A surface temperature downscaling method based on thermal infrared radiation transfer model

Abstract

This invention discloses a surface temperature downscaling method based on a thermal infrared radiation transfer model. Based on an analysis of the physical mechanisms of ground-air radiation transfer in the thermal infrared band, this method develops a surface temperature downscaling method that couples high- and low-spatial-resolution thermal infrared radiation transfer models. Based on low-spatial-resolution radiance downscaling, this method uses a low-spatial-resolution radiation transfer model to calculate atmospheric parameters. This method, combined with high-spatial-resolution surface emissivity calculation and spectral conversion, solves the surface temperature using a high-spatial-resolution thermal infrared radiation transfer equation, and downscales the low-spatial-resolution surface temperature product to a higher spatial resolution. Based on the physical mechanism, this invention effectively obtains surface temperature products with higher spatial resolution through remote sensing, meeting the demand for high-temporal- and spatial-resolution surface temperature data in fields such as urban thermal environment research. The method has important scientific significance and application value.

Description

Technical Field

[0001] The present invention relates to a surface temperature downscaling method based on a thermal infrared radiation transmission model, and belongs to the technical field of surface temperature downscaling. Background Art

[0002] Land surface temperature (LST) is a key parameter in Earth's surface physical processes. It is a crucial parameter in the energy exchange between the Earth's surface and the atmosphere, and a key indicator of Earth's surface energy balance and global climate change. It plays a vital role in economic, social, and environmental research. However, due to limitations in thermal infrared sensor imaging and the influence of weather conditions, LST products derived from thermal infrared bands generally suffer from inconsistencies in temporal and spatial resolution. Therefore, downscaling high-temporal-resolution LST data is an effective method for obtaining high-temporal-resolution LST data.

[0003] Based on a comprehensive analysis of the current research status at home and abroad, scholars have carried out a lot of research work on surface temperature downscaling, and the research methods are mainly divided into two categories: spatiotemporal fusion and statistical regression.

[0004] The spatiotemporal fusion method effectively fuses low-spatial-resolution, high-temporal-resolution images with high-spatial-resolution, low-temporal-resolution images. Using similar pixels within the spatiotemporal neighborhood of the pixel to be fused, a model is constructed to generate fused pixels based on a comprehensive weighting of spectral, temporal, and spatial information, yielding high-spatial-resolution land surface temperature. This method is susceptible to spatial and temporal variations in pixel radiance, resulting in higher accuracy in areas with uniform land cover. However, in urban areas with high spatial heterogeneity, significant fusion errors may occur when projecting low-spatial-resolution images onto high-spatial-resolution images due to the presence of heavily mixed pixels.

[0005] Statistical regression methods (also known as kernel-driven models) establish statistical relationships between surface temperature and influencing factors, such as surface parameters, at low spatial resolution. These relationships are assumed to be constant across spatial scale, and then high-resolution surface parameters are input to estimate high-resolution surface temperature. This method has the advantage of simple computation. However, in urban areas with high spatial heterogeneity, the statistical relationships vary with spatial location, resulting in generally lower downscaling accuracy than in areas with uniform land cover types.

[0006] Based on the above analysis, compared with homogeneous surfaces with simple surface cover types, the accuracy of these two types of surface temperature downscaling methods is usually lower in urban areas with high spatial heterogeneity, and both methods ignore the physical relationship between the radiation brightness of thermal infrared bands with different spatial resolutions. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a surface temperature downscaling method based on the thermal infrared radiation transfer model. Starting from the physical mechanism of the ground-air radiation transfer process in the thermal infrared band, the low spatial resolution surface temperature is downscaled to a higher spatial resolution by coupling the high and low spatial resolution thermal infrared radiation transfer equations.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] A surface temperature downscaling method based on a thermal infrared radiation transfer model includes the following steps:

[0010] Step 1: For the low spatial resolution of the surface temperature to be downscaled The thermal infrared band remote sensing image under the thermal infrared channel is obtained The radiance and radiance influencing factor data are used to train the random forest downscaling model. The trained random forest downscaling model is used to convert the low spatial resolution Downscaling of the radiance to a high spatial resolution Then, the corrected high spatial resolution is obtained. Radiance under

[0011] Step 2: Use the thermal infrared band radiation transfer model to represent the low spatial resolution The thermal infrared channel radiance of the thermal infrared band remote sensing image is calculated, and the low spatial resolution atmospheric parameters in the thermal infrared band radiation transfer model are calculated using the least squares method. Atmospheric parameters at high spatial resolution Atmospheric parameters under

