Urban heat island effect ai identification grading method and system for urban-rural joint areas

Abstract

The application provides a city heat island effect AI identification grading method and system combined with an urban-rural joint area, belongs to the technical field of machine learning and city heat island effect, and comprises the following steps: S1, acquiring basic data and preprocessing; S2, constructing a sample set; S3, constructing an identification model; S4, heat island identification in the whole region; S5, vector extraction; and S6, heat island intensity grading. The application takes machine learning as a technical means, selects natural geographical elements and factors representing human activities and social economic characteristics, identifies and extracts the spatial distribution of the city heat island effect from a land surface temperature anomaly field. The method utilizes the advantage of machine learning in solving nonlinear relationships, intelligently identifies the spatial distribution of the city heat island effect combined with the urban-rural joint area, and realizes heat island intensity grading.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of machine learning and urban heat island effect, and in particular to an urban heat island effect AI identification and grading method and system in combination with an urban-rural joint area. BACKGROUND

[0002] The frequent and expanding trend of economic activities and human production activities in mountainous towns makes the local heat island effect gradually prominent. Heat island effect monitoring must consider the urban-rural joint area. However, the traditional urban-rural dichotomy needs to divide the urban and rural areas and calculate the average temperature of the rural area, which leads to the inability to identify and extract the potential heat island effect area of the urban-rural joint area. This method is easily affected by the strong warming of the local terrain in the rural area, and the results have great uncertainty. There is an urgent need for a heat island effect monitoring method that adapts to current social development and includes the urban-rural joint area.

[0003] The urban-rural dichotomy is essentially a threshold division method based on the urban-rural temperature difference, and there is a lack of unified standard for dividing the boundaries of urban and rural areas. For example, impervious surface data is used as the basis for dividing the urban-rural boundary; the old city area and the area within the third ring are divided into urban areas, and non-urban and water areas are divided into rural areas. The inconsistency of various division methods leads to significant differences in research results, and it is difficult to accurately identify the heat island phenomenon in the urban-rural joint area. Study The method of laying observation instruments in urban and rural areas is used for calculation, but this method does not fully consider the influence of terrain factors on warming. Obviously, in complex terrain conditions, the warming effect caused by natural geographical environment will have a significant impact on the accuracy of the regional heat island identification results.

[0004] Some studies have revealed that the heat island effect is driven by natural geographical elements and human activities, providing a reference for AI identification of heat island effect. For example, by constructing a regression model, the influence of spatial form elements such as building density, sky view factor (SVF), and floor area ratio (FAR) on urban heat island effect is analyzed; the PLUS model is used to construct a comprehensive index system including 7 social and economic indicators such as population density, GDP, and road density, and 5 natural geographical indicators such as elevation, soil type, and annual average temperature, to systematically study the driving mechanism of urban heat island effect. However, these studies are still limited to the main urban area, and have not paid attention to the heat island distribution in the urban-rural joint area. Only a few researchers have attempted to introduce the XGBoost-SHAP interpretable machine learning model to study the land surface temperature. However, this study does not systematically propose how to identify the urban heat island effect. SUMMARY

[0005] The application provides an urban heat island effect AI identification grading method and system for urban-rural combined areas, which is a technical means of machine learning, selects 10 factors representing human activities and social economic characteristics, such as night light remote sensing data, population density, GDP, normalized vegetation index (NDVI), digital elevation model (DEM), and terrain slope, and identifies and grades the spatial distribution of urban heat island effect. The method fully utilizes the data-driven advantage of machine learning and can effectively identify the nonlinear relationship between factors and their comprehensive influence on urban heat island effect. The AI identification and grading of urban heat island effect areas in urban-rural combined areas are realized from the land surface temperature anomaly field by using the machine learning method. The method greatly reduces the uncertainty caused by the need to clearly divide urban and rural areas in the traditional urban-rural dichotomy, and improves the stability and uniformity of multi-period monitoring of urban heat island effect in a large area.

[0006] To achieve the above object, the application adopts the following technical scheme:

[0007] The urban heat island effect AI identification grading method for urban-rural combined areas comprises the following steps:

[0008] S1. Obtain basic data and preprocess; collect multi-year monthly average land surface temperature data, perform grid average value operation to obtain historical land surface temperature field annual average grid data; collect current land surface temperature field grid data, perform grid subtraction operation on the current land surface temperature field grid data and the historical land surface temperature field annual average grid data to obtain current land surface temperature anomaly grid data; collect digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data, perform surface analysis on the digital elevation model data to obtain slope grid data and slope direction grid data, perform time and spatial scale unified processing on the impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data, and project all the data into a coordinate system consistent with the current land surface temperature anomaly grid data to obtain standardized multi-factor grid data;

[0009] S2. Construct a sample set; extract urban areas based on land use data, combine the current land surface temperature field grid data, and identify main urban heat island effect pixels using the urban-rural dichotomy, randomly select pixels from the main urban heat island effect pixels to form a target sample set, and randomly generate sample points from non-urban areas, sample the current land surface temperature anomaly grid data and the standardized multi-factor grid data to obtain a training sample set;

