Urban heat island intensity analysis method and system based on remote sensing inversion

Abstract

The invention discloses an urban heat island intensity analysis method and system based on remote sensing inversion, and the method comprises the steps: collecting multispectral remote sensing image data through a satellite sensor, carrying out the geometric correction and atmospheric correction of an image, and obtaining a corrected remote sensing image for the subsequent temperature inversion; a normalized vegetation index is calculated according to the corrected remote sensing image, and a vegetation index distribution map is obtained through calculation according to the difference value and sum value ratio of the red light band and the near-infrared band to reflect the vegetation coverage density; identifying a high-density building area from the corrected remote sensing image by adopting a building extraction algorithm, and obtaining a building density distribution diagram through combination of texture features and spectral features to represent a building coverage proportion; and judging the temperature gradient change in the accurate surface temperature map, if the temperature gradient exceeds a preset threshold, marking the temperature gradient as a heat island boundary line, and smoothing the boundary through a spatial interpolation method to obtain a heat island boundary contour for intensity distribution analysis.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and system for analyzing urban heat island intensity based on remote sensing inversion. Background Technology

[0002] The field of urban heat island research focuses on revealing the phenomenon of heat accumulation during urbanization. This phenomenon directly affects residents' quality of life and energy consumption patterns, and has become a crucial aspect that cannot be ignored in environmental management. Current monitoring methods mainly rely on ground-based stations to collect temperature data. While these stations can provide accurate local readings, their uneven distribution leads to gaps in heat mapping, especially in rapidly expanding urban fringe areas where the heat island outline is often distorted.

[0003] While remote sensing imagery can cover vast areas and retrieve surface temperatures, temperature values ​​are easily affected by surface cover type. For example, low reflectivity in vegetated areas can lead to cooler temperature readings, while high-density building areas with higher emissivity amplify the heat signal. This cover difference distorts the temperature distribution within the same image, thus blurring the delineation of the heat island boundary. Vegetation index and building density, as key factors influencing radiation, further complicate temperature correction. Variations in vegetation density affect building shadows, resulting in incomplete separation of local radiation superposition effects, ultimately making it difficult to maintain the spatial continuity of heat island intensity. Therefore, accurately integrating radiation corrections based on vegetation index and building density into broad remote sensing coverage is crucial for achieving accurate prediction of the spatial distribution of heat island intensity. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method and system for analyzing urban heat island intensity based on remote sensing inversion.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for analyzing urban heat island intensity based on remote sensing inversion includes: Multispectral remote sensing image data is acquired by satellite sensors, and geometric and atmospheric corrections are performed on the images to obtain corrected remote sensing images for subsequent temperature inversion. The normalized vegetation index is calculated based on the corrected remote sensing image. The vegetation index distribution map is obtained by calculating the difference and sum ratio between the red band and the near-infrared band to reflect the vegetation cover density. High-density building areas are identified from the corrected remote sensing images. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio. If the vegetation index is lower than the preset threshold, a building density weighting factor is added to determine the degree of radiation interference and correct the initial value of the surface temperature. The corrected surface temperature distribution is obtained, the initial temperature is retrieved from the thermal infrared band of the corrected remote sensing image using a single-channel algorithm, and then adjusted by applying a radiation interference model to obtain an accurate surface temperature map showing the heat accumulation area. The temperature gradient changes in the accurate surface temperature map are determined. If the temperature gradient exceeds a preset threshold, it is marked as the boundary line of the heat island. The boundary is smoothed by spatial interpolation method to obtain the heat island boundary profile for intensity distribution analysis. By training on accurate surface temperature maps and heat island boundary contours, and by inputting temperature characteristics and boundary location data, the intensity distribution is predicted, resulting in a spatial distribution map of heat island intensity that reflects the overall thermal environment pattern.

[0006] As a preferred approach, a building extraction algorithm is used to identify high-density building areas from the corrected remote sensing image. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio.

[0007] As a preferred approach, a radiation interference model is constructed using vegetation index distribution maps and building density distribution maps. If the vegetation index is lower than a preset threshold, a building density weighting factor is superimposed to determine the degree of radiation interference and correct the initial value of the surface temperature.

[0008] As a preferred approach, a random forest model is used to train accurate surface temperature maps and heat island boundary contours. By inputting temperature features and boundary location data, the intensity distribution is predicted, resulting in a spatial distribution map of heat island intensity that reflects the overall thermal environment pattern.

[0009] This invention also provides a system for analyzing urban heat island intensity based on remote sensing inversion, comprising: The first processing module is used to acquire multispectral remote sensing image data through satellite sensors, perform geometric correction and atmospheric correction on the images, and obtain corrected remote sensing images for subsequent temperature inversion. The second processing module is used to calculate the normalized vegetation index based on the corrected remote sensing image. It calculates the vegetation index distribution map by using the difference and sum ratio of the red band and the near-infrared band to reflect the vegetation cover density. The third processing module is used to identify high-density building areas from the corrected remote sensing images. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio. The fourth processing module is used to add a building density weighting factor if the vegetation index is lower than a preset threshold, and to determine the degree of radiation interference in order to correct the initial value of the surface temperature. The fifth processing module is used to obtain the corrected surface temperature distribution. It retrieves the initial temperature from the thermal infrared band of the corrected remote sensing image using a single-channel algorithm and applies a radiation interference model to adjust it, thereby obtaining an accurate surface temperature map that shows the heat accumulation area. The sixth processing module is used to determine the temperature gradient change in the accurate surface temperature map. If the temperature gradient exceeds the preset threshold, it is marked as the heat island boundary line. The boundary is smoothed by spatial interpolation method to obtain the heat island boundary profile for intensity distribution analysis. The seventh processing module is used to train the accurate surface temperature map and the heat island boundary contour. By inputting temperature characteristics and boundary location data, it predicts the intensity distribution and obtains a spatial distribution map of heat island intensity that reflects the overall thermal environment pattern.

