A method for constructing an urban heat island network based on multi-level thermal field recursive decomposition

Abstract

The application discloses a kind of urban heat island network construction method based on multilevel thermal field recursive decomposition in the technical field of urban climate analysis, including obtaining seamless summer average ground surface temperature grid data of initial area, the multidimensional environmental driving factor data required for building thermal resistance surface;Based on mean-standard deviation threshold method, the continuous ground surface temperature field of the initial area is multilevel recursive decomposition, and the main level heat island source is identified, and the nested secondary heat island source is recursively identified inside the main level heat island source, and the hierarchical nested heat island network structure is constructed.The application realizes the accurate decomposition of multilevel heat island source by using mean, standard deviation threshold value for recursive calculation of continuous ground surface temperature field, can effectively identify cross-scale heat island connection pinch point and heat island connection barrier, can realize the hierarchical construction of urban heat island network, and is conducive to improving the management efficiency of urban heat island.

Description

Technical Field

[0001] This invention relates to the field of urban climate analysis technology, specifically a method for constructing urban heat island networks based on multi-level thermal field recursive decomposition. Background Technology

[0002] With the continuous advancement of urbanization, the underlying surface structure of cities is becoming increasingly complex. Extensive impermeable surfaces and high-intensity human activities have led to a significant increase in urban surface temperature, making the urban heat island (UHI) effect increasingly prominent. Urban heat islands not only exacerbate summer heat risks but also adversely affect residents' health, energy consumption, and the stability of urban ecosystems. Therefore, accurately identifying the spatial structure of urban heat islands and their connectivity pathways has become a key technical issue in urban planning and thermal environment management.

[0003] Current research on urban heat islands mainly focuses on the spatial distribution characteristics of surface temperature or single-scale heat island identification methods, such as dividing heat island regions based on fixed thresholds or clustering methods. These methods typically treat heat islands as static, single-level spatial phenomena, making it difficult to reflect the nested structural characteristics of the urban thermal environment at different spatial scales. Furthermore, some studies have begun to introduce network analysis concepts, attempting to understand heat island connectivity processes by constructing connectivity channels or barrier structures. However, these studies often rely on empirical assignments or constructing resistance surfaces using single factors, lacking a quantitative characterization of the combined effects of multiple environmental factors.

[0004] Traditional machine learning models are often considered "black boxes," struggling to explain the varying effects of different environmental factors in different spatial locations, thus limiting their application value in planning practice. Furthermore, existing research often trains models at a single spatial scale, neglecting the differences in the formation mechanisms of urban heat islands at different scale levels. It also suffers from problems such as scale insensitivity to heat island networks, strong subjectivity in resistive surface construction, and difficulty in revealing internally nested thermal structures, all of which hinder the improvement of urban heat island governance efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing urban heat island networks based on multi-level recursive decomposition of thermal fields. By identifying heat island source areas through multi-level recursion, introducing spatially interpretable machine learning to quantify the differences in the effects of multi-source environmental factors at different levels, and constructing a multi-level urban heat island network based on circuit theory, this method can effectively identify cross-scale heat island connectivity pinch points and heat island connectivity barriers, enabling hierarchical construction of urban heat island networks and improving the efficiency of urban heat island governance.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for constructing urban heat island networks based on multi-level recursive thermal field decomposition includes: Acquire seamless summer mean surface temperature raster data of the initial region and construct multi-dimensional environmental driving factor data required for thermal resistance surface; The continuous surface temperature field of the initial region is recursively decomposed into multiple levels based on the mean-standard deviation threshold method, the primary-level heat island source areas are identified, and nested secondary-level heat island source areas are recursively identified within the primary-level heat island source areas to construct a hierarchical nested heat island network structure. Machine learning models with surface temperature as the prediction target are trained for different heat island levels in the spatial structure of the heat island. A spatially interpretable method that takes geospatial location as a joint game participant is adopted to quantify the hierarchical dependence contribution of each environmental factor and construct a hierarchically adaptive thermal drag surface accordingly. Based on the heat island source areas and thermal resistance surfaces at corresponding levels, a multi-level urban heat island network is constructed, and key cross-level heat island connectivity nodes are identified.