[0010] S3. Construct an identification model; input the target sample set and the training sample set, perform random forest model training, and obtain an urban heat island effect identification model;

[0011] S4. Full-area heat island identification; input the current land surface temperature anomaly grid data and the standardized multi-factor grid data into the urban heat island effect identification model to obtain full-area urban heat island effect region grid data;

[0012] S5. Vector extraction; perform raster-to-vector processing on the full-area urban heat island effect region grid data to obtain urban heat island region vector data including the urban-rural combined area;

[0013] S6. Heat island intensity classification; perform overlay analysis on the current land surface temperature anomaly grid data and the urban heat island region vector data including the urban-rural combined area, and based on a preset temperature gradient threshold, classify the heat island intensity into four levels of significant heat island, strong heat island, moderate heat island, and slight heat island, to obtain heat island intensity level distribution data.

[0014] In the specification, in S1, the multi-year monthly average land surface temperature data is MODIS monthly average land surface temperature product data for at least 20 years, and the current land surface temperature field grid data is current land surface temperature grid data or average land surface temperature grid data within a preset time range.

[0015] In the specification, in S1, if the current land surface temperature field grid data has null values due to cloud cover, the average value of the 5x5 region pixels around the null pixel is used for filling.

[0016] In the specification, in S2, the process of identifying main urban heat island effect pixels by urban-rural dichotomy includes: establishing a buffer zone for the urban area, erasing the water body area in the buffer zone to obtain a rural area, calculating the average land surface temperature of the rural area, subtracting the average land surface temperature of the rural area from the land surface temperature of the urban area to obtain land surface temperature difference grid data, and determining whether the pixels in the land surface temperature difference grid data are heat island effect pixels based on a preset threshold.

[0017] In the specification, in S3, the parameters of the random forest algorithm include the number of classification trees and the calculation depth of each tree, and the optimal values of the number of classification trees and the calculation depth of each tree are determined through multiple experiments.

[0018] In the specification, in S3, the area under the receiver operating characteristic curve is used to evaluate the model effect, and when the area value is in the interval of 0.9-1, it is determined that the model training is completed.

[0019] In the specification, in S6, the preset temperature gradient threshold is set according to the seasonal natural breakpoint method, with a gradient of 0.7°C in summer and a gradient of 0.6°C in winter.

[0020] In the specification, in S1, the spatial resolution of the digital elevation model data, impervious surface data, nighttime light data, population density data, normalized vegetation index data, GDP data, etc. can be any size that can be collected.

[0021] In the specification, in S2, the target sample set and the training sample set are divided into a model training set and a test set in a ratio of 8:2, the model training set is used for model training, and the test set is used for model effect verification.

[0022] The AI identification grading system for urban heat island effect in the urban-rural joint area, the AI identification grading method for urban heat island effect in the urban-rural joint area according to any one of the above, comprises:

[0023] An acquisition and preprocessing module is configured to:

[0024] The multi-year monthly average ground surface temperature data is collected, and a grid average value operation is performed to obtain historical annual average grid data of the ground surface temperature field. The current ground surface temperature field grid data is collected, and a grid subtraction operation is performed on the current ground surface temperature field grid data and the historical annual average grid data of the ground surface temperature field to obtain current ground surface temperature anomaly grid data. The digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data are collected. The digital elevation model data is subjected to surface analysis to obtain slope grid data and aspect grid data. The impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data are subjected to time and spatial scale unified processing, and all the data are projected into a coordinate system consistent with the current ground surface temperature anomaly grid data to obtain standardized multi-factor grid data.

[0025] A sample set construction module is configured to:

[0026] The urban area is extracted based on the land use data, and the current ground surface temperature field grid data is combined to identify the main urban heat island effect pixels using the urban-rural dichotomy. Pixels are randomly selected from the main urban heat island effect pixels to form a target sample set. Sample points are randomly generated from the non-urban area, and the current ground surface temperature anomaly grid data and the standardized multi-factor grid data are sampled to obtain a training sample set.

[0027] A recognition model construction module is configured to input the target sample set and the training sample set, train a model using a random forest algorithm, and obtain an urban heat island effect recognition model.

[0028] A full-area heat island recognition module is configured to input the current ground surface temperature anomaly grid data and the standardized multi-factor grid data into the urban heat island effect recognition model to obtain full-area urban heat island effect regional grid data.

[0029] A vector extraction module is configured to perform a grid-to-vector conversion on the full-area urban heat island effect regional grid data to obtain urban heat island area vector data in the urban-rural joint area.

[0030] The heat island intensity grading module is used for superimposed analysis of current land surface temperature anomaly grid data and urban heat island area vector data of the urban-rural combined area, and based on a preset temperature gradient threshold, the heat island intensity is divided into four levels of significant heat island, strong heat island, moderate heat island and slight heat island, and heat island intensity level distribution data is obtained.

[0031] In summary, the present application has at least the following beneficial effects:

[0032] Compared with the traditional urban-rural contrast method, the present application effectively avoids errors caused by different urban area division standards. It can more carefully identify the heat island phenomenon of the main urban area and the connected urban-rural transition zone. At the same time, the machine learning method in the field of artificial intelligence replaces the traditional threshold method to identify heat island effect pixels, realizing the AI identification technology of heat island effect pixels. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 It is a schematic diagram of the AI identification grading method of the urban heat island effect of the urban-rural combined area.

