Bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellation

Abstract

The invention discloses a bidirectional multi-factor XGBoost surface temperature reconstruction method based on a virtual constellation. The method comprises the steps of data preparation, data preprocessing, surface temperature reconstruction, precision verification and the like. According to the method, the problem of obtaining the surface temperature of the virtual constellation is successfully solved through the bidirectional multi-factor XGBoost model, accurate reconstruction of the surface temperature of the virtual constellation is realized, the excellent performance of obtaining the surface temperature with high temporal-spatial resolution in a short time is shown, and the application range of remote sensing data is expanded.

Description

Technical Field

[0001] The present invention relates to a bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellation. Background Art

[0002] Land surface temperature is a key physical parameter with important significance in fields such as climatology, ecology, and forest fire monitoring. It plays a vital role in energy exchange and surface-atmosphere interactions, and also reflects physical processes at regional and global scales. Currently, relatively accurate observational data of land surface temperature can be obtained at point scales through ground-based instruments (such as infrared thermometers and radiometers). Using satellite observation data (such as MODIS, Landsat, and Sentinel-3), remote sensing technology provides a reliable means for estimating land surface temperature at regional scales. However, due to the inherent contradiction between high spatial resolution and high temporal resolution of satellite thermal infrared sensors, it is difficult for a single satellite sensor to accurately estimate land surface temperature at high spatiotemporal resolution. Even if the frequency of land surface temperature estimation is increased through multi-source remote sensing data, differences in spectral, spatial, and radiometric resolutions may still affect the reliability of time series land surface temperature analysis.

[0003] One way to address this problem is to combine sensors with similar characteristics into so-called virtual constellations. Virtual constellations have been applied in monitoring ecological and environmental variables, such as land cover and chlorophyll-a. Therefore, the use of virtual constellations is expected to improve the spatiotemporal resolution of surface temperature. For example, Landsat 8 / 9 data has a medium-high spatial resolution of 100-meter thermal infrared band and 30-meter multispectral band, which can be used to estimate surface temperature with a 16-day revisit period. Sentinel-2A / 2B has a medium-high spatial resolution of 10-meter multispectral band and a relatively frequent revisit capability, but is limited by the lack of thermal infrared band information. Therefore, if the thermal infrared information of Sentinel-2A / 2B can be reconstructed, the joint observations of Landsat 8 / 9 and Sentinel-2A / 2B will be able to achieve frequent revisits of 2-3 days at medium-high spatial resolution, thereby inverting surface temperature at high spatiotemporal resolution.

[0004] Recently, the National Aeronautics and Space Administration (NASA) released a virtual constellation multispectral reflectance dataset provided by the Harmonized Landsat and Sentinel-2 (HLS) project. The HLS project is a program initiated by NASA to obtain seamless surface reflectance data from the land imagers and multispectral instruments of Landsat-8 / 9 and Sentinel-2A / B satellites. In the HLS project, the multispectral reflectance and spatial resolution of Sentinel-2 have been adjusted to be consistent with Landsat 8 / 9.

[0005] Thermal infrared bands are crucial in obtaining surface temperature. Although Landsat-8 / 9 provides thermal infrared bands, their temporal resolution is low. On the contrary, Sentinel-2A / B has a higher temporal resolution but lacks thermal infrared bands. Summary of the invention

[0006] The object of the present invention is to provide a bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellation with good reliability.

[0007] The technical solution of the present invention is:

[0008] A bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellation is characterized by comprising the following steps:

[0009] Step 1: Data preparation

[0010] Step 2: Data preprocessing

[0011] Obtain HLS remote sensing images and Landsat 8 surface temperature data in the study area and perform necessary preprocessing; preprocessing includes image registration and image cropping; first, all HLS and Landsat 8 images are mutually registered to ensure that they can be accurately superimposed in the same geographic coordinate system; second, in order to reduce the amount of calculation, the vector boundary data of the study area is used to uniformly crop the registered images so that they only cover the study area;