[0007] As a further aspect of the present invention: the key heat island connectivity node includes heat island connectivity pinch points and heat island connectivity barriers.

[0008] As a further aspect of the present invention: acquiring seamless summer average land surface temperature (LST) raster data of the initial area, including: Acquire multi-temporal remote sensing images of the initial region; Cloud-contaminated pixels in the multi-temporal remote sensing images are removed using cloud masking bands. A missing value imputation algorithm based on the rate of change of time is used to generate continuous summer average land surface temperature raster data, resulting in seamless summer average land surface temperature raster data.

[0009] As a further aspect of the present invention: the multi-temporal remote sensing image is a multi-temporal Landsat remote sensing image.

[0010] As a further aspect of the present invention: the continuous surface temperature field of the initial region is recursively decomposed into multiple levels based on the mean-standard deviation threshold method to identify primary-level heat island source areas, and nested secondary-level heat island source areas are recursively identified within the primary-level heat island source areas to construct a hierarchically nested heat island network structure, including: At the initial regional scale, a threshold is set based on the mean and standard deviation of surface temperature, and areas with temperatures higher than the sum of the mean and a preset multiple of the standard deviation are defined as heat island patches. The core region of the heat island patch was extracted by applying morphological spatial pattern analysis, and core patches with an area greater than a preset threshold were selected as candidate source areas. Calculate the overall connectivity index and potential connectivity index values ​​of each candidate source area, and select patches that meet the preset connectivity threshold as primary-level heat island source areas; One or more typical heat island patches are selected from the primary-level heat island source areas. The mean and standard deviation of the local surface temperature within each patch are recalculated. Based on the set threshold, candidate source areas, and primary-level heat island source areas, the area threshold is adjusted to a smaller value to identify nested secondary-level heat island source areas. Typical heat island patches include the largest continuous heat island region, and the smaller value includes 0.1 km². 2 .

[0011] As a further aspect of the present invention: by calculating the overall connectivity index and potential connectivity index values ​​of each candidate source area, patches that meet the preset connectivity threshold are selected as primary-level heat island source areas, thus completing the landscape connectivity assessment. As a further aspect of this invention: the standard deviation is represented by σ, the mean refers to the average surface temperature within the corresponding level range, morphological spatial pattern analysis is abbreviated as MSPA, and the threshold is 1 km. 2 The overall connectivity index is abbreviated as dIIC, and the potential connectivity index is abbreviated as dPC. The preset connectivity thresholds are dPC>0.2 and dIIC>0.1.

[0012] As a further aspect of this invention: Machine learning models with surface temperature as the prediction target are trained for different heat island levels in the spatial structure of the heat island. A spatially interpretable method that treats geographic location as a joint game participant is employed to quantify the hierarchical dependency contribution of each environmental factor. Based on this, a hierarchically adaptive thermal drag surface is constructed, including: Multiple environmental factors are obtained to obtain characteristic variables, including land use type, normalized vegetation index (NDVI), improved normalized water index (MNDWI), digital elevation model (DEM), slope, building height (BH), normalized building index (NDBI), nighttime light (NL) and road network density (RD). Using surface temperature as the prediction target, we trained a primary-level hierarchical machine learning model and a secondary-level hierarchical machine learning model. Using each resistance factor as the independent variable and LST as the dependent variable, the target machine learning model is obtained by training the primary level and the secondary level separately. The GeoSHAPley spatial interpretability method is adopted, which takes geospatial location as a joint game participant and decomposes the contribution of the model prediction results to obtain the main effect contribution of each environmental factor and its interaction effect contribution as spatial location changes. Based on the sign and magnitude of the GeoSHAPley main effect contribution, each driving factor is assigned a graded value. Factors with a strong contribution to heating are assigned a lower resistance value, while factors with a strong contribution to cooling are assigned a higher resistance value. A hierarchical adaptive thermal resistance surface is generated by weighted superposition.

[0013] As a further aspect of the present invention, the target machine learning model obtained through training can systematically characterize the hindering effect of environmental factors such as urban morphology, distribution of anthropogenic heat sources, and infrastructure on urban heat island connectivity.