[0012] Step 3: Obtain the surface emissivity spectral curve data of different types of ground objects and fit the low spatial resolution Thermal infrared channel and with The closest high spatial resolution Thermal infrared channel The wide-band surface emissivity of the ground is determined by the hybrid pixel decomposition method, and a linear relationship model is established between the two. The corresponding surface emissivity is converted to high spatial resolution Lower thermal infrared channel The corresponding surface emissivity;

[0013] Step 4: High spatial resolution The corrected radiance, atmospheric parameters and surface emissivity are substituted into the high spatial resolution Thermal infrared band radiation transfer model under high spatial resolution The surface temperature under the condition of γ is reduced to achieve surface temperature downscaling.

[0014] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0015] 1. This paper analyzes the ground-air radiation transmission process in the thermal infrared band and designs a surface temperature downscaling method that couples high- and low-spatial-resolution thermal infrared radiation transmission models. This is a further development and expansion of existing surface temperature downscaling methods.

[0016] 2. The surface temperature downscaling method proposed in this invention can downscale the surface temperature of low spatial resolution to obtain surface temperature data with higher precision and high spatial resolution, which can provide support for related fields such as urban thermal environment research that meet the demand for surface temperature data with higher temporal and spatial resolution.

[0017] 3. The present invention uses 1km resolution MODIS surface temperature downscaling as an example and is applied in Hefei, Anhui Province, China. The results show that the downscaling accuracy of the proposed method is higher than that of traditional random forest statistical downscaling. In addition, the spatial distribution of surface temperature in urban areas with high spatial heterogeneity is more reasonable and more consistent with the spatial distribution of Landsat TIRS surface temperature at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a surface temperature downscaling method based on a thermal infrared radiation transfer model of the present invention;

[0019] Figure 2 This is the surface temperature distribution in Hefei City, the test area, at 10:40 on October 17, 2023 in the embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be interpreted as limiting the present invention.

[0021] This paper proposes a surface temperature downscaling method based on a thermal infrared radiation transfer model. This method analyzes the ground-air radiation transfer process in the thermal infrared band from a physical perspective. By coupling high- and low-spatial-resolution thermal radiation transfer models, the method downscales low-spatial-resolution surface temperature data to a higher-spatial-resolution one. The specific steps are as follows:

[0022] 1) Low spatial resolution radiance statistical downscaling

[0023] Statistical regression downscaling methods are based on the principle of scale invariance and are widely used in downscaling research. The present invention trains a random forest relationship model between radiance and influencing factors at low spatial resolution, inputs high spatial resolution influencing factors, and obtains radiance downscaling results at high spatial resolution. Random forest is an ensemble learning algorithm proposed by Breiman in 2001. This model is based on decision trees and uses bootstrap resampling to randomly extract samples from the training set to train multiple trees. When a decision tree node splits, the algorithm randomly selects a feature subset, and the final prediction result is determined by voting on the prediction results of all trees.

[0024] At low spatial resolution, considering the factors that affect the output radiance of the sensor's thermal infrared band, surface remote sensing indices such as the Normalized Difference Water Index (MNDWI), Normalized Difference Building Index (NDBI), Normalized Difference Moisture Index (NDMI), and Normalized Difference Vegetation Index (NDVI) are calculated using surface reflectance data. These remote sensing indices and elevation are used as independent variables, and the low spatial resolution radiance is used as the dependent variable to train a random forest downscaling model, as shown in the following formula:

[0025] (1)

[0026] Among them, the subscript Indicates low spatial resolution; is the thermal infrared channel radiance at low spatial resolution, 、 、 and They are the low spatial resolution normalized difference vegetation index, building index, moisture index and water body index, is the elevation at low spatial resolution.

[0027] At low spatial resolution, the coefficient of determination (R 2 ) and the root mean square error (RMSE) are used to evaluate the model training accuracy and select the optimal hyperparameters of the model. Using a high-resolution image with the closest imaging date to the low-resolution image to be downscaled, various high-resolution remote sensing indices are calculated based on geometric and radiometric correction of the remote sensing image. These indices are then input into Equation (2) to calculate the initial values ​​of the high-resolution radiometric brightness downscaling.

[0028] (2)

[0029] Among them, the subscript Indicates high spatial resolution.

