A Method for Handling Spatiotemporal Inconsistencies in GlobeLand30 Global Land Cover Data

Abstract

This invention discloses a method for handling spatiotemporal inconsistencies in GlobeLand30 global land cover data. The method includes preprocessing multi-source auxiliary land cover data, performing reprojection, cropping, and category conversion on different land cover data; calculating the local structural similarity and neighborhood spatial similarity of pixels with inconsistent edges across different time phases, making judgments and corrections to eliminate geometric inconsistencies among the three phases of GlobeLand30 data; utilizing reliable category information from multi-source auxiliary land cover data to establish discrimination and correction rules to handle classification confusion between similar land cover features, and performing morphological processing on necessary areas to reduce inconsistencies in various land cover categories across different time phases. This invention designs consistency processing rules based on the principle of mutual verification of multi-source data, effectively reducing inconsistencies while avoiding the introduction of classification errors from other land cover data into the processed data, ensuring the reliability of inconsistency handling.

Description

Technical Field

[0001] This invention relates to a method for handling spatiotemporal inconsistencies in GlobeLand30 global land cover data, belonging to the field of mapmaking technology. Background Technology

[0002] The spatial distribution and dynamic changes of global land cover are of great significance for studying global change, geospatial infrastructure construction, and sustainable development. Researchers both domestically and internationally have released numerous single-type and multi-type global land cover products at various resolutions. Among them, my country developed the world's first 30-meter resolution land cover dataset, GlobeLand30, including versions for 2000, 2010, and 2020 (primarily covering 2015 and 2017), which has been widely used in various fields. Subsequently, other 30-meter (or higher) resolution global land cover datasets, such as FROM-GLC and GLC_FCS, have been released and are expected to continue to be updated.

[0003] In these data sets, significant spatiotemporal inconsistencies exist between different time-phase data within the same dataset due to seasonal differences in image data sources, registration errors, and classification errors, severely limiting their application potential. These inconsistencies are even more pronounced between different datasets due to differences in data sources, classification systems, and classification methods. Current research on the inconsistency of global land cover data, both domestically and internationally, mainly focuses on the coordinated processing of different time-phase data within the same dataset, the fusion of different datasets within a single time phase, and the fusion of multiple datasets to generate a single-category product. Currently, there are no methods specifically addressing the consistency processing of the GlobeLand 30-phase global land cover data. Spatial inconsistencies between consecutive time-phase data exist within the GlobeLand 30-phase global land cover data due to geometric registration errors, manifesting as a 1-2 pixel offset between adjacent time-phase data. Temporal inconsistencies also exist in some areas, meaning that the actual land features have not changed, but the categories are inconsistent across multiple time-phase data. The inconsistency among the GlobeLand 30-phase global land cover data is a key issue restricting its application. Summary of the Invention

[0004] The purpose of this invention is to provide a method for handling spatiotemporal inconsistencies in GlobeLand30 global land cover data, which is not only structurally robust and difficult to separate, but also convenient to construct.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] Step 1: Preprocess the GlobeLand30 Phase III data and other auxiliary global land cover data to ensure consistency in projection method, extent, and classification system. The other auxiliary global land cover data includes two types of 30m all-category land cover data: FROM-GLC data and GLC_FCS2015 data; 30m single forest cover data: Global ForestChange data; and 300m resolution ESACCI-LC-L4-LCCS data. The FROM-GLC data includes FROM2015 and FROM2017; the Global Forest Change data includes GFC2000, GFC2010, and GFC2017; and the ESACCI-LC-L4-LCCS data includes CCI2000, CCI2010, and CCI2015.

[0007] Step 2: Spatial inconsistency handling between the three phases of GlobeLand30 data. This step addresses the inconsistency between two consecutive phases of data caused by geometric registration errors in the three phases of GlobeLand30 data. During the process, GlobeLand2010 is used as the reference. If there is spatial inconsistency between GlobeLand2000 and GlobeLand2020, the boundary pixels are corrected to align them with GlobeLand2010.

[0008] Step 3: The missing islands in the GlobeLand30 data are processed by supplementing them with other auxiliary global land cover data from the same period.

[0009] Step 4: Use other auxiliary global land cover data for cross-validation to obtain reliable water body and wetland category results, and address the inconsistencies in water bodies and wetlands in the GlobeLand30 data.

[0010] Step 5: Cross-validate GFC data and other auxiliary global land cover data with GlobeLand30 data, process inconsistencies in forest, shrub, grassland, cultivated land and wetland in GlobeLand30 data, and filter out small patches that may exist in the processed forest, shrub, grassland and cultivated land categories.

[0011] Step 6: Cross-validate CCI data and other auxiliary global land cover data with GlobeLand30 data to address inconsistencies in tundra, bare land, and snow and ice land cover types in GlobeLand30 data.

[0012] Step 7: Address inconsistencies in artificial land surfaces. If GlobeLand2000 and GlobeLand2020 are both artificial land surfaces, and GlobeLand2010 is farmland, woodland, or shrubland, then correct GlobeLand2010 to be an artificial land surface.

[0013] Step 8: Save the processed data from the three phases into image files according to the projection method, resolution, range, and data format of the original data to complete the consistency processing.

[0014] Preferably, the specific steps for preprocessing the GlobeLand30 Phase III data and other auxiliary global land cover data in step 1 are as follows:

[0015] Step 1.1: Crop the GlobeLand30 Phase III land cover data, which have different data ranges, to unify the data range.

