Surface temperature estimation method, system and device and storage medium

Abstract

The invention discloses a surface temperature estimation method, system and device and a storage medium, and relates to the technical field of surface temperature estimation, and the method comprises the steps: obtaining a low-spatial-resolution day-by-day surface temperature data set and a high-spatial-resolution underlying surface data set; re-projecting the day-by-day surface temperature data set with the low spatial resolution to obtain a new day-by-day surface temperature data set; processing the spatial resolutions of the new day-by-day surface temperature data set and the underlying surface data set to be consistent by adopting a sliding window cutting method; new day-by-day surface temperature data and underlying surface data which are processed to be consistent serve as feature data to be input into a surface temperature estimation model, and high-spatial-resolution day-by-day surface temperature data are obtained; when the model is used for resource environment monitoring, reliable long-time-sequence and high-spatial-resolution surface temperature data can be provided.

Description

Technical Field

[0001] The present invention relates to the technical field of surface temperature estimation, and in particular to a surface temperature estimation method, system, device and storage medium. Background Art

[0002] The surface temperature is closely related to the interaction between the surface and the atmosphere and the energy exchange, and is one of the key parameters in the land surface system process at the regional and global scales. As the most important and basic physical parameter of the earth's surface, high-quality surface temperature is widely used in various disciplines such as environmental research, hydrological research, ecological research, and urban heat island research.

[0003] In the agricultural field, comparative analysis of long-term surface temperature data can provide a basis for judging crop growth conditions and selecting suitable crops. Moreover, surface temperature affects the evaporation of soil moisture and the transpiration of crops. By analyzing the long-term surface temperature data, the water requirements of soil and crops can be determined to adjust the irrigation amount and frequency. In addition, long-term surface temperature data can be used to predict crop yields under different surface temperature conditions, providing important data support for agricultural production planning, food security management, and crop insurance assessment. In terms of natural disaster monitoring and prediction (especially drought and forest fires), high-temporal and high-spatial resolution surface temperature data and water resources data can be used to predict the possible time of drought and allocate water resources as early as possible. In addition, the change of surface temperature data before the occurrence of forest fires can be used to analyze the possibility of forest fires and the specific areas where they may occur. Therefore, high-temporal and spatial resolution surface temperature data can provide important data support for agriculture, meteorology, and the environment.

[0004] However, the existing surface temperature data have the problem that long time series and high spatial resolution are difficult to coexist, resulting in a lack of reliable long time series and high spatial resolution surface temperature data when conducting resource and environmental monitoring, and therefore cannot meet the needs of application scenarios in agriculture, meteorology, and the environment. Summary of the invention

[0005] In view of the problem that long time series and high spatial resolution are difficult to coexist in the surface temperature data directly obtained by the prior art, resulting in a lack of reliable long time series and high spatial resolution surface temperature data when conducting resource and environmental monitoring, the present invention proposes a surface temperature estimation method, system, device and storage medium. By exploring the nonlinear relationship between long time series and low spatial resolution surface temperature data and high spatial resolution underlying surface information, a long time series and high spatial resolution surface temperature data estimation model is constructed to improve the time and spatial resolution of surface temperature data, thereby solving the problems existing in the prior art.

[0006] A method for estimating land surface temperature comprises the following steps:

[0007] Obtain a daily surface temperature dataset with low spatial resolution, an underlying surface dataset with high spatial resolution, and a surface temperature dataset with high spatial resolution obtained in the 10th band of Landsat8 OLI; the low spatial resolution ranges from 100 meters to 1000 meters, and the high spatial resolution ranges from within 30 meters;

[0008] The low spatial resolution daily surface temperature dataset is scaled and reprojected to obtain a new daily surface temperature dataset with the same resolution as the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI.

[0009] The sliding window clipping method is used to make the spatial resolution of the new daily surface temperature dataset and the underlying surface dataset consistent;

[0010] The high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI is used as the label dataset, and the new daily surface temperature dataset and underlying surface dataset after processing are used as the feature dataset. The random forest model is trained with the label dataset and the feature dataset to obtain the surface temperature estimation model.

[0011] The underlying surface data to be estimated and the new daily surface temperature data with the same spatial resolution are input into the surface temperature estimation model to obtain the daily surface temperature data with high spatial resolution.

[0012] Furthermore, the daily surface temperature dataset with low spatial resolution is obtained by selecting MOD11A1 data with cloud cover less than 0.5% from the NASA database.

