Remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree

Abstract

The present invention discloses a remote sensing surface temperature spatial downscaling method based on a CatBoost symmetric decision tree. The method comprises the following steps: obtaining low-resolution passive microwave remote sensing surface temperature data and obtaining high-resolution remote sensing visible light and near-infrared band images; calculating a high-resolution normalized water index and a normalized difference mountain vegetation index using the high-resolution visible light and near-infrared band images; calculating high-resolution CIE color coordinates using the high-resolution remote sensing visible light band images; obtaining DEM data, performing spatial aggregation, and obtaining high-resolution DEM data; performing spatial downscaling on the low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscaled data; constructing and training a correction model based on a CatBoost symmetric decision tree algorithm, and correcting the passive microwave remote sensing surface temperature downscaled data using the correction model to obtain high-resolution remote sensing surface temperature data.

Description

Technical Field

[0001] The present invention relates to the field of downscaling technology, and in particular to a remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree regression modeling. Background Art

[0002] The traditional spatial downscaling process of passive microwave remote sensing surface temperature data generally only considers auxiliary variables such as NDVI (Normalized Difference Vegetation Index), DEM (Digital Elevation Model), and NDWI (Normalized Difference Water Index).

[0003] However, in areas with complex terrain, the reflectivity of mountain vegetation on sunny and shady slopes differs significantly, and conventional NDVI cannot eliminate or suppress the impact of terrain. The Normalized Difference Mountain Vegetation Index (NDMVI) can effectively eliminate or suppress the impact of terrain, has a wider dynamic range than NDVI, and has stronger remote sensing identification capabilities for ground objects.

[0004] In addition, the surface temperature is closely related to the color physical characteristics of the remote sensing Red, Green, and Blue bands. The remote sensing RGB color physical characteristics can be expressed by the CIEx and CIEy color coordinates of the CIE color space; the traditional passive microwave remote sensing surface temperature spatial downscaling method does not consider the influence of the normalized difference mountain vegetation index and RGB color physical characteristics, which will cause accuracy loss in the prediction results. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects and shortcomings of the existing technology and provide a remote sensing surface temperature spatial downscaling method based on a CatBoost symmetric decision tree. The method combines the normalized water index, the normalized difference mountain vegetation index, the CIEx color coordinates, the CIEy color coordinates and DEM data, and performs regression modeling based on the CatBoost symmetric decision tree to obtain accurate high-resolution remote sensing surface temperature data in complex terrain areas.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree includes the following steps:

[0008] S1: Acquire low-resolution passive microwave remote sensing surface temperature data, obtain high-resolution remote sensing visible light and near-infrared band images, and process the data and images;

[0009] S2: Using high-resolution remote sensing visible light and near-infrared band images, high-resolution normalized water index and normalized difference mountain vegetation index are calculated respectively;

[0010] Using high-resolution remote sensing visible light band images and combining them with CIE color space theory, high-resolution CIEx color coordinates and CIEy color coordinates are calculated respectively;

[0011] S3: Obtain high-precision DEM data with a set resolution, perform spatial aggregation by weighted averaging, and obtain high-resolution and high-precision DEM data;

[0012] S4: Use the algorithm to spatially downscale the low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscaled data;

[0013] S5: Based on the CatBoost symmetric decision tree algorithm, a correction model for passive microwave remote sensing surface temperature downscaling data is constructed. Hyperparameters of the correction model are tuned, cross-validation is performed, and the optimal parameters are selected.

[0014] S6: Retrain the correction model based on the optimal parameters, and use the correction model to correct the passive microwave remote sensing surface temperature downscaling data to obtain high-resolution remote sensing surface temperature data.

[0015] Furthermore, in step S1, low-resolution passive microwave remote sensing surface temperature data is obtained, high-resolution remote sensing visible light and near-infrared band images are obtained, and the data and images are processed, specifically as follows:

[0016] Obtain low-resolution passive microwave remote sensing surface temperature data in time series, and use ordinary Kriging interpolation function to perform spatial interpolation on missing values ​​to obtain low-resolution passive microwave remote sensing surface temperature data in time and space;

[0017] Obtain high-resolution visible light and near-infrared band images at the corresponding time point, perform atmospheric correction and radiation correction preprocessing; and extract high-resolution red band, green band, blue band, near-infrared band and short-wave infrared band image reflectance information.