[0035] Figure 2 It is a schematic diagram of the identification result of the present application.

[0036] Figure 3 It is a schematic diagram of the identification result of the urban-rural dichotomy.

[0037] Figure 4 It is a schematic diagram of the urban heat island area vector extraction map (Kunming summer) of the urban-rural combined area.

[0038] Figure 5 It is a schematic diagram of the urban heat island area vector extraction map (Qujing summer) of the urban-rural combined area.

[0039] Figure 6 It is a schematic diagram of the urban heat island area vector extraction map (Yuxi summer) of the urban-rural combined area.

[0040] Figure 7 It is a schematic diagram of the urban heat island area vector extraction map (Chuxiong summer) of the urban-rural combined area.

[0041] Figure 8 It is a schematic diagram of the urban heat island area vector extraction map (Kunming winter) of the urban-rural combined area.

[0042] Figure 9 Vector extraction map of urban heat island area together with urban-rural fringe area (Yuxi winter).

[0043] Figure 10 Vector extraction map of urban heat island area together with urban-rural fringe area (Yuxi winter).

[0044] Figure 11 Vector extraction map of urban heat island area together with urban-rural fringe area (Yuxi winter).

[0045] Figure 12 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0046] Figure 13 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0047] Figure 14 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0048] Figure 15 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0049] Figure 16 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0050] Figure 17 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0051] Figure 18 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0052] Figure 19 Schematic diagram of the identification and classification results of Kunming urban heat island effect area in summer daytime.

[0053] Figure 20 Schematic diagram of the classification statistics of urban heat island effect intensity of four cities in central Yunnan in summer daytime.

[0054] Figure 21 Schematic diagram of the classification statistics of urban heat island effect intensity of four cities in central Yunnan in winter daytime. DETAILED DESCRIPTION

[0055] In the following, only certain exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present application. Therefore, the drawings and the description are considered to be exemplary in nature rather than limiting.

[0056] The following disclosure provides many different embodiments or examples for implementing different structures of embodiments of the application. For the purpose of simplicity, the elements and settings of particular examples are described in the following disclosure. Of course, they are merely examples and the purpose is not to limit the embodiments of the application. In addition, the embodiments of the application can refer to reference numerals and / or reference letters in different examples, and such repetition is for the purpose of simplification and clarity, which does not indicate the relationship between the various embodiments and / or settings discussed.

[0057] The embodiments of the application are described in detail below with reference to the accompanying drawings.

[0058] As Figure 1 shown, the present embodiment provides a urban heat island effect AI identification grading method for urban-rural combined areas, comprising:

[0059] S1. Obtain basic data and pre-process; collect multi-year monthly average land surface temperature data, perform grid average value operation to obtain historical land surface temperature field annual average grid data; collect current land surface temperature field grid data, perform grid subtraction operation with the historical land surface temperature field annual average grid data to obtain current land surface temperature anomaly grid data; collect digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data, GDP data, perform surface analysis on the digital elevation model data to obtain slope grid data and slope direction grid data, perform time and spatial scale unified processing on the impervious surface data, night light data, population density data, land use data, normalized vegetation index data, GDP data, and project all data into a coordinate system consistent with the current land surface temperature anomaly grid data to obtain standardized multi-factor grid data;

[0060] S2. Construct a sample set; extract urban areas based on land use data, combine current land surface temperature field grid data, and identify main urban heat island effect pixels using urban-rural dichotomy, randomly select pixels from the main urban heat island effect pixels to form a target sample set; randomly generate sample points from non-urban areas, sample the current land surface temperature anomaly grid data and the standardized multi-factor grid data to obtain a training sample set;

[0061] S3. Construct an identification model; input the target sample set and the training sample set for random forest model training to obtain a urban heat island effect identification model;

[0062] S4. Whole area heat island identification; input the current land surface temperature anomaly grid data and the standardized multi-factor grid data into the urban heat island effect identification model to obtain whole area urban heat island effect regional grid data;

[0063] S5. Vector extraction; perform raster-to-vector processing on the urban heat island area raster data of the whole region to obtain urban heat island area vector data including the urban-rural combined area;

[0064] S6. Heat island intensity classification; superimpose and analyze the current land surface temperature anomaly raster data and the urban heat island area vector data including the urban-rural combined area, divide the heat island intensity into four levels of significant heat island, strong heat island, moderate heat island, and slight heat island based on a preset temperature gradient threshold, and obtain heat island intensity level distribution data.

[0065] In some embodiments, in S1, the multi-year monthly average land surface temperature data is MODIS monthly average land surface temperature product data for at least 20 years, and the current land surface temperature field raster data is current land surface temperature raster data or average land surface temperature raster data in a preset time range. The preset time range can be a day, a week, a month, a season, half a year, a year, or other time ranges or other time ranges set according to actual conditions (needs) or any time period, such as several months, a season, or a custom average.

[0066] In some embodiments, in S1, if the current land surface temperature field raster data has null values due to cloud cover, the average value of the 5x5 region around the null pixel is used to fill in the null values.