[0010] As a preferred option, the third processing module uses a building extraction algorithm to identify high-density building areas from the corrected remote sensing image. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio.

[0011] As a preferred option, the fourth processing module constructs a radiation interference model using vegetation index distribution map and building density distribution map. If the vegetation index is lower than a preset threshold, a building density weighting factor is superimposed to determine the degree of radiation interference and correct the initial value of the surface temperature.

[0012] As a preferred option, the seventh processing module uses a random forest model to train the accurate surface temperature map and heat island boundary contour. By inputting temperature features and boundary location data, it predicts the intensity distribution and obtains a spatial distribution map of heat island intensity that reflects the overall thermal environment pattern.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires multispectral remote sensing image data using satellite sensors, performs geometric and atmospheric corrections on the images, and obtains corrected remote sensing images for subsequent temperature inversion. Based on the corrected remote sensing images, a normalized vegetation index (NVI) is calculated. The vegetation index distribution map, reflecting vegetation cover density, is obtained by calculating the ratio of the difference and sum of the red and near-infrared bands. A building extraction algorithm is used to identify high-density building areas from the corrected remote sensing images. By combining texture and spectral features, a building density distribution map representing the building cover ratio is obtained. Temperature gradient changes are assessed in the precise surface temperature map. If the temperature gradient exceeds a preset threshold, it is marked as a heat island boundary. The boundary is smoothed using spatial interpolation to obtain the heat island boundary contour for intensity distribution analysis. This invention achieves comprehensive correction and intensity prediction of vegetation-building radiation interference, significantly improving the accuracy of heat island boundary identification. Attached Figure Description

[0014] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the urban heat island intensity analysis method based on remote sensing inversion according to an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Example 1 like Figure 1 As shown, this invention provides a method for analyzing urban heat island intensity based on remote sensing inversion, comprising: S101. Collect multispectral remote sensing image data through satellite sensors, perform geometric correction and atmospheric correction processing on the images, and obtain corrected remote sensing images for subsequent temperature inversion.

[0019] Multispectral remote sensing image data is acquired using satellite sensors. Geometric correction is performed on the multispectral remote sensing image data to obtain a geometrically corrected image. Atmospheric correction is then performed on the geometrically corrected image to obtain a corrected image. Spectral reflectance values ​​are obtained from the corrected image. Surface temperature values ​​are determined using a split-window algorithm based on the spectral reflectance values. If abnormal reflectance is found based on the surface temperature values, the abnormal areas are marked to obtain a marked image. Temperature values ​​for normal areas are then obtained from the marked image.

[0020] For example, multispectral remote sensing image data can be acquired using satellite sensors, such as the OLI sensor on the Landsat 8 satellite, which acquires data in the visible, near-infrared, and thermal infrared bands with a resolution of 30 meters. The sensor captures surface reflection and radiation information, forming image data containing multiple bands.

[0021] For example, for a farmland area, the acquired imagery contains blue, green, red, and near-infrared bands for subsequent analysis. Performing geometric correction processing on multispectral remote sensing images can eliminate geometric distortions caused by topography, sensor tilt, etc.

[0022] Specifically, ground control points are used in conjunction with digital elevation models, and the images are projected onto a standard geographic coordinate system using a polynomial correction method.

[0023] For example, in a mountainous area, the image may have pixel shifts due to terrain undulations. After correction, the pixel positions can be matched with the actual ground features, improving the accuracy to the sub-pixel level, which helps to improve the accuracy of subsequent analysis.

[0024] In one possible implementation, atmospheric correction processing obtains a true surface reflectance image by removing the effects of atmospheric scattering and absorption.

[0025] For example, using the 6S radiative transfer model and inputting parameters such as atmospheric water vapor content and aerosol optical thickness, the reflectance values ​​of the corrected image are closer to the true spectral characteristics of the Earth's surface. The corrected image can clearly reflect the health status of farmland vegetation, facilitating subsequent spectral analysis. When obtaining spectral reflectance values ​​from the corrected image, the reflectance of each pixel in different spectral bands can be extracted.

[0026] For example, healthy vegetation has a higher reflectance in the near-infrared band, approximately 0.4–0.6, while it has a lower reflectance in the red band, approximately 0.1–0.2. These values ​​can be used to calculate the vegetation index or surface temperature.

[0027] Specifically, the split-window algorithm uses thermal infrared band data to estimate surface temperature.

[0028] For example, the Landsat 8's TIRS sensor provides data in bands 10 and 11. By analyzing the difference in radiation between the two bands and combining atmospheric correction parameters, surface temperature can be estimated. Calculations for a certain farmland area show that the temperature in healthy areas is approximately 25°C, while in arid areas it reaches as high as 35°C, reflecting water stress. For identifying areas with abnormal reflectivity based on surface temperature values, a threshold can be set to mark anomalies.