[0016] Step 1.2 involves stitching, reprojecting, and cropping auxiliary global land cover data with different map sheet rules to ensure that their projection method and data range are completely consistent with the GlobeLand30 land cover data of the corresponding map sheet.

[0017] Step 1.3: Based on the meaning of the categories, map the categories of the auxiliary global land cover data whose original category system differs from GlobeLand30, so that their category system and category codes are completely consistent with GlobeLand30 data.

[0018] Preferably, the specific steps for handling spatial inconsistencies among the three phases of GlobeLand30 data in step 2 are as follows:

[0019] Step 2.1: Compare the pixels with inconsistent categories obtained from GlobeLand2000 and GlobeLand2010 data.

[0020] Step 2.2: Find the boundary pixels among the inconsistent pixels as candidate spatially inconsistent pixels.

[0021] Step 2.3: Calculate the structural similarity and neighborhood occurrence probability of the boundary pixels in the two temporal phases to determine whether they belong to spatially inconsistent pixels. For spatially inconsistent pixels, perform category correction, that is, for any candidate spatially inconsistent pixel... ,in The set of pixels with inconsistent candidate spaces is determined using the following sub-steps.

[0022] Step 2.3.1: For spatially inconsistent pixels, there is only a 1-2 pixel offset between two temporal data points, indicating local structural similarity. The local structural similarity is calculated using template matching. and Calculate similarity separately :

[0023]

[0024]

[0025] In the formula, Indicates 5 5 neighborhoods Indicates in 3 3. The percentage of pixels of the same category within the neighborhood. Represents pixels, the The specific formula is as follows:

[0026]

[0027] In the formula, A value of 1 indicates that two pixels belong to the same category, while a value of 0 indicates that they belong to different categories.

[0028] Pick and The smallest similarity is taken as the final similarity:

[0029]

[0030] Step 2.3.2, calculate the pixels respectively. Category labels in two time phases and exist 3 3 Percentage within the neighborhood and :

[0031]

[0032]

[0033] Step 2.3.3, determine according to the following rules Determine if a pixel is spatially inconsistent, and modify the spatially inconsistent pixel accordingly:

[0034] if ,

[0035] but .

[0036] Step 2.4, repeat steps 2.1 to 2.3.

[0037] Step 2.5: Repeat the processing methods from Step 2.1 to Step 2.4 to process the GlobeLand2020 data.

[0038] Preferably, the inconsistency in water body and wetland categories among the three phases of GlobeLand30 data in step 4 is handled as follows:

[0039] Step 4.1: Obtain reliable water body and wetland category results through cross-validation between different auxiliary cover data, and correct any missed water bodies or wetlands in GlobeLand2020. That is, for a pixel in GlobeLand2020, correct it according to the following rules:

[0040] if

[0041]

[0042] In the formula, 50 represents wetlands and 60 represents water bodies;

[0043] but ;

[0044] The pixel category is corrected, and the modified pixels are marked.

[0045] Step 4.2: Utilize cross-validation between GlobeLand30 data and other auxiliary global land cover data to obtain reliable water body and wetland classification results. Correct any omissions of water bodies or wetlands in GlobeLand2020. Specifically, for a given pixel, if GlobeLand2000 and GlobeLand2010 are water bodies or wetlands, but GlobeLand2020 is not, then the following rule applies:

[0046] if

[0047]

[0048] Then the pixel category is corrected, and the modified pixels are marked. The value is the category label corresponding to water bodies or wetlands in FROM2017, FROM2015, and FCS2015.

[0049] Step 4.3: Utilize cross-validation between different auxiliary global land cover data to obtain reliable water body and wetland classification results. Correct any misclassifications of water bodies or wetlands in GlobeLand2020. Specifically, for a given pixel, if neither GlobeLand2000 nor GlobeLand2010 is a water body or wetland, but GlobeLand2020 is, then the following rule applies:

[0050] if

[0051]

[0052] In the formula, 20 represents woodland, 30 represents grassland, and 40 represents shrubland;

[0053] Then the pixel category is corrected, and the modified pixels are marked. The values ​​are the most common category labels in FROM2017, FROM2015, and FCS2015.

[0054] Step 4.4: Perform morphological closing operations on the corrected pixels and their neighboring pixels that have been summarized and marked in the GlobeLand2020 data.

[0055] Step 4.4.1: Morphological processing applies only to the pixels marked in the previous steps. First, determine the mask area for morphological processing, taking 10 of all marked pixels. The 10-neighborhood set is used as the mask region.

[0056] Step 4.4.2: For the pixels marked in the previous steps, within the mask area, use 5 The morphological operator 5 is used to perform dilation operations, that is, for any pixel within a mask region... :

[0057] If it exists ,

[0058] but ;

[0059] In the formula, Indicates 5 5 neighborhoods Represents pixels Category tags.

[0060] Step 4.4.3: For the pixels marked in step 4.4.2, within the mask area, use 5 Erosion is performed using the morphological operator 5, that is, for any pixel within a mask region. :

[0061] If it exists ,

[0062] but ;

[0063] In the formula, Indicates 5 5 neighborhoods Represents pixels in the original data Category tags

[0064] Steps 4.4.1 to 4.4.3 are used to process the GlobeLand30 data for the three time phases respectively.