[0067] In some embodiments, in S2, the process of identifying urban heat island effect pixels by urban-rural dichotomy includes: establishing a buffer zone in the urban area, erasing the water area in the buffer zone to obtain a rural area, calculating the average land surface temperature of the rural area, subtracting the average land surface temperature of the rural area from the land surface temperature of the urban area to obtain land surface temperature difference value raster data, and determining whether the pixels in the land surface temperature difference value raster data are heat island effect pixels based on a preset threshold.

[0068] In some embodiments, in S3, the parameters of the random forest algorithm include the number of classification trees and the calculation depth of each tree, the number of classification trees is 80, and the calculation depth of each tree is 8.

[0069] In some embodiments, in S3, the area under the receiver operating characteristic curve is used to evaluate the model effect, and when the area value is in the interval of 0.9-1, it is determined that the model training is completed.

[0070] In some embodiments, in S6, the preset temperature gradient threshold is set according to the seasonal natural breakpoint method, with a gradient of 0.7°C in summer and a gradient of 0.6°C in winter.

[0071] In some embodiments, in S1, the spatial resolution of the digital elevation model data is 30 m, the spatial resolution of the impervious surface data is 30 m, the spatial resolution of the night light data is about 500 m, and the spatial resolution of the population density data, the normalized vegetation index data and the GDP data is 1 km.

[0072] In some embodiments, in S2, the target sample set and the training sample set are divided into a model training set and a test set in a ratio of 8:2, the model training set is used for model training, and the test set is used for model effect verification.

[0073] The technical concept of the present application is as follows:

[0074] To solve the problem of identifying urban heat island effect, an AI identification and grading method and system for urban heat island effect region together with urban and rural combined areas are proposed based on the machine learning method and the land surface temperature anomaly field, and the specific content is as follows:

[0075] Obtain the annual average grid data of the historical land surface temperature field.The computer downloads and stores the monthly average land surface temperature products for multiple years, and calculates the annual average grid data of the historical land surface temperature field by using the conventional grid average value operation.

[0076] Obtain the current land surface temperature anomaly grid data.The computer downloads and stores the current land surface temperature field grid data, and subtracts the current land surface temperature field grid data from the historical land surface temperature field annual average grid data by using the conventional grid subtraction operation, and obtains the difference field grid data, which is the current land surface temperature anomaly grid data.

[0077] Obtain the slope, aspect, impervious surface, night light, population density, land use, normalized vegetation index, and GDP grid data.Download the DEM data, and extract the slope and aspect grid data by using the conventional GIS surface analysis tool.Download the impervious surface, night light, population density, land use, normalized vegetation index, and GDP grid data for multiple years, and unify the time and spatial scales, and project them into the coordinate system of the current land surface temperature anomaly grid data.

[0078] Obtain the target and training sample set for heat island effect identification.First, identify the heat island effect pixels in the main urban area by using the conventional urban and rural bisection method, second, randomly select m pixels in the heat island effect pixels in the main urban area by using the conventional GIS random point tool to form a target sample set, denoted as y, and third, randomly generate n sample points in the non-urban area by using the conventional GIS random point tool, and sample the current land surface temperature anomaly grid data, the slope, aspect, impervious surface, night light, population density, land use, normalized vegetation index, and GDP grid data to obtain the training sample set for heat island effect identification, denoted as x.

[0079] The process of identifying the heat island effect area of the main urban area by the conventional urban-rural dichotomy is as follows: the urban area extracted from the land use data is split into a single city area; a buffer zone is established for the city area, and the water area in the buffer zone is erased; finally, the remaining non-water buffer zone is taken as the rural area; the rural area surface temperature is extracted by batch mask; the average value of the rural area surface temperature is calculated; the surface temperature difference between the city and the countryside is obtained by subtracting the average value of the rural area surface temperature from the city area surface temperature, and is recorded as T1; whether the pixel of the surface temperature difference grid data has heat island effect is judged by threshold method; the gray value of the pixel judged as heat island effect is redefined as 1, and the gray value of other non-heat island pixels is defined as Nodata to obtain the heat island effect area grid data of the main urban area; the heat island effect area grid data of the main urban area is converted into vector polygon data by using the conventional GIS raster-to-vector tool to obtain the heat island effect area of the main urban area.

[0080] The threshold method judgment process is as follows: a threshold TH_k is predefined, and the pixels of the surface temperature difference grid data are traversed; if T1> TH_k, the pixel is judged as the main urban area heat island effect pixel; if T1<= TH_k, the pixel is not judged as the main urban area heat island effect pixel.

[0081] A random forest algorithm is used to construct a computer automatic identification model of urban heat island effect. The RandomForestClassifier function in the sklearn library of Python language is used to determine the parameters n_estimators, max_depth of the random forest modeling based on the above target sample set y and training sample set x as learning sample data, and the random forest model is constructed. The values of n_estimators and max_depth are obtained by adjusting the model effect through multiple experiments. The area under the receiver operating characteristic (ROC) curve (AUC) value is used to judge the model effect. The larger the AUC value, the better the model effect. When the AUC value is between 0.8 and 0.9, the model is better, and when the AUC value is between 0.9 and 1, the model is excellent.

