] Furthermore, the Forest Fire Disturbance Difference Ratio Index (FDDR) is calculated according to the following formula:

[0008] In the formula, FDDR i It is the FDDR of the i-th month (-1 ≤ FDDR) i ≤ 0); EVI b,i and EVI u,i The EVIs correspond to the burning and unburned pixels in the i-th month, respectively.

[0009] Furthermore, the optimal threshold range of the dNBR is determined according to the following steps: 1) Extract a random sample set of size n from the forest fire interference sample space and extract the dNBR corresponding to the random sample set; 2) Perform a normality test on the sample group. If the dNBR of the random sample set follows a normal distribution, calculate the mean m and standard deviation δ of the random sample set. Otherwise, repeat the above steps. 3) Define multiple probability threshold ranges for the study area; 4) Determine the optimal dNBR probability threshold range using the dNBR efficiency index E; 5) Generate a coarse extraction distribution map of forest fire disturbance traces using the optimal dNBR probability threshold range. In the coarse extraction distribution map, the value of fire disturbance traces is 1, and the value of non-fire disturbance traces is 0.

[0010] Furthermore, the optimal threshold range for FDDR is determined according to the following steps: Multiple FDDR probability threshold ranges are defined based on the mean m and standard deviation δ of the FDDR distribution during the reference period; Calculate the FDDR efficiency index e using the following formula and determine the optimal threshold range for FDDR:

[0011] In the formula, m i and n i These correspond to the total number of pixels and the number of valid pixels in the i-th (i=1,2,3,...,n) FDDR threshold range, respectively; b i and e i These correspond to the effective pixel ratio and efficiency index of the i-th (i=1,2,3,...,n) FDDR threshold range, respectively; b max and e max These correspond to the largest values ​​of b and e, respectively; if e i When the value is close to 0, the corresponding FDDR threshold range exhibits optimal performance.

[0012] Furthermore, the generation of the spatial distribution map of forest fire disturbance in the region is determined according to the following steps: Using the sliding window algorithm: Set a window of size Cn×Cn pixels, and scan the entire coarsely extracted forest fire interference area from left to right and top to bottom with a step size of Cn; FDDR is obtained by averaging the EVI of the corresponding interference pixels and the EVI of the reference pixels randomly selected in each sliding window. If the FDDR (Further Data Count) of the coarsely extracted forest fire disturbance pixels is within the optimal FDDR threshold range, it is confirmed as the final forest fire disturbance pixel and assigned a value of 1; the remaining pixels are assigned a value of No Data. A spatial distribution map is then generated based on the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area. The specific calculation formula is as follows:

[0013] In the formula, BA i N represents the area affected by forest fire in year i; i is the number of interfering pixels identified by the paired point method in year i; P is the size of a single pixel, and CBA is the long-term cumulative forest fire interference area.

[0014] Furthermore, the optimal pairing group of the interfering pixel and the reference pixel is determined according to the following steps: In space, interference points and reference points are set up. Interference points refer to pixels that are affected by forest fires, and reference points refer to pixels that are not affected by forest fires within a certain distance. Correspondingly, an interference period and a reference period are set in terms of time. The interference period refers to a certain period after the forest fire interference, and the reference period refers to the period before the forest fire interference. Based on the normality test, determine whether it meets the normal distribution. If the FDDR of each paired point in the reference period follows a normal distribution, then determine the FDDR threshold range of the study area; otherwise, reselect the FDDR paired point group.

[0015] Furthermore, the optimal threshold range for FDDR is determined according to the following steps: The FDDR probability threshold range is defined based on the mean m and standard deviation δ of the FDDR distribution during the reference period; The optimal threshold range for FDDR is determined using the FDDR efficiency index e; The final forest fire disturbance sites are identified by calculating the average FDDR value of the coarsely extracted forest fire disturbance site pixels and comparing it with the optimal FDDR threshold range. During the disturbance period, if the average FDDR value of the coarsely extracted fire disturbance site pixels falls within the optimal FDDR threshold range, it is finally identified as a forest fire disturbance site pixel and assigned a value of 1, while the remaining pixels are assigned a value of No Data. A spatial distribution map is generated based on the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area.