[0030] According to the imaging mechanism of thermal infrared remote sensing images, the radiance at the same position on images with different spatial resolutions is equal, which satisfies the principle of energy conservation. Based on this, the initial value of the radiance downscaling is corrected. The specific correction process is: Aggregate upscaling to low spatial resolution , calculate the original low spatial resolution radiance and The residual between ), resample it to high spatial resolution and smooth it with Gaussian filtering to obtain the resampled residual with high spatial resolution Finally, the smoothed residual is added to the initial value of the radiance downscaling to obtain the corrected high spatial resolution radiance ,Right now:

[0031] (3)

[0032] 2) Calculation of atmospheric parameters based on low spatial resolution thermal radiation transfer model

[0033] According to the ground-air radiation transmission process in the thermal infrared band, the thermal infrared band radiation brightness received by the sensor at low spatial resolution can be expressed as follows:

[0034] (4)

[0035] in, is the low spatial resolution surface emissivity; Temperature Blackbody radiation corresponding to the surface; is the true surface temperature at low spatial resolution; 、 and They are respectively the atmospheric downward radiation, atmospheric upward radiation and atmospheric transmittance corresponding to low spatial resolution.

[0036] In thermal infrared surface temperature inversion, it is usually assumed that the atmospheric distribution level is uniform throughout the entire study area, that is, the atmospheric parameters at different pixel locations are the same. For low spatial resolution sensors, the radiance of a thermal infrared channel is , surface emissivity , and surface temperature Given, input surface temperature and the central wavelength of the thermal infrared channel, which can be calculated using Planck's law According to the thermal infrared radiation transmission model, the surface temperature of different pixels with low spatial resolution, the radiation brightness of the thermal infrared channel and the surface emissivity are substituted into formula (4). The atmospheric upward radiation corresponding to the thermal infrared channel of the low spatial resolution sensor can be estimated using the least squares method. , atmospheric downward radiation and atmospheric transmittance Three atmospheric parameters.

[0037] Based on the assumption that the atmospheric distribution level is uniform throughout the test area, it can be considered that the atmospheric parameters do not change with spatial scale, that is, the atmospheric parameters at low spatial resolution and high spatial resolution are equal, which can be expressed using the following formula:

[0038] (5)

[0039] 3) High spatial resolution surface emissivity estimation and its spectral conversion

[0040] When calculating high-spatial-resolution surface temperature based on the thermal infrared band radiation transfer equation, this method requires obtaining high-spatial-resolution emissivity data corresponding to the spectrum of the low-spatial-resolution thermal infrared channel. Because the spectral response of the thermal infrared channels of high- and low-spatial-resolution sensors differ, it is necessary to convert the high-spatial-resolution surface emissivity to the spectral emissivity corresponding to the low-spatial-resolution thermal infrared channel.

[0041] Download the spectral emissivity data of various typical ground objects from the UCSB website, and use the convolution operation method to calculate the spectral response function of the low spatial resolution thermal infrared channel and the similar channel of the high spatial resolution sensor. , high spatial resolution sensor channels Broadband emissivity 、 :

[0042] (6)

[0043] Among them, the subscript Representative Channel or ; is the wavelength Position corresponding to or spectral response data of the channel; For a typical feature at wavelength emissivity at ; 、 Represents channels or The lower and upper limits of the wavelength range.

[0044] On this basis, a linear conversion relationship between the two is established, as shown in the following formula, which can convert the emissivity of the high spatial resolution thermal infrared channel into the emissivity corresponding to the low spatial resolution thermal infrared channel.

[0045] (7)

[0046] Taking into account that the surface emissivity changes little with time in a short period of time, high spatial resolution remote sensing image data close to the time of low spatial resolution surface temperature imaging to be downscaled can be selected. On the basis of necessary atmospheric correction and radiation correction of the image, the hybrid pixel decomposition method is used to determine the surface emissivity corresponding to the high spatial resolution thermal infrared channel.

[0047] Using high-resolution visible and near-infrared reflectance images, supervised classification was used to classify the test area's surface into four categories: urban surface, natural surface, bare soil, and water. The emissivity of water was set to 0.99. Natural surface pixels can be considered as mixed pixels composed of vegetation and bare soil in proportion, and are calculated using Equation (8). Urban surface pixels are primarily composed of various building surfaces and the vegetation distributed within them, and are calculated using Equation (9).