[0082] The random forest model is used to identify the urban heat island area in the whole area. The current surface temperature anomaly grid data, the slope, the aspect, the impervious surface, the night light, the population density, the land use, the normalized vegetation index, and the GDP grid data are used as the prediction variables, which are substituted into the random forest model to carry out prediction, and the urban heat island effect area identification result in the whole area is obtained, that is, the urban heat island effect area grid data of the urban-rural combination area, recorded as Grid_t.

[0083] Vector_t, which is the vector data of the urban heat island area together with the urban-rural joint area.

[0084] Heat island intensity grade division. With the current land surface temperature anomaly grid data as the reference for heat island intensity grade judgment, the current land surface temperature anomaly grid data is superimposed and analyzed with the above-mentioned urban heat island area vector data Vector_t together with the urban-rural joint area, a pre-defined gradient TH_j of grade division is used to divide the heat island intensity in the identified urban heat island area into four grades, and the final heat island intensity grade distribution data of the urban heat island area together with the urban-rural joint area is obtained. The four grades are defined in order of intensity from strong to weak as four grades of significant heat island, strong heat island, moderate heat island and slight heat island.

[0085] Taking the recognition and extraction of the urban heat island effect area in winter and summer in Yunnan Province of China as an example. Influenced by complex terrain and urbanization, the heat island effect of cities in Yunnan Province is continuously enhanced, and the heat island effect has a certain impact on the urban ecological system and livability index.

[0086] (1) Obtain the annual average grid data of MODIS historical land surface temperature field.

[0087] Obtain the MODIS monthly average land surface temperature product data in Yunnan Province in the past 20 years. The product data is in HDF format, the time resolution is 8 d, and the spatial resolution is 1 km. In the ENVI5.3 environment, use MRT tools for format conversion and coordinate conversion and other pretreatments. Then, the monthly average land surface temperature is obtained by using the map algebra tool of ArcGIS to calculate the average value of each month's 4 scenes of land surface temperature product. The annual average land surface temperature field grid data is calculated by using the average value operation of grid. The null value of land surface temperature data caused by the influence of cloud cover is filled by taking the average value of the pixels in the 5x5 region around the pixel.

[0088] (2) Obtain winter and summer land surface temperature anomaly grid data

[0089] Take June-August as summer, and December to January and February of the next year as winter. The seasonal average land surface temperature grid data of MODIS in winter daytime and summer daytime is calculated by using the average value operation of grid, and then the land surface temperature anomaly grid data of MODIS in winter daytime and summer daytime is calculated, that is, the land surface temperature anomaly field of Yunnan Province in summer and winter.

[0090] ;

[0091] In the formula, is the seasonal land surface temperature anomaly, is the seasonal average land surface temperature, is the average land surface temperature in the past 20 years.

[0092] (3) Obtain slope, aspect, impervious surface, night light, population density, land use, normalized vegetation index, and GDP raster data.

[0093] Slope and aspect data: Download publicly shared digital elevation model (DEM) data from the Copernicus Digital Elevation Model, with a resolution of about 30 m. Then, crop the Yunnan province range, with data format tif and data coordinates GCS_WGS_1984. Finally, use ArcGIS pro software to extract slope and aspect data from DEM.

[0094] Impervious surface data: Obtain the same period and publicly shared impervious surface data, with data format tif and 30 m spatial resolution.

[0095] Night light data: Obtain the same period and publicly shared night light data, with a spatial resolution of about 500 m and data format tif.

[0096] Population density data: Obtain the same period and publicly shared population density data, projected to geographic coordinate system WGS84, with spatial resolution of 30 arc seconds, about 1 km land use data.

[0097] Land use data: Obtain the same period and publicly shared land use data, select urban land as the urban entity area, with data format tif and spatial resolution of 30 m.

[0098] Normalized vegetation index: Obtain the same period and publicly shared normalized vegetation index (NDVI), with a spatial resolution of 1 km.

[0099] GDP raster data: Obtain the same period and publicly shared GDP data, with a spatial resolution of 1 km and coordinate system WGS_1984_Albers Conical Equal Area.

[0100] Unify the above data in time and spatial scale, and project to the unified GCS_WGS_1984 coordinate system.

[0101] (4) Obtain the target and training sample set of heat island effect identification

[0102] The heat island effect pixels of the main urban area are identified by using the conventional urban-rural dichotomy. The urban area extracted from the land use data is divided into a single city area. A 3km buffer zone is established for the city area, and the water area in the buffer zone is erased. Finally, the remaining non-water buffer zone is used as the rural area. The rural area surface temperature is extracted by batch masking. The average surface temperature of the rural area is calculated. The surface temperature difference between the city and the rural area is obtained by subtracting the average surface temperature of the rural area from the surface temperature of the city. Whether the surface temperature difference is greater than 0℃ is used to determine whether the pixel of the surface temperature difference grid data has heat island effect. The pixels of the surface temperature difference grid data are traversed. If it is greater than 0℃, it is determined that the pixel of the main urban area has heat island effect. If it is not greater than 0℃, it is determined that the pixel of the main urban area does not have heat island effect. The gray value of the pixel determined to have heat island effect is redefined as 1, and the gray value of other non-heat island pixels is defined as Nodata. The heat island effect area raster data of the main urban area is obtained. The heat island effect area raster data of the main urban area is converted into vector polygon data using the conventional GIS raster-to-vector tool, and the heat island effect area of the main urban area is obtained.