[0029] For example, areas with temperatures above 33°C and near-infrared reflectance below 0.3 may exhibit abnormal vegetation due to drought or pests and diseases, and are marked as red warning areas. Marked images visually demonstrate the distribution of anomalies, aiding in precision agriculture management.

[0030] In one possible implementation, the temperature values ​​of normal regions are extracted from the marked image, and normal region pixels can be filtered through a masking operation.

[0031] For example, temperatures in normal farmland areas are concentrated between 23-27°C, indicating good crop growth. Analyzing these values ​​can guide irrigation decisions, optimize water resource utilization, and increase crop yields.

[0032] For example, the above process forms a complete chain, from image acquisition to temperature analysis, with rigorous logic and mutual support. Geometric correction ensures accurate location, atmospheric correction guarantees spectral accuracy, the split-window algorithm provides precise temperature, and anomaly markers and normal area extraction intuitively reflect the surface condition. These technologies are highly effective, enhancing the application value of remote sensing data in agricultural monitoring.

[0033] S102. Calculate the normalized vegetation index based on the corrected remote sensing image. The vegetation index distribution map is obtained by calculating the difference and sum ratio between the red band and the near-infrared band to reflect the vegetation cover density.

[0034] Red band and near-infrared band data are obtained from corrected remote sensing imagery. The difference between the red and near-infrared band data is calculated to obtain band sum data. The ratio of the band difference and band sum data is then calculated to obtain the normalized vegetation index (NVI) data. A vegetation index distribution map is generated based on the NVI data, and vegetation cover density data is extracted from the vegetation index distribution map. For example, in the field of remote sensing image processing, acquiring red and near-infrared band data based on corrected multispectral remote sensing images is a core step in vegetation monitoring. The red band, typically in the 0.6-0.7 micrometer range, reflects the chlorophyll absorption characteristics of vegetation; the near-infrared band, in the 0.7-1.1 micrometer range, has high vegetation reflectivity and is suitable for assessing vegetation health. A common method for acquiring this band data is to extract the spectral values ​​of the corresponding channels using satellite sensors such as Landsat or Sentinel-2.

[0035] For example, the fourth band of Landsat8 corresponds to red light, and the fifth band corresponds to near-infrared light. The data is stored in the form of digital quantized values ​​and needs to be converted into reflectance for subsequent calculations.

[0036] Specifically, when calculating band difference data, the reflectance value of the near-infrared band can be subtracted from the reflectance value of the red band.

[0037] For example, assuming a pixel has a red band reflectance of 0.2 and a near-infrared band reflectance of 0.5, the difference is 0.2 - 0.5 = -0.3. This difference reflects the contrast of the vegetation's spectral characteristics, helping to highlight the difference between vegetated and non-vegetated areas. The calculation of the band sum data involves adding the two values ​​together, i.e., 0.2 + 0.5 = 0.7, for normalization in subsequent ratio calculations.

[0038] In one possible implementation, the Normalized Difference Vegetation Index (NDVI) is calculated based on the ratio of the band difference to the sum, i.e., NDVI = (Near Infrared - Red) / (Near Infrared + Red). Using previous data as an example, NDVI = (-0.3) / 0.7 ≈ -0.429. NDVI values ​​typically range from -1 to 1, with higher values ​​indicating denser vegetation.

[0039] For example, areas with an NDVI close to 0.6 may be healthy forests, while areas close to 0 may be bare soil or sparsely vegetated. NDVI data can be automatically calculated using remote sensing processing software such as ENVI to generate raster-format datasets.

[0040] For example, when generating a vegetation index distribution map, NDVI data can be mapped to a color gradient map, where green represents a high NDVI value and red or yellow represents a low NDVI value.

[0041] For example, in a farmland area, an NDVI distribution map can visually show the boundary between crop-growing areas and wasteland, with an NDVI of 0.8 in the central area and decreasing to 0.3 at the edges. Such distribution maps can be generated using tools like ArcGIS, facilitating intuitive analysis of spatial changes in vegetation cover.

[0042] Specifically, when extracting vegetation cover density data, classification can be performed based on the NDVI threshold.

[0043] For example, an NDVI > 0.4 can be defined as high-density vegetation, 0.2-0.4 as medium-density, and < 0.2 as low-density or no vegetation. Taking a forest monitoring scenario as an example, the analysis results might show that 60% of the area has high-density vegetation, 30% medium-density, and 10% low-density. This classification helps quantify vegetation cover ratios and assists in ecological monitoring or agricultural management.

[0044] It should be noted that the threshold setting needs to be adjusted according to the specific landform and vegetation type. For example, the NDVI thresholds for tropical rainforests and grasslands are quite different.

[0045] In one possible implementation, the extraction of vegetation cover density data can also be combined with ground verification.

[0046] For example, in monitoring a national park, NDVI data extracted by remote sensing was compared with field measurements to ensure the accuracy of vegetation cover density results. Ground measurements may involve the percentage of vegetation cover; for instance, an NDVI of 0.5 in a certain area corresponds to an actual vegetation cover of approximately 70%. This approach improves data reliability and is helpful for long-term ecological change monitoring.

[0047] S103. A building extraction algorithm is used to identify high-density building areas from the corrected remote sensing image. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio.

[0048] Texture and spectral features are obtained from corrected remote sensing imagery. High-density building areas are identified from these features using a building extraction algorithm. If the pixel value of a high-density building area exceeds a preset threshold, the building coverage ratio is determined. Building density is obtained from the building coverage ratio using a density calculation method. A distribution map is generated based on the building density. Areas with varying building density are extracted from the distribution map.

