A multi-source data fusion urban and rural residential area accurate identification method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF GEOGRAPHICAL SCI & NATURAL RESOURCE RES CAS
- Filing Date
- 2024-09-23
- Publication Date
- 2026-07-03
Smart Images
Figure CN119313946B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for accurate identification of urban and rural residential areas, and more particularly to a method for accurate identification of urban and rural residential areas through multi-source data fusion. Background Technology
[0002] A clear understanding of the spatiotemporal dynamics of global urban and rural settlements is crucial for the sustainability of the planet. Settlements, including urban and rural settlements, represent the spatial information of human presence on Earth, and their expansion is one of the most direct and enduring land change processes affecting ecosystems. By 2050, the global population is projected to increase by approximately 1.8 billion, further driving settlement expansion. Moreover, with declining settlement density, the encroachment of global settlement expansion on farmland and ecological habitats will become even more severe. Global settlement governance aimed at achieving Sustainable Development Goals is urgently needed. However, global settlement governance needs to be carried out separately for urban and rural settlements because they have different dominant economic forms, architectural styles, and human activity patterns. Therefore, delineating urban and rural settlements and monitoring their expansion is fundamental to achieving refined global settlement governance to mitigate their ecological and environmental impacts.
[0003] Global maps of urban and rural residential areas are fundamental to studying the dynamics of urban and rural development and their environmental impacts. For urban areas, a wealth of long-term global urban extent data products based on nighttime light data or traditional remote sensing imagery have been developed. Methods for determining urban extent can be categorized into two types based on the type of input remote sensing observation data. The first type is based on traditional remote sensing products such as MODIS, primarily determining urban extent based on landscape morphology and texture information. For example, since cities are concentrated areas of impermeable surfaces, the "proportion of impermeable surfaces" is used to determine urban extent. Urban features are extracted from existing global data products, and a global urban area (MGUP) is drawn based on MODIS data using a locally adaptive method. At higher resolutions, based on the large-area contiguous nature of urban areas, continuous urban boundaries (GUBs) without hollow areas are obtained using kernel density estimation, cellular automata, and dilatation-erosion methods based on 30-meter GAIA data. However, this method pursues morphological continuity and smoothness, thus failing to accurately characterize urban boundaries. The second type is based on nighttime light (NTL) data such as DMSP-OLS and VIIRS, primarily determining urban extent based on the coupling information of nighttime light intensity and location. Unlike traditional remote sensing products, NTL data has unique significance in characterizing urbanized areas and urbanization activities. Urban boundaries can be identified based on nighttime light change gradients in urban and rural areas. Global urbanized areas (NTL-UE) have been mapped using global DMSP-OLS-like NTL images from 1992–2020, combined with quantization and parabolic strategies. Furthermore, the fusion of NTL data with other multi-source data (such as Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) data) can also effectively map urban boundaries.
[0004] However, the resolution of NTL data (500 / 1000 meters) is relatively low, resulting in noticeable jagged edges in the derived city boundaries. Overbrightness and saturation effects associated with DMSP-OLS NTL data also reduce the accuracy of the identification results. Overall, global urban residential data products are mainly concentrated in the low-resolution range of 250-1000 meters. There is an urgent need for higher-resolution urban extent products to accurately capture the expansion dynamics of urban residential areas.
[0005] Research on rural residential area identification is limited and mainly focuses on local areas; a global rural residential area map is still lacking. Specifically, most rural residential area identification studies are conducted on local areas based on high-resolution remote sensing imagery such as SPOT-5 (2.5 meters) or Sentinel-1 / 2 (10 meters). However, these remote sensing images have short coverage periods and cannot provide long-term remote sensing monitoring of rural residential areas. In medium-resolution remote sensing images of 30-100 meters, urban and rural residential areas differ significantly from permeable areas, but they are very similar and cannot be directly distinguished through remote sensing interpretation. Therefore, several datasets of impermeable surfaces or human settlements exist, such as GAIA (Global Artificial Impervious Area), GAUD (Global Annual Urban Dynamics), GISA2.0 (Global Impervious Surface Area 2.0), GISD30 (Global 30m Impervious Surface Dynamic Dataset), and GHS-B (GHS-BUILT-S R2023A), but these do not delineate urban and rural residential areas within them. However, localized research indicates that combining medium-resolution Landsat imagery or residential data with supplementary geospatial data (e.g., distance to water sources, seasonal vegetation signal changes, distance to recent active fires, POI data, and NTL data) can accurately identify rural residential areas. The core idea is to first identify urban and rural residential patches, then eliminate urban patches based on area and NTL to identify rural residential areas. The average overall accuracy of this approach is 85%, demonstrating the effectiveness of distinguishing urban and rural residential areas based on area and human activity intensity. Currently abundant global residential data, crowdsourced Open Street Map (OSM) data, and long-term NPP-VIIRS-like NTLs make global urban-rural residential delineation possible. Summary of the Invention
[0006] To improve the resolution of global urban residential area maps and identify the boundaries of rural residential areas, thereby supporting urban-rural comparative studies and updating the urban-rural knowledge system, this invention proposes a multi-source data fusion method for accurate identification of urban and rural residential areas. By fusing human residential area data, OSM data, and NTL data, dynamic thresholds for area and brightness between urban and rural residential areas are calculated. Based on this, a global urban and rural residential area dataset (GURS) with a resolution of 100 meters from 2000 to 2020 can be generated.