[0048] (8)

[0049] (9)

[0050] in, According to the proportion of vegetation composition, hour, ,otherwise ; The vegetation coverage was estimated using a pixel binary model based on NDVI; 、 and They represent the surface emissivity of vegetation, building, and bare soil pure pixels, with values ​​of 0.970, 0.986, and 0.972, respectively; 、 、 represent the temperature ratios of vegetation, buildings and bare soil respectively, and are calculated by formula (10):

[0051] (10)

[0052] The calculated high spatial resolution thermal infrared channel emissivity data is input into formula (7), and the corresponding low spatial resolution thermal infrared channel emissivity at high spatial resolution can be calculated: .

[0053] 4) Surface temperature estimation based on high spatial resolution thermal radiation transfer equation

[0054] At high spatial resolution, the thermal infrared band radiation transfer equation can be expressed as:

[0055] (11)

[0056] Substituting the radiance, atmospheric parameters, and surface emissivity at high spatial resolution calculated by equations (3), (5), and (7) into equation (11), we can calculate , and then use the deformed form of Planck function to solve the high spatial resolution surface temperature , that is, the downscaling processing of low spatial resolution surface temperature is achieved.

[0057] (12)

[0058] in, and is the radiation constant.

[0059] The following describes the downscaling of MODIS surface temperature as an example. First, based on the random forest machine learning method, the radiance of the MODIS thermal infrared band with a resolution of 1km is downscaled. Then, through the low spatial resolution thermal infrared radiation transfer model, the least squares method is used to solve the atmospheric parameters, and then the Landsat TIRS surface emissivity is calculated and converted into the surface emissivity corresponding to the MODIS channel. Finally, the various parameters with a resolution of 30m are substituted into the high spatial resolution thermal infrared radiation transfer model, achieving the goal of downscaling the 1km resolution MODIS surface temperature to 30m. It is compared with the Landsat TIRS inversion surface temperature synchronized with the MODIS imaging time and the traditional random forest statistical downscaling method to illustrate the advantages and effects of the method of the present invention.

[0060] This invention has been applied in Hefei, Anhui Province. The results show that the MODSI surface temperature downscaling effect based on the thermal infrared radiation transfer model is better than the traditional statistical downscaling method. The spatial distribution of the surface temperature obtained after downscaling is more consistent with the TIRS inverted surface temperature at the same time. Figure 1 shown.

[0061] 1) Test Area Selection: Hefei, Anhui Province, was selected as the test area, located between 116.68° and 117.96° east longitude and 30.95° and 32.54° north latitude. Hefei is located in the mid-latitude zone and has a subtropical monsoon humid climate with moderate annual precipitation and relatively flat terrain with an average elevation of approximately 25 meters. The main surface types in the test area are towns, vegetation, water bodies, and bare soil. MODIS and Landsat data for October 17, 2023, were selected. The surface temperature retrieved by Landsat TIRS was used to evaluate the MODIS surface temperature downscaling results. Landsat OLI data for October 9, 2023, were selected as MODIS data for the adjacent date and used to calculate surface emissivity.

[0062] 2) Data Download and Preprocessing: The data required for downloading include Terra MODIS data, DEMs, and spectral emissivity data for typical land features. The MODIS data required are the surface temperature product, the surface emissivity product for channel 31, and the radiance data, which can be downloaded from the Google Earth Engine website (https: / / earthengine.google.com / ) and LAADSDAAC (https: / / ladsweb.modaps.eosdis.nasa.gov / ), respectively. Landsat and DEM data can be downloaded from the USGS website (https: / / earthexplorer.usgs.gov / ). Spectral emissivity data for typical land features can be downloaded from https: / / www.icess.ucsb.edu / modis / EMIS / html / em.html. All downloaded data will be preprocessed, including resampling, image cropping, radiometric calibration, atmospheric correction, coordinate conversion, and remote sensing index calculation, to obtain the data necessary for surface temperature downscaling.

[0063] 3) Low spatial resolution radiance downscaling: At a spatial resolution of 1 km, the radiance of the 31st channel of MODIS in the experimental area is used as the dependent variable, and the auxiliary factors NDVI, NDBI, NDMI, MNDWI, and DEM are used as independent variables. The samples are randomly divided into a training set and a test set with a division ratio of 80% and 20%, respectively. The random forest nonlinear relationship model between radiance and influencing factors is obtained through training (Equation (1)). The auxiliary factors at a resolution of 30 m are then input into the trained random model (Equation (2)) to obtain the initial radiance value at a resolution of 30 m. Finally, the initial radiance value after downscaling is upscaled to 1 km, and the residual with the original 1 km radiance is calculated. The residual is resampled to 30 m resolution and processed using the Gaussian filter smoothing method to take into account both energy balance and visual effects. The smoothed residual is then added to the initial radiance downscaling value (Equation (3)) to obtain the corrected radiance downscaling result at a resolution of 30 m.