[0013] Furthermore, the underlying surface dataset with high spatial resolution includes a normalized difference vegetation index NDVI dataset and a land cover dataset; wherein the NDVI dataset is obtained by selecting a Landsat8-9OLI / TIRS Collection2Level-2 dataset with a cloud coverage rate of less than 5% from the USGS database, and the SR_B4.TIF and SR_B5.TIF band data in the Level-2SurfaceReflectanceBands are used to calculate the NDVI, and the calculation formula is:

[0014]

[0015] Among them, B4 is the red light band of Landsat8OLI, and B5 is the near-infrared band.

[0016] Furthermore, the high spatial resolution surface temperature dataset obtained by the 10th band of the Landsat8 OLI is obtained by selecting the Landsat8-9OLI / TIRS Collection 2 Level-2 dataset with a cloud coverage rate of less than 5% in the USGS database, and the high spatial resolution surface temperature dataset LST is calculated using the Level-2 Surface Temperature Bands, and the calculation formula is:

[0017] LST=0.00341802×B 10 +149

[0018] Among them, B 10 is the 10th band in the Landsat8-9OLI / TIRS Collection2Level-2 dataset. Further, the surface temperature estimation model is expressed as:

[0019] Y tij =K[NDVI ti , LCD t , MODIS tij ]

[0020] Among them, Y tij is the daily surface temperature data with high spatial resolution on the jth day of the ith month of the tth year, K is the integrated machine learning algorithm obtained by training, and NDVI ti is the NDVI data of the ith month of the tth year, LCD t is the land cover data for year t, MODIS tij The new daily surface temperature data for the jth day of the ith month of the tth year after processing to be consistent with the spatial resolution of the underlying surface data.

[0021] Furthermore, it also includes the use of mean absolute percentage error MAPE, root mean square error RMSE, mean absolute error MAE, determination coefficient R 2 , Adjusted R 2 The performance of the surface temperature estimation model is evaluated by using six indicators: mean square error (MSE); among which:

[0022] The mean absolute percentage error MAPE expression is:

[0023]

[0024] where y i is the actual value of the surface temperature, y′ i is the predicted value of the surface temperature, n is the number of samples of the surface temperature;

[0025] The root mean square error RMSE expression is:

[0026]

[0027] The mean absolute error MAE expression is:

[0028]

[0029] The coefficient of determination R 2 The expression is:

[0030]

[0031] in is the average surface temperature;

[0032] The adjusted coefficient of determination Adjusted R 2 The expression is:

[0033]

[0034] Where R 2 is the coefficient of determination, p is the number of independent variables;

[0035] The mean square error MSE expression is:

[0036]

[0037] The present invention also includes a surface temperature estimation system, comprising:

[0038] An acquisition module is used to acquire a daily surface temperature dataset with low spatial resolution, an underlying surface dataset with high spatial resolution, and a surface temperature dataset with high spatial resolution acquired in the 10th band of Landsat8 OLI; the low spatial resolution ranges from 100 meters to 1000 meters, and the high spatial resolution ranges from within 30 meters;

[0039] The reprojection module is used to scale the daily surface temperature dataset with low spatial resolution, and reproject the scaled daily surface temperature dataset to obtain a new daily surface temperature dataset with the same resolution as the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI;

[0040] A processing module, used to process the spatial resolution of the new daily surface temperature dataset and the underlying surface dataset to be consistent by using a sliding window clipping method;

[0041] The model building module is used to use the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI as the label dataset, and the new daily surface temperature dataset and underlying surface dataset after processing as the feature dataset, and train the random forest model through the label dataset and the feature dataset to obtain the surface temperature estimation model;

[0042] The estimation module is used to input the underlying surface data to be estimated and the new daily surface temperature data consistent with its spatial resolution into the surface temperature estimation model to obtain the daily surface temperature data with high spatial resolution.

[0043] The present invention also includes a computer device for estimating ground surface temperature, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the ground surface temperature estimation method when executing the computer program.

[0044] The present invention also includes a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the surface temperature estimation method.

