A fire point inversion method based on object-oriented and local extreme value sliding search

Abstract

This invention discloses a fire origin inversion method based on object-oriented and local extremum sliding search. The method includes: optimizing the spatiotemporal attributes of burn sites using MODIS burn site and VIIRS fire point remote sensing products; constructing a corresponding burn date grid based on the optimized burn sites; obtaining burn site boundary coding data based on object-oriented thinking; and using a local extremum sliding search algorithm to perform a local minimum iterative search on each coded object to determine the fire origin location and burn date. This method can quickly and automatically identify the spatiotemporal attributes of large-scale fire origins, improving the monitoring capabilities of large-scale fires and providing strong data support for fire origin attribution.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing geoscience application optimization technology, and in particular to a fire point inversion method based on object-oriented and local extreme value sliding search. Background Art

[0002] In 2021, the IPCC released its climate science report, "Climate Change 2021: The Physical Science Basis," which shows that climate change has already affected extreme weather and climate events worldwide. Global warming has led to frequent combined heat and drought events, altering the fire disturbance conditions on the land surface. Fire-prone areas and biomass burning intensity are increasing. Each year, an area equivalent to the size of the European Union is burned by fire. According to the fourth edition of the Global Fire Emissions Database, global biomass burning averaged 2.2 billion tons of carbon per year between 1997 and 2016. Fire disturbances, enhanced by climate change, have altered surface properties and energy fluxes (such as evapotranspiration, albedo, and vegetation composition), which in turn have a significant feedback effect on climate change through the carbon cycle. As more and more governments transform "carbon neutrality" into national strategies, large-scale fire dynamic monitoring and fire attribution research have become important topics. Given the important role of global fire disturbances on climate, environment, and human systems, accurately understanding the spatiotemporal properties of fire points in large-scale burn areas is crucial for scientifically assessing the causes of fires. It can also provide scientific reference for forest and agricultural fire monitoring, evaluation, and management.

[0003] Currently, spatiotemporal data on the origins of large burns is scarce, severely hindering the global classification of fire causes. Furthermore, obtaining fire information based on traditional field surveys is extremely limited and costly. A key advantage of satellite remote sensing is its strong detection capability, making it relatively easy and cost-effective to capture surface data over large areas. Multi-source Earth observation satellites regularly observe fire disturbances, and the continuously updated and historically archived global-scale remotely sensed fire disturbance products make it possible to infer the origins of large burns. Existing research has demonstrated that fire-affected pixels can be clustered based on spatial and temporal proximity. In the vast grassland biomes, boreal forests, and tundra ecosystems worldwide, due to their remoteness, one or a few fire sources often burn unchecked, resulting in numerous large burns. However, existing clustering algorithms that rely on spatiotemporal similarity between pixels are limited in their inability to separate individual fire sources within large burns containing multiple origins, hindering fire monitoring. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a fire point inversion method based on object-oriented and local extreme value sliding search. By combining mainstream remote sensing burn sites and fire point products, different burn sites are coded one by one based on object-oriented thinking, and the fire points are traversed and searched using a dynamically adjusted search window radius, thereby quickly and automatically identifying the spatiotemporal attributes of large-scale fire starting points, thereby improving the monitoring capability of large-scale fires.

[0005] Technical solution: The fire point inversion method based on object-oriented and local extreme value sliding search described in the present invention specifically includes the following steps:

[0006] (1) Determine the study area, read the original data of multi-source remote sensing fire products and perform preprocessing; the multi-source remote sensing fire products include: MODIS annual fire scar product MCD64A1, VIIRS daily fire point product VNP14IMG; the preprocessing is to add a daily value index DOY to the VNP14IMG data and assign values ​​based on the date field;

[0007] (2) Construct a spatial grid of size m × m, and use the spatial relationship between VNP14IMG and MCD64A1 to screen out VNP14IMG fire points within the n-meter buffer zone of the MCD64A1 burn scar. After merging, perform spatial grid intersection analysis to obtain the spatially optimized burn scar grid.