[0065] Step 4.5: Using the time-varying relationships of water body and wetland categories in GlobeLand30 data from different time phases, GlobeLand2010 is corrected, namely:

[0066] if

[0067]

[0068] but .

[0069] Preferably, the inconsistencies in the categories of forest land, shrubland, grassland, cultivated land, and wetland among the three phases of GlobeLand30 data in step 5 are handled as follows:

[0070] Step 5.1, using GFC data and other auxiliary global land cover data, corrects inconsistencies in the GlobeLand30 data caused by confusion between forest land and similar land cover types such as shrubland, grassland, and cultivated land, including the following sub-steps:

[0071] Step 5.1.1: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the woodland over-allocation of GlobeLand2000 using the following rule:

[0072] if

[0073]

[0074] but .

[0075] Step 5.1.2: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2000 using the following rule:

[0076] if

[0077]

[0078] but .

[0079] Step 5.1.3: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2020 using the following rule:

[0080] if

[0081]

[0082] but .

[0083] Step 5.1.4: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2010 using the following rule:

[0084] if

[0085]

[0086] but .

[0087] Step 5.1.5, based on the FROM2017, FROM2015, FCS2015, and FCS data, correct the misclassification of grassland as woodland in GlobeLand2020 using the following rules:

[0088] if

[0089]

[0090] but .

[0091] Step 5.1.6: Based on the FROM2017, FROM2015, FCS2015, and FCS data, the following rules are used to correct the misclassification of shrubs as woodland in GlobeLand2020, namely:

[0092] if

[0093]

[0094] but .

[0095] Step 5.2, using GFC data and other auxiliary global land cover data, corrects inconsistencies in the GlobeLand30 data caused by confusion between similar land cover types such as forest, shrubland, grassland, and cultivated land and wetlands. This includes the following sub-steps:

[0096] Step 5.2.1: Based on FCS2000 data, correct the confusion between cultivated land, wetlands, and forest land in GlobeLand2000 using the following rules:

[0097] if

[0098]

[0099] but .

[0100] Step 5.2.2: Based on the FCS2000 data, correct the misclassification of forest land in GlobeLand2000 using the following rules:

[0101] if

[0102]

[0103] but .

[0104] Step 5.2.3: Based on the FROM2017, FROM2015, and FCS2015 data, and combined with previous time-phase data, the following rules are used to correct the under-allocation of cultivated land in GlobeLand2020:

[0105] if

[0106]

[0107]

[0108] but .

[0109] Step 5.2.4: Based on the data from FROM2017, FROM2015, and FCS2015, correct the under-allocated farmland in GlobeLand2020 using the following rules:

[0110] if

[0111]

[0112]

[0113] but .

[0114] Preferably, the inconsistencies between tundra, bare land, and snow and ice data in the three phases of GlobeLand30 are handled as follows in step 6:

[0115] Step 6.1, based on CCI data and the relationship between multiple periods of data, correct the confusion between tundra, bare land, and grassland in GlobeLand2000 using the following rules:

[0116] if

[0117]

[0118] but .

[0119] Step 6.2, based on CCI data and the relationship between data from multiple periods, correct the confusion between tundra and grassland in GlobeLand2020 using the following rules:

[0120] if

[0121]

[0122] but .

[0123] Step 6.3, based on CCI data and the relationship between data from multiple periods, correct the confusion between bare land and tundra / grassland in GlobeLand2000 using the following rules:

[0124] if (GlobeLand2020==90)

[0125]

[0126] but .

[0127] Step 6.4, based on CCI data and the relationship between multiple periods of data, correct the confusion between tundra, bare land, and grassland in GlobeLand2000 using the following rules:

[0128] if (GlobeLand2020==70)

[0129]

[0130] but .

[0131] Step 6.5: Based on CCI data and the relationship between data from multiple periods, the following rules are used to correct the confusion between bare land, grassland, and tundra in GlobeLand2000:

[0132] if

[0133]

[0134] but .

[0135] The advantages of this invention are: it provides a simple and effective method for automatically processing spatiotemporal inconsistencies in GlobeLand30 multi-period land cover data, thereby improving the reliability of GlobeLand30 data in various applications. Other existing consistency processing methods are not designed specifically for the inconsistencies in GlobeLand30 multi-period land cover data, nor do they consider the reliability of auxiliary data or the uneven distribution of data accuracy.

[0136] This invention uses multi-source land cover data as an aid, cross-validates multiple periods of GlobeLand30 land cover data with auxiliary multi-source land cover data, and establishes reliable rules to avoid introducing new errors while correcting inconsistencies. This enables rapid and accurate consistency processing of GlobeLand30 multi-period land cover data, thereby improving its application in many fields such as global change analysis and environmental monitoring. Attached Figure Description

[0137] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0138] Figure 1 This is a schematic diagram of the data inconsistency handling process according to an embodiment of the present invention.

[0139] Figure 2 This is an example of spatial inconsistency phenomena and their handling in an embodiment of the present invention.

[0140] Figure 3 Examples of island errors or omissions and their handling in embodiments of the present invention.

[0141] Figure 4 This is an example of water body and wetland consistency treatment and morphological treatment in the embodiments of the present invention.

[0142] Figure 5 This is a schematic diagram illustrating the spatial inconsistency phenomenon and its processing effect in an embodiment of the present invention.

[0143] Figure 6 This is an example of the consistent treatment effect on wetlands in an embodiment of the present invention.