[0103] 200 pixels are randomly selected in the heat island effect pixels of the main urban area using the conventional GIS random point tool to form a target sample set. Then, 200 sample points are randomly generated in the non-urban area using the conventional GIS random point tool. The winter and summer surface temperature anomaly grid data, slope, aspect, impervious surface, night light, population density, land use, normalized vegetation index, and GDP grid data are sampled. 400 groups of training sample sets for heat island effect identification in winter and summer are obtained.

[0104] (5) Constructing a computer automatic recognition model for urban heat island effect using random forest algorithm

[0105] The sample set is randomly divided into 80% training set for constructing random forest model, and the remaining 20% is test set. The random forest model is constructed using the RandomForestClassifier function in the sklearn library of Python language. After several adjustments, the parameters of the random forest modeling are finally confirmed as the value of n_estimators is 80 and the value of max_depth is 8, which means the number of classification trees in the random forest is 80 and the calculation depth of each tree is 8. The area under the receiver operating characteristic (ROC) curve (AUC) value is used to analyze the effect of the model. The larger the AUC value, the better the model effect. The AUC value between 0.8 and 0.9 indicates a good model, and the AUC value between 0.9 and 1 indicates an excellent model. The AUC result shows that the AUC value of the summer daytime random forest model is 0.95, and the AUC value of the winter daytime random forest model is 0.98, indicating that the model effect is good.

[0106] (6) Identify the urban heat island area in the whole region using the random forest model above

[0107] The grid data of surface temperature anomaly in winter and summer, slope, aspect, impervious surface, night light, population density, land use, normalized vegetation index, and GDP grid data are used as prediction variables. The random forest model is used to predict the urban heat island effect area in Yunnan Province. The result better identifies the urban heat island effect pixels in the urban-rural combined area outside the main urban area. As shown in Figure 2 and Figure 3 , compared with the traditional urban-rural dichotomy, the result of the present embodiment better identifies the heat island effect area in the urban-rural combined area around Kunming. The urban-rural heat island effect area is mainly concentrated in the southern and eastern narrow belts of Kunming, which are new urban areas around the main urban area.

[0108] (7) Extraction of urban heat island area vector together with urban-rural combined area

[0109] The urban heat island effect area grid data in the whole region is converted to vector polygon data using the conventional GIS grid-to-vector tool to obtain the urban heat island area vector data together with the urban-rural combined area. The extraction result of the four cities in central Yunnan is shown in Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 .

[0110] (8) Heat island intensity classification

[0111] Based on the winter and summer surface temperature anomaly fields, the heat island intensity classification standard suitable for Yunnan cities is established. Through multiple threshold tests, it is found that a gradient of 0.7°C in summer and 0.6°C in winter is more appropriate. The heat island intensity is divided into four levels, and the intensity is defined as significant heat island, strong heat island, moderate heat island, and slight heat island from strong to weak. The classification results of the urban heat island effect intensity in summer and winter in Yunnan are shown in Figure 12 , Figure 13 , Figure 14 , Figure 15 , Figure 16 , Figure 17 , Figure 18 , Figure 19 .

[0112] Taking Kunming, Chuxiong, Qujing, and Yuxi in central Yunnan as an example, the statistics after classification are shown in Figure 20 and Figure 21The significant urban heat island area of Kunming mainly concentrates in the impervious surface area and high-density population distribution area, and presents a strip distribution in the urban center, with an area of 38 km 2 ; the slight heat island mainly distributes in the peripheral area with low population density, with an area of 239 km 2 , especially in the east and south areas. The significant urban heat island area of Qujing also concentrates in the high-density population area in the urban center, and the slight heat island mainly exists in the area with small population density in the north. The urban heat island intensity of Yuxi and Chuxiong is relatively low, the urban area of Yuxi is mainly the strong heat island and moderate heat island, and the slight heat island mainly distributes in the non-urban area in the north and west; the overall heat island intensity of Chuxiong is low, mainly the moderate heat island and slight heat island. These results show the effectiveness of extracting the urban heat island effect based on the anomaly field, and also provide scientific basis and reference value for subsequent research.

[0113] In the present application, on the one hand, the surface temperature anomaly is used as a variable to identify the heat island, and on the other hand, the surface temperature anomaly is used as a reference to measure the intensity of the heat island effect to classify the intensity of the heat point effect. This part avoids the technical architecture of the commonly used urban-rural dichotomy, thereby ensuring the identification and extraction of the urban-rural combination area and the intensity classification. It is the key to solve the current urban-rural dichotomy problem.

[0114] The technical solution of the present application is relatively prospective. It is generally believed that the heat island of mountainous towns is weak. The urban-rural combination area has intense human activities and obvious warming phenomenon, which needs to be concerned and analyzed in depth. Unlike plain topography, the mountainous town is not divided by the boundary between the city and the countryside.

