Estimation method for predicting urban surface temperature based on urban green land information acquired by remote sensing

Abstract

The invention discloses an estimation method for predicting urban surface temperature based on urban green land information acquired by remote sensing. The method is characterized by comprising the following steps: acquiring remote sensing image data of a target city; preprocessing the remote sensing image data, constructing a label data set, calculating the surface temperature of the target city, and extracting the built-up area range of the target city; urban green land is extracted in the built-up area range; calculating a vegetation index and a landscape index based on the urban green land distribution range; the vegetation index and the landscape index serve as independent variables, the surface temperature serves as a dependent variable, an optimal independent variable is obtained through screening, three machine learning models are constructed through the optimal independent variable to serve as prediction models, and the prediction precision of the three prediction models is compared to obtain an optimal prediction model; the method has the advantages that the surface temperature is reflected quantitatively through the urban green land information, a basis for green land planning is provided for livable environment construction of a city, and the stability of a prediction model is kept.

Description

Technical Field

[0001] The present invention relates to the technical field of inverted ground feature information from remote sensing images, and in particular to an estimation method for predicting urban surface temperature based on urban green space information acquired by remote sensing. Background Art

[0002] With the acceleration of urbanization, the urban heat island effect is becoming increasingly severe. Extremely high temperatures not only affect urban environmental quality but also pose significant health risks to residents. Urban green spaces play a vital role in regulating urban temperatures and mitigating the urban heat island effect. However, traditional research on the relationship between green space and temperature has been limited to local areas and primarily relies on air temperature sensors or ground-based observations. This has spatial scale limitations and makes it difficult to fully reflect the overall impact of urban green spaces. Studies have shown that green space characteristics such as vegetation density, coverage, and distribution patterns have a significant impact on ambient temperature, particularly in areas such as parks and green spaces. However, existing research often focuses on small-scale areas, often qualitatively analyzing the relationship between characteristics such as vegetation density and temperature, and lacks in-depth analysis of the quantitative relationship between green space distribution and urban temperature at larger scales. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an estimation method for predicting urban surface temperature based on urban green space information obtained by remote sensing. The method can perform quantitative analysis between urban green space information and urban surface temperature, and provide a basis for green space planning in the construction of a livable environment in the city.

[0004] The technical solution adopted by the present invention to solve the above technical problems is: an estimation method for predicting urban surface temperature based on urban green space information obtained by remote sensing, comprising the following specific steps:

[0005] (1) Obtain remote sensing image data of the target city;

[0006] (2) Preprocess the remote sensing image data, construct a label dataset, calculate the surface temperature of the target city, and extract the built-up area of ​​the target city;

[0007] (3) Extract urban green space within the built-up area;

[0008] (4) Calculate the vegetation index and landscape index based on the distribution range of urban green spaces within the built-up area;

[0009] (5) Taking vegetation index and landscape index as independent variables and surface temperature as dependent variable, the independent variables are screened and the optimal independent variables are obtained through screening. Then, three machine learning models are constructed as prediction models based on the optimal independent variables. The prediction accuracy of the three prediction models is compared to obtain the optimal prediction model. The optimal prediction model is used to estimate the urban surface temperature based on urban green space information.

[0010] Furthermore, in the step (1), the remote sensing image data of the target city is obtained through GaoFen-2 satellite data, Landsat8-TIRS satellite data and Luojia 1 satellite data, and the cloud cover of the downloaded satellite data is less than 10%, and the target area is not cloud-obstructed, wherein: GaoFen-2 satellite data is high spatial resolution data, which is used to obtain green space information of the target city; Landsat8-TIRS satellite data is thermal infrared data, which is used to obtain the surface temperature of the target city; Luojia 1 satellite data is night light data, which is used to obtain the built-up area of ​​the target city.

[0011] Furthermore, in the step (2), the GaoFen-2 satellite data in the remote sensing image data is pre-processed by radiometric calibration, atmospheric correction, orthorectification and image fusion.

[0012] Furthermore, in step (2), the specific process of constructing the label data set is as follows:

[0013] (2-1) Based on the true color data provided by GaoFen-2 satellite data, combined with Google Earth images and Baidu image base maps, the distribution range of urban green space (UGS) was visually interpreted. At the same time, with the help of false color synthesis of the band information of GaoFen-2 satellite data and the vegetation index of GaoFen-2, the distribution range of urban green space was auxiliary interpreted. The outline of the urban green space in the target area was outlined. When drawing the vector, if any information that could not be interpreted was encountered, the distribution range of urban green space was further interpreted by combining Baidu Street View Map and using panoramic pictures;

[0014] (2-2) The interpreted urban green space distribution range is sliced ​​with a size of 256×256 and a sampling interval of 128 to produce a labeled dataset.