[0012] Step 3: Surface temperature reconstruction

[0013] In order to ensure full coverage of the study area and prevent the loss of important information, the method of manually selecting sample points was adopted, and a variety of factors were considered to build the model; terrain parameters include digital elevation model (DEM), aspect (Aspect) and slope (Slope); atmospheric parameters include atmospheric water vapor (Water Vapor); radiation parameters involve shortwave downlink radiation (RSD); index parameters include normalized difference vegetation index (NDVI), normalized desert index (NDSI), normalized building index (NDBI) and normalized water index (NDWI); by comprehensively considering these factors, the surface temperature can be predicted more accurately; on this basis, based on the HLS L30 data of two known periods, two independent multi-factor XGBoost models were constructed for different land cover types, as shown below:

[0014] LST1=f1(factors F ) (1)

[0015] Among them, factorsF Including LC F NDVI F NDSI F NDWI F NDBI F 、DEM F 、Aspect F 、Slope F 、waterVapor F 、RSD F and land surface temperature LST1, which are the first period HLS L30 data used to construct the training set f1;

[0016] LST2=f2(factors B ) (2)

[0017] Among them, factors B Including LC B NDVI B NDSI B NDWI B NDBI B 、DEM B 、Aspect B 、Slope B 、waterVapor B 、RSD B and land surface temperature LST2, which are the HLS L30 data of the later period used to construct the training set f2;

[0018] The two obtained training sets were applied to train the HLS S30 factor;

[0019] LST F =f1(factors S ) (3)

[0020] LST B =f2(factors S ) (4)

[0021] Among them, factors S Including LC S NDVI S NDSI S NDWI S NDBI S 、DEM S 、Aspect S 、Slope S 、waterVapor S 、RSD Sis a predictor of HLS S30;

[0022] The final surface temperature is determined by the following formula:

[0023] LST=LST F P F +LST B P B (5)

[0024] Where P F and P B is the weight, and the known data farther from the prediction point in the prediction period are given a greater weight; P F and P B It can be calculated by formula (6)-(7):

[0025] P F =d / T (6)

[0026] P B =(Td) / T (7)

[0027] Where T is the repetition period of the HLS L30 satellite, which is 16 days; d represents the position of S30 within the L30 period;

[0028] The shortwave downlink radiation is calculated by formula (8):

[0029] R sd =(I0 / d 2 )cosθτ sw (8)

[0030] Where I0 is the solar constant, d is the distance between the sun and the earth, θ is the solar incidence angle; τ sw is the atmospheric transmittance, which can be calculated by the following formula (9):

[0031]

[0032] Among them, K t is the correction factor (0<K t ≤1.0), in clear sky K t is 1.0, in polluted air K t is 0.5; θ hor is the horizontal incident angle of the sun, W is the atmospheric water vapor content (mm), which can be calculated using formula (10):

[0033] W=0.14e0·P (10)

[0034]

[0035] Among them, q is the specific humidity, P is the near-surface air pressure;

[0036] The normalized vegetation index is calculated by formula (12):

[0037] NDVI=(ρ Nir -ρ Red ) / (ρ Nir +ρ Red ) (12)

[0038] Among them, ρ Nir and ρ Red are the reflectances of the near-infrared and red bands of HLS, respectively;

[0039] The normalized desert index is calculated by the following formula (13):

[0040] NSDI=(ρ Red –ρ coastal Aerosal ) / (ρ Red +ρ coastal Aerosal ) (13)

[0041] where ρ coastal Aerosal is the reflectivity of band 1 of HLS;

[0042] The normalized water index is calculated by the following formula (14):

[0043] NDWI=(ρ Green -ρ Nir ) / (ρ Green +ρ Nir ) (14)

[0044] where ρ Green is the reflectance of the green band of HLS;

[0045] The normalized building index can be calculated using the following formula (15):

[0046] NDBI=(ρ SWIR1 -ρ Nir ) / (ρ SWIR1 +ρ Nir ) (15)

[0047] where ρ SWIR1 is the reflectivity of the shortwave infrared band of HLS;

[0048] From the prediction results, the spatial distribution of HLS S30 surface temperature predicted by the bidirectional multi-factor XGBoost model is highly consistent with the HLS L30 surface temperature results. The temperature in the desert area is higher, usually between 325-335K, while the temperature in the oasis area is relatively low, about 305-315K. It can also be observed that the surface temperature shows an increasing and decreasing trend over time.