[0115] The variable selection of AI modeling is reasonable. Human activities are the main cause of urban heat island effect. With the increase of population density, the increase of urban built-up area and the development of GDP, the heat island effect is continuously enhanced. The city center is the place with high-density population, and the business is prosperous, so the GDP value is high, and the heat island intensity is high. The population density of the city periphery is low, and the economic development is not as good as the city center, which leads to the significant gradient decrease of the urban heat island intensity from the city center to the periphery. Among them, the influence of the slope factor on the urban heat island effect has significant seasonal differences. In winter, the role of the terrain slope factor is more prominent, because the heat preservation effect of the mountainous city area is enhanced, which increases the temperature difference between the city and the surrounding area. In summer, due to the direct solar radiation and the influence of the generally high temperature in the whole area, the heat preservation effect of the terrain is significantly weakened. According to the influence of these factors, the invention selects the slope, slope direction, impervious surface, night light, population density, land use, normalized vegetation index, GDP and other factors as the measurement variable of AI modeling.

[0116] The technical effects of the present application are: the identification results of the main urban area are consistent with the results of the traditional method, and the heat island effect area of the urban-rural joint area that cannot be identified by the traditional method can be effectively identified. By comparing with the traditional urban-rural comparison method, the spatial distribution patterns of the different urban heat island intensity results identified by the anomaly and the urban-rural comparison method are consistent in the urban area, indicating that the identification results of the present application method have good consistency with the traditional urban-rural comparison method. In Kunming, the area of the slight intensity heat island identified by the traditional urban-rural comparison method is 62 km 2 , and the method reaches 239 km 2 , which is significantly different. This difference mainly comes from the newly built area around the main urban area. Through the identification of the slight heat island intensity of the urban-rural joint area around Kunming, it is shown that the method proposed in the present application is more sensitive and comprehensive in heat island effect identification.

[0117] The random forest model in the present application can be replaced by other models in the field of machine learning and deep machine learning.

[0118] The AI identification and grading system of the urban heat island effect of the urban-rural joint area, the AI identification and grading method of the urban heat island effect of the urban-rural joint area according to any one of the above, the AI identification and grading system of the urban heat island effect of the urban-rural joint area comprises:

[0119] An acquisition and preprocessing module for:

[0120] Collecting multi-year monthly average land surface temperature data, performing grid average value operation to obtain historical land surface temperature field annual average grid data; collecting current land surface temperature field grid data, performing grid subtraction operation on the historical land surface temperature field annual average grid data to obtain current land surface temperature anomaly grid data; collecting digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data, GDP data, performing surface analysis on the digital elevation model data to obtain slope grid data and slope direction grid data, performing time and spatial scale unified processing on the impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data, and projecting all data into a coordinate system consistent with the current land surface temperature anomaly grid data to obtain standardized multi-factor grid data;

[0121] A sample set construction module for:

[0122] Based on the land use data, the urban area is extracted, combined with the current land surface temperature field grid data, the urban-rural two-division method is used to identify the main urban heat island effect pixel, the pixel is randomly selected from the main urban heat island effect pixel to form a target sample set; the sample points are randomly generated from the non-urban area, the current land surface temperature anomaly grid data and the standardized multi-factor grid data are sampled to obtain a training sample set;

[0123] The recognition model module is constructed, which is used as input with the target sample set and the training sample set, and the random forest algorithm is used for model training to obtain the urban heat island effect recognition model;

[0124] The whole area heat island recognition module is used to input the current land surface temperature anomaly grid data and the standardized multi-factor grid data into the urban heat island effect recognition model to obtain the whole area urban heat island effect region grid data;

[0125] The vector extraction module is used for raster to vector processing of the whole area urban heat island effect region grid data to obtain the urban heat island area vector data including the urban-rural combined area;

[0126] The heat island intensity grading module is used for superimposed analysis of the current land surface temperature anomaly grid data and the urban heat island area vector data including the urban-rural combined area, based on the preset temperature gradient threshold, the heat island intensity is divided into four grades of significant heat island, intensity heat island, moderate heat island and slight heat island, to obtain the heat island intensity grade distribution data.

