A method for downscaling surface temperature in complex terrain based on remote sensing images

Abstract

The present invention relates to the technical field of surface temperature downscaling, and discloses a surface temperature downscaling method for complex terrain based on remote sensing images, which obtains image data and measured data from meteorological stations within the scope of a study area; determines the vegetation coverage type of the study area, uses the kernel normalized vegetation index kNDVI to replace the normalized vegetation index NDVI for application in the TsHARP downscaling algorithm, and establishes a relationship model between kNDVI and LST; uses the measured data from meteorological stations to correct the surface temperature results obtained by the improved TsHARP downscaling algorithm; and evaluates its downscaling accuracy through meteorological station observation data. The present invention improves the traditional TSHARP downscaling algorithm, replaces the normalized vegetation index NDVI with the improved vegetation index kernel normalized vegetation index kNDVI to deal with problems such as NDVI saturation effect, introduces vegetation coverage types, and performs surface temperature downscaling on each vegetation coverage type to obtain a more accurate initial downscaling result.

Description

Technical Field

[0001] The present invention relates to the technical field of surface temperature downscaling, and in particular to a method for downscaling surface temperature of complex terrain based on remote sensing images. Background Art

[0002] Land surface temperature (LST) is a key parameter in the physical processes of surface energy and water balance from local to global scales. Surface temperature products of different scales play a vital role in various fields of environmental science research. Existing applications mainly include climate change monitoring, evapotranspiration mapping, surface energy balance, drought monitoring, vegetation dynamics monitoring and surface carbon inversion. Accurate surface temperature information at different spatiotemporal scales helps promote these environmental monitoring studies. However, the spatial resolution of existing surface temperature products is low, resulting in thermal mixing effects, which limits the application of surface temperature data. Previous studies have shown that spatial downscaling of surface temperature is an effective way to solve this problem. One of the key technologies for obtaining high-resolution surface temperature data is the surface temperature downscaling method. These methods are mainly divided into two categories: dynamic downscaling and statistical downscaling.

[0003] Dynamical downscaling technology combines the low-resolution output of the global climate model with the high-resolution details of the regional climate model to provide initial conditions for the regional climate model, thereby obtaining more detailed regional climate characteristics. The advantage of this method is that it has a clear physical basis and can be generally applied without being restricted by specific observational data; however, it also has some limitations, such as high computational cost and being greatly affected by the boundary conditions of the global climate model, which may lead to large systematic errors. In contrast, the statistical downscaling method relies on abundant observational data to establish a statistical association between large-scale climate conditions and regional climate elements. This relationship is verified by independent observational data and then applied to large-scale climate data to simulate climate elements in specific regions. The advantages of this method are low computational cost, simple operation, ability to correct systematic errors of the global climate model, and no influence of boundary conditions on the prediction results.

[0004] However, the traditional TsHARP downscaling algorithm mainly relies on the regression relationship between the Normalized Difference Vegetation Index (NDVI) and the Land Surface Temperature (LST) to improve the spatial resolution of the land surface temperature data. However, there are still some drawbacks:

[0005] 1. Nonlinear saturation problem: When using NDVI to downscale land surface temperature, you may encounter the problem of nonlinear saturation. This means that in areas with high vegetation coverage, changes in NDVI values ​​may no longer sensitively reflect changes in land surface temperature, thus affecting the accuracy of the downscaling results.

[0006] 2. Lack of field monitoring data: The high-resolution surface temperature distribution obtained by statistical downscaling methods often lacks corresponding field observation data for verification and correction. This makes the final surface temperature spatial distribution may not be accurate enough and has certain uncertainties. Summary of the invention

[0007] The purpose of the present invention is to provide a method for downscaling surface temperature of complex terrain based on remote sensing images, so as to solve at least one technical problem raised in the background technology.

