Surface temperature downscaling method and system based on feature interaction optimization and multi-model screening mechanism

Abstract

The invention discloses a surface temperature downscaling method and system based on feature interaction optimization and a multi-model screening mechanism, and belongs to the technical field of surface temperature data processing, and the method comprises the following steps: screening spectral bands, remote sensing indexes and topographic features; resampling the topographic feature parameters to a resolution matched with the optical wave band and the remote sensing index; variables significantly related to the surface temperature are screened through correlation analysis, interaction features are constructed, and high-contribution features are dynamically screened based on an SHAP value; constructing a regression model based on double frameworks of a linear model and a machine learning model, and verifying the regression model; sHAP optimization features extracted from Sentinel-2 data are fused with SRTM topographic data, a verified regression model is input, and surface temperature data with the resolution of 10 meters are generated; according to the method, the high-order nonlinear relation is mined through combined feature interaction, the prediction RMSE in the uniform earth surface area is greatly reduced, the data scale effect is reserved, the prediction is more suitable for the physical process, and the space consistency is better.

Description

Technical Field

[0001] The present invention belongs to the technical field of surface temperature data processing, and specifically relates to a surface temperature downscaling method and system based on feature interactive optimization and multi-model screening mechanism. Background Art

[0002] Against the backdrop of global climate change, accelerated urbanization, and increasingly prominent ecological and environmental problems, it is crucial to accurately obtain high-resolution surface temperature data. As a basic climate variable determined by the Global Climate Observing System, surface temperature plays an indispensable role in many fields. In climate change monitoring, accurate surface temperature data can reflect regional and even global heat change trends, helping scientists to conduct in-depth research on the mechanisms and impacts of climate warming. In the assessment of urban heat island effects, surface temperature can intuitively present differences in temperature distribution within cities, providing a key basis for urban planning and the formulation of strategies to mitigate the heat island effect. In agricultural precision irrigation, surface temperature can assist in determining the water needs of crops and achieve efficient use of water resources. In ecosystem health analysis, it helps to assess the energy balance and stability of ecosystems.

[0003] At present, thermal infrared remote sensing is the main means of obtaining regional-scale surface temperature data, but there are problems with long revisit cycles and high data acquisition costs: taking traditional thermal infrared satellite data acquisition as an example, its revisit cycle may be as long as weeks or even months. In some rapidly changing areas, such as urban expansion areas or ecologically fragile areas, it is impossible to capture the dynamic changes of surface temperature in a timely manner; the high data acquisition cost also limits its application in large-scale, long-term monitoring, and it is difficult to meet the urgent demand for high-temporal and spatial resolution surface temperature data.

[0004] Among the existing surface temperature downscaling methods, traditional empirical or semi-empirical methods are computationally efficient but heavily rely on regional experience. Faced with complex and diverse surface cover, such as mixed areas of various land features such as buildings, vegetation, and water bodies in cities, and complex mountainous areas, it is difficult to accurately model them, resulting in poor downscaling effects. Although geographically weighted regression can capture spatial heterogeneity to a certain extent, its accuracy is often limited to the order of hundreds of meters and cannot meet the demand for high-precision surface temperature data. Although deep learning models, with their multi-layer nonlinear network architecture, perform well in mining complex spatial correlations between multi-source remote sensing data and can achieve high-precision resolution and large-scale generalization, they still have shortcomings in enhancing spatial feature representation capabilities and optimizing model computational efficiency.

[0005] At the same time, machine learning models are constrained in surface temperature downscaling studies by the selection of characteristic variables and model applicability: existing methods often directly apply commonly used characteristic variables (spectral index, remote sensing vegetation index, urbanization index and terrain index) or rely on field experience screening, which means that the complex nonlinear relationships and interaction effects between variables are not fully explored, resulting in insufficient collaborative utilization of multi-source data; at the same time, different models perform differently in different scenarios. The random forest model has high accuracy, but in complex heterogeneous surface environments, it has limited ability to capture high-order interactions between variables and its predictive effectiveness is limited; deep learning models have strong capabilities in processing nonlinear relationships, but their high computational complexity and "black box" characteristics have become bottlenecks hindering their widespread application in practical applications in areas with limited computing resources or real-time monitoring scenarios;

[0006] Therefore, a high-precision and high-stability surface temperature downscaling method is particularly important. Summary of the Invention

[0007] Problem to be solved

[0008] In response to the problems raised in the existing background technology, the present invention provides a surface temperature downscaling method and system based on feature interaction optimization and multi-model screening mechanism.