[0014] As a further aspect of the present invention: contribution decomposition of the model prediction results includes: Based on the game theory framework, the interaction effects between the geographical location and other non-spatial features of the heat island spatial structure hierarchy are calculated, and three types of contribution effects are explicitly separated.

[0015] As a further aspect of the present invention, the three types of contribution effects include intrinsic spatial effects, position-invariant feature effects, and spatial-feature interaction effects.

[0016] As a further aspect of the present invention: the intrinsic spatial effect represents the fixed influence of location itself on LST; the location-invariant characteristic effect represents the average effect of the factor across the entire domain; and the spatial-characteristic interaction effect represents the heterogeneity of the influence of the same factor on LST in different locations.

[0017] As a further aspect of the present invention: based on the corresponding levels of heat island source areas and thermal resistance surfaces, a multi-level urban heat island network is constructed, and key cross-level heat island connectivity nodes are identified, including: Using primary-level and secondary-level heat island source areas as nodes and corresponding thermal resistance surfaces as conduction media, a spatial network analysis tool based on circuit theory is used to simulate the heat island connection path, thereby constructing a primary-level heat island network and nested secondary-level heat island networks. Run the Pinchpoint Mapper module to extract high-density regions as heat island connectivity pinch points based on cumulative current density analysis. Run the Barrier Mapper module to extract regions with high improvement scores as heat island connectivity barriers based on connectivity improvement potential analysis. By spatially overlaying and integrating key nodes at multiple levels, cross-level collaborative identification can be achieved.

[0018] As a further aspect of the present invention, the spatial network analysis tool is the Linkage Mapper module.

[0019] As a further aspect of the present invention: the target machine learning model is an ensemble learning model based on decision trees, and is repeatedly trained in each level of the heat island to adapt to the local thermal field characteristics.

[0020] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves accurate decomposition of multi-level heat island source areas by recursively calculating the continuous surface temperature field using mean and standard deviation thresholds. It can decompose the nested structure of heat islands and construct a multi-level heat island network by using a spatially interpretable method that takes geographical location as a joint game participant to build objective resistance surfaces at each level. This successfully reveals the scale nested structure: the recursive decomposition method not only identifies macro heat islands at the urban scale, but also further identifies more refined secondary heat island clusters within high-intensity built-up areas, providing a multi-level perspective for urban thermal environment analysis.

[0021] 2. By introducing a spatially interpretable machine learning method, this invention can quantitatively reveal the differentiated action mechanisms of multi-source environmental factors at different spatial levels and locations, overcome the shortcomings of traditional subjective weighting, realize the objectification and spatial explicitness of the resistance surface, and quantify the spatial heterogeneity of the contributions of each driving factor.

[0022] 3. This invention generates a multi-level heat island network and a key node map, providing an actionable planning basis that can transform complex urban thermal environment problems into clear spatial connections and barriers. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the method steps of the multi-level urban heat island network construction method provided by the present invention; Figure 2 This is a schematic diagram illustrating the principle of multi-level thermal field recursive decomposition provided by the present invention. Figure 3 This is a map showing the summer surface temperature inversion results in Shanghai as an example in this embodiment of the invention; Figure 4 This is a graph showing the GeoSHAPley analysis results of the resistance factor in an embodiment of the present invention, taking Shanghai as an example. (a) is a summary graph of the GeoSHAPley values ​​of the primary resistance factor, (b) is a summary graph of the GeoSHAPley values ​​of the secondary resistance factor, (c) is a ranking graph of the global contribution of the GeoSHAPley values ​​of the primary resistance factor, and (d) is a ranking graph of the global contribution of the GeoSHAPley values ​​of the secondary resistance factor. Figure 5 This is a diagram showing the construction result of the resistance surface in Shanghai as an example in this embodiment of the invention, where (a) is the main level resistance surface and (b) is the secondary level resistance surface; Figure 6 This is a multi-level urban heat island network result diagram constructed using Shanghai as an example in this embodiment of the invention, wherein (a) is a schematic diagram of hierarchical network decomposition and nesting, (b) is a hierarchical diagram of the main level heat island network, (c) is a hierarchical diagram of the secondary level heat island network, and (d) is a result diagram of the hierarchical nested superimposed heat island network. Figure 7This is a distribution map of cross-level key heat island connectivity nodes (pinch points and barriers) identified in this embodiment of the invention, taking Shanghai as an example. Detailed Implementation

[0024] 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.