[0016] Furthermore, the accuracy evaluation is performed according to the following steps: Various levels of precision were calculated and extracted from forest fire disturbance areas using standard methods. Randomly select forest burning pixels and non-forest burning pixels from the spatial distribution map of forest fire interference, and calculate various levels of accuracy by constructing a confusion matrix; The RMSE method is compared with the standard method to obtain overall accuracy and pixel accuracy.

[0017] The mountain fire identification system that integrates remote sensing and ecological process mechanisms provided by the present invention includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the above-mentioned method.

[0018] The beneficial effects of this invention are as follows: This invention provides a method and system for identifying mountain forest fires that integrates remote sensing and ecological process mechanisms. It is applicable to the identification of forest fires in mountainous areas over long time periods and large areas. It eliminates the influence of factors such as terrain shadows and differences in vegetation recovery capacity in the identification of forest fires in mountainous areas, thereby improving the accuracy and speed of forest fire identification in mountainous areas and further supporting the long-term dynamic assessment of forest carbon sinks in mountainous areas.

[0019] The method provided by this invention is for monitoring the long-term dynamics of forest fires in mountainous areas. It can effectively and stably identify the long-term disturbance area of ​​forest fires in mountainous areas on a large spatial scale. This provides necessary support for forest fire management, forest carbon assessment, and forest ecosystem restoration. The paired-point method provided by this invention constructs a new forest fire disturbance difference ratio index (FDDR) to optimize the traditional fire difference index (dNBR), eliminating the influence of factors such as complex terrain and vegetation response differences to climate on fire identification, and improving the identification accuracy of forest fires in mountainous areas.

[0020] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0021] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 Flowchart of a method for identifying mountain fires that integrates remote sensing and ecological process mechanisms; Figure 2 This is a schematic diagram of FDDR calculation using the sliding window method; Figure 3 This is a diagram comparing the effects of the three methods; Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0023] Example 1

[0024] like Figure 1 As shown in the figure, this embodiment provides a method for identifying mountain fires by integrating remote sensing and ecological process mechanisms, which includes the following steps: Selection of forest fire disturbance sampling points: Based on multi-source data and visual interpretation methods, forest fire traces samples from the years in which forest fires occurred were selected, and the scope of forest fire disturbance was roughly defined. Coarse extraction of forest fire disturbance traces: Randomly select dNBR sequence samples that conform to a normal distribution, calculate the normalized burn ratio (NBR) of the disturbance year and the year before the disturbance, calculate the differential normalized burn ratio (dNBR) of the sample points, and determine the optimal threshold range of dNBR. The optimal threshold range of dNBR is used to generate a coarse extraction distribution map of forest fire disturbance traces. In the coarse extraction distribution map, the value of fire disturbance traces is 1, and the value of non-fire disturbance traces is 0. Refined extraction of forest fire disturbance traces: The forest fire disturbance difference ratio index (FDDR) is calculated, and paired point groups are set up. The paired point groups are obtained according to the following steps: In space, interference points and reference points are set up. Interference points refer to pixels that are affected by forest fires, and reference points refer to pixels that are not affected by forest fires within a certain distance. Correspondingly, an interference period and a reference period are set in terms of time. The interference period refers to a certain period after the forest fire interference, and the reference period refers to the period before the forest fire interference. Multiple pairs of interference and reference pixels were set up based on the forest fire traces.

[0025] Based on the normality test, determine whether it meets the normal distribution. If the FDDR of each paired point in the reference period follows a normal distribution, then determine the FDDR threshold range of the study area; otherwise, reselect the FDDR paired point group. The FDDR probability threshold range is defined based on the mean m and standard deviation δ of the FDDR distribution during the reference period; The optimal threshold range for FDDR is determined using the FDDR efficiency index e; The final forest fire disturbance traces are identified by calculating the average value of the coarsely extracted forest fire disturbance trace pixels and comparing it with the optimal threshold range of FDDR. During the disturbance period, if the average value of the coarsely extracted fire trace pixels falls within the optimal threshold range of FDDR, they are finally identified as forest fire disturbance trace pixels.