[0008] A method for downscaling surface temperature of complex terrain based on remote sensing images, comprising the following steps:

[0009] S1. Data acquisition and preprocessing: obtain Landsat-8 and Sentinel-2 image data and meteorological station measured data within the study area, and preprocess the data;

[0010] S2. Establish a regression model to determine the vegetation cover type in the study area and use the kernel normalized vegetation index Alternative Normalized Difference Vegetation Index It is applied in the TsHARP downscaling algorithm, and then based on each vegetation cover type, a Relationship model with LST;

[0011] S3, regression algorithm, uses the measured data from the meteorological station to correct the surface temperature results obtained by the improved TsHARP downscaling algorithm;

[0012] S4. Evaluate the downscaling accuracy through meteorological station observation data;

[0013] The relationship model establishment process is as follows:

[0014] First, according to the prior relationship, The threshold is 0.1, and the 30m and LST, the regression model calculation formula is as follows:

[0015] ;

[0016] in: , , All are regression coefficients; and The study areas The maximum and minimum values ​​of ;

[0017] Then downscaling is performed to establish the The regression model was applied to 10m In terms of data, we establish a connection between the two, add regression error, and obtain , as follows:

[0018] ；

[0019] ；

[0020] in: and Respectively represent 30m resolution and surface temperature, is the regression function, is the regression error.

[0021] Preferably, the preprocessing in S1 includes correction, resampling and cropping of remote sensing images, and surface parameters of different resolutions required in subsequent steps.

[0022] Preferably, the measured data in S3 is the temperature data of the location of the weather station. Based on the continuity of the temperature data, the grid point temperature data at a resolution of 10 meters where the weather station data is located is expanded.

[0023] Preferably, the correction process is as follows:

[0024] S31. Based on the downscaling process, calculate the ratio coefficient between the measured temperature data of the expanded meteorological station and the downscaled temperature at the corresponding location ;

[0025] ;

[0026] Where: After the expansion The weather station temperature at each site; It is Improved TsHARP downscaled air temperature at each station location;

[0027] S32, As the dependent variable, the surface topographic parameters of the same location were taken as independent variables to construct a surface temperature correction model based on the random forest algorithm, and the model was applied to the entire study area to obtain the corrected surface temperature downscaling results.

[0028] Preferably, the indicators for evaluating the downscaling accuracy in S4 include determination coefficient, root mean square error and mean absolute error.

[0029] Beneficial effects of the present invention: The present invention reduces the vegetation index in the traditional TsHARP downscaling algorithm by Replaced with improved vegetation index kernel normalized vegetation index To deal with The saturation effect and other limitations are eliminated. At the same time, the surface temperature is downscaled on each vegetation cover type to obtain the initial surface temperature downscale product. Then, based on the characteristics of the continuity of meteorological data, the meteorological station measured data is expanded to obtain more meteorological station measured data. Then, based on the random forest regression algorithm, the expanded meteorological station measured data is used to correct the results obtained by the improved TsHARP downscaling algorithm to obtain the final downscaling results. By using field meteorological observation data to correct the results obtained by the improved TsHARP downscaling algorithm, the accuracy of the downscaling results is greatly improved.

[0030] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0032] Figure 1 is a flow chart of the surface temperature downscaling method of the present invention;

[0033] Figure 2 It is a schematic diagram of the expansion of the measured data of the weather station of the present invention;

[0034] Figure 3 30m LST result diagram of Landsat 8 inversion of surface temperature downscaling in an embodiment of the present invention;

[0035] Figure 4 This is a result map of the Landsat 8 downscaling of the surface temperature in an embodiment of the present invention with a resolution of 10m. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Embodiment 1: Figure 1 As shown, the present invention uses a complex terrain surface temperature downscaling method based on remote sensing images:

[0038] The specific steps include:

[0039] S1. Data acquisition and preprocessing: obtain Landsat-8 and Sentinel-2 image data and meteorological station measured data within the study area, and preprocess the data;

[0040] S2. Establish a regression model to determine the vegetation cover type in the study area and use the kernel normalized vegetation index Alternative Normalized Difference Vegetation Index It is applied in the TsHARP downscaling algorithm, and then based on each vegetation cover type, a Relationship model with LST;