[0049] For example, in building density analysis based on corrected remote sensing images, texture features can be obtained using the gray-level co-occurrence matrix method.

[0050] Specifically, the gray-level co-occurrence matrix (GLCM) can capture the spatial distribution characteristics of pixels in an image. For example, architectural areas typically exhibit regular geometric textures, while natural areas are more chaotic. By analyzing texture indicators such as contrast, correlation, and uniformity of the image, architectural and non-architectural areas can be effectively distinguished.

[0051] In one possible implementation, co-occurrence matrices for four gray levels can be extracted from the corrected multispectral image, and their contrast values ​​can be calculated. For example, in images of urban areas, the contrast value may reach above 0.8, while in vegetated areas it is typically below 0.4. This method helps in the initial separation of built-up areas. The extraction of spectral features depends on the reflectance differences in different bands.

[0052] For example, building materials such as concrete typically exhibit high reflectivity in the near-infrared band and low reflectivity in the visible light band. A spectral feature vector can be constructed by extracting reflectivity values ​​from the red and near-infrared bands.

[0053] In one embodiment, assuming an image has a red band reflectance of 0.3 and a near-infrared band reflectance of 0.7, a feature vector can be generated using the ratio or difference between these two values ​​for subsequent classification. This spectral feature effectively assists texture features, improving the accuracy of building area recognition. For building extraction algorithms, machine learning-based classification methods can be employed.

[0054] For example, support vector machines can be used to fuse and classify texture features and spectral features.

[0055] Specifically, the texture index of the gray-level co-occurrence matrix and the spectral reflectance data are used as input features to train the model to distinguish between built and non-built areas.

[0056] In one possible implementation, assuming that building areas comprise 60% of the pixels in the training data, the model can identify 80% of the building areas in the image after classification, with a false negative rate below 5%. This method can accurately delineate high-density building areas. A pixel value threshold can be set for determining high-density building areas.

[0057] For example, assuming that the pixel value range of building areas in the classified image is 0 to 1, and the threshold is set to 0.75, then areas with pixel values ​​higher than 0.75 are identified as high-density building areas.

[0058] In one embodiment, the pixel proportion of high-density building areas in a city image is 40%, indicating that the building coverage ratio of the area is relatively high. Then, the building coverage ratio can be converted into a building density value using a density calculation method.

[0059] For example, assuming an image resolution of 1 meter and a coverage ratio of 40%, the building area per square kilometer is approximately 400,000 square meters. This density value can intuitively reflect the building density of a region. When generating a building density distribution map, the density value can be mapped to a color gradient.

[0060] For example, low-density areas are represented in green, and high-density areas are represented in red.

[0061] In one possible implementation, a distribution map of a city's imagery shows the central area in red, indicating high building density, while the suburbs are in green, indicating lower density. This visualization method facilitates intuitive analysis of the city's layout. When extracting areas of varying building density from the distribution map, areas with large density gradients can be identified using edge detection algorithms.

[0062] For example, the boundary between the city center and the suburbs may show a significant change in density, manifested as a rapid transition in color from red to green. Extracting such areas of change can help analyze urban expansion trends.

[0063] It should be noted that the advantage of the above method lies in its ability to efficiently identify and present the spatial distribution characteristics of building density through multi-feature fusion and visualization analysis, providing data support for urban planning.

[0064] S104. Construct a radiation interference model using vegetation index distribution map and building density distribution map. If the vegetation index is lower than the preset threshold, add a building density weighting factor to determine the degree of radiation interference and correct the initial value of the surface temperature.

[0065] Data on vegetation index distribution and building density distribution are acquired, and the impact of environmental factors is determined through spatial distribution analysis. If the vegetation index is below a preset threshold, a building density weighting factor is calculated to obtain the overlay data. Based on the overlay data, a radiation interference model is constructed to determine the radiation interference level. A temperature correction coefficient is generated based on the radiation interference level. The initial surface temperature value is corrected using the temperature correction coefficient to obtain the corrected surface temperature value. Based on the corrected surface temperature value, the spatial distribution characteristics of the environmental factor impact are analyzed to determine the final radiation interference distribution.

[0066] For example, when acquiring vegetation index distribution and building density distribution data, multispectral images can be analyzed using remote sensing image processing software to extract the Normalized Difference Vegetation Index (NDVI) as the vegetation index. This, combined with the building density distribution map mentioned earlier, generates a spatial distribution dataset. Assuming a city area with NDVI values ​​ranging from 0.2 to 0.8 and building density distribution expressed as a percentage ranging from 10% to 90%, spatial analysis tools, such as GIS software, can be used to overlay these two types of data, generating a spatial distribution map reflecting the influence of environmental factors and identifying areas with low vegetation cover and high building density.

[0067] Specifically, if the vegetation index is below a preset threshold (e.g., NDVI < 0.4), it indicates poor vegetation cover, which may be affected by urbanization. In this case, the building density weighting factor is calculated.

[0068] For example, in a region with an NDVI of 0.3 and a building density of 80%, a weighting factor can be set as a linear function of the building density. Assuming a factor value of 0.8, this reflects the amplifying effect of buildings on environmental pressure. After overlay processing, a comprehensive dataset is generated, containing information on the interaction between vegetation and buildings.