Claims

1. A method for identifying and classifying urban heat island effect (AI) in conjunction with urban-rural fringe areas, characterized by, The method comprises the following steps: S1. Obtain basic data and pre-process; collect multi-year monthly average ground temperature data, perform grid average value operation to obtain historical ground temperature field annual average grid data; collect current ground temperature field grid data, perform grid subtraction operation on the current ground temperature field grid data and the historical ground temperature field annual average grid data to obtain current ground temperature anomaly grid data; collect digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data and GDP data, perform surface analysis on the digital elevation model data to obtain slope grid data and slope direction grid data, perform time and space scale unified processing on the impervious surface data, night light data, population density data, land use data, normalized vegetation index data and GDP data, and project the time and space scale unified processed data, slope grid data and slope direction grid data into a coordinate system consistent with the current ground temperature anomaly grid data to obtain standardized multi-factor grid data; S2. Construct a sample set; extract a town area based on the land use data, identify main urban heat island effect pixels by using urban-rural dichotomy in combination with the current ground temperature field grid data, randomly select pixels from the main urban heat island effect pixels to form a target sample set, randomly generate sample points from a non-town area, and sample the current ground temperature anomaly grid data and the standardized multi-factor grid data to obtain a training sample set; S3. Construct an identification model; input the target sample set and the training sample set to perform random forest model training to obtain a city heat island effect identification model; S4. Identify heat islands in the whole region; input the current ground temperature anomaly grid data and the standardized multi-factor grid data into the city heat island effect identification model to obtain whole-region city heat island effect regional grid data; S5. Vector extraction; perform grid-to-vector processing on the whole-region city heat island effect regional grid data to obtain city heat island area vector data including urban-rural combined areas; S6. Heat island intensity grading; perform overlay analysis on the current ground temperature anomaly grid data and the city heat island area vector data including urban-rural combined areas, divide the heat island intensity into four grades of significant heat island, strong heat island, moderate heat island and slight heat island based on a preset temperature gradient threshold, and obtain heat island intensity grade distribution data. 2.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S1, the multi-year monthly average ground temperature data is MODIS monthly average ground temperature product data for at least 20 years, and the current ground temperature field grid data is current time ground temperature grid data or average ground temperature grid data in a preset time range. 3.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S1, if the current ground temperature field grid data has null values due to cloud cover, the average value of the pixels in a 5*5 region around the null pixels is used for filling. 4.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S2, the process of identifying main urban heat island effect pixels by using urban-rural dichotomy comprises the following steps: establishing a buffer zone in the town area, erasing water body areas in the buffer zone to obtain a rural area, calculating the average ground temperature of the rural area, subtracting the average ground temperature of the rural area from the ground temperature of the town area to obtain ground temperature difference grid data, and determining whether the pixels in the ground temperature difference grid data are heat island effect pixels based on a preset threshold. 5.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S3, the parameters of the random forest algorithm include the number of classification trees and the calculation depth of each tree, and the optimal values of the number of classification trees and the calculation depth of each tree are determined through multiple experiments. 6.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S3, the area value under the receiver operating characteristic curve is used to evaluate the model effect, and when the area value is in the interval of 0.9-1, it is determined that the model training is completed. 7.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S6, the preset temperature gradient threshold is set according to the season combined with the natural breakpoint method, and the gradient of 0.7℃ is used in summer and the gradient of 0.6℃ is used in winter. 8.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is determined by using a deep learning algorithm. In S1, the spatial resolution of the digital elevation model data, the impervious surface data, the night light data, the population density data, the normalized vegetation index data, and the GDP data can be any size that can be collected. 9.The method of claim 1, wherein the AI identification of the urban heat island effect in the urban fringe area is performed by using a deep learning algorithm. The target sample set and the training sample set are used as learning sample data, and are randomly divided into a model training set and a test set in a ratio of 8:2, the model training set is used for model training, and the test set is used for model effect verification.

10. An urban heat island effect AI identification grading system in conjunction with an urban-rural fringe area, characterized by, The urban heat island effect AI identification grading system according to any one of claims 1-9, comprising: An acquisition and preprocessing module for: Collecting multi-year monthly average land surface temperature data, performing grid average value operation to obtain historical land surface temperature field annual average grid data; collecting current land surface temperature field grid data, performing grid subtraction operation on the current land surface temperature field grid data and the historical land surface temperature field annual average grid data to obtain current land surface temperature anomaly grid data; collecting digital elevation model data, impervious surface data, night light data, population density data, land use data, normalized vegetation index data, and GDP data, performing surface analysis on the digital elevation model data to obtain slope grid data and slope direction grid data, performing time and spatial scale unified processing on the impervious surface data, the night light data, the population density data, the land use data, the normalized vegetation index data, and the GDP data, and projecting the time and spatial scale unified processed data, the slope grid data, and the slope direction grid data into a coordinate system consistent with the current land surface temperature anomaly grid data to obtain standardized multi-factor grid data; A sample set construction module for: Extracting urban areas based on land use data, combining current land surface temperature field grid data, and identifying main urban heat island effect pixels using urban-rural dichotomy, randomly selecting pixels from the main urban heat island effect pixels to form a target sample set; randomly generating sample points from non-urban areas, and sampling the current land surface temperature anomaly grid data and the standardized multi-factor grid data to obtain a training sample set; A recognition model construction module for inputting the target sample set and the training sample set, performing random forest model training, and obtaining an urban heat island effect recognition model; A full-area heat island recognition module for inputting the current land surface temperature anomaly grid data and the standardized multi-factor grid data into the urban heat island effect recognition model to obtain full-area urban heat island effect regional grid data; A vector extraction module for performing grid-to-vector processing on the full-area urban heat island effect regional grid data to obtain urban heat island area vector data together with urban-rural combined areas. The heat island intensity grading module is used for superimposing and analyzing the current land surface temperature anomaly grid data and the urban heat island area vector data together with the urban-rural joint area, dividing the heat island intensity into four grades of significant heat island, intensity heat island, moderate heat island and slight heat island based on a preset temperature gradient threshold, and obtaining heat island intensity grade distribution data.

Images