[0025] Example: The initial area of ​​this embodiment is a portion of the continuous built-up area of ​​Shanghai, covering 15 districts: Huangpu, Xuhui, Changning, Jing'an, Putuo, Hongkou, Yangpu, Minhang, Baoshan, Jiading, Pudong New Area, Fengxian, Songjiang, Jinshan, and Qingpu (excluding Chongming Island), with a total area of ​​approximately 5438 km². 2 This region has a high degree of urbanization and a significant heat island effect, making it an ideal case for validating the method of this invention. However, the method of this invention is not limited to this and can be applied to thermal environment analysis in other megacities or at different scales.

[0026] like Figure 1 As shown in the figure, this embodiment provides a method for constructing an urban heat island network based on multi-level thermal field recursive decomposition, including the following steps: S1: Obtain seamless summer average surface temperature raster data for the initial region and construct the multi-dimensional environmental driving factor data required for the thermal resistance surface.

[0027] All data used in this embodiment are from publicly available platforms. Specific types, sources, resolutions, and years are shown in Table 1. All data were uniformly projected to the WGS_1984_UTM_Zone_51N coordinate system in ArcGIS Pro 3.1 software and resampled to a spatial resolution of 30 meters to ensure spatial alignment.

[0028] Table 1. List of data used in the embodiments .

[0029] Step S1, acquiring initial seamless summer mean land surface temperature (LST) raster data for the region, includes the following sub-steps: S11. Process Landsat 8 Collection 2 Level 2 (C2 L2) surface reflectance and surface temperature products based on the Google Earth Engine (GEE) platform (dataset ID: LANDSAT / LC08 / C02 / T1_L2). S12. Apply a Quality Assessment (QA) mask and use the QA_PIXEL and QA_RADSAT bands to remove low-quality pixels affected by clouds, cirrus clouds, cloud shadows, snow, fill value, and saturation. S13. For the surface reflectance band (SR_B*), the standard Collection 2 scaling parameters (multiplicative factor: 0.0000275; additive offset: -0.2) are used to convert it into a physical quantity; for the Kelvin temperature value represented by the infrared band (ST_B10), the scaling parameters of the GEE platform (multiplicative factor: 0.00341802; additive offset: 149.0) are applied for calibration before further conversion to Celsius for analysis. This scaling step is indispensable because although the C2 L2 data is atmospherically corrected, it is stored in GEE in a scaled integer format. S14. To address the persistent gap caused by cloud contamination, calculate the long-term monthly LST variability of each pixel during the summer months (June, July, and August) in the historical period (2017-2022) as a reconstruction constraint. The specific formula is as follows: In the formula: and These represent the average surface temperature differences for June-July, June-August, and July-August, respectively. , and Represents pixels LST values ​​for June, July, and August of the corresponding year; Indicates the number of valid observations.

[0030] S15. Employ a priority-based imputation algorithm: prioritize retaining the original LST values ​​of valid observation pixels for the target year (2022); for cloud-contaminated pixels, reconstruct missing values ​​using historical inter-month relationships, with the specific formula as follows: In the formula: , and These are the pixels from 2022. exist , and The original observations. , and They are pixels Place and , and , and The difference in average LST between them; , and Representing pixels exist , and Fill the gaps with LST values; S16. Finally, calculate the seamless average LST value of each pixel in the summer (June-August) of the target year (2022), using the following formula: In the formula: Represents summer month pixels The average surface temperature.

[0031] like Figure 3 As shown, a spatially continuous seamless LST grid is ultimately generated. The results show that the average surface temperature in the built-up area of ​​Shanghai in the summer of 2022 was 43.78°C, with a maximum temperature of 71.85°C, indicating significant spatial heterogeneity.

[0032] In this embodiment, the multi-temporal remote sensing image is a multi-temporal Landsat remote sensing image.

[0033] S2: Based on the mean-standard deviation threshold method, the continuous surface temperature field of the initial region is recursively decomposed into multiple levels to identify the primary-level heat island source areas. Nested secondary-level heat island source areas are then recursively identified within the primary-level heat island source areas to construct a hierarchical nested heat island network structure.