[0015] Furthermore, in step (2), the surface temperature of the target city is calculated as follows:

[0016] First, perform radiation calibration, surface emissivity calculation and radiation brightness calculation on the Landsat8-TIRS satellite data, and then calculate the surface temperature T using the radiation transfer equation method. s , the radiation transfer equation is:

[0017] L sen =B(T sen )=(εB(T s )+(1-ε)L d )τ+L u ,

[0018] Where: L senis the radiation brightness value obtained from the Landsat8 satellite sensor, B is the Planck function, T sen is the brightness temperature on the Landsat8 satellite sensor, ε is the surface emissivity of the earth, τ is the transmittance of radiation energy in the atmosphere, T s is the surface temperature of the target city, L d L is the radiant brightness of the atmosphere reaching the Landsat8 satellite after radiating to the surface. u It is the radiance transmitted upward by the atmosphere itself and reaching the Landsat 8 satellite;

[0019] In order to eliminate the influence of the atmosphere, the relationship is:

[0020] Among them: K1 and K2 are radiation constants, and the accurate surface temperature T is obtained s .

[0021] Furthermore, in step (2), the method for extracting the built-up area of ​​the target city is:

[0022] The built-up area of ​​the target city is superimposed with the Luojia-1 satellite data from high to low values ​​until it is consistent with the officially designated built-up area. The range formed by the built-up area at this time is the built-up area range.

[0023] Furthermore, in step (3), the extraction process of urban green space is as follows:

[0024] (3-1) Rotate the original data of the label dataset by 90° three times to augment the label data;

[0025] (3-2) Calculate the Normalized Difference Vegetation Index (NDVI) within the built-up area based on GaoFen-2 satellite data. Among them: NIR is the near infrared band, RED is the red band, and the raster expressed in the near infrared band and the normalized difference vegetation index NDVI are added as new labels to the label dataset to form a new label dataset;

[0026] (3-3) Select 90% of the new labeled dataset as training data and 10% as test data, and extract the urban green space (UGS) within the built-up area of ​​the GaoFen-2 satellite data using the U-Net semantic segmentation model;

[0027] (3-4) After urban green space extraction, the accuracy, F1 score, and similarity coefficient Dice are used as evaluation indicators to determine the final extraction accuracy. The relationship is:

[0028]

[0029] Among them: TP means that the sample is judged to be positive and is actually a positive sample; TN means that the sample is judged to be negative and is actually a negative sample; FP means that the sample is judged to be positive but is actually a negative sample; FN means that the sample is judged to be negative but is actually a positive sample; the similarity coefficient Dice is one of the evaluation criteria in the segmentation process, and the Dice value ranges from 0 to 1. The best segmentation result is 1, and the worst segmentation result is 0; the F1 score is defined as the harmonic mean of the precision and recall.

[0030] Furthermore, in step (5), the importance of the independent variables is evaluated using the aggregate boosting tree model, and three machine learning models are constructed using the index of the relative influence percentage of the independent variables greater than 0, and the adjusted determination coefficient is used. The root mean square error (RMSE), residual prediction deviation (RPD), and P-value were used to compare the prediction accuracy of the three models. The prediction model with the highest value and the smallest RMSE is the best prediction model.

[0031] Furthermore, in step (5), the three machine learning models are support vector machine, random forest and gradient boosting regression tree, and all three models are regression models.

[0032] Compared with the prior art, the advantages of the present invention are:

[0033] (1) This method uses the calculated vegetation index and landscape index as independent variables to construct a machine learning model for inverting surface temperature from urban green space information. It quantitatively reflects the surface temperature through urban green space information, providing a basis for green space planning in the construction of a livable environment in cities.

[0034] (2) This method uses the aggregation boosting tree model to select the independent variables, so that when there are few independent variables, the urban green space information can still maintain the predictive ability of the surface temperature, realize the screening of important urban green space information variables for inverting the urban surface temperature, and further maintain the stability of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A diagram showing the process of creating a label dataset of the present invention;

[0036] Figure 2 The surface temperature inversion and surrounding environment map of the present invention;

[0037] Figure 3 This is the built-up area extraction map of the present invention;

[0038] Figure 4 This is a schematic diagram of label data augmentation of the present invention;

[0039] Figure 5 The urban green space extraction map within the built-up area of ​​the present invention;