[0007] The purpose of this invention is to propose a method for accurate identification of urban and rural residential areas through multi-source data fusion, comprising the following steps:
[0008] A. Identify potential urban and rural residential patches: Extract potential residential grids from GHS-B data and aggregate spatially adjacent grids into a residential patch with an area. Then, calculate the area and brightness attributes of the residential patch based on NTL data.
[0009] B. Differentiate urban and rural residential areas based on two-dimensional attributes of area and brightness: Extract urban and rural residential area patch samples according to OSM, and determine the urban and rural classification threshold of area or brightness according to the intersection of the data distribution of urban and rural residential area patch samples in the area or brightness dimension. Based on the threshold and classification scheme, divide all residential area patches into urban and rural areas and generate an initial residential area map.
[0010] C. Correct and update the multi-temporal urban and rural residential area classification results, and update the initial map according to the evolution logic of urban and rural residential areas.
[0011] Preferably, the step of extracting potential residential area grids includes: using 30-meter GAIA data as a reference, gradually increasing the built-up area percentage threshold and comparing it with GAIA data, and selecting the built-up area percentage when the two areas are closest.
[0012] Preferably, step B further includes: repeating the process of "extracting urban and rural residential patch samples based on OSM, and determining the urban-rural classification threshold for area or brightness based on the intersection of the data distribution of urban and rural residential patch samples in the area or brightness dimension" multiple times, and taking the average value to obtain a robust urban-rural residential classification threshold.
[0013] Preferably, in step B, the step of extracting patch samples from urban and rural residential areas includes:
[0014] Based on the United Nations Statistics Division's classification of the world, the excessively large residential data groups were split into 9 sub-regions;
[0015] Extract urban and rural residential areas from OSM-Place point data;
[0016] Spatial matching of residential patches and residential points; successfully matched urban and rural residential patches are the basic samples with urban and rural residential attributes.
[0017] The residential area and the number of basic samples of urban / rural residential patches in the nine sub-regions were counted respectively, and the minimum density of basic urban / rural residential patches was calculated.
[0018] Nine sub-regions were randomly selected based on the minimum density of patches to form a statistical sample of urban / rural residential patches.
[0019] Preferably, in step B, the step of determining the urban-rural residential area classification threshold includes:
[0020] Take the natural logarithm of the sample data to reduce the absolute value of the data;
[0021] Plot the cumulative frequency curve of area / brightness of urban residential patch samples and the inverse cumulative frequency curve of area / brightness of rural residential patch samples;
[0022] The intersection of the cumulative frequency curve and the inverse cumulative frequency curve is defined as the classification threshold for urban and rural residential areas, thus obtaining the classification thresholds for urban and rural residential areas in two dimensions: area and brightness.
[0023] Preferably, in step B, the classification scheme includes: using brightness as the x-axis, area as the y-axis, and the classification threshold as the origin, dividing residential patches into four types: contiguous urban areas, contiguous rural areas, scattered rural areas, and high-brightness enclaves. Contiguous urban areas and scattered urban areas are classified as urban residential areas, and contiguous rural areas and scattered rural areas are classified as rural residential areas. Specifically, contiguous urban areas include cities and developed towns; contiguous rural areas include underdeveloped towns and contiguous rural areas; scattered rural areas are scattered small villages; and high-brightness enclaves include scattered cities and scattered rural areas near urban areas and highway light sources. High-brightness enclaves within a certain distance range are classified as scattered cities, and other high-brightness enclaves are classified as scattered rural areas.