[0018] Furthermore, in step S2, high-resolution remote sensing visible light and near-infrared band images are used to calculate high-resolution normalized water index and normalized difference mountain vegetation index, respectively, as follows:

[0019] The high-resolution normalized water index is calculated using the reflectance information of the high-resolution green band and near-infrared band images. :

[0020] ;

[0021] Where, Indicates the green band image reflectance; Indicates the near-infrared band image reflectance;

[0022] The high-resolution normalized difference mountain vegetation index is calculated using the reflectance information of high-resolution near-infrared and red band images. :

[0023] ;

[0024] Where, Indicates the red band image reflectance; Indicates the minimum value of the red band image reflectance; Indicates the minimum value of the near-infrared band image reflectance.

[0025] Furthermore, in step S2, high-resolution remote sensing visible light band images are used in combination with CIE color space theory to calculate high-resolution CIEx color coordinates and CIEy color coordinates, specifically:

[0026] Using the high-resolution red, green, and blue band image reflectance information, high-resolution CIEx color coordinates and CIEy color coordinates are calculated respectively:

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] Where, Indicates the red band image reflectance; Indicates the green band image reflectance; Indicates the reflectance of the blue band image; 、 、 、 、 、 、 、 、 Represent different coefficients respectively.

[0033] Furthermore, in step S3, high-precision DEM data of a set resolution is obtained, and spatial aggregation is performed by weighted averaging to obtain high-resolution high-precision DEM data, specifically:

[0034] The high-precision DEM data uses the Copernicus digital elevation model dataset provided by the European Space Agency with a spatial resolution of 30m or 90m. Spatial aggregation is performed through weighted averaging to obtain high-resolution and high-precision DEM data.

[0035] Furthermore, in step S4, the low-resolution passive microwave remote sensing surface temperature data is spatially downscaled using an algorithm to obtain high-resolution passive microwave remote sensing surface temperature downscaled data, specifically:

[0036] The surface-to-point Kriging interpolation algorithm is used to spatially downscale the spatiotemporally continuous low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscale data. .

[0037] Furthermore, in step S5, based on the CatBoost symmetric decision tree algorithm, a correction model for the passive microwave remote sensing surface temperature downscaling data is constructed, hyperparameters of the correction model are tuned, and cross-validation is performed to select the optimal parameters, specifically:

[0038] Using high-resolution passive microwave remote sensing surface temperature downscaling data as the dependent variable, and high-resolution normalized water index, normalized difference mountain vegetation index, CIEx color coordinates, CIEy color coordinates and DEM data at the corresponding time points as independent variables, a correction model for passive microwave remote sensing surface temperature downscaling data was constructed using the CatBoost symmetric decision tree algorithm, and the results were:

[0039] ;

[0040] Where, express t Passive microwave remote sensing surface temperature downscaled data at the moment; express t Normalized water index corresponding to the moment; express t Normalized difference mountain vegetation index corresponding to the time; express t CIEx color coordinates at the moment; express t CIEy color coordinates at the moment;

[0041] By constructing a correction model for passive microwave remote sensing surface temperature downscaling data, the error correction of high-resolution passive microwave remote sensing surface temperature downscaling data obtained by spatially downscaling low-resolution passive microwave remote sensing surface temperature data using the surface-to-point Kriging interpolation algorithm is performed.

[0042] Furthermore, in step S5, cross validation is performed, specifically:

[0043] The cross-validation method provided by the CatBoost symmetric decision tree algorithm was used to perform a five-fold cross-validation on each set of hyperparameter combinations. In each cross-validation, the dataset was divided into five parts, four of which were used for training and one for validation, and this was repeated five times. The evaluation metric was specified as accuracy, and the best model was saved when the best performance was achieved. A fixed random seed was used to ensure consistent results. For all 720 parameter combinations, the average loss value under cross-validation was recorded.

[0044] Furthermore, in step S5, the optimal parameters are selected, specifically:

[0045] Based on the average loss value of five-fold cross-validation, the performance of all parameter combinations was compared, and the parameter combination that produced the lowest average loss value was selected as the optimal hyperparameter configuration; the correction model was retrained using the optimal parameter combination for the correction task of passive microwave remote sensing surface temperature downscaling data.

[0046] Furthermore, Python language is used to implement data processing, model construction and model training.