[0049] Step 4: Accuracy Verification

[0050] In order to evaluate the accuracy of the two-way multi-factor XGBoost model in predicting land surface temperature, ground station data were used to verify the land surface temperature results in the study area.

[0051] The accuracy of the surface temperature inversion results was evaluated by using indicators such as root mean square error, bias, coefficient of determination, and relative error. The validation work was carried out at three sites representing different underlying surface types: Daman (vegetation), Huazhaizi Desert (desert), and Zhangye Wetland (wetland). Overall, the S30 results of Daman, Huazhaizi Desert, and Zhangye Wetland performed best in terms of prediction accuracy. Compared with L30 and the overall results, S30 had the smallest deviation (-3.16 W / m 2 、-1.59W / m 2 、-3.63W / m 2 ), the RMS error is low (3.67W / m 2 , 4.12W / m 2 、3.95W / m 2 ), with the highest coefficient of determination (0.98, 0.92, and 0.98, respectively), and the relative error is also small (1.05%, 1.07%, and 1.20%, respectively). This shows that the consistency and accuracy between the prediction results of S30 and the measured values ​​at each station are higher. Although the deviation and root mean square error of the L30 results are relatively high, its prediction effect is still trustworthy and shows a certain degree of reliability. The overall result has a high coefficient of determination, indicating that the model has a good fit at different stations and a strong prediction ability. The results of the scatter plot further verify the consistency of the S30 results with the L30 results in terms of prediction accuracy and spatial distribution. The coefficient of determination of all three stations exceeds 0.92, indicating that the model has a high degree of fit. The prediction results are highly consistent with the surface temperature values ​​of the ground observation stations, further verifying the reliability of the model. These results also prove the reliability of the bidirectional multi-factor XGBoost model in predicting the surface temperature of the virtual constellation HLS.

[0052] The present invention predicts the HLS surface temperature of the virtual constellation with good reliability and is easy to implement. The present invention successfully solves the problem of obtaining the surface temperature of the virtual constellation through a bidirectional multi-factor XGBoost model, realizes the accurate reconstruction of the surface temperature of the virtual constellation, demonstrates the superior performance in obtaining the surface temperature with high spatiotemporal resolution in a short time, and expands the application scope of remote sensing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0054] Figure 1 is a roadmap for the method of the present invention;

[0055] Figure 2 is a location map of the study area provided by the present invention;

[0056] Figure 3 It is the research data provided by the present invention;

[0057] Figure 4 It is the verification site data provided by the present invention;

[0058] Figure 5 is the surface temperature inversion result provided by the present invention;

[0059] Figure 6 It is a scatter plot for evaluating the accuracy of the surface temperature inversion result provided by the present invention;

[0060] Figure 7 It is the accuracy evaluation index of the surface temperature inversion result provided by the present invention. DETAILED DESCRIPTION

[0061] A bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellation includes the following steps:

[0062] Step 1: Data preparation

[0063] This paper takes Zhangye Oasis and its surrounding desert as the experimental area, and the scope of the research area is as follows: Figure 2 The main data sources include satellite images provided by the Harmonized Landsatand Sentinel-2 (HLS) project, a collaboration between NASA and the U.S. Geological Survey (USGS). The project generates two data products at a resolution of 30 meters: the L30 data product based on Landsat 8 data and the S30 product based on Sentinel-2 data. In addition, it also includes Landsat 8 30-meter resolution surface temperature satellite images. Figure 3In order to verify the prediction results, the present invention selected three ground observation sites with different environmental conditions in the Heihe Ecohydrological Remote Sensing Experiment (HiWATER) as verification data. Figure 3 shown.

[0064] Step 2: Data preprocessing. After the data preparation is completed, the HLS remote sensing images and Landsat8 surface temperature data in the study area are preprocessed as necessary. The preprocessing steps include image registration and image cropping to ensure data consistency and computational efficiency. First, image registration is performed. In this study, all HLS images and Landsat 8 surface temperature images need to be registered with each other to ensure that they have consistent geographical locations in the same coordinate system. Next, image cropping is performed. Since the study area is limited to the main urban area of ​​Zhangye, in order to reduce the amount of calculation and storage requirements, the registered images are uniformly cropped. The cropping process is based on the vector boundary data of the study area to ensure that all processed images only cover the geographical area within the study area.