[0024] Preferably, in step C, updating the multi-temporal urban and rural residential area classification results includes: extracting coastal residential area patches not covered by NPP-VIIRS-like NTL data, and reclassifying coastal residential area patches located within 1 kilometer of existing urban residential area patches as urban residential areas.
[0025] Preferably, the method for accurate identification of urban and rural residential areas further includes the step of evaluating the accuracy of a global urban and rural residential area dataset based on an independent verification sample set constructed from publicly available data and high-resolution remote sensing images.
[0026] Based on the above technical solution, the advantages of the present invention are:
[0027] This invention presents a multi-source data fusion method for accurate identification of urban and rural residential areas. By fusing GHS-BUILT-S R2023A (GHS-B) data, NPP-VIIRS-like nighttime light (NTL) data, and OpenStreetMap (OSM) data, it can construct a 100-meter resolution global urban and rural residential area dataset (GURS) from 2000 to 2020. First, potential residential patches are extracted from the GHS-B data, and their area and brightness attributes are statistically analyzed. Second, urban and rural residential patch samples are extracted based on the OSM dataset, a two-dimensional urban-rural classification threshold of "area-brightness" is determined, and an initial residential area map is generated. Finally, the initial map is updated according to the evolutionary logic of urban and rural residential areas.
[0028] The urban and rural residential area identification method of this invention, based on an accuracy assessment of 31,474 independent samples, shows that the overall accuracy of GURS is 92.39%, with a Kappa coefficient of 0.84. Comparison with nine multi-scale reference datasets shows that GURS can accurately characterize the scope of global urban and rural residential areas from 2000 to 2020. GURS provides new insights into global urban and rural residential areas, supporting comparative studies of urban and rural areas in various fields such as socio-economic characteristics, environmental impact, and governance models, and is of great significance for the refined and sustainable management of residential areas. Attached Figure Description
[0029] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0030] Figure 1 Framework diagram for a method of accurate identification of urban and rural residential areas based on multi-source data fusion;
[0031] Figure 2 a to f are scatter plots of GHSL and GAIA under different critical building area percentages;
[0032] Figure 3a Thresholds for dividing urban and rural residential areas;
[0033] Figure 3b Thresholds for dividing urban and rural residential areas;
[0034] Figure 4 Spatial distribution of some samples for accuracy verification from 2000 to 2020;
[0035] Figure 5 A step-by-step diagram of a method for accurate identification of urban and rural residential areas using multi-source data fusion. Detailed Implementation
[0036] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0037] This invention proposes a method for accurate identification of urban and rural residential areas through multi-source data fusion, such as... Figure 5 As shown, it includes the following steps:
[0038] A. Identify potential urban and rural residential patches: Extract potential residential grids from GHS-B data and aggregate spatially adjacent grids into a residential patch with an area. Then, calculate the area and brightness attributes of the residential patch based on NTL data.
[0039] B. Differentiate urban and rural residential areas based on two-dimensional attributes of area and brightness: Extract urban and rural residential area patch samples according to OSM, and determine the urban and rural classification threshold of area or brightness according to the intersection of the data distribution of urban and rural residential area patch samples in the area or brightness dimension. Based on the threshold and classification scheme, divide all residential area patches into urban and rural areas and generate an initial residential area map.
[0040] C. Correct and update the multi-temporal urban and rural residential area classification results, and update the initial map according to the evolution logic of urban and rural residential areas.
[0041] Specifically, in the technical solution of this invention, the data used includes:
[0042] (1) Global human settlement data
[0043] This invention uses GHS-B as the foundational data for mapping the extent of urban and rural settlements globally. This dataset identifies "any roofed structure built on the ground," i.e., human settlements, and provides estimated 100-meter resolution distributions of human settlement area every five years from 1975 to 2030. This data was obtained through spatiotemporal interpolation of 30-meter data from 1975, 1990, 2000, and 2014, and 10-meter data from 2018.
[0044] The specific interpolation method is as follows: Symbolic Machine Learning (SML) is used to extrapolate from the 2018 10-meter data to the past or future. Corrections are made if empirical evidence from specific historical 30-meter data proves the extrapolation incorrect. Finally, predictions of changes in building surfaces are summarized using 100-meter grid cells. The 2018 10-meter data, generated based on a multi-source integrated learning set (GHS-BUILT-S2 R2020A, Facebook residential area mapping, Microsoft and OpenStreetMap (OSM) building mapping) and 10-meter resolution Sentinel-2 images, significantly improved the accuracy of the GHS-BUILT-S R2023A dataset, providing support for identifying rural residential areas. Accuracy validation results show an overall accuracy of 82.6% across all 250,000 reference points and 90.5% across 155,649 reference points in the "High Confidence Consistency (JAHC)" sample set. Large-sample, high-confidence visual interpretation results demonstrate the accuracy of the GHS-B dataset. Although long-term series residential area data such as GAIA, GISD30, GISA, and GAUD, with a resolution of 30 meters, have higher resolution, the fragmentation caused by high resolution is unsuitable for the patch-based urban-rural division method of this invention. Therefore, this invention selects the 100-meter resolution GHS-B dataset to map the basic extent of global urban and rural residential areas and to statistically analyze the "area-human agglomeration scale" information of global urban and rural residential areas.