[0041] S3, regression algorithm, uses the measured data from the meteorological station to correct the surface temperature results obtained by the improved TsHARP downscaling algorithm;

[0042] S4. Evaluate the downscaling accuracy through meteorological station observation data;

[0043] The relationship model establishment process is as follows:

[0044] ;

[0045] in: , , All are regression coefficients; and The study areas The maximum and minimum values ​​of ;

[0046] Then downscaling is performed to establish the The regression model was applied to 10m In terms of data, we establish a connection between the two, add regression error, and obtain , as follows:

[0047] ；

[0048] ；

[0049] in: and Respectively represent 30m resolution and surface temperature, is the regression function, is the regression error.

[0050] By improving the traditional TSHARP downscaling algorithm, the normalized vegetation index in the traditional algorithm is Replaced with improved vegetation index kernel normalized vegetation index To deal with In order to solve the problems such as saturation effect, the vegetation cover types were introduced, and the surface temperature was downscaled on each vegetation cover type to obtain a more accurate initial downscaling result.

[0051] As an implementation mode of the present invention, specifically, the preprocessing in S1 includes correction, resampling, and cropping of remote sensing images, surface parameters of different resolutions required in subsequent steps, and surface parameters of different resolutions required in subsequent steps also need to be processed, as well as surface temperature inversion of the 10th band of the Landsat-8 TIRS thermal infrared using a single window algorithm to obtain surface temperature data at a resolution of 30m.

[0052] As an implementation mode of the present invention, specifically, the measured data in S3 is the temperature data of the location of the meteorological station. Based on the continuity of the temperature data, the grid point temperature data at a resolution of 10 meters where the meteorological station data is located is expanded, wherein the expansion method can be selected as: expanding the grid point temperature data at a resolution of 10 meters where the meteorological station data is located to the meteorological station data of eight grid points around the grid point where the meteorological station is located, thereby realizing the expansion of the measured data of the meteorological station and solving the problem of overfitting in machine learning due to less data.

[0053] As an implementation mode of the present invention, specifically, the correction process is as follows:

[0054] S31. Based on the downscaling process, calculate the ratio coefficient between the measured temperature data of the expanded meteorological station and the downscaled temperature at the corresponding location ;

[0055] ;

[0056] Where: After the expansion The temperature of the weather station at each site; It is Improved TsHARP downscaled air temperature at each station location;

[0057] S32, As the dependent variable, the surface topographic parameters of the same location were taken as independent variables to construct a surface temperature correction model based on the random forest algorithm, and the model was applied to the entire study area to obtain the corrected surface temperature downscaling results.

[0058] Based on the random forest regression algorithm, the initial results obtained by the improved TsHARP downscaling algorithm were corrected using the expanded measured data from meteorological stations, resulting in more accurate surface temperature downscaling results, which solved the problem of lack of field observation data correction for surface temperature downscaling.

[0059] As an implementation manner of the present invention, specifically, the indicators for evaluating the downscaling accuracy in S4 include determination coefficient, root mean square error and mean absolute error.

[0060] Embodiment 2: This embodiment is carried out in a certain area of ​​Shangyu District, Zhejiang Province, and the method steps of the present application are executed to verify the present invention.

[0061] Step 1: Data acquisition and preprocessing:

[0062] The satellite remote sensing image data of Landsat8 and Sentinel-2 and the measured data of the meteorological station in the study area of ​​Shangyu District, Zhejiang Province were obtained, and the data were preprocessed. The preprocessing included image correction, resampling, and cropping. It also included the surface parameters of different resolutions required in the subsequent steps and the single window algorithm for the 10th band of Landsat-8 TIRS thermal infrared to invert the surface temperature, and obtain the surface temperature at a resolution of 30m, such as Figure 3 shown.