[0009] Technical Solution

[0010] To solve the above problems, the present invention adopts the following technical solutions.

[0011] A surface temperature downscaling method based on feature interaction optimization and multi-model screening mechanism is proposed. The steps are as follows:

[0012] S1. Preprocessing of multi-source remote sensing data: Screening of spectral bands, remote sensing indices, and terrain features, including preprocessing of optical bands, remote sensing indices, and terrain characteristic parameters of Landsat 9's 100-meter resolution data and Sentinel-2's 10-meter resolution data. The optical bands include the red, green, and blue primary color bands; the remote sensing indices include the Enhanced Vegetation Index, the Normalized Building Index, and the Normalized Moisture Index; and the terrain characteristic parameters include elevation, slope, and aspect. The terrain characteristic parameters are then resampled to a resolution that matches the optical bands and remote sensing indices.

[0013] S2. Variable screening: Variables significantly correlated with surface temperature were screened through correlation analysis, interactive features were constructed, and high-contribution features were dynamically screened based on SHAP values.

[0014] S3. Establish a regression model: Based on the dual framework of linear model and machine learning model, build a regression model and verify the regression model;

[0015] S4. High-resolution surface temperature generation: SHAP optimized features extracted from Sentinel-2 data are fused with SRTM terrain data and input into a validated regression model to generate surface temperature data with a resolution of 10 meters.

[0016] Preferably, the method for constructing the interactive feature in step S2 is:

[0017] The selected variables are divided into four categories: spectral bands, remote sensing indices, terrain characteristics and nonlinear transformation characteristics. Through non-repetitive combinations in combinatorial mathematics, intra-class interaction features and cross-class interaction features are generated, including interactions between spectral bands, bands and remote sensing indices, bands and terrain characteristics, bands and nonlinear exchange characteristics, between remote sensing indices, remote sensing indices and terrain characteristics, remote sensing indices and nonlinear exchange characteristics, between terrain characteristics, terrain characteristics and nonlinear exchange characteristics, and between nonlinear exchange characteristics. Finally, 105 groups of independent interaction features are formed, and the interactive operation retains the scale effect of the original data.

[0018] Furthermore, the specific method of SHAP dynamic screening in step S2 is:

[0019] The contribution of interactive features is ranked based on SHAP values, and the top 10 highly contributing features are selected as model inputs. Through the model independence and interpretability mechanism of SHAP, the core driving factors in nonlinear relationships are identified and redundant variables are eliminated.

[0020] Preferably, the training framework of the machine learning model in step S3 is:

[0021] The model was trained at a 100-meter resolution and directly migrated to the 10-meter resolution data of Sentinel-2 for prediction. Outliers were removed through post-processing to ensure the spatial consistency of the high-resolution surface temperature data.

[0022] Furthermore, in step S3, when verifying the regression model, the specific method is as follows:

[0023] The stability and generalization ability of the regression model were evaluated using 10-fold cross validation: three evaluations were performed for each validation. 2 ) quantifies the degree to which the prediction features explain the surface temperature variation, the root mean square error (RMSE) measures the deviation between the predicted and measured temperature, and the mean absolute error (MAE) intuitively reflects the level of prediction error; models that meet all three criteria are judged to be valid.

[0024] Furthermore, the model effectiveness evaluation criteria in step S3 include:

[0025] Coefficient of determination (R 2) quantifies the degree to which the prediction features explain the surface temperature variation, the root mean square error (RMSE) measures the deviation between the predicted and measured temperature, and the mean absolute error (MAE) reflects the level of prediction error, meeting the R 2 Models with a C-value > 0.4, RMSE < 2°C, and MAE < 1.5°C were considered valid.

[0026] A surface temperature downscaling system based on feature interaction optimization and multi-model screening mechanism, including:

[0027] Data preprocessing module: used to obtain multi-source remote sensing data and terrain data, and perform preprocessing operations such as extraction and resampling on the data to ensure the spatial consistency of the data.

[0028] Variable screening module: includes correlation analysis unit, interactive feature construction unit and SHAP score screening unit, which is used to screen high-contribution features.

[0029] Model building module: used to build a dual framework of linear models and machine learning models, train and verify the models, and select effective models.

[0030] High-resolution generation module: used to fuse the optimized features with terrain data, input the effective model, and generate high-resolution surface temperature data.

[0031] Furthermore, the data preprocessing module also includes a data quality control unit for performing radiation calibration, atmospheric correction and other processing on the multi-source remote sensing data to improve data quality.