[0045] The present invention provides a method, system, device and storage medium for estimating surface temperature, which have the following features:

[0046] Beneficial effects:

[0047] The present invention estimates the surface temperature by selecting long-time series daily surface temperature data as input data, thereby ensuring the continuity and reliability of the data at the source; by reprojecting the long-time series low-spatial-resolution surface temperature data, the spatial-resolution surface temperature data is sampled as new surface temperature data with the same resolution as the high-spatial-resolution surface temperature data obtained by the 10th band of the OLI of Landsat8, and the spatial resolution of the data is processed to be consistent with the spatial resolution of the underlying surface data through a sliding window clipping method, and a surface temperature data estimation model is obtained based on a random forest model through specific high-spatial-resolution underlying surface information and long-time series low-spatial-resolution surface temperature data, thereby predicting and obtaining long-time series, high-spatial-resolution and high-precision surface temperature data, so that reliable long-time series, high-spatial-resolution surface temperature data can be provided when performing resource and environmental monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the structure of a method for estimating ground surface temperature in an embodiment of the present invention;

[0049] Figure 2 This is a 30m surface temperature map in an embodiment of the present invention;

[0050] Figure 3 This is the surface temperature map at 990m in the embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of 30mNDVI in an embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram of 30m land cover data in an embodiment of the present invention;

[0053] Figure 6 is the R of the RandomForest model in the embodiment of the present invention 2 Schematic diagram of the results;

[0054] Figure 7 is the R of the XG-Boost model in the embodiment of the present invention 2 Schematic diagram of the results;

[0055] Figure 8 is the R of the CatBoost model in the embodiment of the present invention 2 Schematic diagram of the results;

[0056] Fig. 9 This is a schematic diagram of the mean absolute error (MAE) results of the RandomForest model in an embodiment of the present invention; Fig. 9 (a) is a schematic diagram of the density distribution results of MAE. Fig. 9 (b) is a schematic diagram of the proportion of different MAE intervals;

[0057] Fig.10 This is a schematic diagram of the mean absolute error (MAE) results of the XG-Boost model in an embodiment of the present invention; Fig.10 (a) is a schematic diagram of the density distribution results of MAE. Fig.10 (b) is a schematic diagram of the proportion of different MAE intervals;

[0058] Fig.11 Schematic diagram of mean absolute error (MAE) results of the CatBoost model in an embodiment of the present invention; Fig.11 (a) is a schematic diagram of the density distribution results of MAE. Fig.11 (b) is a schematic diagram of the proportion of different MAE intervals;

[0059] Fig.12 Schematic diagram of mean square error (MSE) result of RandomForest model in an embodiment of the present invention; Fig.12 (a) is a schematic diagram of the density distribution results of MSE. Fig.12 (b) is a schematic diagram of the proportion of different intervals of MSE;

[0060] Fig.13Schematic diagram of mean square error (MSE) results of the XG-Boost model in an embodiment of the present invention; Fig.13 (a) is a schematic diagram of the density distribution results of MSE. Fig.13 (b) is a schematic diagram of the proportion of different intervals of MSE;

[0061] Fig.14 Schematic diagram of mean square error (MSE) result of CatBoost model in an embodiment of the present invention; Fig.14 (a) is a schematic diagram of the density distribution results of MSE. Fig.14 (b) is a schematic diagram of the proportion of different intervals of MSE;

[0062] Fig.15 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in January 2022 using the RandomForest model in an embodiment of the present invention;

[0063] Fig.16 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in February 2022 using the RandomForest model in an embodiment of the present invention;

[0064] Fig.17 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in March 2022 using the RandomForest model in an embodiment of the present invention;

[0065] Fig.18 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in April 2022 using the RandomForest model in an embodiment of the present invention;

[0066] Fig.19 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in May 2022 using the RandomForest model in an embodiment of the present invention;

[0067] Fig. 20 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in June 2022 using the RandomForest model in an embodiment of the present invention;

[0068] Fig.21 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in July 2022 using the RandomForest model in an embodiment of the present invention;

[0069] Fig. 22 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in August 2022 using the RandomForest model in an embodiment of the present invention;

[0070] Fig.23This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in September 2022 using the RandomForest model in an embodiment of the present invention;

[0071] Fig.24 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in October 2022 by the RandomForest model in an embodiment of the present invention;

[0072] Fig.25 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in November 2022 by the RandomForest model in an embodiment of the present invention;

[0073] Fig.26 This is a schematic diagram of the prediction results of geothermal data with a spatial resolution of 30 meters in December 2022 by the RandomForest model in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0075] The present invention proposes a method for estimating land surface temperature. First, low spatial resolution data that can provide long time series support are selected as input data to ensure the long time series of predicted land surface temperature. Then, a variety of underlying surface information related to land surface temperature is integrated to ensure the high spatial resolution of the prediction model. Finally, a variety of machine learning models are used to construct a high temporal and spatial resolution land surface temperature estimation model to improve the temporal and spatial resolution of existing land surface temperature data. Taking a certain city as the research object, long-time series low spatial resolution land surface temperature data (MODIS daily land surface temperature data) and high spatial resolution underlying surface information (NDVI and land cover data) with strong correlation between the spatiotemporal distribution of land surface temperature are used as input variables, and high temporal and high spatial resolution land surface temperature data are used as output variables. By constructing different machine learning land surface temperature models and comparing their estimation accuracy differences, the temporal and spatial resolution of land surface temperature data can be improved. Figure 1 As shown, the specific steps include:

[0076] S1. Data preparation: Obtain long-term, low-spatial-resolution surface temperature data and high-spatial-resolution underlying surface data, as well as high-spatial-resolution surface temperature data obtained in the 10th band of the Landsat8 OLI.

[0077] S1.1. Acquisition of long-term, low-spatial-resolution land surface temperature data: The long-term, low-spatial-resolution land surface temperature data selected in the present invention is a daily land surface temperature data set (LST990m) with a spatial resolution of 1000 meters. The data is downloaded from the MOD11A1 data with a cloud cover of less than 0.5% in the NASA database.

[0078] S1.2. Obtaining high spatial resolution underlying surface data: The high spatial resolution underlying surface data selected in the present invention include NDVI data with a spatial resolution of 30 meters (NDVI30m) and land cover data with a spatial resolution of 30 meters (2022Type30m). The NDVI data is selected from the Landsat8-9OLI / TIRS Collection2Level-2 dataset with a cloud coverage rate of less than 5% in the USGS database, and the NDVI is calculated using the SR_B4.TIF and SR_B5.TIF band data in the Level-2Surface Reflectance Bands (13 files). The land cover data uses the national 30-meter land cover data released by the team of Professors Yang Jie and Huang Xin of Wuhan University.

[0079] S1.3. Surface temperature data with high spatial resolution: The surface temperature data with a spatial resolution of 30 meters obtained in the present invention is used for parameter adjustment and accuracy verification in the process of building the surface temperature model. The data is selected from the Landsat8-9OLI / TIRS Collection2Level-2 dataset with a cloud coverage rate of less than 5% in the USGS database, and the 30-meter spatial resolution surface temperature data (LSTY30m) is calculated using Level-2SurfaceTemperatureBands (14files).

[0080] S2. Data preprocessing.

[0081] S2.1. Use the Level-2 Surface Temperature Bands (14 files) data in the Landsat8-9OLI / TIRS Collection2 Level-2 dataset to calculate the surface temperature data (LSTY30m) with a spatial resolution of 30 meters. The calculation formula is as follows:

[0082] LST=0.00341802×B 10 +149

[0083] Specific multiplication factor B 10 This can be found in the data manual published by USGS.

[0084] S2.2. Use the Level-2 Surface Reflectance Bands (13 f.les) SR_B4.TIF and SR_B5.TIF bands in the Landsat8-9OLI / TIRS Collection2Level-2 dataset to calculate the NDVI data with a spatial resolution of 30 meters. The calculation formula is as follows:

[0085]

[0086] Wherein, B4 is the red light band of Landsat8 OLI, and B5 is the near-infrared band.

[0087] S2.3. First, the daily surface temperature dataset with a spatial resolution of 1000 meters (LST990m) was scaled to make the unit uniform to K (Kelvin). Secondly, the data was reprojected to make the projection method consistent with Landsat8-9OLI / TIRS. The data was again clipped by sliding window (i.e., a rectangular area was determined as the scope of the study area, and all data were clipped according to the rectangular range), and its spatial resolution was resampled to 30m, so that the spatial resolution of all input and output data was unified to 30 meters.

[0088] LST 990m =0.02×Data 990m

[0089] Where Data 990m is the extracted original data pixel value.

[0090] For example, the spatial resolution of MODIS is lower than that of Landsat, so the area covered by a single MODIS pixel contains multiple Landsat pixel data, resulting in the inability to fully align data with different resolutions. Therefore, a region extraction method is proposed. First, all data projections are unified to avoid the inability to correctly match pixels due to different projections. Secondly, the MODIS data is resampled to an integer multiple of the Landsat data resolution (990m), and then a single pixel is divided into 33 pixels with the same value, so that the resolution of the MODIS data pixel is consistent with that of the Landsat data (all 30m), achieving the purpose of unifying the spatial position and resolution of the data. Then, the sliding window processing method is used for unification: first, a rectangular area is determined as the study area, and the coordinates of the upper left corner point of the rectangle and the pixel index containing the point are obtained; secondly, the number of rows and columns of pixels required for the rectangular area is determined (the number of rows and columns is: 2343, 2112); finally, different data are cropped by the pixel index and the number of rows and columns in the upper left corner. The above operations unify the projection, spatial position and total number of pixels of all data, reducing the model training error.