[0026] Generation of regional forest fire disturbance spatial distribution map: Using the sliding window algorithm, based on the coarsely extracted burning pixels, the final forest fire disturbance traces are determined by combining the optimal threshold range of FDDR. If the FDDR of the coarsely extracted forest fire disturbance trace pixels is within the optimal threshold range of FDDR, it is confirmed as the final forest fire disturbance trace and assigned a value of 1, while the remaining pixels are assigned a value of No Data. A spatial distribution map is generated based on the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area.

[0027] The specific steps of the district fire identification method are described in detail below: Step 1: Data Preparation (1) Download the MODIS surface radiation data product dataset (MODIS MOD09A1, spatial resolution 500 m) and land cover data product dataset (MODIS MCD12Q1, spatial resolution 500 m) for the growing season (May to October) of the year of fire disturbance and the reference year (the year before the disturbance); download the SRTM DEM data product (SRTMDEMUTM, spatial resolution 90 m) for the study area; download the monthly time series precipitation and temperature data of national standard meteorological stations in the study area; download the Landsat satellite image data (spatial resolution 30 m) for the study area; download the vegetation change dataset (Tree Cover Loss, spatial resolution 30 m) for the study area; download the dynamic fire point data (MCD14DL, spatial resolution 1 km) for the study area.

[0028] (2) Using the ArcGIS resampling tool, the spatial resolution of the DEM data of the study area was resampled to 500 m, and the precipitation and temperature data of the study area were resampled to 500 m.

[0029] Step Two: Identification of Forest Fire Disturbance (3) Selection of forest fire interference samples. Using a combination of multi-source data (such as news reports, Landsat TM, Google Earth imagery, and MCD12Q1 and MCD14DL), forest fire traces from the years in which the forest fires occurred were selected through visual interpretation, and the extent of forest fire interference was roughly defined. These forest fire interference sample points will be used in subsequent pre-extraction and refined paired-point extraction of forest fire interference.

[0030] (4) Coarse extraction of forest fire disturbance areas. Based on the MOD09A1 surface emissivity product, the normalized burn ratio (NBR) for every 8 days of the growing season in the year of disturbance and the year preceding the disturbance (reference year) was calculated, as shown in Equation (1-1). The differential normalized burn ratio (dNBR) for the sample points was further calculated, as shown in Equation (1-2):

[0031] In the formula, NIR and SWIR3 represent the reflectivity of the near-infrared and short-wave infrared bands, respectively; NBRpre and NBRpost represent the NBR before and after the fire, respectively.

[0032] According to the Standard Method for Forest Fire Identification (hereinafter referred to as the "Standard Method"), the dNBR value of the burned area is below 1350 (dNBR of the burned area magnified 1000 times). The optimal threshold range for dNBR is further determined through the following steps: 1) Extract a random sample set of size n from the forest fire interference sample space and extract the dNBR corresponding to the random sample set; 2) Use the Kolmogorov-Smirnov (KS) test or Shapiro-Wilk (SW) test to test the normality of the sample group. If the dNBR of the random sample set follows a normal distribution (significance greater than 0.05), calculate its mean (m) and standard deviation (δ). Otherwise, repeat the above steps. 3) Define the five probability threshold ranges for the study area as: [m-2.5δ, 1350], [m-2δ, 1350], [m-1.5δ, 1350], [m-δ, 1350], [m-0.5δ, 1350]; 4) Use the dNBR efficiency index (E) as shown in equations (3) and (4) to determine the optimal dNBR probability threshold range; 5) Generate a coarse extraction distribution map of forest fire disturbance traces using the optimal dNBR probability threshold range. In the coarse extraction distribution map, the value of fire disturbance traces is 1, and the value of non-fire disturbance traces is 0.