[0034] In this embodiment, step S2 includes: S21: At the initial regional scale, based on the mean and standard deviation of surface temperature, a threshold is set, and areas exceeding the sum of the mean and a preset multiple of the standard deviation are defined as heat island patches. Specifically, the LST classification criteria and value ranges are shown in Table 2 below: Table 2 LST Classification Criteria and Value Range .

[0035] Standardized surface temperature; : Average standardized surface temperature; σ: Standard deviation.

[0036] S22: Using the heat island patch area as the foreground and the non-heat island area as the background, morphological spatial pattern analysis (MSPA) was performed using GuidosToolbox 3.3 software to extract the "core" type patches.

[0037] S23: Select areas larger than 1 km² 2For the core patches, the overall connectivity index (dIIC) and potential connectivity index (dPC) values ​​of each candidate source region were calculated using Conefor 2.6 software. Patches that met the preset connectivity thresholds (e.g., dPC>0.2, dIIC>0.1) were selected as primary-level heat island source regions. The specific formula is as follows: In the formula: n represents the total number of patches in the landscape, a i and a j Let i and j represent the areas of patch i and patch j, respectively, in km. 2 ;nl ij A represents the number of connections between patch i and patch j; L It is the total landscape area, in km² 2 ; I represents the maximum product probability of all paths between patch i and j; I is the connectivity index value of a certain landscape, which in this paper refers to the overall connectivity index IIC or the potential connectivity index PC; Iremove is the connectivity index value of the landscape after removing patch I from the landscape.

[0038] This embodiment identified a total of 43 primary-level heat island source areas, with a total area of ​​1220.23 km². 2 The largest contiguous source area spans 11 administrative districts in Puxi, covering an area of ​​946.50 km². 2 It is defined as a continuous giant heat island region (CMUHI).

[0039] S24: Using the area within the aforementioned continuous giant heat island region (CMUHI) as the new analysis scope, recalculate the local mean and standard deviation, recursively apply the same threshold and screening process, but adjust the area threshold to 0.1 km². 2 It identified 164 nested secondary heat island source areas, with a total area of ​​131.98 km². 2 This process revealed a more refined local thermal heterogeneity structure within the continuous giant heat island region.

[0040] In this embodiment, by calculating the overall connectivity index and potential connectivity index values ​​of each candidate source area, patches that meet the preset connectivity threshold are selected as primary-level heat island source areas. In this embodiment, the standard deviation is represented by σ, the mean refers to the average surface temperature within the corresponding level range, morphological spatial pattern analysis is abbreviated as MSPA, and the threshold is 1 km. 2 The overall connectivity index is abbreviated as dIIC, and the potential connectivity index is abbreviated as dPC. The preset connectivity thresholds are dPC>0.2 and dIIC>0.1.

[0041] S3: For different heat island levels in the spatial structure of the heat island, machine learning models with surface temperature as the prediction target are trained respectively. A spatially interpretable method that takes geospatial location as a joint game participant is adopted to quantify the hierarchical dependence contribution of each environmental factor, and a hierarchically adaptive thermal resistance surface is constructed accordingly.

[0042] In this embodiment, step S3 includes: S31: Obtain multi-source environmental factors and obtain characteristic variables. Among them, multi-source environmental factors include land use type, normalized vegetation index, improved normalized water index, digital elevation model, slope, building height, normalized building index, nighttime light and road network density. S32: Using surface temperature as the prediction target, train both primary-level and secondary-level machine learning models. Specifically, for the macro-level (global) and secondary-level (CMUHI region) models, divide the corresponding environmental factor datasets into training and testing sets according to a preset ratio (e.g., 8:2). Train various preset machine learning regression models (including but not limited to XGBoost, CatBoost, LightGBM, and Random Forest) based on the training set. Then, train the models according to the evaluation metrics of the testing set (e.g., coefficient of determination R). 2 It automatically selects the best-performing model for each level as the land surface temperature prediction model for that level.