[0008] (3) Count the minimum DOY values ​​of MCD64A1 and VNP14IMG in each grid of the burn site as the grid value; identify the grid without assigned grid value and search for the VNP14IMG data closest to the grid without assigned grid value, and assign the DOY value in the VNP14IMG data to the grid without assigned grid value; convert all DOY fields into raster data to obtain the burning date raster; MCD64A1 data contains DOY field;

[0009] (4) Convert the burned area grid into raster data and perform binarization processing, perform single-pixel mathematical morphological expansion on the binary image, and then perform unique value encoding on the expanded raster to obtain the boundary coding data of the burned area;

[0010] (5) For each coded object in the boundary coding data of the burn site, a sliding radius window is used to perform an iterative search for the local minimum value on the burning date grid on each coded object in turn, and the pixel coordinates corresponding to the local minimum value obtained by the search are the coordinates of the ignition point; if multiple identical local minimum values ​​are obtained, the centroid positions of adjacent pixels are calculated as the coordinates of the ignition point; when encountering a large burn site, the search center of the sliding radius window will automatically split and the local minimum value iterative search will be performed on the burning date grid on each coded object in turn; the large burn site is a burn site whose range exceeds the range set by the sliding radius window.

[0011] Furthermore, quality judgment is performed based on the acquired fire point coordinate data, that is, the erroneous fire points with spatial standard deviations exceeding the set threshold range and caused by broken spots are removed to obtain the final fire point coordinates and combustion date.

[0012] Furthermore, the n-meter buffer zone described in step (2) is specifically set to a 5 km buffer zone.

[0013] Furthermore, in the process of obtaining the grid value in step (3), the original DOY value of MCD64A1 outside the VNP14IMG data neighborhood of 1 km needs to be retained.

[0014] Furthermore, the unique value encoding of the expanded grid in step (4) specifically includes grid segmentation based on a four-neighborhood connection rule, and unique value encoding of the grid connected by the four neighborhoods.

[0015] Furthermore, the local sliding window extreme value search in step (5) is developed based on the Python platform, and multiple arrays are constructed to store and determine the local minimum values ​​within the sliding window.

[0016] Furthermore, the centroid position in step (5) is calculated specifically by performing a weighted average operation on the pixels corresponding to multiple identical local minimum values.

[0017] The present invention also provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program. A computer-readable storage medium having the computer program stored thereon, wherein the computer program implements the steps of the above method when executed by the processor, is provided.

[0018] Beneficial effects: Compared with the existing technology, the method of the present invention has the following significant advantages: by combining the MODIS burn site MCD64A1 and VIIRS fire point VNP14IMG remote sensing products, the spatiotemporal attributes of the burn site are optimized; based on the object-oriented concept and the local extreme value sliding search algorithm, local sliding searches are performed one by one for different coded objects, achieving accurate inversion of the starting point of the burn site on a large spatial scale, and the size of the sliding window can be dynamically adjusted based on the size of the burn site, thereby enhancing the robustness of the algorithm. The present invention helps to compile a detailed, timely and effectively updated list of fire starting points, providing strong data support for the attribution of fire origins, and also providing a scientific reference for forest and agricultural fire monitoring, evaluation and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Shown is a flow chart of the method of the present invention;

[0020] Figure 2 The figure shows the omission error diagram of the original MCD64A1 product in this embodiment;

[0021] Figure 3 The figure shows the grid diagram of the burnt area after supplementation and optimization by VNP14IMG in this embodiment;

[0022] Figure 4 Shown is a schematic diagram of the burning date grid in this embodiment;

[0023] Figure 5 Shown is a schematic diagram of the combustion boundary object in this embodiment;

[0024] Figure 6 The figure shows a schematic diagram of a local extreme value sliding window in this embodiment;

[0025] Figure 7 Shown is a schematic diagram of the inversion of the fire point coordinates and daily values ​​in this embodiment;

[0026] Figure 8 Shown is a display of the locations and dates of all fire points in the study area in 2020 in this embodiment. DETAILED DESCRIPTION

[0027] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0028] like Figure 1 As shown, the present invention provides a fire point inversion method based on object-oriented and local extreme value sliding search, which specifically includes the following steps:

[0029] Step 1: Determine the area to be studied, prepare multi-source remote sensing fire products, and read and preprocess the raw data.

[0030] The data collected in this example are the MODIS burn scar product (MCD64A1) and the VIIRS active fire product (VNP14IMG) covering the Yakutia region of Russia. The data spans from January to December 2020. Specifically, it includes:

[0031] (1) Download the MODIS annual burn scar product (MCD64A1) and VIIRS daily fire point product (VNP14IMG) covering the study area, with spatial resolutions of 500 m and 375 m, respectively; reproject them to the WGS_1984_North_Pole_LAEA_Russia coordinate system in ArcGIS software and clip them according to the study area.

[0032] (2) In ArcGIS, a DOY field was added to the VNP14IMG data and assigned a natural day value based on the date field. The DOY range was set to 1–365 (days). Batch numbering was implemented using a Python script, and the UpdateCursor function was used to update the fields sequentially.