[0144] Figure 7 This is an example of the processing of woodland, grassland, etc. in the embodiments of the present invention. After processing, not only are the incorrect classification results corrected, but abnormal seams in the original data are also successfully eliminated.

[0145] Figure 8 This is another example of consistent treatment of woodland, grassland, etc. in the embodiments of the present invention.

[0146] Figure 9 This is an example of the consistent treatment effect on ice, snow, bare ground, and grassland in the embodiments of the present invention.

[0147] Figure 10 This is an example of the consistent treatment effect on tundra, grassland, bare land, etc. in the embodiments of the present invention. Detailed Implementation

[0148] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0149] GlobeLand30 is the world's first 30-meter resolution land cover dataset, which is of great significance for studying global change, geospatial infrastructure construction, and sustainable development. However, the spatiotemporal inconsistency between its multiple periods of data is a key problem restricting its application. This invention provides a method that uses multi-period GlobeLand30 global land cover data and other auxiliary global land cover data as input, and achieves rapid and accurate multi-period data consistency processing by cross-validating the multi-period GlobeLand30 land cover data with auxiliary multi-source land cover data and establishing reliable rules to correct inconsistencies. The method of this invention includes the following steps:

[0150] Step 1 involves preprocessing the GlobeLand30 Phase III data and other auxiliary global land cover data to ensure consistency in projection method, extent, and classification system. This includes the following sub-steps:

[0151] Step 1.1: Crop the GlobeLand30 Phase III land cover data with different data ranges to unify the data range;

[0152] Step 1.2: For auxiliary global land cover data with different map sheet rules, stitch, reproject, and crop them to make their projection method and data range completely consistent with the GlobeLand30 land cover data of the corresponding map sheet;

[0153] Step 1.3, based on the meaning of the categories, maps the categories of auxiliary global land cover data whose original category system differs from GlobeLand30, ensuring that their category system and category codes are completely consistent with GlobeLand30 data. This includes the following sub-steps:

[0154] Step 1.3.1: The encoding method of the FROM dataset is similar to that of GlobeLand30. Although the FROM2015 category contains secondary categories, the primary category is the same as that of GlobeLand30. The FROM2017 category encoding number and the GlobeLand30 encoding number have a simple 10-fold relationship, and a simple correspondence mapping can be made directly.

[0155] Step 1.3.2: The encoding method of the FROM-GLC dataset is similar to that of GlobeLand30. Although the FROM2015 category contains secondary categories, the primary category is the same as that of GlobeLand30. The category encoding numbers of FROM2017 and GlobeLand30 have a simple 10-fold relationship, which can be directly mapped.

[0156] Step 1.3.3, FCS2015 is mapped to GlobeLand30 category labels using the following rules: dry land (including grass dry land and tree dry land) and irrigated land are mapped to cultivated land (10); evergreen broad-leaved forest, deciduous broad-leaved forest (including open deciduous broad-leaved forest and closed deciduous broad-leaved forest), evergreen coniferous forest (including open evergreen coniferous forest and closed evergreen coniferous forest), deciduous coniferous forest (including open deciduous coniferous forest and closed deciduous coniferous forest), and mixed forest are mapped to... The map is mapped to woodland (20); grassland is mapped to grassland (30); evergreen shrubland and deciduous shrubland are mapped to shrubs (40); wetland is mapped to wetland (50); water body is mapped to water body (60); lichen and moss are mapped to tundra (70); impermeable surface is mapped to artificial surface (80); hard bare ground, non-hard bare ground, and sparse vegetation (including sparse grass vegetation and sparse shrub vegetation) are mapped to bare ground (90); permanent ice and snow are mapped to permanent ice and snow (100).

[0157] Step 1.3.4: The three phases of CCI data are mapped to GlobeLand30 category labels using the following rules: dryland (including grass dryland and tree dryland), irrigated land, and mixed vegetation of cultivated land are mapped to cultivated land (10); evergreen broad-leaved forest, deciduous broad-leaved forest (including open deciduous broad-leaved forest and closed deciduous broad-leaved forest), evergreen coniferous forest (including open evergreen coniferous forest and closed evergreen coniferous forest), deciduous coniferous forest (including open deciduous coniferous forest and closed deciduous coniferous forest), mixed forest, and mixed tree and grass (trees greater than 50%) are mapped to forest land. (20); Grassland and mixed tree-grass (grass greater than 50%) are mapped to grassland (30); Evergreen shrubland and deciduous shrubland are mapped to shrubland (40); Wetland, aquatic forest, aquatic shrub, and aquatic grassland are mapped to wetland (50); Water body is mapped to water body (60); Lichen and moss are mapped to tundra (70); Impermeable surface is mapped to artificial surface (80); Hard bare ground, non-hard bare ground, and sparse vegetation (including sparse grass vegetation and sparse shrub vegetation) are mapped to bare ground (90); Permanent ice and snow are mapped to permanent ice and snow (100).

[0158] Step 1.3.5: Global Forest Cover Data (GFC) contains only tree information and corresponding forest land categories. Areas with a canopy density greater than 0.1 in GFC2000, GFC2010, and GFC2017 are mapped to forest land (20), and the remaining pixels are labeled as 0.