[0043] S33: Calculate the spatially interpretable contribution: Apply the GeoSHAPley spatially interpretable framework to the optimal models selected in step S32. Use geographic coordinates (latitude and longitude) and environmental factors as joint feature inputs to the interpreter to decompose the model predictions, such as... Figure 4 As shown, the output shows the location-invariant main effect values ​​of each environmental factor (reflecting the global fundamental contribution of the factor to surface temperature) and the spatial interaction effect values.

[0044] S34: Perform adaptive resistance grading and resistance surface generation.

[0045] Table 3. Primary Level Resistance Factors and Their Weights .

[0046] Table 4. Secondary Resistance Factors and Their Weights .

[0047] In this embodiment, the contribution decomposition of the model prediction results includes: Based on the game theory framework, the interaction effects between the geographical location and other non-spatial features of the heat island spatial structure hierarchy are calculated, and three types of contribution effects are explicitly separated.

[0048] In this embodiment, the three types of contribution effects include intrinsic spatial effects, position-invariant feature effects, and spatial-feature interaction effects.

[0049] In this embodiment, the intrinsic spatial effect represents the fixed influence of location itself on LST; the location-invariant characteristic effect represents the average effect of the factor across the entire domain; and the spatial-characteristic interaction effect represents the heterogeneity of the influence of the same factor on LST in different locations.

[0050] The specific assignment method is as follows: based on the positive and negative values ​​and magnitudes of the GeoSHAPley main effect contribution, each drag factor is assigned a graded value, with factors that contribute more to heating being assigned a lower drag value and factors that contribute more to cooling being assigned a higher drag value, and a hierarchical adaptive thermal drag surface is generated by weighted superposition.

[0051] S35: Based on the joint distribution of GeoSHAPley main effect values ​​(reflecting the net impact of factors on surface temperature) and original factor values, adaptive classification is performed: For categorical variables, the average main effect value for each class is calculated; for continuous variables, the average main effect value for each interval is calculated using a binning method (e.g., equal-frequency binning); for variables with a large number of zero values, the zero values ​​are processed separately and the remaining values ​​are binned; all categories or intervals are sorted according to their average main effect values ​​and assigned drag levels from low to high; the total contribution value of each factor (including main effects and interaction effects) is extracted, normalized, and used as an objective weight for weighted superposition to generate a comprehensive thermal drag surface.

[0052] In this embodiment, a target machine learning model is trained based on a primary-level hierarchical machine learning model and a secondary-level hierarchical machine learning model, which can systematically characterize the hindering effect of environmental factors such as urban morphology, distribution of anthropogenic heat sources, and infrastructure on urban heat island connectivity.

[0053] Furthermore, the S3-level adaptive thermal resistance surface construction process of the present invention can be represented by the following pseudocode flow, as shown in Tables 5-6 below: Table 5 Table 6 The S34 resistance factor grading assignment method of the present invention can be represented by the pseudocode flow shown in Tables 7-8 below: Table 7 Table 8 .

[0054] S4: Based on the heat island source areas and thermal resistance surfaces at the corresponding levels, construct a multi-level urban heat island network and identify key heat island connection nodes across levels.

[0055] In this embodiment, the key heat island connectivity nodes include heat island connectivity pinch points and heat island connectivity barriers.

[0056] Specifically, step S4 includes the following sub-steps: S41. Using the Linkage Mapper toolkit, with 43 primary-level source sites and 164 secondary-level source sites as "nodes" and the corresponding resistance surfaces generated in step S4 as "conduction media," the minimum cost path and random walk probability of heat island connectivity are simulated based on circuit theory. This ultimately constructs a macroscopic heat island network containing 103 corridors and a nested secondary-level heat island network containing 450 corridors. Figure 6 The macro-network corridors are long and sparse, forming the framework for regional heat island connectivity; the secondary network corridors are short and dense, reflecting the complex heat exchange paths within the high-intensity built-up area. S42. Run the Pinchpoint Mapper module ("all-to-one" mode) to generate a cumulative current density map. Divide the current density values ​​into 5 levels using the natural breakpoint method, and extract the top two levels as heat island connectivity pinch points. A pinch point region of 16.09 km was identified in the macroscopic network. 2 It is mainly distributed along the Huangpu River and other heat flow convergence channels; a pinch area of ​​41.86 km was identified in the secondary network. 2 They are densely distributed in the core areas of the old city, such as Jing'an and Hongkou. S43. Run the Barrier Mapper module and use moving window analysis to calculate the potential improvement in network connectivity if each cell were removed (i.e., transformed into a low-resistance cell). The improvement scores are categorized into 5 levels using the natural break method, and the top two levels are extracted as heat island connectivity barriers. Barrier regions totaling 183.98 km were identified in both the macro and secondary networks. 2 and 249.29 km 2 ,like Figure 7 As shown, they often overlap with large parks, bodies of water, or existing green space systems; S44. By spatially superimposing and integrating the results of the two levels, a cross-scale key thermal regulation node system is formed, which provides a clear spatial target area for targeted intervention and supports the coordinated regulation strategy of "blocking macro corridors and dissipating local networks".