[0047] Compared with the existing technology, the present invention combines the CIE color physics characteristics with the CatBoost symmetric decision tree algorithm to perform spatial downscaling on low-resolution passive microwave remote sensing surface temperature data, fully considering the impact of complex terrain mountain vegetation, the color space characteristic information of the remote sensing RGB band on the surface temperature, and also considering the impact of water bodies and altitude on the surface temperature.

[0048] The present invention combines the Normalized Difference Mountain Vegetation Index (NDMVI), CIEx color coordinates, CIEy color coordinates, and DEM data, and performs regression modeling based on the CatBoost symmetric decision tree to obtain accurate high-resolution remote sensing surface temperature data in complex terrain areas, which has important theoretical and practical significance and promotion and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flowchart of the remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree. DETAILED DESCRIPTION

[0050] The remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0051] See also Figure 1 The present invention discloses a remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree, comprising the following steps:

[0052] S1: Acquire low-resolution passive microwave remote sensing surface temperature data, obtain high-resolution remote sensing visible light and near-infrared band images, and process the data and images;

[0053] S2: Using high-resolution remote sensing visible light and near-infrared band images, high-resolution normalized water index and normalized difference mountain vegetation index are calculated respectively;

[0054] Using high-resolution remote sensing visible light band images and combining them with CIE color space theory, high-resolution CIEx color coordinates and CIEy color coordinates are calculated respectively;

[0055] S3: Obtain high-precision DEM data with a set resolution, perform spatial aggregation by weighted averaging, and obtain high-resolution and high-precision DEM data;

[0056] S4: Use the algorithm to spatially downscale the low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscaled data;

[0057] S5: Based on the CatBoost symmetric decision tree algorithm, a correction model for passive microwave remote sensing surface temperature downscaling data is constructed. Hyperparameters of the correction model are tuned, cross-validation is performed, and the optimal parameters are selected.

[0058] S6: Retrain the correction model based on the optimal parameters, and use the correction model to correct the passive microwave remote sensing surface temperature downscaling data to obtain high-resolution remote sensing surface temperature data.

[0059] Step S1: Obtain low-resolution passive microwave remote sensing surface temperature data, obtain high-resolution remote sensing visible light and near-infrared band images, and process the data and images.

[0060] Passive microwave remote sensing surface temperature data with a temporal and spatial resolution of 25 km were obtained from the Microwave Radiation Imager (MWRI) on China's Fengyun-3C satellite. Due to the influence of striping, the passive microwave remote sensing surface temperature data had missing values. Ordinary Kriging interpolation function was used to spatially interpolate the missing values ​​to obtain passive microwave remote sensing surface temperature data with a temporal and spatial resolution of 25 km.

[0061] Visible and near-infrared images provided by the Visible Infrared Scanning Radiometer (VIRR) on China's Fengyun-3C satellite at the corresponding time point with a spatial resolution of 1 km were obtained and preprocessed with atmospheric correction and radiometric correction. The reflectance information of the red band (0.58-0.68μm), green band (0.53-0.58μm), blue band (0.48-0.53μm), near-infrared band (NIR, 0.84-0.89μm), and short-wave infrared band (SWIR, 1.58-1.64μm) images with a spatial resolution of 1km were extracted.

[0062] In step S2, high-resolution remote sensing visible light and near-infrared band images are used to calculate high-resolution normalized water index and normalized difference mountain vegetation index respectively.

[0063] The normalized water index with a spatial resolution of 1 km was calculated using the green band and near-infrared band image reflectance information of the visible infrared scanning radiometer of China Fengyun-3C satellite with a spatial resolution of 1 km. :

[0064] ;

[0065] Where, represents the green band image reflectivity of the visible infrared scanning radiometer on China's Fengyun-3C satellite; Represents the near-infrared band image reflectivity of the visible infrared scanning radiometer of China's Fengyun-3C satellite.

[0066] The normalized difference mountain vegetation index with a spatial resolution of 1 km was calculated using the near-infrared and red band image reflectance information of the visible infrared scanning radiometer of China's Fengyun-3C satellite with a spatial resolution of 1 km. :

[0067] ;

[0068] Where, represents the red band image reflectivity of the visible infrared scanning radiometer on China's Fengyun-3C satellite; It represents the minimum value of the red band image reflectivity of the visible infrared scanning radiometer of China's Fengyun-3C satellite; Indicates the minimum value of the near-infrared band image reflectivity of the visible infrared scanning radiometer of China's Fengyun-3C satellite.