[0045] (2) Global Nighttime Light Data
[0046] Globally extended NPP-VIIRS-like NTL data with 500-meter resolution provides crucial "brightness" information for identifying urban and rural residential areas worldwide. This data, based on DMSP-OLS NTL data (2000–2012) and monthly NPP-VIIRS NTL data (2013–2018), utilizes a novel cross-sensor calibration method based on vegetation indices and an autoencoder model, exhibiting excellent spatial patterns and temporal consistency. Compared to the consistent global DMSP OLS-like NTL dataset 1992–2020, this data offers higher spatial resolution and lower overbrightness and saturation effects; therefore, this invention uses it to assign "brightness-human activity intensity" information to urban and rural residential areas globally.
[0047] (3) Global Urban and Rural Residential Area Label Data
[0048] The Place information provided by OSM is used to determine the urban / rural attributes of residential area samples. OSM is a free, open, and editable world map database containing a rich collection of geographic objects (e.g., roads, locations, buildings, land use, points of interest) from around the world. Experts, geoscientists, cartographers, volunteers, and the general public can all contribute data to OSM. Extensive public participation makes OSM an important source of spatial information on residential areas. This invention extracts points representing residential areas, such as city, suburb, town, village, hamlet, etc., from OSM-Place spatial point data and uses them as label data for classifying urban and rural residential areas.
[0049] (4) Relevant residential area data
[0050] This invention utilizes three types of global impervious surface (human settlement) data covering almost the entire period from 2000 to 2020: GAIA (1985-2018), GISA2.0 (1972-2019), and GISD30 (1985-2020); and three types of global urban settlement data: GUB (Global Urban Boundaries, 2000-2018), MGUP (MODIS Global UrbanExtent Product, 2001-2018), and NTL-UE-2020 (Global Annual Urban Extents from Harmonized NTL, 1992-2020). The invention comprehensively analyzes the performance of the data in characterizing global human settlements and urban settlements. GAIA is a 30-meter resolution global impervious surface dataset from 1985 to 2018 generated using a combination of exclusion-inclusion and time-checking methods. GISA2.0 is a global 30m resolution impervious surface dataset for the period 1972-2019, generated based on existing impervious surface datasets and manually interpreted samples trained on a model. GISD30 (Global 30m Impervious Surface Dynamic Dataset) is a global 30m impervious surface dataset for the period 1985-2020, generated using an automated method that combines the advantages of spectral generalization and automatic sample extraction strategies.
[0051] The three higher-resolution datasets mentioned above can be used to compare the performance of the data from this invention in mapping human settlements and local rural settlements. GUB, MGUP, and NTL-UE-2020 represent city boundaries mapped using morphological methods, traditional remote sensing data, and nighttime light data, respectively, and are used to compare the performance of the data from this invention in identifying urban settlements.
[0052] Currently, there is no global-scale map of rural settlements. Therefore, this invention evaluates the performance of its data in identifying rural settlements by comparing it with the national-scale CLUD (China's Land Use / cover Dataset), which is the land use dataset with the longest time span, largest spatial scope, and distinguishes between urban and rural construction land. CLUD identifies three types of construction land: urban land, rural settlements, and industrial and mining land outside cities. This dataset was generated using a unified technical process and classification system in a human-machine digital environment and is widely used in comparative studies of urban and rural areas.
[0053] Global urban and rural residential areas are a collective term for all roofed buildings used for human habitation and activity, encompassing both urban and rural residential areas. Urban and rural residential areas typically differ significantly in size (area) and intensity of human activity (light intensity). Cities are usually hubs of industry and services, possessing more developed infrastructure. The economic structure, facilities, and services of cities encourage large populations to congregate and engage in high-intensity activities. Rural areas, on the other hand, rely primarily on agriculture and natural resource development, resulting in significantly lower population density and activity intensity compared to cities. Therefore, this invention proposes an object-oriented approach, such as... Figure 1 As shown, based on the "area-brightness" two-dimensional attribute urban and rural residential area identification method framework, global urban and rural residential areas (2000-2020) with a resolution of 100 meters were drawn.