[0033] The dNBR efficiency index (E) can be calculated from the effective number of pixels (N) and the total number of pixels (M) for each dNBR threshold range. Furthermore, the effective pixel ratio (B) and efficiency index (E) for each dNBR threshold range can be calculated. The smallest dNBR efficiency index E is considered the optimal index, as shown in equations (1-3)-(1-4):

[0034] In the formula, M i and N i These correspond to the total number of pixels and the number of effective pixels in the i-th (i=1,2,3,...,n) dNBR threshold range, respectively; Bi and Ei correspond to the effective pixel ratio and efficiency index in the i-th (i=1,2,3,...,n) dNBR threshold range, respectively; B max and E max These correspond to the largest B and E respectively; if E i When E equals 0, the corresponding dNBR threshold range exhibits optimal performance; in practice, E can be... i Set it to a value close to 0, such as 0.1, or a smaller number; (5) Refined Extraction of Forest Fire Disturbance Areas: The paired-point method, based on the Forest Fire Disturbance Differential Ratio Index (FDDR) calculated using the Enhanced Vegetation Index (EVI), can further improve the accuracy of coarse extraction of forest fire disturbance areas. The paired-point method spatially sets up disturbance points and reference points. Disturbance points refer to pixels affected by forest fire disturbance, while reference points refer to pixels within a certain distance that are not affected by forest fire disturbance. Correspondingly, temporal disturbance periods and reference periods are set up. The disturbance period refers to a certain period after the forest fire disturbance, while the reference period refers to the period before the forest fire disturbance.

[0035] Based on the forest fire traces identified in step (4), multiple paired groups of burning (interference) and unburned (reference) pixels are established. The rules for setting up the paired groups are as follows: 1) The distance between unburned pixels and burned pixels should be within 1 to 3 pixels to ensure that paired pixels have more than 80% similarity in climate, soil and vegetation type; in practice, the unburned pixels should be close enough to the burned pixels. 2) The distance between the burning pixel and the core area of ​​combustion is within the range of 1 to 3 pixels. In practice, the burning pixel should be close to the core area of ​​combustion to avoid selecting pixels that are not completely burned. 3) Paired groups of burned and unburned pixels should satisfy α>0.05 in terms of monthly average precipitation, monthly average temperature, and average DEM; and satisfy the condition that there are no significant differences; where α represents the significance level; 4) During the reference period, the EVI between burning and unburned pixels satisfies α>0.05, indicating no significant difference, and they have similar EVI time-series change trends; after forest fire disturbance, the EVI difference between burning and unburned pixels becomes larger, and the EVI of burning pixels is lower than that of unburned pixels.

[0036] EVI is calculated using equation (1-5) based on the MOD09A1 surface radiation product.

[0037]

[0038] In the formula, NIR, RED, and BLUE represent the surface reflectance of the near-infrared, visible red light, and blue light bands, respectively, corresponding to the 2nd, 1st, and 3rd bands in the MOD09A1 surface radiation product.

[0039] The calculation of the forest fire disturbance difference ratio index (FDDR) is shown in equation (1-6):

[0040] In the formula, FDDR iIt is the FDDR of the i-th month (-1 ≤ FDDR) i ≤ 0); EVI b,i and EVI u,i The EVIs of the burning and unburned pixels in the i-th month, respectively; According to the normality test of Kolmogorov-Smirnov (KS) test or Shapiro-Wilk (SW) test, if the monthly FDDR of each paired point in the reference period follows a normal distribution (significance greater than 0.05), it can be used to determine the FDDR threshold range of the study area; otherwise, the FDDR paired point group should be reselected.

[0041] Similar to the determination of the dNBR threshold range in (4), five FDDR probability threshold ranges can be defined based on the mean (m) and standard deviation (δ) of the FDDR distribution during the reference period: [-1, m-3δ], [-1, m-2.5δ], [-1, m-2δ], [-1, m-1.5δ], [-1, m-δ].

[0042] Similarly, the optimal threshold range for FDDR can be determined using the FDDR efficiency index (e), as shown in equations (1-7)-(1-8), and the formulas are as follows:

[0043] In the formula, m i and n i The total number of pixels and the number of valid pixels correspond to the i-th (i=1,2,3,...,n) FDDR threshold range, respectively; b i and e i The effective pixel ratio and efficiency index correspond to the i-th (i=1,2,3,...,n) FDDR threshold range, respectively; b max and e max These correspond to the largest values ​​of b and e, respectively; if e i When the value is 0, the corresponding FDDR threshold range exhibits optimal performance. In fact, e can be... i Set it to a value close to 0, such as 0.1, or a smaller number.