[0040] Figure 6 This is the result diagram of screening independent variables using the Aggregate Boosting Tree Model (ABT) in the present invention;

[0041] Figure 7 This is a comparison chart of the prediction accuracy of the prediction models of the present invention. It can be seen from the figure that the gradient boosted regression tree GBRT model is the best prediction model. DETAILED DESCRIPTION

[0042] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

[0043] As shown in the figure, the estimation method for predicting urban surface temperature based on urban green space information obtained by remote sensing includes the following specific steps:

[0044] (1) Obtain remote sensing image data of the target city through GaoFen-2 (GF-2) satellite data, Landsat8-TIRS satellite data and Luojia 1 (LJ1-01) satellite data. The cloud cover of the downloaded satellite data is less than 10%, and the target area is not cloud-obstructed. The remote sensing image time of the target city obtained from the GF-2 satellite data and Landsat8-TIRS satellite data is the summer season when vegetation grows densely and urban thermal effects are significant. Among them: GF-2 satellite data is high spatial resolution data, which is used to obtain green space information of the target city; Landsat8-TIRS satellite data is thermal infrared data, which is used to obtain the surface temperature of the target city; LJ1-01 satellite data is night light data, which is used to obtain the built-up area of ​​the target city;

[0045] (2) Preprocess the remote sensing image data, construct a label dataset, calculate the surface temperature of the target city, and extract the built-up area of ​​the target city; specifically:

[0046] (2-1) Preprocessing of GF-2 satellite data in remote sensing image data by radiometric calibration, atmospheric correction, orthorectification and image fusion;

[0047] (2-2) Based on the true color data provided by the GF-2 satellite data, combined with Google Earth images and Baidu image base maps, the distribution range of urban green spaces was visually interpreted. At the same time, with the help of false color synthesis of the band information of the GF-2 satellite data and the vegetation index of GF-2, the distribution range of urban green spaces was auxiliary interpreted. The outline of the urban green space in the target area was outlined by drawing. When drawing the vectors, if any information that could not be interpreted was encountered, the distribution range of urban green spaces was further interpreted by combining Baidu Street View Maps and using panoramic images;

[0048] (2-3) The interpreted urban green space distribution range is sliced ​​with a size of 256×256 and a sampling interval of 128 to produce a labeled dataset;

[0049] (2-4) Perform radiation calibration, surface emissivity calculation and radiation brightness calculation on Landsat8-TIRS satellite data, and then calculate the surface temperature T using the radiation transfer equation (RTE) method. s , the radiation transfer equation is:

[0050] L sen =B(T sen )=(εB(T s )+(1-ε)L d )τ+L u ,

[0051] Where: L sen is the radiation brightness value obtained by the Landsat8 satellite sensor (unit: W·m -2 ·sr -1 μm -1 ), B is the Planck function, T sen is the brightness temperature on the Landsat 8 satellite sensor (unit: K), ε is the surface emissivity of the earth, τ is the transmittance of radiation energy in the atmosphere, T s is the surface temperature of the target city (unit: K), L d It is the radiant brightness of the atmosphere reaching the Landsat 8 satellite after radiating to the surface (unit: W·m -2 ·sr -1 μm -1 ), L u It is the radiance transmitted upward by the atmosphere itself and reaching the Landsat 8 satellite (unit: W·m -2 ·sr -1 μm -1 );

[0052] In order to eliminate the influence of the atmosphere, the relationship is:

[0053] Among them: K1 and K2 are radiation constants, and the accurate surface temperature T is obtained s ;

[0054] (2-5) Overlay the built-up area of ​​the target city using the LJ1-01 satellite data from high to low values ​​until it matches the officially designated built-up area. The range formed by the built-up area at this point is the built-up area range;

[0055] (3) Extract urban green space within the built-up area, specifically:

[0056] (3-1) Rotate the original data of the label dataset by 90° three times to augment the label data;

[0057] (3-2) Calculate the Normalized Difference Vegetation Index (NDVI) within the built-up area based on GaoFen-2 satellite data. Among them: NIR is the near infrared band, RED is the red band, and the raster expressed in the near infrared band and the normalized difference vegetation index NDVI are added as new labels to the label dataset to form a new label dataset;

[0058] (3-3) Select 90% of the new labeled dataset as training data and 10% as test data, and extract urban green spaces within the built-up area of ​​GF-2 satellite data using the U-Net semantic segmentation model;

[0059] (3-4) After urban green space extraction, the accuracy, F1 score, and similarity coefficient Dice are used as evaluation indicators to determine the final extraction accuracy. The relationship is:

[0060]

[0061] Among them: TP means that the sample is judged to be positive and is actually a positive sample; TN means that the sample is judged to be negative and is actually a negative sample; FP means that the sample is judged to be positive but is actually a negative sample; FN means that the sample is judged to be negative but is actually a positive sample; the similarity coefficient Dice is one of the evaluation criteria in the segmentation process, and the Dice value ranges from 0 to 1. The best segmentation result is 1, and the worst result is 0; the F1 score is defined as the harmonic mean of the precision and recall.