[0069] In one embodiment, when constructing a radiation interference model, the urban heat island effect can be analyzed based on overlaid data combined with surface albedo and thermal radiation data. Assuming a region has high building density, low NDVI, an albedo of only 0.15, and high thermal radiation, it can be classified as a high-level radiation interference area. The model output is divided into three levels: low, medium, and high, corresponding to the intensity of radiation interference.

[0070] For example, high-grade areas may be concentrated in commercial districts, with radiation levels reaching 30W / m².

[0071] For example, when generating temperature correction coefficients, the initial surface temperature can be adjusted according to the radiation interference level. Assuming the initial temperature is 32°C and the correction coefficient for high-interference areas is 1.2, the corrected temperature will be 38.4°C. The corrected surface temperature distribution map can more accurately reflect the spatial variation of the heat island effect.

[0072] Specifically, when analyzing the spatial distribution characteristics of the impact of environmental factors, spatial autocorrelation analysis can be used to identify the correlation between surface temperature and vegetation and building density.

[0073] For example, in a certain area, the temperature in the commercial district is significantly higher than in the suburbs, NDVI is negatively correlated with temperature, and building density is positively correlated with temperature. The final radiation interference distribution map can visually show high-interference areas, such as city centers, assisting urban planning in optimizing the layout of green spaces.

[0074] S105. Obtain the corrected surface temperature distribution, retrieve the initial temperature from the corrected remote sensing image thermal infrared band using a single-channel algorithm, and apply a radiation interference model for adjustment to obtain an accurate surface temperature map showing the heat accumulation area.

[0075] Acquire corrected remote sensing image data, extract thermal infrared band data from it, and determine the corrected thermal infrared data. Invert the corrected thermal infrared data using a single-channel algorithm to obtain the initial surface temperature. If the initial surface temperature has a deviation, a radiation interference model is used, combined with preset adjustment parameters, to optimize the initial surface temperature and obtain the accurate surface temperature. Based on the accurate surface temperature, a temperature distribution map is generated to determine the heat distribution characteristics. For the heat distribution characteristics, high-value areas in the temperature distribution map are analyzed to identify heat accumulation areas. Using the heat accumulation area data, a spatial distribution map of the heat accumulation areas is generated, yielding the final display result.

[0076] For example, when acquiring corrected remote sensing image data, thermal infrared band data can be extracted through multispectral remote sensing image processing. The correction process typically includes radiometric correction and geometric correction to ensure that the image data accurately reflects the surface features.

[0077] In one possible implementation, Landsat 8 TIRS band data can be used, and after radiometric calibration, digital values ​​of the thermal infrared band can be obtained and converted into radiance values. Assuming that the corrected thermal infrared data for a certain area shows a radiance value of 10 W / m²·sr·μm, this provides a reliable basis for subsequent surface temperature retrieval.

[0078] Specifically, when using a single-channel algorithm to retrieve surface temperature, the surface emissivity can be estimated based on thermal infrared band data and atmospheric correction parameters.

[0079] For example, using Landsat 8 band 10, the initial surface temperature is estimated using a single-channel algorithm. Suppose the inversion result for a certain urban area shows an initial surface temperature of 32°C, but the result may be higher due to estimation errors in atmospheric transmittance or emissivity. To address this error, a radiation interference model can be introduced to optimize the temperature value. The radiation interference model can incorporate vegetation cover and urban building distribution to adjust the initial temperature.

[0080] For example, if a certain area has a low vegetation coverage of 0.2 and a high building density, the model can be optimized by pre-setting adjustment parameters, such as adding a correction value of 0.5°C, to obtain an accurate surface temperature of 31.5°C.

[0081] In one embodiment, a temperature distribution map is generated based on the precise surface temperature, and GIS software can be used to map the temperature data into a spatial distribution map.

[0082] For example, a temperature distribution map of a city shows high temperatures in the central urban area and lower temperatures in the suburbs, with high-value areas concentrated in industrial zones, reaching a maximum temperature of 35°C. Analyzing the heat distribution characteristics reveals a significant heat island effect in the high-value areas, with heat accumulation primarily located in densely populated factory areas and high-rise building clusters. For these heat accumulation areas, a spatial distribution map can be further generated.

[0083] For example, using ArcGIS tools, areas of heat accumulation can be represented by color gradients, with red indicating areas with temperatures above 34°C, accounting for 15% of the urban area, making it easy to visually display the spatial characteristics of heat distribution.

[0084] For example, when analyzing areas of heat accumulation, land use type can be considered to determine the causes of heat accumulation. Suppose an industrial area experiences temperatures 3°C higher than its surroundings due to large amounts of heat emitted from concrete surfaces and machinery. Spatial distribution maps reveal that the heat accumulation areas highly overlap with major transportation routes and industrial facilities. This type of analysis helps identify key areas of the urban heat island effect, providing data support for subsequent environmental planning.

[0085] Preferably, the spatial distribution map can be overlaid with topographic data to further analyze the impact of topography on heat distribution.

[0086] For example, in a low-lying area, due to poor ventilation, heat accumulates more significantly, and the temperature is 1°C higher than in higher areas.

[0087] Understandably, the generated spatial distribution map not only visually displays the distribution of heat, but also provides a basis for urban planning.

[0088] For example, by analyzing heat accumulation area distribution maps, green spaces or water bodies can be prioritized in high-temperature areas to mitigate the heat island effect. This method employs multi-level analysis, progressing step by step from data extraction to spatial visualization, ensuring the scientific rigor and practicality of the results.