[0032] Furthermore, the interactive feature construction unit of the variable screening module can generate intra-class interactive features and cross-class interactive features according to feature classification rules, and the feature classification rules divide features into spectral bands, remote sensing indices, terrain features and nonlinear transformation features.

[0033] Beneficial effects

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) Improved accuracy and stability to adapt to complex scenarios: By using combined features to interactively mine high-order nonlinear relationships, the RMSE of predictions in uniform surface areas is significantly reduced, and the data scale effect is retained to make the predictions more consistent with the physical process and achieve better spatial consistency.

[0036] (2) Optimize computation and generalization capabilities and eliminate redundant features: Use the SHAP framework to filter features, reduce the computational load of model training, improve efficiency, enhance the generalization of the model in different scenarios, and avoid overfitting;

[0037] (3) High-resolution products are highly practical: the generated 10-meter resolution surface temperature data meets the accuracy index, providing strong data support for refined scenarios such as urban heat island control and agricultural drought warning;

[0038] (4) Good technical compatibility and scalability: The multi-model library can add new models and feature types, integrate new technologies to expand research dimensions, and adapt to ever-changing research and application needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or exemplary descriptions. Obviously, the drawings described below are only some embodiments of the present application and should not be regarded as limiting the scope. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0040] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0042] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0043] Example 1

[0044] like Figure 1 As shown in the figure, a surface temperature downscaling method based on feature interaction optimization and multi-model screening mechanism includes:

[0045] Data acquisition and preprocessing:

[0046] Landsat 9 data with 100-meter resolution in the study area was obtained from the official website of the United States Geological Survey. The data includes optical bands, enhanced vegetation index, normalized building index, normalized moisture index, as well as terrain characteristic parameters such as elevation, slope, and aspect. Sentinel-2 data with 10-meter resolution in the corresponding area was obtained from the Copernicus Open Access Center of the European Space Agency, which also covers the above-mentioned optical bands and remote sensing indices.

[0047] Use geographic information processing software such as ENVI or ArcGIS to resample terrain feature parameters.

[0048] For Landsat 9 data, elevation, slope, and aspect were resampled to a 100-meter resolution to match the optical band and remote sensing index resolution of Landsat 9. For Sentinel-2 data, terrain feature parameters were resampled to a 10-meter resolution. Radiometric calibration and atmospheric correction were also performed on all data to eliminate the effects of atmospheric scattering and absorption, thereby improving data quality.

[0049] Variable screening:

[0050] Using Python's pandas and scipy libraries, we calculated the Spearman correlation coefficients between all satellite bands, vegetation indices, terrain indices, and land surface temperature. Assuming there are 100 sample points in the study area, we calculated the correlation coefficient between each variable and LST and selected variables with a significance level of P < 0.001, such as the red band, EVI, and slope.

[0051] The selected variables were categorized into four groups: spectral bands, remote sensing indices, topographic features, and nonlinear transformation features. The Python itertools library was used to perform non-repeated combination calculations, generating intra-class and cross-class interaction features, ultimately resulting in 105 independent interaction features. During the calculation process, the original data were directly multiplied to preserve data scale effects.

[0052] The SHAP library was used to calculate the SHAP value of each interaction feature. Based on the SHAP value, the 105 interaction features were ranked and the top 10 contributing features were selected. For example, the SHAP values for features such as EVI × elevation and NDBI × slope were found to be high, and these were used as model inputs to eliminate redundant variables.

[0053] Build a regression model:

[0054] A linear model was constructed using 11 basic features, while a machine learning model was also constructed using high-contribution features optimized using SHAP. For example, an XGBoost model was trained using the 100-meter resolution Landsat 9 data training framework.

[0055] The stability and generalization ability of the model were evaluated by 10-fold cross validation: 100 sample point data were randomly divided into 10 parts, 9 of which were used as training sets and 1 as test sets each time; in each validation, the coefficient of determination, root mean square error and mean absolute error were calculated; after multiple cross validations, if a model met the R 2 >0.4, RMSE <2°C, and MAE <1.5°C, the model is considered valid.

[0056] High-resolution land surface temperature generation:

[0057] Extract SHAP-optimized features from Sentinel-2 data and fuse them with 30-meter-resolution SRTM terrain data; for example, combine the selected optimized features such as EVI and NDBI with the SRTM elevation data.

[0058] The fused data is input into a proven regression model to generate surface temperature data with a resolution of 10 meters. The generated data is post-processed by setting a temperature threshold range to eliminate outliers and ensure the spatial consistency of the data.