[0159] Step 2, handling spatial inconsistencies between the three GlobeLand30 datasets, addressing inconsistencies between consecutive datasets caused by geometric registration errors (e.g., ...). Figure 2 As shown), the processing uses GlobeLand2010 as the reference. That is, if there is a spatial inconsistency between GlobeLand2000 and GlobeLand2020, the boundary pixels are corrected to align it with GlobeLand2010. This includes the following sub-steps:

[0160] Step 2.1: Compare and obtain pixels with inconsistent categories from GlobeLand2000 and GlobeLand2010 data;

[0161] Step 2.2: Find the boundary pixels among the inconsistent category pixels as candidate spatially inconsistent pixels;

[0162] Step 2.3: Calculate the structural similarity and neighborhood occurrence probability of the boundary pixels in the two temporal phases to determine whether they belong to spatially inconsistent pixels. For spatially inconsistent pixels, perform category correction, that is, for any candidate spatially inconsistent pixel... The following sub-steps are used to make the judgment.

[0163] Step 2.3.1: For spatially inconsistent pixels, there is only a 1-2 pixel offset between two temporal data points, indicating local structural similarity. The local structural similarity is calculated using template matching. and Calculate separately:

[0164]

[0165]

[0166] In the formula, Indicates 5 5 neighborhoods express In 3 3. Percentage of pixels of the same category within a neighborhood:

[0167]

[0168] In the formula, A value of 1 indicates that two pixels belong to the same category, while a value of 0 indicates that they belong to different categories.

[0169] Pick and The smallest similarity is taken as the final similarity:

[0170]

[0171] Step 2.3.2, calculate the pixels respectively. Category labels in two time phases and exist 3 Percentage within 3 neighborhoods:

[0172]

[0173]

[0174] Step 2.3.3, determine according to the following rules Determine if a pixel is spatially inconsistent, and modify the spatially inconsistent pixel accordingly:

[0175]

[0176]

[0177] Step 2.4: Repeat steps 2.1 to 2.3 to eliminate spatially inconsistent pixels, as shown in the image. Figure 2 As shown;

[0178] Step 2.5: Repeat the processing methods from Step 2.1 to Step 2.4 to process the GlobeLand2020 data.

[0179] Step 3: Address any potential islanding errors or omissions in certain time phases of the GlobeLand30 data by supplementing them with other auxiliary data from the same time phase (e.g., ...). Figure 3 (As shown), including the following sub-steps,

[0180] Step 3.1: If GlobeLand2020 has missing islands, use FROM2015, FROM2017, and FCS2015 to correct it. If the three are of the same category, assign the corresponding pixel to GlobeLand2020; if they are different, select the category that is the same as GlobeLand2010 among the three.

[0181] Step 3.2: If GlobeLand2010 has a missing island issue, use CCI2010 to correct it;

[0182] Step 3.3: If GlobeLand2000 has a missing island issue, use CCI2000 to correct it;

[0183] Step 4, addressing inconsistencies in water bodies and wetlands within the GlobeLand30 data, includes the following sub-steps:

[0184] Step 4.1: Obtain reliable water body and wetland category results through cross-validation between different auxiliary cover data, and correct any missed water bodies or wetlands in GlobeLand2020. That is, for a pixel in GlobeLand2020, correct it according to the following rules:

[0185]

[0186]

[0187]

[0188] Correct the pixel category and mark the modified pixels;

[0189] Step 4.2: Utilize the cross-validation between GlobeLand30 data and auxiliary overlay data to obtain reliable water body and wetland category results. Correct any omissions of water bodies or wetlands in GlobeLand2020. Specifically, for a given pixel, if GlobeLand2000 and GlobeLand2010 are water bodies or wetlands, but GlobeLand2020 is not, then the following rule applies:

[0190]

[0191]

[0192]

[0193] The pixel category is corrected, and the modified pixels are marked. It is the category label corresponding to water bodies or wetlands in FROM2017, FROM2015, and FCS2015.

[0194] Step 4.3: Obtain reliable water body and wetland category results through cross-validation between different coverage data, and correct any misclassifications of GlobeLand2020 as water bodies or wetlands. That is, for a certain pixel, if neither GlobeLand2000 nor GlobeLand2010 is a water body or wetland, but GlobeLand2020 is a water body or wetland, then the following rule applies:

[0195]

[0196]

[0197]

[0198] The pixel category is corrected, and the modified pixels are marked. It is the most common category label in FROM2017, FROM2015, and FCS2015.

[0199] Step 4.4: Since the auxiliary data may be generated based on pixel classification methods, the water bodies and wetlands corrected using the auxiliary data may also exhibit the salt-and-pepper phenomenon found in pixel classification, such as... Figure 4 As shown, therefore, morphological closing operations are performed on the corrected pixels marked in the GlobeLand2020 data summary and their neighboring pixels.

[0200] Step 4.4.1: Morphological processing applies only to the pixels marked in the previous steps. First, determine the mask area for morphological processing, taking 10 of all marked pixels. 10-neighborhood set as mask region ;

[0201] Step 4.4.2: For the pixels marked in the previous steps, in the mask area... Within the range, using 5 Dilation is performed using the morphological operator 5, that is, for any pixel... :

[0202]

[0203]

[0204] In the formula, Indicates 5 5 neighborhoods Represents pixels Category tags

[0205] Step 4.4.3, for the pixels marked in step 4.4.2, in the mask area Within the range, using 5 Erosion operation is performed using the morphological operator 5, that is, for any pixel :

[0206]

[0207]

[0208] In the formula, Indicates 5 5 neighborhoods Represents pixels in the original data Category tags

[0209] Steps 4.4.1 to 4.4.3 are used to process the GlobeLand30 data for the three time phases respectively.