[0033] Step 2: Combine MODIS burn scars (MCD64A1) and VIIRS fire points (VNP14IMG) to spatially optimize MCD64A1.

[0034] Existing research shows that the MCD64A1 burnt area product has large omission errors, which affects the characterization of the fire process. Figure 2 The higher detection rate of the VIIRS sensor fire point product effectively complements the area coverage of the MCD64A1 product. The following operations were all completed in ArcGIS.

[0035] (1) A 500 × 500 m grid was constructed in the study area using the Create Fishnet tool and projected into the WGS_1984_North_Pole_LAEA_Russia spatial coordinate system.

[0036] (2) Since there are other types of thermal anomalies (industrial heat sources, volcanoes, etc.) in the VNP14IMG product, the Select By Location tool is used to filter out the fire point data within 5 km of the burn scar data based on the adjacent spatial relationship between VNP14IMG and MCD64A1 to supplement the original burn scar. The Intersect tool is used to filter out the grids that contain both MCD64A1 and VNP14IMG, hereinafter referred to as Grid 2. Figure 3 As shown in the figure, the area of ​​the burned areas after merging in the study area increased by 44% compared with the original MCD64A1.

[0037] Step 3: Construct the burning date raster.

[0038] Existing research indicates that the DOY values ​​of the MCD64A1 burn scar product contain significant errors, necessitating the assignment of VNP14IMG DOY values ​​to the grids. However, due to cloud obscuration, the VNP14IMG fire point product also contains omission errors, necessitating the retention of the original MCD64A1 DOY values ​​beyond 1 km of the VNP14IMG data neighborhood.

[0039] (1) Use the Spatial Join tool in ArcGIS to spatially link MCD64A1 and VNP14IMG to grid 2, and calculate the minimum value of DOY in the grid as the value of grid 2.

[0040] (2) Write a script based on Arcpy to identify the grid 2 with no assigned value, search for the VNP14IMG data closest to it, and obtain its DOY value as the grid value.

[0041] (3) Use the Feature to Raster tool in ArcGIS to convert the DOY field of grid 2 into raster data to obtain the burning date raster, as shown in the following example: Figure 4 shown.

[0042] Step 4: Obtain the boundary coding data of the burned area.

[0043] Since there are a large number of burnt areas in the study area, in order to improve the efficiency of the algorithm, the global search of the algorithm is optimized to a local search. Therefore, object-oriented thinking is adopted to segment the burnt areas into objects with unique coding values. The following operations are all completed in the ArcGIS platform.

[0044] (1) Use the Feature to Raster tool to convert Grid 2 into burnt area raster data;

[0045] (2) Use the Raster Calculator tool and write Con("raster">0,1,0) to convert the burned area into a binary raster;

[0046] (3) Since there are isolated patches with only one or a few pixels on the boundaries of large continuous burn areas, these fragmented patches are caused by the omission errors of MCD64A1. These fragmented patches need to be included in the large burn area. Therefore, the Expand tool is used to perform mathematical morphological expansion of single pixels.

[0047] (4) For the expanded grid, use the Region Group tool to encode the four neighboring grids into unique values ​​to obtain the boundary code data of the burned area, such as Figure 5 The burned area is divided into two coding areas, namely burn boundary object 1 and burn boundary object 2.

[0048] Step 5: Use the local extreme value sliding search algorithm to search for each coded object and determine the coordinates of the fire point.

[0049] By writing a Python program, we write a local extreme value sliding search algorithm LocalMinSearch, which performs an iterative search for the burnt area boundary code object, loops through the burning date grid in each object, and inverts the fire point, such as Figure 6 shown.

[0050] (1) Construct the noMin array to test pixels that are not local minima, construct the potMin array to store potential minima to be tested, construct the trueMinCluster to store true local minima, and construct the result array to store a list of all local minima.

[0051] (2) Construct a sliding window with a search radius of 10 kilometers. Read the coordinates and DOY value of each pixel in the test window box, add potential minima to the existing potMin, and check potMin. If any local minimum has a pixel with an earlier burn date within its search radius, then the potential local minimum is not a true minimum. The radius threshold of the sliding window can be dynamically adjusted and optimized based on the size of the burned area. In this example, 10 kilometers is used.

[0052] (3) All pixels in the box that are not local minima are added to noMin. All tested pixels and all pixels in the non-minimum list are added to the trash bin and not tested again. This cycle is repeated to obtain the local minimum and store it in trueMinCluster. The pixel coordinates corresponding to the local minimum stored in trueMinCluster are the coordinates of the fire starting point. The standard deviation ellipse is generated based on the fire starting point coordinates and the positioning accuracy.