[0089] S106. Determine the temperature gradient change in the accurate surface temperature map. If the temperature gradient exceeds the preset threshold, mark it as the heat island boundary line. Smooth the boundary using spatial interpolation to obtain the heat island boundary profile for intensity distribution analysis.

[0090] Surface temperature distribution data is acquired, and a temperature distribution matrix is ​​generated using a gridding method to obtain the surface temperature distribution. For the surface temperature distribution, the temperature difference between adjacent grid points is calculated, and the temperature gradient is calculated using the gradient formula to obtain the gradient distribution. If the gradient value of a grid point in the gradient distribution exceeds a preset threshold, that point is marked as a heat island boundary point, resulting in a set of boundary points. For the boundary point set, spatial interpolation is performed using the Kriging interpolation method to generate a continuous sequence of boundary points, resulting in smooth boundary points. Based on the smooth boundary points, the boundary is fitted using a spline curve method to generate the heat island boundary contour, obtaining boundary contour data. For the boundary contour data, the temperature statistical characteristics within the contour are calculated, and the intensity distribution is generated using a kernel density estimation method, resulting in the heat island intensity distribution.

[0091] For example, when acquiring surface temperature distribution data, a gridded method can be constructed by combining thermal infrared band data from remote sensing images with measured temperatures from ground stations.

[0092] Specifically, assuming the study area is a 100-square-kilometer urban area, the area can be divided into 1-kilometer × 1-kilometer grid cells, with each grid point corresponding to a temperature value. Using interpolation algorithms, such as inverse distance weighting, the thermal infrared data from the remote sensing imagery is mapped to the grid points, generating a temperature distribution matrix. The temperature value at each grid point reflects the surface temperature characteristics of the area.

[0093] In one possible implementation, when calculating the temperature difference between adjacent grid points, the temperature of each grid point can be compared with the temperature of its eight neighboring grid points.

[0094] For example, a grid point has a temperature of 32.5 degrees Celsius, while a neighboring grid point has a temperature of 31.2 degrees Celsius, a difference of 1.3 degrees Celsius. Based on these differences, a gradient formula is used to calculate the temperature gradient, generating a gradient distribution map. Areas with high gradient values ​​typically indicate drastic temperature changes and may correspond to the edge of an urban heat island.

[0095] Specifically, when marking the boundary points of the heat island, a gradient threshold can be set to 1.5 degrees Celsius per kilometer. If the gradient value of a grid point exceeds this threshold, such as 1.8 degrees Celsius per kilometer, it is marked as a boundary point of the heat island.

[0096] For example, assuming that the gradient values ​​are generally high in the urban center area, the set of boundary points can initially outline the contours of the heat island area.

[0097] For example, when generating a continuous sequence of boundary points using kriging interpolation, the spatial distribution characteristics of the boundary point set can be utilized. Kriging interpolation takes into account the spatial correlation between points, enabling smooth transitions in the region.

[0098] For example, if the distance between boundary points is 500 meters, interpolation can be used to generate a new point every 50 meters, forming a continuous sequence. This method can effectively reduce abrupt changes between boundary points and generate smooth boundary lines.

[0099] In one possible implementation, when fitting the contour of the heat island boundary, the spline curve method can generate a smooth curve based on a continuous sequence of boundary points.

[0100] For example, the boundary of an urban heat island may be an irregular ellipse, and spline curves can accurately fit its shape to generate boundary contour data. This data can be used for subsequent spatial analysis.

[0101] Specifically, when calculating the temperature statistics within the contour, the average temperature and maximum temperature within the heat island area can be statistically analyzed.

[0102] For example, the average temperature in a certain heat island region is 34.2 degrees Celsius, with a maximum temperature reaching 36.8 degrees Celsius. Based on these statistical characteristics, a heat island intensity distribution map is generated using the kernel density estimation method, highlighting the core areas with the highest temperatures, such as urban commercial or industrial areas.

[0103] For example, kernel density estimation methods can control the smoothness of the distribution through bandwidth parameters. Assuming a bandwidth of 200 meters, the intensity distribution map can clearly show the intensity changes in the core region of the heat island. Such distribution maps help identify high-risk areas for heat accumulation, providing data support for urban planning or thermal environment management.

[0104] S107. A random forest model is used to train the accurate surface temperature map and heat island boundary contour. By inputting temperature features and boundary location data, the intensity distribution is predicted, and a spatial distribution map of heat island intensity is obtained to reflect the overall thermal environment pattern.

[0105] A training dataset is constructed using surface temperature data and the location of the heat island boundary, generating a standardized feature matrix. A random forest model is then used to train the standardized feature matrix, resulting in a heat island intensity prediction model. Based on this model, surface temperature data and boundary locations are input to generate a heat island intensity distribution. The intensity value at each point within the spatial grid is calculated using this distribution, yielding spatial intensity distribution data. If outliers exist in the spatial intensity distribution data, median filtering is applied to generate a smoothed intensity distribution. This smoothed intensity distribution is then mapped to a geographic coordinate system to generate a spatial pattern map. The overall thermal environment is analyzed using this spatial pattern map, producing thermal environment analysis results.

[0106] For example, when constructing a training dataset, key features can be extracted from surface temperature data and the location of heat island boundaries. Surface temperature data is typically stored in a grid format, with each grid point containing a temperature value, such as 30.5°C or 32.8°C. The location of heat island boundaries is represented by the coordinates of boundary points, such as latitude and longitude pairs. During feature extraction, factors such as mean temperature, maximum temperature difference, and boundary point density can be selected as input features.