[0044] Furthermore, by calculating the average monthly FDDR of the coarsely extracted forest fire disturbance plot pixels and comparing it with the optimal FDDR threshold range, the final forest fire disturbance plots can be confirmed. During the disturbance period, if the average FDDR of the coarsely extracted fire plot pixels falls within the optimal FDDR threshold range, they are ultimately confirmed as forest fire disturbance plot pixels.

[0045] Step 3: Generation of Spatial Distribution Map of Forest Fire Disturbance Sites (6) The spatial distribution map of regional forest fire interference is generated to extract forest fire traces. Using the sliding window algorithm, based on the coarsely extracted burning pixels, the final forest fire interference traces are determined by combining the optimal threshold range of FDDR in step (5). A window with a size of Cn×Cn pixels is set, and the entire coarsely extracted forest fire interference traces are scanned sequentially from left to right and from top to bottom with a step size of Cn. FDDR is obtained by calculating and averaging the EVI of the corresponding burning (interference) pixels and the EVI of the randomly selected unburned (reference) pixels in each sliding window, such as... Figure 2 As shown, Figure 2 This is a schematic diagram of FDDR calculation using the sliding window method.

[0046] If the FDDR of the forest fire disturbance trace pixels extracted in step (4) is within the optimal FDDR threshold range, it is confirmed as the final forest fire disturbance trace and assigned a value of 1; the remaining pixels are assigned the value "No Data". A spatial distribution map is generated based on the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area, as shown in equations (1-9) to (1-10). The calculation of the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area is shown in equations (1-9) and (1-10):

[0047] In the formula, BA i N represents the area affected by forest fire in year i (in hectares); i P represents the number of interfering pixels identified by the paired-point method in year i; P is the size of a single pixel (500m in this case). CBA represents the long-term cumulative forest fire disturbance area (in hectares).

[0048] Step 4: Accuracy Evaluation (7) Accuracy Evaluation. The RMSE method is evaluated using three methods: overall accuracy, pixel accuracy, and field verification. The RMSE method is compared with the standard method to obtain overall accuracy and pixel accuracy. The key steps of the standard method for extracting forest fire disturbance areas include: 1) Calculate the NBR and dNBR of the interference period and reference period using MOD09A1 data; 2) Select the pixels in the range [100, 1300] of dNBR as the forest fire area.

[0049] A large number of forest-burning and non-forest-burning pixels were randomly selected from the spatial distribution maps of forest fire interference generated by the two methods, and their locations were marked on Landsat imagery, Google Earth imagery, MCD14DL data products, and the TreeCover Loss dataset. The correctness of the burning and non-burning pixels was then confirmed by visual interpretation. The confusion matrix was determined by calculating the overall accuracy, user accuracy, production accuracy, and Kappa coefficient.

[0050] Pixel accuracy was assessed based on fire cases reported by local forestry authorities and forest fire records obtained from Landsat imagery. Fire boundaries corresponding to forest fire events identified visually during periods of interference were compared with fire boundaries identified using both methods. Effective pixels, invalid pixels, spatial integrity, and forest fire continuity were assessed for each forest fire event.

[0051] Field verification further evaluated the accuracy of the two methods: First, representative plots were selected from the spatial distribution map of forest fire disturbance, each containing pixels to be verified. Then, a field survey was conducted to confirm whether the pixels to be verified were indeed forest fire-affected area pixels, based on vegetation and soil characteristics of the fire-affected areas and interview records with local forestry professionals.

[0052] Example 2

[0053] This embodiment uses the 2010 forest fire traces in Southwest my country as an example to extract forest fire tracks. Southwest my country is one of the hardest-hit areas by forest fires in my country. The Southwest region is characterized by rugged terrain, with altitudes ranging from 46 to 7053 meters and slopes from 0° to 64.24°. The mountain fire identification method integrating remote sensing technology and ecological mechanisms includes the following steps: 1. Data Collection Data on climate, land cover, vegetation, topography, fire disturbance, and surface radiation for the study area were used in the example (Table 2-1). Download the MODIS surface radiation data product dataset (MODIS MOD09A1) and land cover data product dataset (MODIS MCD12Q1) for the study area from 2001 to 2017; download the SRTM DEM data product (SRTMDEMUTM) for the study area; download the monthly time-series precipitation and temperature data from national standard meteorological stations within the study area from 2002 to 2017; download Landsat satellite imagery data for the study area in 2010 and 2017; download the vegetation change dataset (Tree Cover Loss) for the study area in 2010; download the annual dynamic fire point data for the study area in MCD14DL (2010, 2017), FireCCI51 (2010, 2017), and MODIS MCD64A1 (2010, 2017).