[0054] Specifically, firstly, potential residential rasters are extracted from GHS-B data, and spatially adjacent rasters are aggregated into residential patches with an "area." Then, the "brightness" of these patches is statistically analyzed based on NTL data. Secondly, urban and rural residential patch samples are extracted using OSM, and urban / rural classification thresholds for area or brightness are determined based on the data distribution of these samples in both dimensions. Finally, all residential patches are classified as urban or rural based on these thresholds and the classification scheme. Finally, based on the principle that "rural development into urban is an irreversible process," the multi-temporal urban / rural residential classification results are updated to ensure the spatiotemporal consistency and logicality of urban / rural residential evolution.
[0055] Identify potential urban and rural residential patches
[0056] Mapping potential residential patches globally involves three main steps.
[0057] The first step is to extract potential residential area grids. GHS-B uses a continuous value of 0-10000 to represent the built-up area percentage of a pixel. In order to find the critical value for the built-up area percentage that distinguishes between residential and non-residential areas, this invention uses 30-meter GAIA data as a reference, gradually increases the built-up area percentage threshold and compares it with GAIA data, and selects the built-up area percentage when the two are closest in area.
[0058] Specifically, this invention uses a grid of 375 regions provided by GHSL on the GEE platform to statistically analyze the area of GAIA in 2015, as well as the area of GHS-B when the percentage of built-up area exceeds specific values (0, 500, 750, 1000, 1250, 1500). The scatter plot fitting results for both are as follows: Figure 2As shown in a to f, the area of the raster with a built-up area percentage higher than 1000 is closest to the area of GAIA. The study found that classifying raster with a pixel value of 1000 or higher as built-up areas has the lowest omission rate and helps in identifying scattered small rural residential areas. Therefore, in this invention, GHS-B raster with a pixel value greater than or equal to 1000 is extracted as potential residential raster.
[0059] The second step is to aggregate residential patches. Human settlements tend to cluster spatially; therefore, this invention considers spatially adjacent grids as belonging to a single residential patch. This approach classifies rural areas connected to the city on its periphery as part of the city. Although these rural areas retain their rural spatial characteristics, they have close interactions with the city's elements, enjoy the city's free market and public services, and are areas poised for urbanization. Therefore, this invention considers these rural areas as part of the city.
[0060] The third step involves statistically analyzing the basic attributes of residential patches. Area and brightness are the fundamental attributes used in this invention to classify urban and rural residential areas. The area attribute is obtained during the aggregation of residential patches. Subsequently, using NPP-VIIRS-like NTL (2000-2020) data as input, the average light radiation value of the residential patches is statistically analyzed as the brightness attribute. Due to the larger population concentration, higher economic activity intensity, and greater infrastructure demand in cities, the area of a single urban residential area is often much larger than that of a single rural residential area. The area attribute can distinguish the vast majority of urban and rural residential areas. However, large rural residential areas (especially in developing countries) and small urban residential areas (especially in developed countries) still exist. Furthermore, some newly developed buildings and construction projects on the outskirts of cities, spatially distant from the main urban area, may be identified as another small residential patch. The brightness attribute can further distinguish these confused urban and rural residential areas. The complementarity of area and brightness attributes helps this invention accurately classify urban and rural residential areas.
[0061] Urban and rural residential areas are distinguished based on two-dimensional attributes: "area" and "brightness".
[0062] It should be noted that this invention distinguishes between urban and rural residential areas based on a threshold, but unlike existing technologies that pre-set a classification threshold for urban and rural areas based on a certain attribute, such as the proportion of impermeable surfaces, this invention finds the classification threshold between urban and rural residential areas by extracting a large number of actual samples of urban and rural residential areas and based on the statistical distribution of the sample attributes.
[0063] The global classification of urban and rural residential areas mainly involves three steps: The first step is to extract urban and rural residential patch samples separately. OSM provides spatial point data with urban and rural residential area classification attributes, which is matched with the residential patch map to extract urban and rural residential patch samples. The second step is to determine the classification thresholds for urban and rural residential areas. The area / brightness of the urban and rural residential patch samples are statistically analyzed separately, and the threshold is determined based on the intersection of the statistical distributions of urban and rural residential patch samples in the area / brightness dimension. The first and second steps are repeated one hundred times and the average is taken to obtain a robust urban-rural residential area classification threshold. The third step is to classify all residential patches based on the classification threshold to obtain an initial global urban-rural residential area map.