[0059] Example 2:

[0060] A surface temperature downscaling system based on feature interaction optimization and multi-model screening mechanism, including:

[0061] Data preprocessing module:

[0062] Data acquisition unit: Develop a dedicated data acquisition program to automatically download Landsat9, Sentinel-2 data and SRTM terrain data in the study area by calling the API interfaces of USGS and ESA.

[0063] Resampling unit: Use the GDAL library to write resampling code to match terrain feature parameters with optical bands and remote sensing index resolution.

[0064] Data quality control unit: Integrates radiometric calibration and atmospheric correction algorithms, such as the 6S model, to process multi-source remote sensing data and improve data quality.

[0065] Variable screening module:

[0066] Correlation analysis unit: Based on the Python data analysis library, a correlation calculation program was developed to calculate the Spearman correlation coefficients of all variables with LST and screen out the variables that meet the conditions.

[0067] Interaction feature construction unit: Based on the feature classification rules, Python's combined calculation function is used to generate intra-class and cross-class interaction features, forming 105 independent interaction features.

[0068] SHAP scoring screening unit: calls the SHAP library, calculates the SHAP value of the interactive features, sorts the contributions, and screens the top 10 high-contribution features.

[0069] Model building module:

[0070] Model Building Unit: Use machine learning libraries such as Scikit-learn and XGBoost to build linear models and machine learning models.

[0071] Model training unit: Train the model on 100-meter resolution data and set appropriate training parameters.

[0072] Model Validation Unit: Perform 10-fold cross validation and calculate R 2 , RMSE and MAE to determine whether the model is effective.

[0073] High-resolution generation module:

[0074] Feature fusion unit: fuses the SHAP optimized features of Sentinel-2 with SRTM terrain data.

[0075] Temperature generation unit: Input the fused data into the effective model to generate 10-meter resolution surface temperature data.

[0076] Post-processing unit: Write data post-processing programs to remove outliers and output high-quality 10-meter resolution surface temperature data.

[0077] In summary, this application has the following beneficial effects:

[0078] Significantly improved accuracy and stability, and enhanced adaptability to complex scenarios: This method leverages innovative combined feature interaction methods, such as constructing interactive features like EVI×elevation and NDBI×slope, to fully exploit high-order nonlinear relationships between different types of features. Tests conducted on uniform surface areas, such as those covered by vegetation, have shown a significant reduction in the root mean square error (RMS) of predictions compared to traditional machine learning methods. Furthermore, by preserving the scale effects of the original data, the model more closely reflects the physical processes of surface energy exchange. This significantly improves the spatial consistency of predictions and more accurately reflects the distribution of surface temperature.

[0079] Optimizing computational efficiency and generalization, and eliminating redundant features: The SHAP interpretability framework is introduced for dynamic feature screening. Based on quantified feature contributions, it accurately identifies and eliminates redundant features. This reduces the computational workload of model training and improves computational efficiency when processing large-scale multi-source remote sensing data. After SHAP screening, the number of model input features is reduced by approximately 30%-50%, without reducing model accuracy. Instead, the generalization capability in different scenarios is enhanced.

[0080] During the specific implementation, verification was carried out in multiple test areas with different climate zones and different terrains. The stability and adaptability of the model were significantly improved, effectively avoiding overfitting and ensuring stable and reliable operation in complex and changeable actual application scenarios.

[0081] Improving the practicality of high-resolution products: Successfully generated surface temperature data with a 10-meter resolution, meeting the requirements of a coefficient of determination greater than 0.4 and a root mean square error less than 2°C.

[0082] Technical compatibility and scalability: The multi-model library constructed by the present invention has good scalability, allowing the addition of new machine learning models or feature types; it can flexibly integrate new models and data to further improve the performance and application scope of surface temperature downscaling.

[0083] The above-described embodiments merely represent preferred embodiments of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications, improvements, and substitutions without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A surface temperature downscaling method based on feature interaction optimization and multi-model screening mechanism, characterized in that: Here are the steps: S1. Preprocessing of multi-source remote sensing data: Screening of spectral bands, remote sensing indices, and terrain features, including preprocessing of optical bands, remote sensing indices, and terrain characteristic parameters of Landsat 9's 100-meter resolution data and Sentinel-2's 10-meter resolution data. The optical bands include the red, green, and blue primary color bands; the remote sensing indices include the Enhanced Vegetation Index, the Normalized Building Index, and the Normalized Moisture Index; and the terrain characteristic parameters include elevation, slope, and aspect. The terrain characteristic parameters are then resampled to a spatial resolution that matches the optical bands and remote sensing indices. S2. Variable screening: Variables significantly correlated with surface temperature were screened through correlation analysis, interactive features were constructed, and high-contribution features were dynamically screened based on SHAP values. S3. Establish a regression model: Based on the dual framework of linear model and machine learning model, build a regression model and verify the regression model; S4. High-resolution surface temperature generation: SHAP optimized features extracted from Sentinel-2 data are fused with SRTM terrain data and input into a validated regression model to generate surface temperature data with a resolution of 10 meters.