[0054] In ArcGIS, MOD09A1 surface radiance product data for the growing season (May to October) of 2009 and 2010 in Southwest China were stitched together and cropped out every 8 days.

[0055] Table 2-1. Data Sources.

[0056] 2. Rough extraction of forest fire disturbance traces (1) Combining news reports, Landsat remote sensing images, Google Earth, MODIS MOC64A1 and MCD14DL data, 19 sets of forest fire traces in Southwest China in 2010 were identified by visual interpretation, as shown in Table 2-2.

[0057] (2) Calculate the growing season (May to October) NBR based on MOD09A1 surface radiation product data (multiply the calculation results by 1000 times), as shown in Equation (2-1); and synthesize the NBR spatial distribution maps for the growing seasons of 2009 and 2010 according to the mean distribution. Calculate dNBR according to Equation (2-2) and generate the dNBR spatial distribution map for 2009-2010.

[0058] (3) Combine the 2009 forest spatial distribution map (Categories 1-9 of MCD12Q1 data products) to obtain the dNBR spatial distribution map of the forest area in 2010.

[0059]

[0060] In the formula, NIR and SWIR3 represent the reflectivity of the near-infrared and short-wave infrared bands, respectively; NBRpre and NBRpost represent the NBR before and after the fire, respectively.

[0061] Table 2-2 Information on forest fire smuggling sites in Southwest China in 2010

[0062] (4) Since Guizhou has the lowest forest density compared to the other three regions, dNBR threshold ranges were set for forest fire-affected areas in Guizhou and Yunnan-Sichuan-Chongqing regions, respectively. 343 and 26 forest fire pixels were selected in Guizhou and Yunnan-Sichuan-Chongqing regions, respectively, and their dNBR values ​​for 2010 were extracted. The KS test and SW test were used to test the normality of the two sets of data (Table 2-3). The tests showed that both sets of data conformed to a normal distribution.

[0063] Table 2-3 dNBR Normality Test Results for Yunnan, Sichuan, Chongqing and Guizhou Regions

[0064] Note: α = 0.05; (5) Define the five probability threshold ranges for the study area as: [m-2.5δ, 1350], [m-2δ, 1350], [m-1.5δ, 1350], [m-δ, 1350], [m-0.5δ, 1350]. According to the definition, the dNBR threshold range for forest fire scars in the Yunnan-Sichuan-Chongqing region is [205, 1350], [160, 1350], [115, 1350], [70, 1350], [25, 1350]; and the dNBR threshold range for forest fire scars in the Guizhou region is [112, 1350], [101, 150], [89, 1350], [78, 1350], [66, 1350].

[0065]

[0066] (6) Six groups of forest fire-affected areas of equal size were divided by visual interpretation, including four groups in the Yunnan-Sichuan-Chongqing region and two groups in the Guizhou region. Areas 1 and 2 were forest fire-affected areas in Guizhou, while areas 3, 4, and 5 were forest fire-affected areas in the Yunnan-Sichuan-Chongqing region. The dNBR efficiency index for the Yunnan-Sichuan-Chongqing region and the Guizhou region was calculated using equations (2-3) and (2-4). Ultimately, the optimal dNBR threshold ranges for the initial forest fire-affected area pixels in the Yunnan-Sichuan-Chongqing region and the Guizhou region were [160, 1350] and [112, 1350], respectively. Based on the optimal dNBR threshold ranges, the forest fire-affected areas in Southwest China were coarsely extracted, resulting in a coarsely extracted distribution map of the forest fire-affected area sample areas in Southwest China.

[0067] 3. Refined extraction of forest fire disturbance traces (1) Based on MCD12Q1 land cover data and Landsat TM imagery combined with EVI data, a total of 9 pairs of forest fire interference pairing points were selected.