[0210] Step 4.5: Using the time-varying relationships of water body and wetland categories in GlobeLand30 data from different time phases, GlobeLand2010 is corrected, namely:

[0211]

[0212]

[0213]

[0214] Step 5, handle inconsistencies in the GlobeLand30 data such as woodland, shrubland, grassland, cultivated land, and wetland, including the following sub-steps,

[0215] Step 5.1, using GFC data and other auxiliary data, corrects inconsistencies in the GlobeLand30 data caused by confusion between woodland and similar land types such as shrubland, grassland, and cultivated land. This includes the following sub-steps:

[0216] Step 5.1.1: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the woodland over-allocation of GlobeLand2000 using the following rule:

[0217]

[0218]

[0219]

[0220] Step 5.1.2: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2000 using the following rule:

[0221] )

[0222]

[0223]

[0224] Step 5.1.3: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2020 using the following rule:

[0225] )

[0226]

[0227]

[0228] Step 5.1.4: If GlobeLand2000, GlobeLand2000, and GlobeLand2000 are all woodland, shrubland, or grassland, correct the missing woodland allocation for GlobeLand2010 using the following rule:

[0229] )

[0230]

[0231]

[0232] Step 5.1.5, based on the FROM2017, FROM2015, FCS2015, and FCS data, correct the misclassification of grassland as woodland in GlobeLand2020 using the following rules:

[0233] )

[0234]

[0235]

[0236] Step 5.1.6: Based on the FROM2017, FROM2015, FCS2015, and FCS data, the following rules are used to correct the misclassification of shrubs as woodland in GlobeLand2020, namely:

[0237] )

[0238]

[0239]

[0240] Step 5.2, using GFC data and other data, corrects inconsistencies in the GlobeLand30 data caused by confusion between similar land types such as woodland, shrubland, grassland, and cultivated land, wetlands, etc., including the following sub-steps:

[0241] Step 5.2.1: Based on FCS2000 data, correct the confusion between cultivated land, wetlands, and forest land in GlobeLand2000 using the following rules:

[0242]

[0243]

[0244]

[0245] Step 5.2.2: Based on the FCS2000 data, correct the misclassification of forest land in GlobeLand2000 using the following rules:

[0246]

[0247]

[0248]

[0249] Step 5.2.3: Based on the FROM2017, FROM2015, and FCS2015 data, and combined with previous time-phase data, the following rules are used to correct the under-allocation of cultivated land in GlobeLand2020:

[0250]

[0251]

[0252]

[0253]

[0254] Step 5.2.4: Based on the data from FROM2017, FROM2015, and FCS2015, correct the under-allocated farmland in GlobeLand2020 using the following rules:

[0255]

[0256]

[0257]

[0258]

[0259] Step 5.3 involves filtering out any small patches that may exist in the processed woodland, shrubland, grassland, and cultivated land categories. This includes the following sub-steps:

[0260] Step 5.3.1: Perform a regional growth process on the categories of forest land, shrubland, grassland, and cultivated land to obtain category patches;

[0261] Step 5.3.2: Filter out the patches with an area of ​​less than 9 pixels in each of the above categories, that is, change their category code to the category with the largest proportion of adjacent pixels.

[0262] Step 6, addressing inconsistencies in tundra, bare land, and snow / ice landforms in the GlobeLand30 data, includes the following sub-steps:

[0263] Step 6.1, based on CCI data and the relationship between multiple periods of data, correct the confusion between tundra, bare land, and grassland in GlobeLand2000 using the following rules:

[0264]

[0265]

[0266]

[0267] Step 6.2, based on CCI data and the relationship between data from multiple periods, correct the confusion between tundra and grassland in GlobeLand2020 using the following rules:

[0268]

[0269]

[0270]

[0271] Step 6.3, based on CCI data and the relationship between data from multiple periods, correct the confusion between bare land and tundra / grassland in GlobeLand2000 using the following rules:

[0272] (GlobeLand2020==90)

[0273]

[0274]

[0275] Step 6.4, based on CCI data and the relationship between multiple periods of data, correct the confusion between tundra, bare land, and grassland in GlobeLand2000 using the following rules:

[0276] (GlobeLand2020==70)

[0277]

[0278]

[0279] Step 6.5: Based on CCI data and the relationship between data from multiple periods, the following rules are used to correct the confusion between bare land, grassland, and tundra in GlobeLand2000:

[0280]

[0281]

[0282]

[0283] Step 7: Address inconsistencies in artificial land surfaces. If GlobeLand2000 and GlobeLand2020 are both artificial land surfaces, and GlobeLand2010 is farmland, woodland, or shrubland, then correct GlobeLand2010 to be an artificial land surface.

[0284] Step 8: Save the processed data from the three phases into image files according to the projection method, resolution, range, and data format of the original data to complete the consistency processing.