[0091] S3. The surface temperature estimation model is obtained by training the random forest model with the high spatial resolution surface temperature data obtained from the 10th band of Landsat8 OLI as label data and the new surface temperature data and underlying surface data after consistent processing as feature data.

[0092] S3.1. The preprocessed NDVI_30m, Type_30m, and LST_990m data will be used as feature data, and LST_30m will be used as label data. The surface temperature calculation formula with high temporal and spatial resolution can be expressed as:

[0093] Y=AX+B

[0094] In the formula, Y is the high temporal and high spatial resolution surface temperature data that needs to be calculated, A is the trained integrated machine learning model, B is the fitting constant, and X is the collection of three variables: NDVI_30m, Type_30m, and LST_990m, that is:

[0095] X = [x1, x2, x3]

[0096] x1 and x2 are monthly NDVI data (NDVI_30m) and annual land cover data (Type_30m), respectively. x3 is daily MODIS 1km land surface temperature data (LST_990m). In the high-precision land surface temperature data prediction for the jth day of the i-th month of the t-th year, it is expressed as:

[0097] Y tij =K[NDVI ti , LCD t , MODIS tij ]

[0098] Among them, Y tij is the 30m surface temperature data of the jth day of the ith month of the tth year, K is the integrated machine learning algorithm obtained by training, NDVI ti is the NDVI data of the ith month of the tth year, LCD t is the land cover data for year t, MODIS tij It is the 1km surface temperature data of the jth day of the ith month of the tth year.

[0099] S3.2. Construction of surface temperature estimation model.

[0100] S3.2.1. Use the following three machine learning algorithms, including (1) XG-Boost; XG-Boost is the abbreviation of ExtremeGradient Boosting. It is an integrated machine learning algorithm based on decision trees and an improved version of GBDT gradient boosting tree. The two have the same idea, but XG-Boost has made many optimizations, such as using the second-order Taylor formula expansion to optimize the loss function, improve the calculation accuracy, use the regularization term to simplify the model (parameter: reg_lambda), avoid overfitting, use the Blocks storage structure, and can perform parallel calculations (parameter device). (2) CatBoost; CatBoost is similar to XG-Boost and is also an improved version of GBDT. Its characteristics are that the algorithm has fewer parameters, supports categorical variables and has high accuracy. The accuracy of the algorithm can be further improved by modifying the parameters. Compared with the XG-Boost algorithm, which is also a gradient boosting tree, CatBoost has more adjustable parameters, which is convenient for optimization according to one's own needs. (3) RandomForest: Random Forest is an ensemble learning method based on decision trees. It improves the accuracy and stability of the model by building multiple decision trees and aggregating the results. Similar to XG-Boost, the random forest algorithm also uses the idea of ​​ensemble learning, but it introduces randomness when building decision trees. It uses replacement sampling when selecting data points and introduces random selection when selecting split attributes, so as to increase the diversity of the model and reduce the risk of overfitting.

[0101] S3.2.2, according to the feature data and label data, different algorithms select feature data through different rules to calculate a result, and adjust the parameters or scoring rules in the model according to the difference between the label data and the result for the next round of calculation, so as to obtain a result closer to the label data, and obtain the parameters closest to the label data through continuous iteration. The random forest model with the best surface temperature prediction accuracy is obtained by using different tree splitting rules. The model parameter adjustment results are shown in Table 1.

[0102] Table 1 Results of tuning the main parameters of the random forest model

[0103]

[0104] S4. Model accuracy assessment: When comparing the performance of different surface temperature estimation models, six indicators are introduced: mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE), determination coefficient (R) 2 , Adjusted R 2 , mean square error MSE, as follows:

[0105] (1)MAPE (Mean absolute percentage error): MAPE is the mean absolute percentage error, which measures the difference between the predicted value and the actual value. The calculation formula is:

[0106]

[0107] where y i is the actual value of the surface temperature, y i is the predicted value of the surface temperature, and n is the number of samples of the surface temperature.