[0069] In step S2, high-resolution remote sensing visible light band images are used in combination with the International Commission on Illumination (CIE) color space theory to calculate high-resolution CIEx color coordinates and CIEy color coordinates.

[0070] The CIEx color coordinates and CIEy color coordinates with a spatial resolution of 1 km are calculated using the red, green, and blue band image reflectance information of the visible infrared scanning radiometer of China's Fengyun-3C satellite with a spatial resolution of 1 km:

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] Where, represents the red band image reflectivity of the visible infrared scanning radiometer on China's Fengyun-3C satellite; represents the green band image reflectivity of the visible infrared scanning radiometer on China's Fengyun-3C satellite; represents the blue band image reflectivity of the visible infrared scanning radiometer on China's Fengyun-3C satellite; 、 、 、 、 、 、 、 、 Represent different coefficients respectively.

[0077] Step S3: Obtain high-precision DEM data with a set resolution, perform spatial aggregation by weighted averaging, and obtain high-resolution high-precision DEM data.

[0078] The high-precision DEM data uses the Copernicus digital elevation model dataset with a spatial resolution of 30m provided by the European Space Agency. Spatial aggregation is performed through weighted averaging to obtain high-precision DEM data with a spatial resolution of 1km.

[0079] Step S4: spatially downscaling the low-resolution passive microwave remote sensing surface temperature data using an algorithm to obtain high-resolution passive microwave remote sensing surface temperature downscaled data.

[0080] The passive microwave remote sensing surface temperature data with a spatial resolution of 1 km was obtained by spatially downscaling the passive microwave remote sensing surface temperature data with a spatial resolution of 25 km from the microwave imager of China's Fengyun-3C satellite using the surface-to-point Kriging interpolation algorithm. .

[0081] In step S5, based on the CatBoost symmetric decision tree algorithm, a correction model for the passive microwave remote sensing surface temperature downscaling data is constructed, hyperparameters of the correction model are tuned, and cross-validation is performed to select the optimal parameters.

[0082] The passive microwave remote sensing surface temperature downscaling data with a spatial resolution of 1 km was used as the dependent variable, and the normalized water index, normalized difference mountain vegetation index, CIEx color coordinates, CIEy color coordinates and DEM data with a spatial resolution of 1 km at the corresponding time point were used as independent variables. The CatBoost symmetric decision tree algorithm was used to construct a correction model for the passive microwave remote sensing surface temperature downscaling data, and the results were:

[0083] ;

[0084] Where, express t Passive microwave remote sensing surface temperature downscaled data at the moment; express t Normalized water index corresponding to the moment; express t Normalized difference mountain vegetation index corresponding to the time; express t CIEx color coordinates at the moment; express t CIEy color coordinates at the moment;

[0085] By constructing a correction model for passive microwave remote sensing surface temperature downscaling data, the passive microwave remote sensing surface temperature downscaling data with a spatial resolution of 1 km obtained by spatially downscaling the passive microwave remote sensing surface temperature data with a spatial resolution of 25 km using the surface-to-point Kriging interpolation algorithm are corrected.

[0086] Determine the hyperparameter search space for the modified model. The parameters are set as follows:

[0087] Iterations: 500, 600, 700, 800

[0088] depth: 4, 5, 6, 8

[0089] loss_function: Logloss, CrossEntropy, RMSE

[0090] l2_leaf_reg: np.logspace(-20, -19, 3)

[0091] eval_metric: Accuracy

[0092] use_best_model: True

[0093] random_seed: 42

[0094] learning_rate: 0.01, 0.03, 0.05, 0.1, 0.2

[0095] metric_period: 500

[0096] Generate all possible combinations of the hyperparameters iterations (500-800), depth (4-8), loss_function (Logloss / CrossEntropy / RMSE), l2_leaf_reg (three values ​​generated in logarithmic space: such as 1e-20, 3e-20, 1e-19), and learning_rate (0.01-0.2), for a total of approximately 4×4×3×3×5 = 720 parameter configurations.

[0097] In step S5, cross validation is performed.

[0098] Using the cross-validation method provided by the CatBoost symmetric decision tree algorithm, a five-fold cross-validation was performed on each set of hyperparameter combinations. In each cross-validation, the dataset was divided into five parts, four of which were used for training and one for validation, and this was repeated five times. The evaluation metric was specified as accuracy (eval_metric was set to Accuracy), and the best model was saved when the best performance was achieved (use_best_model was set to True). The random seed was fixed to 42 to ensure consistent results. The average loss value under cross-validation was recorded for all 720 parameter combinations.