[0064] The first step is to extract patch samples from urban and rural residential areas:
[0065] Due to the uneven development of global information technology, the amount of OSM-Place data varies significantly across different regions. Directly using global samples will result in classification thresholds biased towards regions with large OSM-Place data volumes, affecting the universality of the thresholds. Therefore, this invention, based on the UN Statistics Division's classification of the world, splits groups with excessively large residential data volumes ("Europe and North America" split into "Europe" and "North America"; "East Asia and Southeast Asia" split into "East Asia" and "Southeast Asia"), and merges groups with excessively small data volumes ("Oceania" and "Australia and New Zealand" merged into "Oceania"), ultimately dividing the world's countries into 9 sub-regions, such as... Figure 4 As shown.
[0066] Based on this, urban residential areas (city, suburb) and rural residential areas (village, hamlet, farm) are extracted from the OSM-Place point data. Spatial matching is performed between residential patches and residential points. Successfully matched urban and rural residential patches become the basic samples with urban / rural residential area attributes. The residential area area and the number of basic urban / rural residential patch samples are counted for each of the nine sub-regions, and the minimum density of urban / rural basic sample residential areas is calculated. Patches are randomly selected from the nine sub-regions according to this density to form statistical samples of urban / rural residential patches. Attribute information is extracted from the statistical samples to obtain four sets of data: "urban residential patch sample area", "rural residential patch sample area", "urban residential patch sample brightness", and "rural residential patch sample brightness".
[0067] The second step is to determine the threshold for classifying urban and rural residential areas.
[0068] This invention identifies classification thresholds for urban and rural residential areas based on the statistical distribution of data. First, the natural logarithm of the sample data is taken to reduce the absolute value of the data. Then, a cumulative frequency curve for the area / brightness ratio of urban residential area patches and an inverse cumulative frequency curve for the area / brightness ratio of rural residential area patches are plotted. The former means that urban residential areas are extremely rare in patches with very small area / brightness ratios, and the cumulative frequency of urban residential areas gradually increases as the area / brightness ratio of the residential patches increases. The latter means that rural residential areas are extremely rare in patches with very large area / brightness ratios, and the cumulative frequency of rural residential areas gradually increases as the area / brightness ratio of the residential patches decreases. To the right of the intersection of the two curves, the frequency of rural residential areas is extremely low, while the frequency of urban residential areas is increasingly high; to the left of the intersection, the frequency of urban residential areas is extremely low, while the frequency of rural residential areas is increasingly high. In this invention, this intersection is defined as the classification threshold for urban and rural residential areas. Ultimately, classification thresholds for urban and rural residential areas in terms of both area and brightness are obtained. To ensure the robustness of the threshold, the steps of "extracting urban and rural residential patch samples based on OSM, and determining the urban / rural classification threshold for area or brightness based on the intersection of the data distributions of urban and rural residential patch samples in the area or brightness dimension" are repeated one hundred times, and the average of the 100 thresholds is taken to obtain the final area-brightness urban / rural residential classification threshold, as follows. Figure 3a , Figure 3b As shown.
[0069] The third step is to divide urban and rural residential areas.
[0070] Based on the two-dimensional attributes of area and brightness of residential patches and the classification thresholds for urban and rural residential areas, this invention proposes the following classification scheme: with brightness as the x-axis, area as the y-axis, and the classification threshold as the origin, residential patches are divided into four types: contiguous urban areas, contiguous rural areas, scattered rural areas, and high-brightness enclaves. Specifically, contiguous urban areas include cities and developed towns; contiguous rural areas include underdeveloped towns and contiguous rural areas; scattered rural areas refer to scattered small villages; and high-brightness enclaves include scattered urban areas and scattered rural areas near urban areas, highways, and other light sources.
[0071] Furthermore, this invention reclassifies high-brightness enclaves based on their distance from contiguous urban areas: high-brightness enclaves within the specified distance range are classified as scattered urban areas, while other high-brightness enclaves are classified as scattered rural areas. This invention sets four distance thresholds: 1000, 800, 500, and 300 meters, and selects the optimal distance threshold based on subsequent accuracy evaluation results. In this invention, 300 meters is the optimal distance threshold based on the accuracy evaluation results. Ultimately, this invention classifies contiguous urban areas and scattered urban areas as urban residential areas, and contiguous rural areas and scattered rural areas as rural residential areas.