[0063] Step 2: Downscaling based on the improved traditional TSHARP algorithm. Based on Landsat8 and Sentinel-2 data, the initial surface temperature product with a high spatial resolution of 10m is obtained by using the improved traditional TSHARP downscaling algorithm.

[0064] The improved TsHARP downscaling algorithm first determines the vegetation cover type of the study area, and then converts the normalized vegetation index (NDVI) in the traditional TsHARP downscaling algorithm into Replaced with improved vegetation index kernel normalized vegetation index To deal with Saturation effect and other issues, and finally establish a The relationship model with LST. According to the prior relationship, The threshold is 0.1, and the 30m and regression models of LST.

[0065] ;

[0066] in: , , All are regression coefficients; and The study areas The maximum and minimum values ​​of .

[0067] Downscaling: The image created at a low spatial resolution of 30 m is The regression model was applied to the high-resolution 10m , establish the connection between the two, add the regression error, and obtain , as follows:

[0068] ；

[0069] ；

[0070] in: and Respectively represent 30m resolution and surface temperature, is the regression function, is the regression error.

[0071] Step 3: Use the measured meteorological station data to expand the measured data of the meteorological stations and establish a random forest regression model to correct the improved TsHARP downscaling results.

[0072] The present invention expands the measured data of the meteorological station. The temperature data measured by the meteorological station usually represents the temperature data of the location point where the meteorological station is located. Based on the continuity of the temperature data, the temperature data of the grid point where the meteorological station data is located at a resolution of 10 meters is expanded to the meteorological station data of the eight grid points around the grid point where the meteorological station is located, so as to realize the expansion of the measured data of the meteorological station. Figure 2 shown.

[0073] Based on the downscaling results of the improved TSHARP downscaling algorithm, the present invention calculates the proportional coefficient between the observed temperature of the expanded meteorological station and the downscaled temperature of its corresponding location .

[0074] ;

[0075] Where: After the expansion The temperature of the weather station at each site; It is Temperature results after downscaling using the improved traditional TsHARP algorithm at 4 locations.

[0076] The ratio coefficient between the observed temperature at a meteorological station and the downscaled temperature at its corresponding location The surface topographic parameters of the same location were used as the dependent variable, and the surface temperature correction model based on the random forest algorithm was constructed. The surface topographic parameters of Shangyu area in Zhejiang Province were input into the surface temperature correction model to obtain the proportional coefficient of the entire study area. , the obtained study area scale coefficient Combined with the temperature results after downscaling using the improved traditional TsHARP algorithm, the corrected temperature downscaling results are obtained. Figure 4 As shown in the figure, the accuracy of the downscaling results is greatly improved by using field meteorological observation data to correct the results obtained by the improved TsHARP downscaling algorithm.

[0077] In the selection of surface parameters, we mainly focus on the spatial distribution of local temperature, which is mainly affected by three types of radiation: direct solar radiation, surface thermal radiation, and reflection from surrounding objects. These radiation effects are the product of the interaction between solar radiation, local topography, and surface cover. In order to capture the characteristics of these three types of radiation, we use the digital elevation model (DEM) to calculate parameters such as slope and aspect that affect solar radiation. In addition, by analyzing Sentinel-2 image data, we further calculated the kernel normalized difference vegetation index (kNDVI), normalized difference vegetation index (NDVI), and normalized snow cover index (NDSI) that reflect the emission and reflection characteristics of the surface. Based on the above analysis, we selected the following surface parameters to correct the downscaling results: normalized difference vegetation index (NDVI), normalized difference snow cover index (NDSI), kernel normalized difference vegetation index (kNDVI), digital elevation data (DEM), slope (Slope) and aspect (Aspect). The selection of these parameters aims to fully reflect the impact of surface characteristics on temperature distribution, thereby improving the prediction accuracy and reliability of the model.