[0064] 4) Calculation of atmospheric parameters based on the low-resolution thermal radiation transfer model: At a resolution of 1 km, the known MODIS surface temperature, the surface emissivity of the 31st channel, and the radiation brightness received by the sensor are input into formula (4). The atmospheric upward radiation, atmospheric downward radiation, and atmospheric transmittance of the 31st channel are calculated by the least squares method. Assuming that the atmosphere in the test area is horizontally uniform, at a resolution of 30 m, the atmospheric upward radiation, atmospheric downward radiation, and atmospheric transmittance of the 31st channel are the same as the atmospheric parameters at a resolution of 1 km. The calculated atmospheric upward radiation, atmospheric downward radiation, and atmospheric transmittance are respectively 、 and 0.93.

[0065] 5) High spatial resolution surface emissivity estimation: Download the common typical ground object emissivity spectral data from the USCB website, calculate the wide-band ground object emissivity corresponding to the MODIS channel 31 and the TIRS channel 10 through the spectral response function of the two channels, and establish a linear fitting formula between the two (Equation (7)). The equation coefficients are obtained by fitting, and the linear equation is obtained. Download Landsat OLI data with a date close to that of MODIS, use the mixed pixel decomposition method to determine the surface emissivity corresponding to the 10th channel of TIRS (Equations (8), (9), and (10)), and convert it to the surface emissivity corresponding to the 31st channel of MODIS using a linear formula.

[0066] 6) Surface temperature estimation based on high spatial resolution thermal radiation transfer equation: Substitute the 30m resolution radiation brightness, surface emissivity, atmospheric upward radiation, atmospheric downward radiation, atmospheric transmittance and other parameters calculated in the above process into equation (11) to calculate , and then use Equation (12) to solve the high spatial resolution surface temperature .

[0067] 7) Traditional random forest surface temperature downscaling: At a spatial resolution of 1 km, the MODIS surface temperature of the experimental area was used as the dependent variable, and the auxiliary factors NDVI, NDBI, NDMI, MNDWI, and DEM were used as independent variables. The samples were randomly divided into a training set and a test set with a division ratio of 80% and 20%, respectively. The random forest nonlinear relationship model between the surface temperature and the influencing factors was obtained through training. The auxiliary factors at a resolution of 30 m were then input and residual correction was performed to obtain the downscaled surface temperature at a resolution of 30 m.

[0068] 8) Analysis of experimental results: Figure 2 This is a comparison chart of surface temperature products in Hefei, the test area, at 10:40 on October 17, 2023. The left picture is the downscaling result of the present invention, the middle picture is the downscaling result of the traditional random forest surface temperature, and the right picture is the surface temperature inverted by Landsat TIRS. Figure 2 , both downscaling methods achieved the downscaling of MODIS surface temperature and significantly improved the spatial texture details of the original surface temperature image; the surface temperature expressed by the two methods is generally consistent with the Landsat TIRS surface temperature in terms of spatial distribution. However, when carefully examining the details of the spatial distribution of surface temperature, such as urban areas with high spatial heterogeneity, the results of the present invention are significantly better than the random forest downscaling results, and are more consistent with the Landsat surface temperature. The traditional random forest surface temperature downscaling method has the erroneous results of overestimation of low temperature areas and underestimation of high temperature areas. In terms of quantitative analysis, compared with Landsat surface temperature, the root mean square error RMSE, mean absolute error MAE, and Pearson correlation coefficient of the method of the present invention are significantly better than those of Landsat surface temperature. They are 1.38K, 1.03K, and 0.89 respectively. The RMSE, MAE, Based on the above qualitative and quantitative evaluation results, the downscaling method proposed in this paper can achieve the goal of downscaling the 1km resolution MODIS surface temperature to 30m with high accuracy, and its results are significantly better than the most commonly used, traditional random forest surface temperature downscaling results.