[0072] Finally, the initial urban and rural residential area classification maps were post-processed to address classification errors caused by spatial extent mismatches in multi-source data and to establish logically sound urban and rural residential area time series data. Because the land extents of GHS-B and NPP-VIIRS-like NTLs do not perfectly match, a small number of scattered residential patches belonging to coastal cities are misclassified as rural residential patches due to the lack of NTL data coverage. To address this issue, this invention first extracts coastal residential patches not covered by NPP-VIIRS-like NTL data (currently all identified as rural residential patches). Subsequently, coastal residential patches located within 1 kilometer of existing urban residential patches are reclassified as urban residential areas.
[0073] To obtain logically sound time-series data, this invention corrects and updates erroneous urban-rural residential area classifications based on contextual spatiotemporal information and urban-rural transformation logic. First, low-probability urban-rural residential area pixels are corrected using a multi-time moving window. Then, based on the fundamental laws of urbanization, this paper establishes the basic logic of urban and rural residential area evolution: the transformation between urban and rural residential areas and non-residential areas is reversible; urban and rural residential areas can be converted back into green spaces, farmland, etc. However, the transformation between urban and rural residential areas is irreversible; once a rural residential area transforms into an urban residential area, it will not regress back to a rural residential area. Based on this, the urban-rural residential area sequence is logically checked, correcting residential patches that do not conform to the basic logic (i.e., regressing from urban to rural residential areas).
[0074] This invention uses an independent validation sample set constructed based on publicly available data and high-resolution remote sensing images to evaluate the accuracy of a global urban and rural residential dataset. First, sample points are randomly generated from the global urban and rural residential map according to the residential area proportions of the nine regions mentioned above. 10,000 sample points are generated in 2020 (if a region has fewer than 500 sample points, they are padded to 500), and 5,000 sample points are generated in each of the other years (if a region has fewer than 300 sample points, they are padded to 300), for a total of 31,474 sample points. Figure 4 As shown in the diagram. Secondly, the city boundaries of the three global long-term series data sets (GUB, MGUP, and NTL-UE) were overlaid. Areas identified as urban by all three sets were considered high-confidence urban residential areas; areas identified as non-urban by all three sets were considered high-confidence potential rural residential areas; and areas with inconsistent classifications across the three sets were considered uncertain areas. Sample points located in high-confidence urban residential areas were classified as urban residential area sample points, sample points located in high-confidence potential rural residential areas were classified as rural residential area sample points, and for sample points located in uncertain areas (a total of 7515), high-resolution remote sensing imagery from Google Earth was used to determine whether they were urban or rural residential area sample points.
[0075] Combining publicly available data with high-resolution remote sensing images to construct a validation sample set can significantly improve validation efficiency. Finally, based on the confusion matrix of the validation samples and results, the overall accuracy (OA), Kappa coefficient, producer accuracy (PA, measuring commissioning error), and user accuracy (UA, measuring omission error) for global urban and rural residential classification are calculated.
[0076] The validation results based on 31,474 samples show that the overall accuracy of the global urban and rural residential data of this invention is 92.39%, and the Kappa coefficient is 0.84. The results are shown in Table 1 below.
[0077] Table 1. GURS Confusion Matrix from 2000 to 2020
[0078]
[0079]
[0080] Note: US represents urban residential area, and RS represents rural residential area.
[0081] Specifically, over the five years (2000, 2005, 2010, 2015, 2020), the overall accuracy rate was above 91% for all years, and the Kappa coefficient was above 0.80. From a user accuracy perspective, rural residential areas had the highest accuracy at 97.76%, because high-brightness and large urban areas are difficult to misclassify as rural. However, rural areas near cities may be misclassified as high-brightness enclaves due to low resolution of nighttime light data or spillover of city lights; therefore, the user accuracy rate for urban residential areas was only 84.91%. From a producer accuracy perspective, both urban and rural residential areas had producer accuracy rates above 90%.
[0082] This invention, by fusing GHS-B, NTL, and OSM data, classifies urban and rural residential areas based on the two-dimensional "area-brightness" attribute of residential patches. It constructs a 100-meter resolution global urban and rural residential area dataset, which effectively characterizes the global rural residential area and improves the accuracy of urban residential area identification to 100 meters. The specific data production process is as follows: First, residential patches with area attributes are extracted from GHS-B data, and brightness attributes are obtained from NTL data. Then, based on the urban / rural labels provided by OSM, the two-dimensional "area-brightness" classification threshold for urban and rural residential areas is determined, and preliminary urban and rural residential areas are divided. Finally, the initial results are updated according to the evolutionary logic of urban and rural residential areas to obtain the final 100-meter global urban and rural residential area map.