[0065] Step 3: Reconstruction of land surface temperature. In order to fully cover the study area and prevent the loss of important information, a manual selection method of sample points was adopted, taking into account multiple factors. Topographic parameters include digital elevation model (DEM), aspect and slope, which provide detailed underlying surface terrain information for the model. By introducing these topographic parameters, the model can more accurately consider the elevation changes, slope and direction of the surface, thereby improving the prediction accuracy of land surface temperature. Atmospheric parameters include atmospheric water vapor, which provides important information about atmospheric humidity. Atmospheric water vapor is an important factor affecting land surface temperature because it affects radiation transfer and heat exchange. By using atmospheric water vapor as model input, the regulation of the atmosphere on the land surface temperature can be more accurately considered, thereby improving the accuracy of land surface temperature prediction. Radiation parameters include shortwave downwelling radiation (RSD), which is the main source of energy and directly affects the surface energy balance. Shortwave downwelling radiation comes from the sun and plays a key role in the surface heating process. By incorporating shortwave downwelling radiation into the model, the impact of solar radiation on the surface can be more accurately simulated, thereby optimizing the estimation of land surface temperature. In addition, changes in shortwave downlink radiation may also be related to seasonal and temporal changes in surface temperature, providing important radiation information for surface temperature inversion. Index parameters include the normalized difference vegetation index (NDVI), the normalized desert index (NDSI), the normalized building index (NDBI), and the normalized water index (NDWI). These indices are used to evaluate different surface characteristics of the study area, where NDVI reflects vegetation coverage and health, NDSI is used to detect the distribution of desert areas, NDBI identifies buildings in urban areas, and NDWI is used to monitor the distribution and changes of water bodies. Comprehensive consideration of these index parameters can improve the accuracy of surface temperature prediction. On this basis, two independent multi-factor XGBoost models were constructed for different land cover types based on HLS L30 data from two known periods. As shown below:

[0066] LST1=f1(factors F ) (1)

[0067] Among them, factors F Including LC F NDVI F NDSI F NDWI F NDBI F 、DEM F 、Aspect F 、Slope F 、waterVapor F 、RSD Fand land surface temperature LST1, which are the first period HLS L30 data used to construct the training set f1;

[0068] LST2=f2(factors B ) (2)

[0069] Among them, factors B Including LC B NDVI B NDSI B NDWI B NDBI B 、DEM B 、Aspect B 、Slope B 、waterVapor B 、RSD B and land surface temperature LST2, which are the HLS L30 data of the later period used to construct the training set f2;

[0070] The two obtained training sets were applied to train the HLS S30 factor;

[0071] LST F =f1(factors S ) (3)

[0072] LST B =f2(factors S ) (4)

[0073] Among them, factors S Including LC S NDVI S NDSI S NDWI S NDBI S 、DEM S 、Aspect S 、Slope S 、waterVapor S 、RSD S is a predictor of HLS S30;

[0074] The final surface temperature is determined by the following formula:

[0075] LST=LST F P F +LST B P B (5)

[0076] Where P F and P Bis the weight, and the known data farther from the prediction point in the prediction period are given a greater weight; P F and P B It can be calculated by formula (6)-(7):

[0077] P F =d / T (6)

[0078] P B =(Td) / T (7)

[0079] Where T is the repetition period of the HLS L30 satellite, which is 16 days; d represents the position of S30 within the L30 period;

[0080] The shortwave downlink radiation is calculated by formula (8):

[0081] R sd =(I0 / d 2 )cosθτ sw (8)

[0082] Where I0 is the solar constant, d is the distance between the sun and the earth, θ is the solar incidence angle; τ sw is the atmospheric transmittance, which can be calculated by the following formula (9):

[0083]

[0084] Among them, K t is the correction factor (0<K t ≤1.0), in clear sky K t is 1.0, in polluted air K t is 0.5; θ hor is the horizontal incident angle of the sun, W is the atmospheric water vapor content (mm), which can be calculated using formula (10):