[0069] 9) Experimental Conclusion: The experimental results of this paper fully demonstrate that the surface temperature downscaling method based on the thermal infrared radiation transfer model can downscale the MODIS surface temperature with a resolution of 1 km to 30 m with high accuracy. This is an effective development of the surface temperature downscaling method and can provide technical support and high-temporal and spatial resolution surface temperature data support for related fields such as urban thermal environment research. It has important scientific significance and application value.

[0070] Based on the same inventive concept, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the steps of the aforementioned surface temperature downscaling method based on the thermal infrared radiation transfer model are implemented.

[0071] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the aforementioned surface temperature downscaling method based on the thermal infrared radiation transfer model.

[0072] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] The present invention is described with reference to flowcharts of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process in the flowcharts, as well as combinations of processes in the flowcharts, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts. Figure 1 A device that specifies functions in a process or multiple processes.

[0074] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0075] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0076] The above embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A surface temperature downscaling method based on a thermal infrared radiation transfer model, characterized in that: The steps include: Step 1: For the low spatial resolution of the surface temperature to be downscaled The thermal infrared band remote sensing image under the thermal infrared channel is obtained The radiance and radiance influencing factor data are used to train the random forest downscaling model. The trained random forest downscaling model is used to convert the low spatial resolution Downscaling of the radiance to a high spatial resolution Then, the corrected high spatial resolution is obtained. Radiance under Step 2: Use the thermal infrared band radiation transfer model to represent the low spatial resolution The thermal infrared channel radiance of the thermal infrared band remote sensing image is calculated, and the low spatial resolution atmospheric parameters in the thermal infrared band radiation transfer model are calculated using the least squares method. Atmospheric parameters at high spatial resolution Atmospheric parameters under Step 3: Obtain the surface emissivity spectral curve data of different types of ground objects and fit the low spatial resolution Thermal infrared channel and with The closest high spatial resolution Thermal infrared channel The wide-band surface emissivity of the ground is determined by the hybrid pixel decomposition method, and a linear relationship model is established between the two. The corresponding surface emissivity is converted to high spatial resolution Lower thermal infrared channel The corresponding surface emissivity; Step 4: High spatial resolution The corrected radiance, atmospheric parameters and surface emissivity are substituted into the high spatial resolution Thermal infrared band radiation transfer model under high spatial resolution The surface temperature under the condition of γ is reduced to achieve surface temperature downscaling.

2. The surface temperature downscaling method based on the thermal infrared radiation transfer model according to claim 1 is characterized in that: The specific process of step 1 is as follows: Step 1.1, determine the influencing factors of the thermal infrared channel radiance of the thermal infrared remote sensing image, including the normalized difference vegetation index, normalized difference building index, normalized difference moisture index, normalized difference water index and elevation; Step 1.2: For the low spatial resolution of the surface temperature to be downscaled The thermal infrared band remote sensing image under the condition of the infrared band is used to obtain the surface reflectance data from the visible light-near infrared band remote sensing image from the same sensor as the thermal infrared band remote sensing image. The low spatial resolution is calculated based on the surface reflectance data. Normalized difference vegetation index, normalized difference building index, normalized difference moisture index and normalized difference water index under; Step 1.3, obtain high spatial resolution Download the elevation data corresponding to the thermal infrared band remote sensing image and resample it to the low spatial resolution corresponding to the thermal infrared band remote sensing image , resulting in low spatial resolution the elevation below; Step 1.4, obtain low spatial resolution The thermal infrared channel radiation brightness data under , and based on steps 1.2 and 1.3, train the random forest downscaling model, namely: ， in, is the thermal infrared channel radiance at low spatial resolution, represents the random forest downscaling model, and Low spatial resolution Normalized difference vegetation index, normalized difference building index, normalized difference moisture index, normalized difference water index and elevation under; Step 1.5: Obtain the high spatial resolution closest to the imaging date of the thermal infrared remote sensing image. Visible-near infrared remote sensing images based on high spatial resolution Computing high spatial resolution of remote sensing images in the visible-near infrared band The radiation brightness influencing factor under is input into the trained random forest downscaling model to obtain the initial value of high spatial resolution radiation brightness downscaling, namely: ， in, is the initial value of high spatial resolution radiance downscaling, and High spatial resolution Normalized difference vegetation index, normalized difference building index, normalized difference moisture index, normalized difference water index and elevation under; Step 1.6: Downscale the initial value of the high spatial resolution radiance Resample to lower spatial resolution , and obtain the resampled low spatial resolution radiance ,calculate and The residual between ,Will Resampling to high spatial resolution , and smoothed with Gaussian filtering to obtain the resampled residual with high spatial resolution ,Will Add to The corrected high spatial resolution radiance is obtained .