[0108] (2) RMSE (Root Mean Square Error): RMSE is the root mean square error, which is the square root of MSE (mean square error). RMSE measures the deviation between the predicted value and the actual value, and the calculation formula is:

[0109]

[0110] (3) MAE (Mean Absolute Error): MAE is the mean absolute error, which is the average of the absolute differences between the predicted value and the actual value. The calculation formula is as follows:

[0111]

[0112] (4)R 2 (R-squared or Coefficient of Determination):R 2 is the coefficient of determination, which measures the proportion of the variability explained by the model to the total variability, R 2 The value of is between 0 and 1. The value close to 1 means that the model can well explain the variability of the data, that is, the model effect is better. The calculation formula is as follows:

[0113]

[0114] in is the average surface temperature.

[0115] (5)Adjusted R 2 : Adjusted R 2 It is R 2 It takes into account the number of independent variables in the model and penalizes the addition of multiple irrelevant variables to the model. 2 Than the normal R 2 More rigorous and more suitable for comparing models with different numbers of independent variables, the calculation formula is as follows:

[0116]

[0117] Where R 2 is the coefficient of determination and p is the number of independent variables.

[0118] (6) MSE (Mean Squared Error): MSE is the mean square error, which measures the average of the squares of the difference between the predicted value and the actual value. MSE is a commonly used indicator to measure the prediction performance of a model, especially in regression analysis. A high MSE indicates that the model prediction is significantly different from the actual result. The calculation formula is as follows:

[0119]

[0120] The present invention selects input data supported by a long time series for surface temperature estimation, thereby ensuring the continuity and reliability of the data at the source; when cleaning the data, sliding window prediction is performed, and a rectangle is drawn according to the required coordinates and length and width, which technically provides the possibility and convenience for splicing at a larger spatial scale; through regression analysis of specific high spatial resolution underlying surface information and long time series low spatial resolution surface temperature data and machine learning methods, a long time series, high spatial resolution and high precision surface temperature data estimation model can be obtained; due to the use of sliding window prediction based on geographic coordinates, this prediction method provides the possibility for large-scale data splicing; the present invention uses long time series low spatial resolution surface temperature data and high spatial resolution underlying surface information data as input variables through a specific data cleaning method, and then combines the adjusted different machine learning methods to establish a regression model for long time series low spatial resolution surface temperature data, uses the model to estimate high spatial resolution surface temperature data, and obtains a long time series high spatial resolution surface temperature estimation method with high precision, strong explanatory power and low overfitting according to six commonly used regression analysis accuracy evaluation methods.

[0121] To prove the practicality of the present invention, part of a city was selected as the study area, and June 16, 2022 was selected for data download (small cloud cover). According to the download method in the technical details, Landsat8OLI data, MODIS11A1 data, and land cover data were downloaded. After data screening, calculation, and cleaning, four types of available data were obtained. These four sets of data are all unified in geographic coordinates, projection, area, and number of pixels. The surface temperature map of the 30m study area ( Figure 2 Surface temperature map of the study area at 30m in the middle) and 990m in the study area ( Figure 3 990m study area surface temperature map), 30m NDVI ( Figure 4 30m NDVI), 30m land cover data ( Figure 5 30m land cover data).

[0122] The four variables are used as data to train the machine learning model, and their accuracy is verified using six accuracy verification methods. The density map is used to print R 2 , MAE, MSE reference model accuracy.

[0123] Goodness of fit of three different machine learning methods: XG-Boost, CatBoost and RandomForest 2 as follows: Figure 6 is the R of the RandomForest model 2 Results diagram, Figure 7 is the R of the XG-Boost model 2 Results diagram, Figure 8 is the R of the CatBoost model 2 Schematic diagram of the results.

[0124] The mean absolute error MAE of XG-Boost, CatBoost and RandomForest is as follows: Fig. 9 This is a schematic diagram of the mean absolute error MAE results of the RandomForest model; Fig. 9 (a) is a schematic diagram of the density distribution results of MAE. Fig. 9 (b) is a schematic diagram of the proportion of different MAE intervals; Fig.10 This is a schematic diagram of the mean absolute error (MAE) results of the XG-Boost model; Fig.10 (a) is a schematic diagram of the density distribution results of MAE. Fig.10 (b) is a schematic diagram of the proportion of different MAE intervals; Fig.11 This is a schematic diagram of the mean absolute error (MAE) results of the CatBoost model; Fig.11 (a) is a schematic diagram of the density distribution results of MAE. Fig.11 (b) is a schematic diagram of the proportion of different MAE intervals.