[0285] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing spatio-temporal inconsistency of GlobeLand30 global land cover data, characterized in that, Comprising the following steps: Step 1, pre-processing GlobeLand30 three-period data and other auxiliary global land cover data to make the projection mode, range, and category system consistent; the other auxiliary global land cover data includes two kinds of 30m full-category land cover data FROM-GLC data, GLC_FCS2015 data, 30m single forest cover data Global Forest Change data, and 300m resolution ESACCI-LC-L4-LCCS data; the FROM-GLC data includes FROM2015 and FROM2017, the Global Forest Change data includes GFC2000, GFC2010, and GFC2017, and the ESACCI-LC-L4-LCCS data includes CCI2000, CCI2010, and CCI2015; Step 2, spatial inconsistency processing between GlobeLand30 three-period data, processing the inconsistency between the front and rear two-period data caused by the geometric registration error of GlobeLand30 three-period data, taking GlobeLand2010 as the reference during the processing, if there is spatial inconsistency between GlobeLand2000 and GlobeLand2020, the boundary pixels are corrected to align with GlobeLand2010; Step 3, processing the island loss phenomenon in GlobeLand30 data, and supplementing it with other auxiliary global land cover data; Step 4, obtaining reliable water body and wetland category results by mutual verification of other auxiliary global land cover data, and processing the inconsistency of water body and wetland in GlobeLand30 data; Step 5, mutual verification of GFC data and other auxiliary global land cover data with GlobeLand30 data to process the inconsistency of forest land, shrub, grassland, cultivated land, and wetland in GlobeLand30 data, and filter out small patches of forest land, shrub, grassland, and cultivated land categories that may exist after processing; Step 6, mutual verification of CCI data and other auxiliary global land cover data with GlobeLand30 data to process the inconsistency of tundra, bare land, and snow and ice land in GlobeLand30 data; Step 7, processing artificial surface inconsistency, if GlobeLand2000 and GlobeLand2020 are both artificial surfaces, and GlobeLand2010 is cultivated land, forest land, and shrub, then correcting GlobeLand2010 to be artificial surface; Step 8, saving the processed three-period data into image files according to the projection mode, resolution, range, and data format of the original data, and completing the consistency processing process.

2. The GlobeLand30 global land cover data spatio-temporal inconsistency processing method according to claim 1, characterized in that, The specific steps of pre-processing GlobeLand30 three-period data and other auxiliary global land cover data in step 1 are as follows: Step 1.1, cutting GlobeLand30 three-period land cover data with different data ranges to unify the data range; Step 1.2, splicing, re-projecting and cutting the auxiliary global land cover data with different framing rules to make the projection mode and data range consistent with the GlobeLand30 land cover data of the corresponding sheet; Step 1.3, according to the meaning of the category, mapping the categories of the auxiliary global land cover data whose original category system is different from GlobeLand30, so that the category system and category code are completely consistent with GlobeLand30 data.

3. The GlobeLand30 global land cover data spatio-temporal inconsistency processing method according to claim 1, characterized in that, The specific steps of the spatial inconsistency processing between GlobeLand30 three-period data in step 2 are as follows: Step 2.1, comparing to obtain the category inconsistent pixels of GlobeLand2000 and GlobeLand2010 data; Step 2.2, finding the boundary pixels as candidate spatial inconsistent pixels in the category inconsistent pixels; Step 2.

3. Calculate the structural similarity of the boundary pixel on two time phases and the neighborhood appearance probability, determine whether it belongs to the spatial inconsistent pixel, and correct the category of the spatial inconsistent pixel, that is, for any candidate spatial inconsistent pixel wherein represents the candidate spatial inconsistent pixel set, and the judgment is performed by the following sub-steps; Step 2.3.1, for the pixels belonging to the spatially inconsistent pixels, there is only 1~2 pixel offset between the two time-series data, and there is local structure similarity between them, the local structure similarity is calculated by using template matching method, and the local structure similarity is calculated by using template matching method and respectively : In the formula, denotes 5 5 neighborhood, denotes in 3 3 neighborhood, the proportion of pixels of the same category, denotes the pixel, said The specific formula is as follows: In the formula, represents two pixels of the same class as 1, and different as 0; Take and the minimum as the final similarity: ； Step 2.3.2, Calculate pixel by pixel Class labels over two time phases and In 3 3-neighborhood and : Step 2.3.3, judging according to the following rules whether it belongs to spatially inconsistent pixels and modifying spatially inconsistent pixels: If , then ; Step 2.4, repeating steps 2.1 to 2.3; Step 2.5, repeating the processing method of steps 2.1 to 2.4 to process GlobeLand2020 data.

4. The GlobeLand30 global land cover data spatio-temporal inconsistency processing method according to claim 1, characterized in that, The water body and wetland category inconsistency processing between GlobeLand30 three-period data in step 4 is as follows: Step 4.1, using the mutual verification between different auxiliary cover data to obtain reliable water body and wetland category results, correcting the missing water body or wetland of GlobeLand2020, that is, for a pixel of GlobeLand2020, through the rule: If In the formula, 50 represents wetland, and 60 represents water body; then ; The category of the pixel is corrected, and the modified pixel is marked; Step 4.2, using the mutual verification between GlobeLand30 data and other auxiliary global land cover data to obtain reliable water body and wetland category results, correcting the missing water body or wetland of GlobeLand2020, that is, for a certain pixel, if GlobeLand2000 and GlobeLand2010 are water body or wetland, GlobeLand2020 is not water body or wetland, then through the following rule: If Then the pixel class is corrected, and the modified pixel is marked, wherein FROM2017, FROM2015, FCS2015, and the corresponding class label when the value belongs to water or wetland; Step 4.3, using the mutual verification between different auxiliary global land cover data to obtain reliable water body and wetland category results, correcting the misclassified water body or wetland of GlobeLand2020, that is, for a certain pixel, if GlobeLand2000 and GlobeLand2010 are not water body or wetland, GlobeLand2020 is water body or wetland, then through the following rule: If In the formula, 20 represents forest land, 30 represents grassland, and 40 represents shrubbery; Then the pixel class is corrected and the modified pixel is marked, wherein The value is the majority class label among FROM2017, FROM2015, and FCS2015. Step 4.4, performing morphological closing operation processing on the modified pixels and their neighborhood pixels of GlobeLand2020 data; Step 4.4.1, morphological processing is only applied to the pixels marked in the previous step. First, the mask region for morphological processing is determined by taking the 10 10 neighborhood union of all the marked pixels; Step 4.4.