[0068] (2) Calculate the FDDR of each pair of paired sample points during the growing season (May to October) of 2009 and 2010 according to Equation (2-5). The FDDR of the paired sample points was tested and found to conform to a normal distribution with a mean of 4.22% and a standard deviation of 6.34% (p>0.05).

[0069] (3) Based on the five defined FDDR probability threshold ranges: [-1, m-3δ], [-1, m-2.5δ], [-1, m-2δ], [-1, m-1.5δ], [-1, m-δ], the FDDR threshold ranges are determined to be: [-1, -2.12%], [-1, -5.30%], [-1, -8.47%], [-1, -11.64%] and [-1, -14.81%].

[0070] (4) Calculate the FDDR efficiency index according to equations (2-6) and (2-7). Finally, the optimal FDDR threshold range for the Southwest region is determined to be [-1, -8.47%].

[0071]

[0072] (5) Set a sliding window with a window size of 5km×5km, calculate the FDDR of the pre-extracted forest fire interference pixels in 2010, and calculate their average value. If the FDDR of the forest fire interference pixels falls within [-1, -8.47%], it is finally determined to be a forest fire interference pixel. Based on this, a spatial distribution map of forest fire traces in Southwest China in 2010 is generated, and a finely extracted spatial distribution map of forest fire traces in Southwest China in 2010 is obtained.

[0073] 4. Accuracy Evaluation (1) Extract forest fire traces according to standard methods. The key steps of extracting forest fire disturbance traces according to standard methods include: 1) Calculate the NBR of the disturbance period and the reference period and the dNBR of the disturbance period using MOD09A1 data; 2) Select the pixels with dNBR in the interval [100, 1300] as the forest fire area.

[0074] (2) 1100 forest fire interference pixels and 1100 non-forest fire interference pixels were randomly selected in the study area. The overall accuracy, production accuracy, user accuracy, and Kappa coefficient were calculated by constructing a confusion matrix, as shown in Table 2-4: Table 2-4. Accuracy Evaluation of RSEM and Standard Methods

[0075] (3) Six forest fire sites were selected in 2010 and 2017 through visual interpretation. The pixel accuracy of forest fire site identification and extraction by different methods was evaluated, such as... Figure 3 As shown, the accuracy of RSEM reached 81.16%; Figure 3 This diagram illustrates the comparison of the effects of the three methods. Figure 3 (a) in the image shows the fire scene marked on Landsat. Figure 3 (b) shows the recognition effect of the RSEM method (the blue area is the pixel identified by RSEM, and the red line area is the actual fire trace area). Figure 3 (c) shows the results of the standard method (the black area represents the pixels identified by the standard method, and the red line area represents the actual fire area). (a) Landsat visual interpretation, (b) RSEM method, (c) standard method. The six fire sites are as follows: Site #1 (102°37′23″E, 24°6′36″N), Site #2 (100°22′17″E, 25°40′16″N), Site #3 (101°16′47″E, 30°11′09″N) are fire sites from 2010; Site #4 (101°09′07″E, 30°22′08″N), Site #5 (99°43′23″E, 26°47′13″N), and Site # (100°33′32″E, 28°32′42″N) are fire sites from 2017. Generate spatial distribution maps of forest fire disturbance: Forest fires are identified using the RSEM method, and spatial distribution maps of forest fire disturbance from 2002 to 2017 are generated, including spatial distribution maps of forest fire disturbance in Southwest China from 2002 to 2017, specifically for 2002 to 2005, 2006 to 2009, 2010 to 2013, and 2014 to 2017.