[0107] In one possible implementation, assume a temperature grid for a certain urban area is 100×100, with grid point temperatures ranging from 28.0°C to 35.0°C. Boundary points are distributed in areas with high temperature gradients, such as densely built-up areas in the city center. By calculating the temperature features of each grid point and the distance features to boundary points, a feature vector containing dimensions such as temperature, distance, and density is generated, forming the training dataset. This method ensures that the dataset comprehensively reflects the characteristics of the heat island, facilitating subsequent model training. In the generation of the standardized feature matrix, the Z-score standardization method can be used to unify features of different dimensions to a range with a mean of 0 and a standard deviation of 1.

[0108] For example, temperature values ​​and distances to boundary points have different dimensions. Standardization can eliminate the influence of dimensions and improve the stability of model training.

[0109] In one possible implementation, assuming a grid point temperature of 33.2°C, a dataset temperature mean of 31.0°C, and a standard deviation of 1.5°C, the standardized temperature feature value is (33.2-31.0) / 1.5=1.47. Other features, such as boundary point distances, are also standardized to generate a standardized feature matrix. This method makes different features comparable, which is beneficial for random forest models.

[0110] For example, during the training of a random forest model, multiple decision trees can be integrated to predict the intensity of the heat island. Random forests reduce the risk of overfitting by constructing decision trees through random selection of samples and features.

[0111] In one possible implementation, the training dataset is assumed to contain 5000 samples, each including five features such as temperature and boundary distance. After model training, the heat island intensity value can be predicted for each grid point, with 0.8 representing a high-intensity heat island region. This method, by integrating multiple decision trees, can capture complex nonlinear relationships and improve prediction accuracy. In the generation of intensity spatial distribution data, the intensity value for each grid point can be calculated based on the prediction model output.

[0112] For example, the predicted intensity value for a certain area ranges from 0.2 to 0.9, with higher intensity values ​​at grid points in the city center (e.g., 0.85) and lower values ​​at suburban areas (e.g., 0.3). This distribution reflects the spatial variation trend of the heat island effect, facilitating subsequent analysis.

[0113] For example, median filtering can effectively smooth intensity distribution when dealing with outliers. Suppose a grid point has an outlier intensity value of 1.5, significantly higher than the surrounding grid points' 0.8. After applying a 3×3 median filter, the intensity value at that point might be adjusted to 0.82, taking the median of the surrounding grid points' intensities. This method preserves spatial distribution characteristics while eliminating outlier interference.

[0114] In one possible implementation, when smoothing the intensity distribution and mapping it to a geographic coordinate system, the intensity values ​​can be correlated with latitude and longitude to generate a spatial pattern map.

[0115] For example, grid points in the city center are mapped to latitude and longitude (120.5, 30.2), with an intensity value of 0.85, and are displayed as red high-intensity areas, while suburban areas are green low-intensity areas. This visualization method intuitively shows the distribution of heat islands and is beneficial for analyzing the spatial characteristics of the urban thermal environment.

[0116] For example, when analyzing the overall thermal environment, high-intensity heat island areas can be identified through spatial pattern maps, such as high-rise areas in city centers and areas with less green space. Combining this with data on population density, building distribution, and other factors, the impact of heat islands on residents' lives can be further analyzed.

[0117] For example, if 30% of the grid points in a certain area have an intensity value higher than 0.8, it indicates a significant thermal environment problem, necessitating optimization of urban planning. This type of analysis provides data support for urban heat island management and helps in developing cooling measures. Example 2 This invention also provides a system for analyzing urban heat island intensity based on remote sensing inversion, comprising: The first processing module is used to acquire multispectral remote sensing image data through satellite sensors, perform geometric correction and atmospheric correction on the images, and obtain corrected remote sensing images for subsequent temperature inversion. The second processing module is used to calculate the normalized vegetation index based on the corrected remote sensing image. It calculates the vegetation index distribution map by using the difference and sum ratio of the red band and the near-infrared band to reflect the vegetation cover density. The third processing module is used to identify high-density building areas from the corrected remote sensing images. By combining texture features and spectral features, a building density distribution map is obtained to represent the building coverage ratio. The fourth processing module is used to add a building density weighting factor if the vegetation index is lower than a preset threshold, and to determine the degree of radiation interference in order to correct the initial value of the surface temperature. The fifth processing module is used to obtain the corrected surface temperature distribution. It retrieves the initial temperature from the thermal infrared band of the corrected remote sensing image using a single-channel algorithm and applies a radiation interference model to adjust it, thereby obtaining an accurate surface temperature map that shows the heat accumulation area. The sixth processing module is used to determine the temperature gradient change in the accurate surface temperature map. If the temperature gradient exceeds the preset threshold, it is marked as the heat island boundary line. The boundary is smoothed by spatial interpolation method to obtain the heat island boundary profile for intensity distribution analysis. The seventh processing module is used to train the accurate surface temperature map and the heat island boundary contour. By inputting temperature characteristics and boundary location data, it predicts the intensity distribution and obtains a spatial distribution map of heat island intensity that reflects the overall thermal environment pattern.

[0118] As one embodiment of the present invention, the third processing module uses a building extraction algorithm to identify high-density building areas from the corrected remote sensing image, and obtains a building density distribution map representing the building coverage ratio by combining texture features and spectral features.