[0062] (4) Calculate the vegetation index and landscape index based on the distribution range of urban green spaces within the built-up area and use them as characteristic information for estimating surface temperature;

[0063] The vegetation index can reflect the surface coverage of vegetation and indirectly reflect the density of vegetation growth. The potential vegetation index is selected as the characteristic information of urban green space and participates in the inversion of surface temperature.

[0064] The vegetation indices are: difference vegetation index DVI, enhanced vegetation index EVI, global environment monitoring index GEMI, green atmospheric resistance index GARI, green difference vegetation index GDVI, green normalized vegetation index GNDVI, green ratio vegetation index GRVI, near infrared percentage vegetation index IPVI, leaf area index LAI, modified nonlinear index MNLI, modified ratio index MSR, nonlinear index NLI, normalized difference snow index NDSI, normalized vegetation index NDVI, optimized soil adjusted vegetation index OSAVI, renormalized vegetation index RDVI, simple ratio index SR, soil adjusted vegetation index SAVI, transformed difference vegetation index TDVI, visible atmospheric resistance vegetation index VARI;

[0065] The landscape index can highly condense the landscape pattern, thereby quantitatively describing the landscape pattern. The landscape pattern of urban green space is used as a type of characteristic information, and then participates in the prediction of surface temperature together with the vegetation index.

[0066] The landscape indices are: total area TA, number of patches NP, patch density PD, maximum patch index LPI, total edge TE, edge density ED, landscape shape index LSI, peripheral area fractal dimension PAFRAC, spread index CONTAG, similar proximity percentage PLADJ, cohesion index COHESION, landscape separation index DIVISION, effective grid area MESH, fragmentation index SPLIT, patch abundance density PRD, Shannon diversity index SHDI, Simpson diversity index SIDI, modified Simpson diversity index MSIDI, Shannon evenness index SHEI, Simpson evenness index SIEI, effective grid area index MSIEI, aggregation index AI;

[0067] (5) Taking vegetation index and landscape index as independent variables and surface temperature as dependent variable, the Aggregate Boosting Tree Model (ABT) was used to evaluate the importance of independent variables. The index with a relative influence percentage greater than 0 (i.e., the optimal independent variable) among the independent variables (i.e., vegetation index and landscape index) was used to construct three machine learning models: support vector machine (SVM), random forest (RF), and gradient boosting regression tree (GBRT) as prediction models. All three prediction models were regression models, and the adjusted coefficient of determination was used. The root mean square error (RMSE), residual prediction deviation (RPD), and P-value were used to compare the prediction accuracy of the three prediction models. The prediction model with the highest value and the smallest RMSE is the best prediction model, and the best prediction model is used to estimate the urban surface temperature based on urban green space information.

[0068] The protection scope of the present invention includes but is not limited to the above embodiments, and its protection scope is subject to the claims. Any replacement, deformation, and improvement of this technology that can be easily thought of by those skilled in the art fall within the protection scope of the present invention.

Claims

1. The estimation method for predicting urban surface temperature based on urban green space information obtained by remote sensing is characterized by The specific steps include: (1) Obtain remote sensing image data of the target city; (2) Preprocess the remote sensing image data, construct a label dataset, calculate the surface temperature of the target city, and extract the built-up area of ​​the target city; (3) Extract urban green space within the built-up area; (4) Calculate the vegetation index and landscape index based on the distribution range of urban green spaces within the built-up area; (5) Taking vegetation index and landscape index as independent variables and surface temperature as dependent variable, the independent variables are screened and the optimal independent variables are obtained through screening. Then, three machine learning models are constructed as prediction models based on the optimal independent variables. The prediction accuracy of the three prediction models is compared to obtain the optimal prediction model. The optimal prediction model is used to estimate the urban surface temperature based on urban green space information.

2. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 1, characterized in that: In the step (1), the remote sensing image data of the target city is obtained through GaoFen-2 satellite data, Landsat8-TIRS satellite data and Luojia 1 satellite data, and the cloud cover of the downloaded satellite data is less than 10%, and the target area is not cloud-obstructed, wherein: GaoFen-2 satellite data is high spatial resolution data, which is used to obtain green space information of the target city; Landsat8-TIRS satellite data is thermal infrared data, which is used to obtain the surface temperature of the target city; Luojia 1 satellite data is night light data, which is used to obtain the built-up area of ​​the target city.

3. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 2, characterized in that: In the step (2), the GaoFen-2 satellite data in the remote sensing image data is preprocessed by radiometric calibration, atmospheric correction, orthorectification and image fusion.

4. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 2, characterized in that: In step (2), the specific process of constructing the label data set is as follows: (2-1) Based on the true color data provided by GaoFen-2 satellite data, combined with Google Earth images and Baidu image base maps, the distribution range of urban green spaces was visually interpreted. At the same time, with the help of false color synthesis of the band information of GaoFen-2 satellite data and the vegetation index of GaoFen-2, the distribution range of urban green spaces was auxiliary interpreted. The outline of the urban green space in the target area was outlined. When outlining the vectors, if any information that could not be interpreted was encountered, the distribution range of urban green spaces was further interpreted by combining Baidu Street View Maps and using panoramic images; (2-2) The interpreted urban green space distribution range is sliced ​​with a size of 256×256 and a sampling interval of 128 to produce a labeled dataset.

5. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 2, characterized in that: In step (2), the surface temperature of the target city is calculated as follows: First, perform radiation calibration, surface emissivity calculation and radiation brightness calculation on the Landsat8-TIRS satellite data, and then calculate the surface temperature T using the radiation transfer equation method. s , the radiation transfer equation is: L sen =B(T sen )=(εB(T s )+(1-e)L d )τ+L u , Where: L sen is the radiation brightness value obtained from the Landsat8 satellite sensor, B is the Planck function, T sen is the brightness temperature on the Landsat8 satellite sensor, ε is the surface emissivity of the earth, τ is the transmittance of radiation energy in the atmosphere, T s is the surface temperature of the target city, L d L is the radiant brightness of the atmosphere reaching the Landsat8 satellite after radiating to the surface. u It is the radiance transmitted upward by the atmosphere itself and reaching the Landsat 8 satellite; In order to eliminate the influence of the atmosphere, the relationship is: Among them: K1 and K2 are radiation constants, and the accurate surface temperature T is obtained s .

6. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 2, characterized in that: In the step (2), the method for extracting the built-up area of ​​the target city is: The built-up area of ​​the target city is superimposed with the Luojia-1 satellite data from high to low values ​​until it is consistent with the officially designated built-up area. The range formed by the built-up area at this time is the built-up area range.

7. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 2, characterized in that: In the step (3), the extraction process of urban green space is as follows: (3-1) Rotate the original data of the label dataset by 90° three times to augment the label data; (3-2) Calculate the Normalized Difference Vegetation Index (NDVI) within the built-up area based on GaoFen-2 satellite data. Among them: NIR is the near infrared band, RED is the red band, and the raster expressed in the near infrared band and the normalized difference vegetation index NDVI are added as new labels to the label dataset to form a new label dataset; (3-3) Select 90% of the new labeled dataset as training data and 10% as test data, and extract urban green spaces within the built-up area of ​​GaoFen-2 satellite data using the U-Net semantic segmentation model; (3-4) After urban green space extraction, the accuracy, F1 score, and similarity coefficient Dice are used as evaluation indicators to determine the final extraction accuracy. The relationship is: Among them: TP means that the sample is judged to be positive and is actually a positive sample; TN means that the sample is judged to be negative and is actually a negative sample; FP means that the sample is judged to be positive but is actually a negative sample; FN means that the sample is judged to be negative but is actually a positive sample; the similarity coefficient Dice is one of the evaluation criteria in the segmentation process, and the Dice value ranges from 0 to 1. The best segmentation result is 1, and the worst segmentation result is 0; the F1 score is defined as the harmonic mean of the precision and recall.

8. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 1, characterized in that: In step (5), the importance of the independent variables is evaluated using the aggregate boosting tree model, and three machine learning models are constructed using the index of the relative influence percentage of the independent variables greater than 0, and the adjusted coefficient of determination is used. The root mean square error (RMSE), residual prediction deviation (RPD), and P-value were used to compare the prediction accuracy of the three models. The prediction model with the highest value and the smallest RMSE is the best prediction model.

9. The method for estimating urban surface temperature based on urban green space information acquired by remote sensing according to claim 1, characterized in that: In step (5), the three machine learning models are support vector machine, random forest and gradient boosting regression tree, and all three models are regression models.

Images