2. For the pixels marked in the previous step, perform an inflation operation using the morphological operator of 5 within the mask region, i.e. for any pixel within a mask region 5, the pixel is marked if there are at least 5 pixels within the mask region that are also within the 5x5 window of the pixel : If present , then ; In the formula, denotes 5 5 neighborhood, denotes the class label of a certain pixel pixel Step 4.4.

3. For the pixels labeled in step 4.4.2, within the mask region, perform an erosion operation using the morphological operator of 5, i.e., for any pixel in the mask region 5, if there are at least 5 pixels in the mask region that are also in the image region ​ If present , then ; In the formula, denotes 5 5 neighborhood, denotes the class label of the pixel in the original data Processing GlobeLand30 data of three phases respectively by using steps 4.4.1~4.4.3; Step 4.5, using the relationship between the changes in water and wetland categories over time in different GlobeLand30 data to infer the relationship, GlobeLand2010 is corrected, that is: If then . 5.The GlobeLand30 global land cover data spatio-temporal inconsistency processing method according to claim 1, characterized in that, The inconsistent categories between forest, shrub, grassland, cultivated land, and wetland in GlobeLand30 three-period data in step 5 are processed as follows: Step 5.1, using GFC data and other auxiliary global land cover data to correct the inconsistencies caused by similar land cover confusion between forest and shrub, grassland, and cultivated land in GlobeLand30 data, including the following sub-steps: Step 5.1.1, if GlobeLand2000, GlobeLand2000, GlobeLand2000 are all forest or shrub or grassland, use the following rules to correct the over-estimation of forest in GlobeLand2000, that is: If then ; Step 5.1.2, if GlobeLand2000, GlobeLand2000, GlobeLand2000 are all forest or shrub or grassland, use the following rules to correct the under-estimation of forest in GlobeLand2000, that is: If then ; Step 5.1.3, if GlobeLand2000, GlobeLand2000, GlobeLand2000 are all forest or shrub or grassland, use the following rules to correct the under-estimation of forest in GlobeLand2020, that is: If then ; Step 5.1.4, if GlobeLand2000, GlobeLand2000, GlobeLand2000 are all forest or shrub or grassland, use the following rules to correct the under-estimation of forest in GlobeLand2010, that is: If then ; Step 5.1.5, based on FROM2017, FROM2015, FCS2015, FCS data, use the following rules to correct the misclassification of grassland as forest in GlobeLand2020, that is: If then ; Step 5.1.6, based on FROM2017, FROM2015, FCS2015, FCS data, use the following rules to correct the misclassification of shrub as forest in GlobeLand2020, that is: If then ; Step 5.2, using GFC data and other auxiliary global land cover data to correct the inconsistencies caused by similar land cover confusion between forest, shrub, grassland and cultivated land, wetland, including the following sub-steps, Step 5.2.1, based on FCS2000 data, use the following rules to correct the confusion between cultivated land, wetland and forest in GlobeLand2000, that is: If then ; Step 5.2.2, based on FCS2000 data, use the following rules to correct the misclassification of forest in GlobeLand2000, that is: If then ; Step 5.2.3, based on FROM2017, FROM2015, FCS2015 data, combined with previous phase data, use the following rules to correct the under-estimation of cultivated land in GlobeLand2020, that is: If then ; Step 5.2.

4. Correct the cultivated land misclassification in GlobeLand2020 based on FROM2017, FROM2015, FCS2015 data using the following rules, i.e.: If then . 6.The method according to claim 1, characterized in that, The inconsistency between the tundra, bare land, and ice and snow in the GlobeLand30 three-period data in step 6 is processed as follows: Step 6.

1. Correct the confusion between tundra and bare land, grassland in GlobeLand2000 based on CCI data and multi-period data relationship using the following rules, i.e.: If then ; Step 6.

2. Correct the confusion between tundra and grassland in GlobeLand2020 based on CCI data and multi-period data relationship using the following rules, i.e.: If then ; Step 6.

3. Correct the confusion between bare land and tundra, grassland in GlobeLand2000 based on CCI data and multi-period data relationship using the following rules, i.e.: If (GlobeLand2020 == 90) then ; Step 6.

4. Correct the confusion between tundra and bare land, grassland in GlobeLand2000 based on CCI data and multi-period data relationship using the following rules, i.e.: If (GlobeLand2020 == 70) then ; Step 6.

5. Correct the confusion between bare land, grassland and tundra in GlobeLand2000 based on CCI data and multi-period data relationship using the following rules, i.e.: If then .

Images