3. The surface temperature downscaling method based on the thermal infrared radiation transfer model according to claim 1 is characterized in that: The specific process of step 2 is as follows: At low spatial resolution In the following example, the thermal infrared channel radiation brightness of the thermal infrared band remote sensing image is expressed by the thermal infrared band radiation transfer model as follows: ， in, is the thermal infrared channel radiance at low spatial resolution, is the low spatial resolution surface emissivity, is the true surface temperature at low spatial resolution, Temperature The blackbody radiation corresponding to the surface of the earth, and They are respectively the atmospheric downward radiation, atmospheric upward radiation and atmospheric transmittance corresponding to low spatial resolution; Low spatial resolution The thermal infrared channel radiance, surface emissivity, surface true temperature and blackbody radiation under the above thermal infrared band radiation transfer model are substituted, and the low spatial resolution is estimated by the least squares method. Atmospheric parameters under the atmospheric conditions include atmospheric downward radiation, atmospheric upward radiation and atmospheric transmittance; The atmospheric distribution level in the target area is set to be uniform, that is, the atmospheric parameters do not change with the spatial scale, and the low spatial resolution High spatial resolution The atmospheric parameters are equal.

4. The surface temperature downscaling method based on the thermal infrared radiation transfer model according to claim 1 is characterized in that: The specific process of step 3 is as follows: Step 3.1, obtain the surface emissivity spectral curve data of different types of ground objects, respectively using low spatial resolution Thermal infrared channel Spectral response function and The closest high spatial resolution Thermal infrared channel The spectral response function is calculated by convolution operation to obtain the spectral response of each object in the channel. and Broadband surface emissivity on ; Step 3.2, low spatial resolution thermal infrared channel Broadband surface emissivity on As the dependent variable, establish With thermal infrared channel Broadband surface emissivity on The linear relationship model between: ， in, All are linear relationship model coefficients; Step 3.3: Use the hybrid pixel decomposition method to use the high spatial resolution closest to the imaging date of the thermal infrared remote sensing image described in step 1. Calculate high spatial resolution of visible-near infrared remote sensing images Lower thermal infrared channel The corresponding surface emissivity is converted to high spatial resolution using the linear relationship model established in step 3.2 Lower thermal infrared channel The corresponding surface emissivity.

5. The surface temperature downscaling method based on the thermal infrared radiation transfer model according to claim 4 is characterized in that: In step 3.3, the hybrid pixel decomposition method is used to calculate the thermal infrared channel The specific process of the corresponding surface emissivity is as follows: The high spatial resolution image closest to the imaging date of the thermal infrared remote sensing image in step 1 The surface pixels corresponding to the visible-near infrared band remote sensing images under the CMOS are divided into four categories: urban surface, natural surface, bare soil and water body pixels. Among them, the surface emissivity of water body pixels is set to 0.99; The surface emissivity of urban surface pixels and natural surface pixels is calculated as follows: ， ， in, and Represent the surface emissivity of urban surface pixels and natural surface pixels respectively, is the vegetation coverage, 、 and Represent the surface emissivity of vegetation, building and bare soil pixels respectively, According to the proportion of vegetation composition, hour, ,otherwise ; 、 and Represents the temperature ratios of vegetation, buildings and bare soil respectively: 。 6. The surface temperature downscaling method based on the thermal infrared radiation transfer model according to claim 1 is characterized in that: The specific process of step 4 is as follows: The high spatial resolution obtained from step 1 to step 3 The corrected radiance, atmospheric parameters and surface emissivity are substituted into the high spatial resolution Thermal infrared band radiation transfer model: ， in, represents the corrected high spatial resolution radiance, and They are atmospheric downward radiation, atmospheric upward radiation, atmospheric transmittance and surface emissivity corresponding to high spatial resolution; Calculated high spatial resolution Blackbody radiation under , high spatial resolution is calculated by the deformation form of Planck function Surface temperature below , to achieve downscaling of low spatial resolution surface temperature; the deformation form of Planck function is as follows: ， in, and is the radiation constant.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the surface temperature downscaling method based on the thermal infrared radiation transfer model according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the surface temperature downscaling method based on the thermal infrared radiation transfer model according to any one of claims 1 to 6 are implemented.

Images