[0083] The validation results, based on 31,474 multi-temporal and spatial validation samples, show that the overall accuracy of the global urban and rural residential area data of this invention is 92.39%, with a Kappa coefficient of 0.84. Overall, the 100-meter GURS data can accurately characterize the extent of global urban and rural residential areas.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.
Claims
1. A method for precise identification of urban and rural settlements through multi-source data fusion, characterized in that: Includes the following steps: A. Identify potential urban and rural residential patches: Extract potential residential grids from GHS-B data and aggregate spatially adjacent grids into a residential patch with an area. Then, calculate the area and brightness attributes of the residential patch based on NTL data. B. Differentiate urban and rural residential areas based on two-dimensional attributes of area and brightness: Extract urban and rural residential area patch samples according to OSM, and determine the urban and rural classification threshold of area or brightness according to the intersection of the data distribution of urban and rural residential area patch samples in the area or brightness dimension. Based on the threshold and classification scheme, divide all residential area patches into urban and rural areas and generate an initial residential area map. The steps for determining the threshold for classifying urban and rural residential areas include: Take the natural logarithm of the sample data to reduce the absolute value of the data; Plot the cumulative frequency curve of area / brightness of urban residential patch samples and the inverse cumulative frequency curve of area / brightness of rural residential patch samples; The intersection of the cumulative frequency curve and the inverse cumulative frequency curve is defined as the classification threshold for urban and rural residential areas, thus obtaining the classification thresholds for urban and rural residential areas in two dimensions: area and brightness. The classification scheme includes: using brightness as the x-axis, area as the y-axis, and the classification threshold as the origin, dividing residential patches into four types: contiguous urban areas, contiguous rural areas, scattered rural areas, and high-brightness enclaves. Contiguous urban areas and scattered urban areas are classified as urban residential areas, and contiguous rural areas and scattered rural areas are classified as rural residential areas. Specifically, contiguous urban areas include cities and developed towns; contiguous rural areas include underdeveloped towns and contiguous rural areas; scattered rural areas are scattered small villages; and high-brightness enclaves include scattered cities and scattered rural areas near urban areas and highway light sources. High-brightness enclaves within a certain distance range are classified as scattered cities, and other high-brightness enclaves are classified as scattered rural areas. C. Correct and update the multi-temporal urban and rural residential area classification results, and update the initial map according to the evolution logic of urban and rural residential areas.
2. The method for accurate identification of urban and rural residential areas according to claim 1, characterized in that: The steps for extracting potential residential area grids include: using 30-meter GAIA data as a reference, gradually increasing the built-up area percentage threshold and comparing it with GAIA data, and selecting the built-up area percentage when the two areas are closest.
3. The method for accurate identification of urban and rural residential areas according to claim 1, characterized in that: Step B also includes: repeating the process of "extracting urban and rural residential patch samples based on OSM, and determining the urban-rural classification threshold for area or brightness based on the intersection of the data distribution of urban and rural residential patch samples in the area or brightness dimension" multiple times, and taking the average value to obtain a robust urban-rural residential classification threshold.
4. The method for accurate identification of urban and rural residential areas according to claim 1, characterized in that: Step B, the steps for extracting urban and rural residential patch samples, include: Based on the United Nations Statistics Division's classification of the world, the excessively large residential data groups were split into 9 sub-regions; Extract urban and rural residential areas from OSM-Place point data; Spatial matching of residential patches and residential points; successfully matched urban and rural residential patches are the basic samples with urban and rural residential attributes. The residential area and the number of basic samples of urban / rural residential patches in the nine sub-regions were counted respectively, and the minimum density of basic urban / rural residential patches was calculated. Nine sub-regions were randomly selected based on the minimum density of patches to form a statistical sample of urban / rural residential patches.
5. The method for accurate identification of urban and rural residential areas according to claim 1, characterized in that: In step C, updating the multi-temporal urban-rural residential area classification results includes: extracting coastal residential area patches not covered by NPP-VIIRS-like NTL data, and reclassifying coastal residential area patches located within 1 kilometer of existing urban residential area patches as urban residential areas.
6. The method for accurate identification of urban and rural residential areas according to claim 1, characterized in that: Also includes: The steps to evaluate the accuracy of a global urban and rural residential dataset by constructing an independent validation sample set based on publicly available data and high-resolution remote sensing imagery.