[0099] In step S5, the optimal parameters are selected.

[0100] The performance of all parameter combinations was compared based on the average loss from five-fold cross-validation. The parameter combination that produced the lowest average loss was selected as the optimal hyperparameter configuration, which maximized the model's generalization ability on unseen data. The model was then retrained using the optimal parameter combination and applied to the correction of passive microwave remote sensing surface temperature downscaling data.

[0101] Through the above steps, a passive microwave remote sensing surface temperature downscaling data correction model based on the CatBoost symmetric decision tree algorithm can be obtained.

[0102] Combined with high-resolution (e.g. spatial resolution of 1km) remote sensing multispectral band reflectance and DEM data, the normalized water index is calculated using the formula , Normalized Difference Mountain Vegetation Index , CIEx color coordinates and CIEy color coordinates, and then spatially downscale the low-resolution (such as spatial resolution of 25km) passive microwave remote sensing surface temperature data to obtain high-resolution (such as spatial resolution of 1km) passive microwave remote sensing surface temperature downscaled data, and construct a correction model for passive microwave remote sensing surface temperature downscaled data based on the CatBoost symmetric decision tree algorithm, and correct the high-resolution (such as spatial resolution of 1km) remote sensing surface temperature downscaled data to obtain high-resolution (such as spatial resolution of 1km) remote sensing surface temperature data.

[0103] The present invention proposes a remote sensing surface temperature spatial downscaling method based on CatBoost symmetric decision tree, which mainly uses Python language to implement data processing, model construction and model training processes.

[0104] In summary, the present invention combines the CIE color physics characteristics with the CatBoost symmetric decision tree algorithm to perform spatial downscaling on low-resolution passive microwave remote sensing surface temperature data. It fully considers the influence of complex terrain mountain vegetation, the color space characteristic information of remote sensing RGB bands on the surface temperature, and the influence of water bodies and altitude on the surface temperature.

[0105] The present invention combines the Normalized Difference Mountain Vegetation Index (NDMVI), CIEx color coordinates, CIEy color coordinates, and DEM data, and performs regression modeling based on the CatBoost symmetric decision tree to obtain accurate high-resolution remote sensing surface temperature data in complex terrain areas, which has important theoretical and practical significance and promotion and application value.

[0106] The above description is a detailed description of the preferred embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. Any equivalent changes or modifications made under the technical spirit disclosed by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree, characterized in that: The following steps are involved: S1: Acquire low-resolution passive microwave remote sensing surface temperature data, obtain high-resolution remote sensing visible light and near-infrared band images, and process the data and images; S2: Using high-resolution remote sensing visible light and near-infrared band images, high-resolution normalized water index and normalized difference mountain vegetation index are calculated respectively; Using high-resolution remote sensing visible light band images and combining them with CIE color space theory, high-resolution CIEx color coordinates and CIEy color coordinates are calculated respectively; S3: Obtain high-precision DEM data with a set resolution, perform spatial aggregation by weighted averaging, and obtain high-resolution and high-precision DEM data; S4: Use the algorithm to spatially downscale the low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscaled data; S5: Based on the CatBoost symmetric decision tree algorithm, a correction model for passive microwave remote sensing surface temperature downscaling data is constructed. Hyperparameters of the correction model are tuned, cross-validation is performed, and the optimal parameters are selected. S6: Retrain the correction model based on the optimal parameters, and use the correction model to correct the passive microwave remote sensing surface temperature downscaling data to obtain high-resolution remote sensing surface temperature data.

2. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 1, characterized in that: Step S1: Obtain low-resolution passive microwave remote sensing surface temperature data, obtain high-resolution remote sensing visible light and near-infrared band images, and process the data and images, specifically as follows: Obtain low-resolution passive microwave remote sensing surface temperature data in time series, and use ordinary Kriging interpolation function to perform spatial interpolation on missing values ​​to obtain low-resolution passive microwave remote sensing surface temperature data in time and space; Obtain high-resolution visible light and near-infrared band images at the corresponding time point, perform atmospheric correction and radiation correction preprocessing; and extract high-resolution red band, green band, blue band, near-infrared band and short-wave infrared band image reflectance information.

3. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 2, characterized in that: In step S2, high-resolution remote sensing visible light and near-infrared band images are used to calculate high-resolution normalized water index and normalized difference mountain vegetation index, respectively. Specifically, The high-resolution normalized water index is calculated using the reflectance information of the high-resolution green band and near-infrared band images. : ； Where, Indicates the green band image reflectance; Indicates the near-infrared band image reflectance; The high-resolution normalized difference mountain vegetation index is calculated using the reflectance information of high-resolution near-infrared and red band images. : ； Where, Indicates the red band image reflectance; Indicates the minimum value of the red band image reflectance; Indicates the minimum value of the near-infrared band image reflectance.

4. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 2, characterized in that: In step S2, high-resolution remote sensing visible light band images are used in combination with CIE color space theory to calculate high-resolution CIEx color coordinates and CIEy color coordinates, specifically: Using the high-resolution red, green, and blue band image reflectance information, high-resolution CIEx color coordinates and CIEy color coordinates are calculated respectively: ； ； ； ； ； Where, Indicates the red band image reflectance; Indicates the green band image reflectance; Indicates the reflectance of the blue band image; 、 、 、 、 、 、 、 、 Represent different coefficients respectively.

5. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 1, characterized in that: Step S3: Obtain high-precision DEM data of a set resolution, perform spatial aggregation by weighted averaging, and obtain high-resolution high-precision DEM data. Specifically: The high-precision DEM data uses the Copernicus digital elevation model dataset provided by the European Space Agency with a spatial resolution of 30m or 90m. Spatial aggregation is performed through weighted averaging to obtain high-resolution and high-precision DEM data.

6. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 1, characterized in that: Step S4: spatially downscaling the low-resolution passive microwave remote sensing surface temperature data using an algorithm to obtain high-resolution passive microwave remote sensing surface temperature downscaled data, specifically: The surface-to-point Kriging interpolation algorithm is used to spatially downscale the spatiotemporally continuous low-resolution passive microwave remote sensing surface temperature data to obtain high-resolution passive microwave remote sensing surface temperature downscale data. .

7. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 6, characterized in that: In step S5, based on the CatBoost symmetric decision tree algorithm, a correction model for passive microwave remote sensing surface temperature downscaling data is constructed, hyperparameters of the correction model are tuned, and cross-validation is performed to select the optimal parameters, specifically: Using high-resolution passive microwave remote sensing surface temperature downscaling data as the dependent variable, and high-resolution normalized water index, normalized difference mountain vegetation index, CIEx color coordinates, CIEy color coordinates and DEM data at the corresponding time points as independent variables, a correction model for passive microwave remote sensing surface temperature downscaling data was constructed using the CatBoost symmetric decision tree algorithm, and the results were: ； Where, express t Passive microwave remote sensing surface temperature downscaled data at the moment; express t Normalized water index corresponding to the moment; express t Normalized difference mountain vegetation index corresponding to the time; express t CIEx color coordinates at the moment; express t CIEy color coordinates at the moment; By constructing a correction model for passive microwave remote sensing surface temperature downscaling data, the error correction of high-resolution passive microwave remote sensing surface temperature downscaling data obtained by spatially downscaling low-resolution passive microwave remote sensing surface temperature data using the surface-to-point Kriging interpolation algorithm is performed.

8. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 1, characterized in that: In step S5, cross validation is performed, specifically: The cross-validation method provided by the CatBoost symmetric decision tree algorithm was used to perform a five-fold cross-validation on each set of hyperparameter combinations. In each cross-validation, the dataset was divided into five parts, four of which were used for training and one for validation, and this was repeated five times. The evaluation metric was specified as accuracy, and the best model was saved when the best performance was achieved. A fixed random seed was used to ensure consistent results. For all 720 parameter combinations, the average loss value under cross-validation was recorded.

9. The method for spatial downscaling of remote sensing surface temperature based on CatBoost symmetric decision tree according to claim 8, characterized in that: In step S5, the optimal parameters are selected, specifically: Based on the average loss value of five-fold cross-validation, the performance of all parameter combinations was compared, and the parameter combination that produced the lowest average loss value was selected as the optimal hyperparameter configuration; the correction model was retrained using the optimal parameter combination for the correction task of passive microwave remote sensing surface temperature downscaling data.

10. The method for spatial downscaling of remotely sensed surface temperature based on CatBoost symmetric decision tree according to any one of claims 1 to 9, characterized in that: Use Python language to implement data processing, model construction and model training.

Images