2. The surface temperature downscaling method based on feature interactive optimization and multi-model screening mechanism according to claim 1 is characterized by: The method for constructing the interactive features in step S2 is: The selected variables are divided into four categories: spectral bands, remote sensing indices, terrain characteristics and nonlinear transformation characteristics. Through non-repetitive combinations in combinatorial mathematics, intra-class interaction features and cross-class interaction features are generated, including interactions between spectral bands, bands and remote sensing indices, bands and terrain characteristics, bands and nonlinear exchange characteristics, between remote sensing indices, remote sensing indices and terrain characteristics, remote sensing indices and nonlinear exchange characteristics, between terrain characteristics, terrain characteristics and nonlinear exchange characteristics, and between nonlinear exchange characteristics. Finally, 105 groups of independent interaction features are formed, and the interactive operation retains the scale effect of the original data.

3. The surface temperature downscaling method based on feature interactive optimization and multi-model screening mechanism according to claim 1 is characterized by: The specific method of SHAP dynamic screening in step S2 is: The contribution of interactive features is ranked based on SHAP values, and the top 10 highly contributing features are selected as model inputs. Through the model independence and interpretability mechanism of SHAP, the core driving factors in nonlinear relationships are identified and redundant variables are eliminated.

4. The surface temperature downscaling method based on feature interactive optimization and multi-model screening mechanism according to claim 1 is characterized by: The training framework of the machine learning model in step S3 is: The model was trained at a 100-meter resolution and directly migrated to the 10-meter resolution data of Sentinel-2 for prediction. Outliers were removed through post-processing to ensure the spatial consistency of the high-resolution surface temperature data.

5. The surface temperature downscaling method based on feature interactive optimization and multi-model screening mechanism according to claim 1 is characterized by: When verifying the regression model in step S3, the specific method is as follows: The stability and generalization ability of the regression model were evaluated using 10-fold cross validation: three evaluations were performed for each validation. 2 ) quantifies the degree to which the prediction features explain the surface temperature variation, the root mean square error (RMSE) measures the deviation between the predicted and measured temperature, and the mean absolute error (MAE) intuitively reflects the level of prediction error; models that meet all three criteria are judged to be valid.

6. The surface temperature downscaling method based on feature interactive optimization and multi-model screening mechanism according to claim 5 is characterized by: The model effectiveness evaluation criteria in step S3 include: Coefficient of determination (R 2 ) quantifies the degree to which the prediction features explain the surface temperature variation, the root mean square error (RMSE) measures the deviation between the predicted and measured temperature, and the mean absolute error (MAE) reflects the level of prediction error, meeting the R 2 Models with a C-value > 0.4, RMSE < 2°C, and MAE < 1.5°C were considered valid.

7. A surface temperature downscaling system based on feature interaction optimization and multi-model screening mechanism, characterized by: include: Data preprocessing module: used to obtain multi-source remote sensing data and terrain data, and perform preprocessing operations such as extraction and resampling on the data to ensure the spatial consistency of the data. Variable screening module: includes correlation analysis unit, interactive feature construction unit and SHAP score screening unit, which is used to screen high-contribution features. Model building module: used to build a dual framework of linear models and machine learning models, train and verify the models, and select effective models. High-resolution generation module: used to fuse the optimized features with terrain data, input the effective model, and generate high-resolution surface temperature data.

8. The surface temperature downscaling system based on feature interactive optimization and multi-model screening mechanism according to claim 7 is characterized by: The data preprocessing module also includes a data quality control unit for performing radiation calibration, atmospheric correction and other processing on multi-source remote sensing data to improve data quality.

9. The surface temperature downscaling system based on feature interactive optimization and multi-model screening mechanism according to claim 8 is characterized in that: The interactive feature construction unit of the variable screening module can generate intra-class interactive features and cross-class interactive features according to feature classification rules. The feature classification rules divide features into spectral bands, remote sensing indices, terrain features and nonlinear transformation features.

Images