[0078] In the process of building and optimizing parameters of the random forest regression model, we focus on three key parameters: the randomness of sample attributes, the number of decision trees, and the number of features considered when splitting nodes. The model constructs multiple regression trees by randomly selecting samples from the data set, but this random sampling method may lead to unbalanced selection of different types of data. In order to achieve the best dimensionality reduction effect, we carefully adjust the parameters of the number of decision trees in the algorithm and the number of features randomly selected for each tree. Generally, increasing the number of decision trees can improve the prediction performance of the model, but when the number of trees reaches a certain level, further increasing the number of trees has limited effect on performance improvement and increases the computational cost. This study uses MATLAB to perform computational analysis on the changes in error with the number of decision trees and the number of features, and comprehensively considers the error size and computational efficiency to determine the optimal number of decision trees and the number of selected features.

[0079] Step 4: Accuracy evaluation and applicability analysis.

[0080] The existing high-resolution surface temperature data and the measured temperature data of the surface meteorological station are used to evaluate the accuracy of the downscaling results. The determination coefficient (R2), root mean square error (RMSE) and mean absolute error (MAE) are used as accuracy evaluation indicators. The formula is as follows:

[0081] Coefficient of determination (R2):

[0082] ;

[0083] Root Mean Square Error (RMSE):

[0084] ;

[0085] Mean Absolute Percent Error (MAE):

[0086] ;

[0087] In the formula, is the total number of samples, The image showing the surface temperature downscaling result is The temperature value of each pixel, Verify the surface temperature on the image The temperature value of each pixel, is the average of the verified values ​​of surface temperature.

[0088] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.

Claims

1. A method for downscaling surface temperature of complex terrain based on remote sensing images, characterized in that: The specific steps include: S1. Data acquisition and preprocessing: obtain Landsat-8 and Sentinel-2 image data and meteorological station measured data within the study area, and preprocess the data; S2. Establish a regression model to determine the vegetation cover type in the study area and use the kernel normalized vegetation index Alternative Normalized Difference Vegetation Index It is applied in the TsHARP downscaling algorithm, and then based on each vegetation cover type, a Relationship model with LST; S3, regression algorithm, uses the measured data from the meteorological station to correct the surface temperature results obtained by the improved TsHARP downscaling algorithm; S4. Evaluate the downscaling accuracy through meteorological station observation data; The relationship model establishment process is as follows: First, according to the prior relationship, The threshold is 0.1, and the 30m and LST, the regression model calculation formula is as follows: ； in: , , All are regression coefficients; and The study areas The maximum and minimum values ​​of ; Then downscaling is performed to establish the The regression model was applied to 10m In terms of data, we establish a connection between the two, add regression error, and obtain , as follows: ； ； in: and Respectively represent 30m resolution and surface temperature, is the regression function, is the regression error.

2. The method for downscaling surface temperature of complex terrain based on remote sensing images according to claim 1, characterized in that: The preprocessing in S1 includes the correction, resampling and cropping of the remote sensing image, and the surface parameters of different resolutions required in the subsequent steps.

3. The method for downscaling surface temperature of complex terrain based on remote sensing images according to claim 1, characterized in that: The measured data in S3 is the temperature data of the location of the weather station. Based on the continuity of the temperature data, the grid point temperature data at a resolution of 10 meters where the weather station data is located is expanded.

4. The method for downscaling surface temperature of complex terrain based on remote sensing images according to claim 3 is characterized in that: The correction process is as follows: S31. Based on the downscaling process, calculate the ratio coefficient between the measured temperature data of the expanded meteorological station and the downscaled temperature at the corresponding location ; ； Where: After the expansion The temperature of the weather station at each site; It is Improved TsHARP downscaled air temperature at each station location; S32, Taking the terrain parameters at the same location as the dependent variable and the independent variables, a surface temperature correction model based on the random forest algorithm was constructed. The model was applied to the entire study area to obtain the corrected surface temperature downscaling results.

5. The method for downscaling surface temperature of complex terrain based on remote sensing images according to claim 1, characterized in that: The indicators for evaluating the downscaling accuracy in S4 include the coefficient of determination, the root mean square error and the mean absolute error.

Images