[0076] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A method for identifying mountain fires that integrates remote sensing and ecological process mechanisms, characterized in that: Includes the following steps: Selecting forest fire disturbance sampling points: Defining the scope of forest fire disturbance based on forest fire trace samples; Coarse extraction of forest fire disturbance traces: Randomly select dNBR sequence samples that conform to a normal distribution, calculate the differential normalized burning ratio dNBR of the sample points, and determine the optimal threshold range of dNBR. Refined extraction of forest fire disturbance traces: Based on the forest fire trace sample, set up the best pairing group of multiple disturbance pixels and reference pixels, calculate the forest fire disturbance difference ratio index FDDR of the reference period of the paired points, and obtain the FDDR threshold range that meets the normal distribution through normality test. Determine the optimal FDDR threshold range, thereby confirming the optimal FDDR threshold range, and finally confirming it as a forest fire disturbance trace pixel. Generation of regional forest fire disturbance spatial distribution map: Using the sliding window algorithm, the FDDR of the coarsely extracted forest fire disturbance pixels is calculated based on the optimal threshold range of dNBR. The finely extracted forest fire disturbance trace is determined by combining the optimal threshold range of FDDR. Finally, the extracted forest fire trace is obtained. Based on the accuracy evaluation, output a long-term forest fire disturbance spatial distribution map that meets the preset requirements; The generation of the spatial distribution map of forest fire disturbance in the region is determined according to the following steps: Using the sliding window algorithm: Set a window of size Cn×Cn pixels, and scan the entire coarsely extracted forest fire interference area from left to right and top to bottom with a step size of Cn; FDDR is obtained by averaging the EVI of the corresponding interference pixels and the EVI of the reference pixels randomly selected in each sliding window. If the FDDR (Further Data Count) of the coarsely extracted forest fire disturbance pixels is within the optimal FDDR threshold range, it is confirmed as the final forest fire disturbance pixel and assigned a value of 1; the remaining pixels are assigned a value of No Data. A spatial distribution map is then generated based on the annual forest fire disturbance area and the long-term cumulative forest fire disturbance area. The specific calculation formula is as follows: In the formula, BA i N represents the area affected by forest fire in year i; i P represents the number of interfering pixels identified by the paired point method in year i; P is the size of a single pixel, and CBA is the long-term cumulative forest fire interference area. The optimal pairing group of the interfering pixel and the reference pixel is determined according to the following steps: In space, interference points and reference points are set up. Interference points refer to pixels that are affected by forest fires, and reference points refer to pixels that are not affected by forest fires within a certain distance. Correspondingly, an interference period and a reference period are set in terms of time. The interference period refers to a certain period after the forest fire interference, and the reference period refers to the period before the forest fire interference. Based on the normality test, determine whether it meets the normal distribution. If the FDDR of each paired point in the reference period follows a normal distribution, then determine the FDDR threshold range of the study area; otherwise, reselect the FDDR paired point group.

2. The mountain fire identification method integrating remote sensing and ecological process mechanisms as described in claim 1, characterized in that: The Forest Fire Disturbance Differential Ratio Index (FDDR) is calculated according to the following formula: In the formula, FDDR i It is the FDDR of the i-th month; EVI b,i and EVI u,i The EVIs correspond to the burning and unburned pixels in the i-th month, respectively.

3. The mountain fire identification method integrating remote sensing and ecological process mechanisms as described in claim 1, characterized in that: The optimal threshold range for the dNBR is determined according to the following steps: 1) Extract a random sample set of size n from the forest fire interference sample space and extract the dNBR corresponding to the random sample set; 2) Perform a normality test on the sample group. If the dNBR of the random sample set follows a normal distribution, calculate the mean m and standard deviation δ of the random sample set. Otherwise, repeat the above steps. 3) Define multiple probability threshold ranges for the study area; 4) Determine the optimal dNBR probability threshold range using the dNBR efficiency index E; 5) Generate a coarse extraction distribution map of forest fire disturbance traces using the optimal dNBR probability threshold range. In the coarse extraction distribution map, the value of fire disturbance traces is 1, and the value of non-fire disturbance traces is 0.

4. The mountain fire identification method integrating remote sensing and ecological process mechanisms as described in claim 1, characterized in that: The accuracy evaluation is performed according to the following steps: Various levels of precision were calculated and extracted from forest fire disturbance areas using standard methods. Randomly select forest burning pixels and non-forest burning pixels from the spatial distribution map of forest fire interference, and calculate various levels of accuracy by constructing a confusion matrix; The RMSE method is compared with the standard method to obtain overall accuracy and pixel accuracy.

5. A mountain fire identification system integrating remote sensing and ecological process mechanisms, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 4.

Images