[0119] As one embodiment of the present invention, the fourth processing module constructs a radiation interference model through a vegetation index distribution map and a building density distribution map. If the vegetation index is lower than a preset threshold, a building density weighting factor is superimposed to determine the degree of radiation interference and correct the initial value of the surface temperature.

[0120] As one embodiment of the present invention, the seventh processing module uses a random forest model to train the accurate surface temperature map and the heat island boundary contour. By inputting temperature features and boundary location data, it predicts the intensity distribution and obtains a heat island intensity spatial distribution map that reflects the overall thermal environment pattern.

[0121] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for analyzing intensity of urban heat island based on remote sensing inversion, characterized in that, The method comprises the following steps: Collecting multispectral remote sensing image data through a satellite sensor, performing geometric correction and atmospheric correction on the image, and obtaining corrected remote sensing image for subsequent temperature inversion; Calculating the normalized vegetation index according to the corrected remote sensing image, and obtaining the vegetation index distribution map reflecting the vegetation coverage density through the ratio of the difference value to the sum value of the red light band and the near-infrared band; Identifying high-density building areas from the corrected remote sensing image, and obtaining the building density distribution map representing the building coverage ratio through the combination of texture features and spectral features; If the vegetation index is lower than a preset threshold, a building density weighting factor is superimposed to judge the degree of radiation interference to correct the initial value of the land surface temperature; Obtaining the corrected land surface temperature distribution, inversing the initial temperature from the corrected remote sensing image thermal infrared band through a single-channel algorithm, and adjusting the radiation interference model to obtain the accurate land surface temperature map to show the heat accumulation area; Judging the temperature gradient change in the accurate land surface temperature map, and if the temperature gradient exceeds a preset threshold, marking it as a heat island boundary line, and smoothing the boundary through a spatial interpolation method to obtain the heat island boundary contour for intensity distribution analysis; Training the accurate land surface temperature map and the heat island boundary contour, predicting the intensity distribution by inputting temperature features and boundary position data, and obtaining the heat island intensity spatial distribution map to reflect the overall heat environment pattern. 2.The remote sensing inversion-based urban heat island intensity analysis method according to claim 1, characterized in that, Identifying high-density building areas from the corrected remote sensing image through a building extraction algorithm, and obtaining the building density distribution map representing the building coverage ratio through the combination of texture features and spectral features. 3.The remote sensing inversion-based urban heat island intensity analysis method according to claim 1, characterized in that, Constructing a radiation interference model through the vegetation index distribution map and the building density distribution map, and if the vegetation index is lower than a preset threshold, a building density weighting factor is superimposed to judge the degree of radiation interference to correct the initial value of the land surface temperature. 4.The remote sensing inversion-based urban heat island intensity analysis method of claim 1, wherein, Training the accurate land surface temperature map and the heat island boundary contour using a random forest model, predicting the intensity distribution by inputting temperature features and boundary position data, and obtaining the heat island intensity spatial distribution map to reflect the overall heat environment pattern. 5.A system for analyzing intensity of urban heat island based on remote sensing inversion, characterized in that, The method comprises the following steps: The first processing module is configured to collect multispectral remote sensing image data through a satellite sensor, perform geometric correction and atmospheric correction on the image, and obtain corrected remote sensing image for subsequent temperature inversion; The second processing module is configured to calculate the normalized vegetation index according to the corrected remote sensing image, and obtain the vegetation index distribution map reflecting the vegetation coverage density through the ratio of the difference value to the sum value of the red light band and the near-infrared band; The third processing module is configured to identify high-density building areas from the corrected remote sensing image, and obtain the building density distribution map representing the building coverage ratio through the combination of texture features and spectral features; The fourth processing module is configured to superimpose a building density weighting factor if the vegetation index is lower than a preset threshold to judge the degree of radiation interference to correct the initial value of the land surface temperature; The fifth processing module is configured to obtain the corrected land surface temperature distribution, inversing the initial temperature from the corrected remote sensing image thermal infrared band through a single-channel algorithm, and adjusting the radiation interference model to obtain the accurate land surface temperature map to show the heat accumulation area; The sixth processing module is configured to judge the temperature gradient change in the accurate land surface temperature map, mark a heat island boundary line if the temperature gradient exceeds a preset threshold, smooth the boundary through a spatial interpolation method, and obtain a heat island boundary contour for intensity distribution analysis. The seventh processing module is configured to train the accurate land surface temperature map and the heat island boundary contour, predict the intensity distribution through input of temperature features and boundary position data, and obtain a heat island intensity spatial distribution map to reflect an overall heat environment pattern. 6.The remote sensing inversion-based urban heat island intensity analysis system of claim 5, wherein, The third processing module adopts a building extraction algorithm to identify a high-density building area from the corrected remote sensing image, combines texture features and spectral features, and obtains a building density distribution map to represent a building coverage proportion.

7. The remote sensing inversion-based urban heat island intensity analysis system of claim 5, wherein, The fourth processing module constructs a radiation interference model through a vegetation index distribution map and a building density distribution map, superimposes a building density weighting factor if the vegetation index is lower than a preset threshold, judges the radiation interference degree to correct the initial value of the land surface temperature.

8. The remote sensing inversion-based urban heat island intensity analysis system of claim 5, wherein, The seventh processing module adopts a random forest model to train the accurate land surface temperature map and the heat island boundary contour, predict the intensity distribution through input of temperature features and boundary position data, and obtain a heat island intensity spatial distribution map to reflect an overall heat environment pattern.

Images