[0053] When the local minimum contains multiple identical pixel values, the centroid positions of these adjacent pixels are taken as the initial fire points, such as Figure 7 The centroid position is obtained by performing a weighted average operation on the pixels corresponding to multiple identical local minima.

[0054] When encountering a large burn area, meaning a burn area that exceeds the set sliding radius window (10 kilometers in this example), the search center automatically splits and searches the burn date grid on each coded object in turn. Each search center is relatively independent, making this approach suitable for inverting larger burn complexes generated by multiple fire points.

[0055] Step 6: Perform quality judgment based on the acquired fire point coordinate data, remove erroneous fire points whose spatial standard deviation exceeds the set threshold range and caused by fragmented spots, and obtain the final fire point coordinates and combustion date.

[0056] Specifically, in ArcGIS, standard deviation ellipses with standard deviations greater than 10 km and fragmented patches at the edge of the burn site were filtered out, and then the Feature to Point tool was used to convert the standard deviation ellipses into point data. Finally, the spatial location and daily value of the fire points in the burn site in the study area in 2020 were obtained, as shown in the figure below: Figure 8 shown.

[0057] In summary, the method described in this paper can automatically and accurately invert large-scale fire origins and determine their spatiotemporal attributes based on mainstream burn scar and fire point products. This allows for scientific assessment of fire causes and provides a sound theoretical foundation for subsequent research on the classification of fire-induced carbon emissions. Furthermore, the extraction and vector conversion processes are implemented using Python programming and the ArcGIS platform, reducing manual intervention and improving inversion efficiency.

Claims

1. A fire point inversion method based on object-oriented and local extreme value sliding search, characterized in that: The following steps are involved: (1) Determine the study area, read the original data of multi-source remote sensing fire products and perform preprocessing; the multi-source remote sensing fire products include: MODIS annual fire scar product MCD64A1, VIIRS daily fire point product VNP14IMG; the preprocessing is to add a daily value index DOY to the VNP14IMG data and assign values ​​based on the date field; (2) Construct a spatial grid of size m × m, and use the spatial relationship between VNP14IMG and MCD64A1 to screen out VNP14IMG fire points within the n-meter buffer zone of the MCD64A1 burn scar. After merging, perform spatial grid intersection analysis to obtain the spatially optimized burn scar grid. (3) Count the minimum DOY values ​​of MCD64A1 and VNP14IMG in each grid of the burn site as the grid value; identify the grid without assigned grid value and search for the VNP14IMG data closest to the grid without assigned grid value, and assign the DOY value in the VNP14IMG data to the grid without assigned grid value; convert all DOY fields into raster data to obtain the burning date raster; MCD64A1 data contains DOY field; (4) Convert the burned area grid into raster data and perform binarization processing, perform single-pixel mathematical morphological expansion on the binary image, and then perform unique value encoding on the expanded raster to obtain the boundary coding data of the burned area; (5) For each coded object in the boundary coding data of the burn site, a sliding radius window is used to perform an iterative search for the local minimum value on the burning date grid on each coded object in turn, and the pixel coordinates corresponding to the local minimum value obtained by the search are the coordinates of the ignition point; if multiple identical local minimum values ​​are obtained, the centroid positions of adjacent pixels are calculated as the coordinates of the ignition point; when encountering a large burn site, the search center of the sliding radius window will automatically split and the local minimum value iterative search will be performed on the burning date grid on each coded object in turn; the large burn site is a burn site whose range exceeds the range set by the sliding radius window.

2. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: Also includes: (6) Perform quality judgment based on the acquired fire point coordinate data, that is, remove the erroneous fire points whose spatial standard deviation exceeds the set threshold range and those caused by broken patches, and obtain the final fire point coordinates and combustion date.

3. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: The n-meter buffer zone described in step (2) is specifically set to a 5 km buffer zone.

4. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: In the process of obtaining the grid value in step (3), the original DOY value of MCD64A1 outside the VNP14IMG data neighborhood of 1 km must be retained.

5. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: The unique value encoding of the expanded grid in step (4) specifically includes grid segmentation based on the four-neighborhood connection rule, and unique value encoding of the grid connected by the four neighborhoods.

6. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: The local sliding window extreme value search in step (5) is developed based on the Python platform, and multiple arrays are constructed to store and determine the local minimum values ​​within the sliding window.

7. The fire point inversion method based on object-oriented and local extreme value sliding search according to claim 1 is characterized in that: The centroid position in step (5) is calculated specifically by performing a weighted average operation on the pixels corresponding to multiple identical local minimum values.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images