[0057] In this embodiment, the target machine learning model is an ensemble learning model based on decision trees, and it is retrained in the sub-level heat island layer to adapt to the local thermal field characteristics.

[0058] This invention achieves accurate decomposition of multi-level heat island source areas by recursively calculating the continuous surface temperature field using mean and standard deviation thresholds. It can decompose the nested structure of heat islands and construct hierarchical objective resistance surfaces by using geographical location as a joint game participant to build a multi-level heat island network. This effectively identifies cross-scale heat island connectivity pinch points and barriers, enabling hierarchical construction of urban heat island networks and improving the efficiency of urban heat island governance. By constructing a multi-level recursive decomposition algorithm for the thermal field, it can identify the hierarchical nested structure of urban heat islands, overcoming the limitations of traditional single-scale heat island analysis. By introducing spatially interpretable machine learning methods, it can quantitatively reveal the differentiated action mechanisms of multi-source environmental factors at different spatial levels and locations, improving the interpretability of thermal environment analysis results. By constructing an urban heat island network based on hierarchical adaptive thermal resistance surfaces and circuit theory, it can effectively identify key heat island connectivity paths and nodes, providing refined spatial guidance for thermal environment regulation. Furthermore, it has relatively simple data requirements, a clear process, strong operability, and is applicable to thermal environment analysis and planning practices at different urban scales, demonstrating good scenario adaptability.

[0059] In summary, this embodiment, using Shanghai as a case study, fully and clearly demonstrates the entire process of this invention, from data preprocessing, recursive source identification, intelligent resistance surface construction to multi-level nested network analysis. The results show that this method can effectively characterize the complex spatial structure of the thermal environment in megacities, and its analytical results can directly serve climate adaptation planning and precise thermal regulation decisions, possessing significant practical application value.

[0060] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing an urban heat island network based on multi-level recursive decomposition of thermal fields, characterized in that, include: Acquire seamless summer mean surface temperature raster data of the initial region and construct multi-dimensional environmental driving factor data required for thermal resistance surface; The continuous surface temperature field of the initial region is recursively decomposed into multiple levels based on the mean-standard deviation threshold method, the primary-level heat island source areas are identified, and nested secondary-level heat island source areas are recursively identified within the primary-level heat island source areas to construct a hierarchical nested heat island network structure. Machine learning models with surface temperature as the prediction target are trained for different heat island levels in the spatial structure of the heat island. A spatially interpretable method that takes geospatial location as a joint game participant is adopted to quantify the hierarchical dependency contribution of each environmental factor and construct a hierarchically adaptive thermal drag surface accordingly. Based on the heat island source areas and thermal resistance surfaces at corresponding levels, a multi-level urban heat island network is constructed, and key cross-level heat island connectivity nodes are identified.

2. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 1, characterized in that, The key heat island connectivity nodes include heat island connectivity pinch points and heat island connectivity barriers.

3. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 2, characterized in that, Obtain seamless summer mean surface temperature raster data for the initial region, including: Acquire multi-temporal remote sensing images of the initial region; Cloud-contaminated pixels in the multi-temporal remote sensing images are removed using cloud masking bands. A missing value imputation algorithm based on the rate of change of time is used to generate continuous summer average land surface temperature raster data, resulting in seamless summer average land surface temperature raster data.

4. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 3, characterized in that, The continuous surface temperature field of the initial region is recursively decomposed into multiple levels based on the mean-standard deviation threshold method. This identifies primary-level heat island source areas and recursively identifies nested secondary-level heat island source areas within these primary-level source areas, constructing a hierarchically nested heat island network structure, including: At the initial regional scale, a threshold is set based on the mean and standard deviation of surface temperature, and areas with temperatures higher than the sum of the mean and a preset multiple of the standard deviation are defined as heat island patches. The core region of the heat island patch was extracted by applying morphological spatial pattern analysis, and core patches with an area greater than a preset threshold were selected as candidate source areas. Calculate the overall connectivity index and potential connectivity index values ​​of each candidate source area, and select patches that meet the preset connectivity threshold as primary-level heat island source areas; One or more typical heat island patches are selected from the primary-level heat island source areas. The mean and standard deviation of the local surface temperature within each patch are recalculated. Based on the set threshold, candidate source areas, and primary-level heat island source areas, the area threshold is adjusted to a smaller value to identify nested secondary-level heat island source areas. Typical heat island patches include the largest continuous heat island region, and the smaller value includes 0.1 km². 2 .

5. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 4, characterized in that, Machine learning models predicting surface temperature are trained for different heat island levels based on their spatial structure. A spatially interpretable method, incorporating geospatial location as a joint game participant, is employed to quantify the hierarchical dependence of various environmental factors. Based on this, a hierarchically adaptive thermal drag surface is constructed, including: Multiple environmental factors are obtained to obtain characteristic variables, wherein the multiple environmental factors include land use type, normalized vegetation index, improved normalized water index, digital elevation model, slope, building height, normalized building index, nighttime light and road network density; Using surface temperature as the prediction target, we trained a primary-level hierarchical machine learning model and a secondary-level hierarchical machine learning model. Using each resistance factor as the independent variable and LST as the dependent variable, the target machine learning model is obtained by training the primary level and the secondary level separately. The GeoSHAPley spatial interpretability method is adopted, which takes geospatial location as a joint game participant and decomposes the contribution of the model prediction results to obtain the main effect contribution of each environmental factor and its interaction effect contribution as spatial location changes. Based on the sign and magnitude of the GeoSHAPley main effect contribution, each driving factor is assigned a graded value. Factors with a strong contribution to heating are assigned a lower resistance value, while factors with a strong contribution to cooling are assigned a higher resistance value. A hierarchical adaptive thermal resistance surface is generated by weighted superposition.

6. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 5, characterized in that, The contribution decomposition of the model prediction results includes: Based on the game theory framework, the interaction effects between the geographical location and other non-spatial features of the heat island spatial structure hierarchy are calculated, and three types of contribution effects are explicitly separated.

7. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 6, characterized in that, The three types of contribution effects include intrinsic spatial effects, location-invariant feature effects, and spatial-feature interaction effects.

8. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 7, characterized in that, The intrinsic spatial effect includes the fixed influence of location itself on LST; the location-invariant characteristic effect includes the average effect of the factor across the entire domain; the spatial-characteristic interaction effect includes the heterogeneity of the influence of the same factor on LST in different locations.

9. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 8, characterized in that, Based on the corresponding levels of heat island source areas and thermal resistance surfaces, a multi-level urban heat island network is constructed, and key cross-level heat island connectivity nodes are identified, including: Using primary-level and secondary-level heat island source areas as nodes and corresponding thermal resistance surfaces as conduction media, a spatial network analysis tool based on circuit theory is used to simulate the heat island connectivity path, thereby constructing a primary-level heat island network and nested secondary-level heat island networks. Run the key point mapper module and extract high-density regions as heat island connectivity pinch points based on cumulative current density analysis. Run the obstacle point identifier module and extract regions with high improvement scores as heat island connectivity barriers based on connectivity improvement potential analysis; By spatially overlaying and integrating key nodes at multiple levels, cross-level collaborative identification can be achieved.

10. The method for constructing an urban heat island network based on multi-level thermal field recursive decomposition according to claim 9, characterized in that, The target machine learning model is an ensemble learning model based on decision trees, and it is repeatedly trained in the heat island range at each level to adapt to the local thermal field characteristics.

Images