[0125] The mean square error (MSE) of XG-Boost, CatBoost and RandomForest is as follows: Fig.12 This is a schematic diagram of the mean square error MSE results of the RandomForest model; Fig.12 (a) is a schematic diagram of the density distribution results of MSE. Fig.12 (b) is a schematic diagram of the proportion of different intervals of MSE; Fig.13 is a schematic diagram of the mean square error (MSE) results of the XG-Boost model; 13 (a) is a schematic diagram of the density distribution results of MSE, Fig.13 (b) is a schematic diagram of the proportion of different intervals of MSE; Fig.14is a schematic diagram of the mean square error (MSE) result of the CatBoost model; 14 (a) is a schematic diagram of the density distribution result of MSE, Fig.14 (b) is a schematic diagram of the proportion of different MSE intervals.

[0126] According to the above machine learning accuracy evaluation indicators, the RandomForest machine learning method with the best performance is selected to predict the ground temperature data with a spatial resolution of 30 meters per month in 2022. The prediction results are as follows: Fig.15 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in January 2022 using the RandomForest machine learning method. Fig.16 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in February 2022 using the RandomForest machine learning method. Fig.17 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in March 2022 using the RandomForest machine learning method. Fig.18 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in April 2022 using the RandomForest machine learning method. Fig.19 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in May 2022 using the RandomForest machine learning method. Fig. 20 This is a schematic diagram of the prediction results of the ground temperature data with a spatial resolution of 30 meters in June 2022 using the RandomForest machine learning method. Fig.21 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in July 2022 using the RandomForest machine learning method. Fig. 22 This is a schematic diagram of the prediction results of the ground temperature data with a spatial resolution of 30 meters in August 2022 using the RandomForest machine learning method. Fig.23 This is a schematic diagram of the prediction results of the ground temperature data with a spatial resolution of 30 meters in September 2022 using the RandomForest machine learning method. Fig.24 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in October 2022 using the RandomForest machine learning method. Fig.25 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in November 2022 using the RandomForest machine learning method. Fig.26 This is a schematic diagram of the prediction results of ground temperature data with a spatial resolution of 30 meters in December 2022 using the RandomForest machine learning method.

[0127] The present invention can obtain a long-time series, high-spatial resolution and high-precision surface temperature data estimation model through regression analysis of specific high-spatial resolution underlying surface information and long-time series low-spatial resolution surface temperature data with machine learning methods; due to the use of sliding window prediction based on geographic coordinates, this prediction method makes it possible to achieve large-scale data splicing.

[0128] Based on the same inventive concept, the present invention also proposes a surface temperature estimation system, comprising:

[0129] The acquisition module is used to acquire a daily surface temperature dataset with low spatial resolution, an underlying surface dataset with high spatial resolution, and a surface temperature dataset with high spatial resolution acquired in the 10th band of Landsat8's OLI; the low spatial resolution ranges from 100 meters to 1000 meters, and the high spatial resolution ranges from within 30 meters.

[0130] The reprojection module is used to scale the low spatial resolution daily surface temperature dataset and reproject the scaled daily surface temperature dataset to obtain a new daily surface temperature dataset with the same resolution as the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI.

[0131] The processing module is used to use the sliding window clipping method to make the spatial resolution of the new daily surface temperature dataset and the underlying surface dataset consistent.

[0132] The model building module is used to use the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8's OLI as a label dataset, and the new daily surface temperature dataset and underlying surface dataset that have been processed consistently as feature datasets. The random forest model is trained through the label dataset and the feature dataset to obtain a surface temperature estimation model.

[0133] The estimation module is used to input the underlying surface data to be estimated and the new daily surface temperature data consistent with its spatial resolution into the surface temperature estimation model to obtain the daily surface temperature data with high spatial resolution.

[0134] Based on the same inventive concept, the present invention also proposes a computer device for estimating ground temperature, comprising: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of the ground temperature estimation method when executing the computer program.

[0135] Based on the same inventive concept, the present invention also proposes a readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the surface temperature estimation method.

[0136] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for estimating ground surface temperature, characterized in that: The following steps are involved: Obtain a daily surface temperature dataset with low spatial resolution, an underlying surface dataset with high spatial resolution, and a surface temperature dataset with high spatial resolution obtained in the 10th band of Landsat8 OLI; the low spatial resolution ranges from 100 meters to 1000 meters, and the high spatial resolution ranges from within 30 meters; The low spatial resolution daily surface temperature dataset is scaled and reprojected to obtain a new daily surface temperature dataset with the same resolution as the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI. The sliding window clipping method is used to make the spatial resolution of the new daily surface temperature dataset and the underlying surface dataset consistent; The high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI is used as the label dataset, and the new daily surface temperature dataset and underlying surface dataset after processing are used as the feature dataset. The random forest model is trained with the label dataset and the feature dataset to obtain the surface temperature estimation model. The underlying surface data to be estimated and the new daily surface temperature data with the same spatial resolution are input into the surface temperature estimation model to obtain the daily surface temperature data with high spatial resolution.

2. A method for estimating ground surface temperature according to claim 1, characterized in that: The low spatial resolution daily surface temperature dataset is obtained by selecting MOD11A1 data with cloud cover less than 0.5% from the NASA database.

3. A method for estimating ground surface temperature according to claim 1, characterized in that: The underlying surface dataset with high spatial resolution includes a normalized difference vegetation index NDVI dataset and a land cover dataset; the NDVI dataset is obtained by selecting the Landsat8-9OLI / TIRS Collection2Level-2 dataset with a cloud coverage rate of less than 5% in the USGS database, and the SR_B4.TIF and SR_B5.TIF band data in the Level-2Surface Reflectance Bands are used to calculate the NDVI, and the calculation formula is: Among them, B4 is the red light band of Landsat8OLI, and B5 is the near-infrared band.

4. A method for estimating ground surface temperature according to claim 1, characterized in that: The high spatial resolution surface temperature dataset obtained by the 10th band of the Lands at8 OLI is calculated by selecting the Landsat8-9OLI / TIRS Collection2Level-2 dataset with a cloud coverage rate of less than 5% in the USGS database and using Level-2Surface TemperatureBands to calculate the high spatial resolution surface temperature dataset LST. The calculation formula is: LST=0.00341802×B 10 +149 Among them, B 10 It is the 10th band in the Landsat8-9OLI / TIRS Collection2Level-2 dataset.

5. A method for estimating ground surface temperature according to claim 3, characterized in that: The surface temperature estimation model is expressed as: Y tij =K[NDVI ti ,LCD t ,MODIS tij ] Among them, Y tij is the daily surface temperature data with high spatial resolution on the jth day of the ith month of the tth year, K is the integrated machine learning algorithm obtained by training, and NDVI ti is the NDVI data of the ith month of the tth year, LCD t is the land cover data for year t, MODIS tij It is the new daily surface temperature data of the jth day of the ith month of the tth year after processing to be consistent with the spatial resolution of the underlying surface data.

6. A method for estimating ground surface temperature according to claim 1, characterized in that: It also includes the mean absolute percentage error MAPE, root mean square error RMSE, mean absolute error MAE, and determination coefficient R 2 , Adjusted R 2 The performance of the surface temperature estimation model is evaluated using six indicators: mean square error (MSE), The mean absolute percentage error MAPE expression is: where y i is the actual value of the surface temperature, y i is the predicted value of the surface temperature, n is the number of samples of the surface temperature; The root mean square error RMSE expression is: The mean absolute error MAE expression is: The coefficient of determination R 2 The expression is: in is the average surface temperature; The adjusted coefficient of determination Adjusted R 2 The expression is: Where R 2 is the coefficient of determination, p is the number of independent variables; The mean square error MSE expression is:

7. A surface temperature estimation system, characterized in that: include: An acquisition module is used to acquire a daily surface temperature dataset with low spatial resolution, an underlying surface dataset with high spatial resolution, and a surface temperature dataset with high spatial resolution acquired in the 10th band of Landsat8 OLI; the low spatial resolution ranges from 100 meters to 1000 meters, and the high spatial resolution ranges from within 30 meters; The reprojection module is used to scale the daily surface temperature dataset with low spatial resolution, and reproject the scaled daily surface temperature dataset to obtain a new daily surface temperature dataset with the same resolution as the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI; A processing module, used to process the spatial resolution of the new daily surface temperature dataset and the underlying surface dataset to be consistent by using a sliding window clipping method; The model building module is used to use the high spatial resolution surface temperature dataset obtained by the 10th band of Landsat8 OLI as the label dataset, and the new daily surface temperature dataset and underlying surface dataset after processing as the feature dataset, and train the random forest model through the label dataset and the feature dataset to obtain the surface temperature estimation model; The estimation module is used to input the underlying surface data to be estimated and the new daily surface temperature data consistent with its spatial resolution into the surface temperature estimation model to obtain the daily surface temperature data with high spatial resolution.

8. A computer device for estimating surface temperature, characterized in that: include: A memory, a processor and a computer program stored in the memory, wherein the processor implements the steps of the method for estimating ground surface temperature according to any one of claims 1 to 6 when executing the computer program.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the surface temperature estimation method according to any one of claims 1 to 6.

Images