[0085] W=0.14e0·P (10)

[0086]

[0087] Among them, q is the specific humidity, P is the near-surface air pressure;

[0088] The normalized vegetation index is calculated by formula (12):

[0089] NDVI=(ρ Nir -ρ Red ) / (ρ Nir +ρ Red ) (12)

[0090] Among them, ρ Nir and ρ Redare the reflectances of the near-infrared and red bands of HLS, respectively;

[0091] The normalized desert index is calculated by the following formula (13):

[0092] NSDI=(ρ Red –ρ coastal Aerosal ) / (ρ Red +ρ coastal Aerosal ) (13)

[0093] where ρ coastal Aerosal is the reflectivity of band 1 of HLS;

[0094] The normalized water index is calculated by the following formula (14):

[0095] NDWI=(ρ Green -ρ Nir ) / (ρ Green +ρ Nir ) (14)

[0096] where ρ Green is the reflectance of the green band of HLS;

[0097] The normalized building index can be calculated using the following formula (15):

[0098] NDBI=(ρ SWIR1 -ρ Nir ) / (ρ SWIR1 +ρ Nir ) (15)

[0099] where ρ SWIR1 is the reflectivity of the shortwave infrared band of HLS;

[0100] From the prediction results, the spatial distribution of HLS S30 surface temperature predicted by the multi-factor XGBoost model is highly consistent with the HLSL30 surface temperature results. The temperature in the desert area is higher, usually between 325-335K, while the temperature in the oasis area is relatively low, about 305-315K. Figure 5 As shown in the figure, it can also be observed that the surface temperature shows an increasing and decreasing trend over time.

[0101] Step 4: Accuracy verification. In order to evaluate the accuracy of the multi-factor XGBoost model in predicting land surface temperature, this study used ground station data to verify the land surface temperature results in the study area. This paper used bias, root mean square error (RMSE), and determination coefficient (R 2) and relative error (RE) are used to evaluate the accuracy of surface temperature prediction results. Bias is used to evaluate the systematic error of the model, revealing the average difference between the predicted value and the actual observed value. The root mean square error reflects the standard deviation between the predicted value and the measured value, and can quantify the overall level of prediction error. The coefficient of determination indicates the consistency between the predicted value and the measured value, reflecting the degree of fit of the model to the actual data. The relative error is expressed in percentage form, which can more intuitively reflect the reliability of the prediction results and the performance of the model. These indicators comprehensively consider the accuracy and consistency of the prediction results, and help determine the reliability and effectiveness of the model in practical applications. By analyzing these accuracy indicators, the model can be further optimized and the accuracy of surface temperature prediction can be improved. The calculation formula is:

[0102]

[0103] The validation work was conducted at three sites with different underlying surface types: Daman (vegetation), Huazhaizi Desert (desert) and Zhangye Wetland (wetland). Figure 6 The scatter plots of surface temperature estimation results and ground observation data are shown, where (a), (d), (g) are from Daman Station, (b), (e), (h) are from Huazhaizi Desert Station, and (c), (f), (i) are from Zhangye Wetland Station. In the figure, (a), (b), (c) show the overall verification results, (d), (e), (f) are from L30 verification results, and (g), (h), (i) are from S30 verification results. Figure 7 The validation accuracy indicators of the three sites are listed, including the overall, L30 and S30 indicators. In general, the S30 results of Daman, Huazhaizi Desert and Zhangye Wetland performed best in terms of prediction accuracy. Compared with the L30 and overall results, the S30 had the smallest bias (-3.16W / m 2 、-1.59W / m 2 , -3.63W / m2), and the RMSE is low (3.67W / m 2 , 4.12W / m 2 、3.95W / m 2 ), R 2The highest (0.98, 0.92, and 0.98, respectively), and the RE is also smaller (1.05%, 1.07%, and 1.20%, respectively). This shows that the consistency and accuracy between the prediction results of S30 and the measured values ​​at various stations are higher. Although the bias and root mean square error of the L30 results are relatively high, especially at the Huazhaizi Desert and Zhangye Wetland stations, its prediction effect is still trustworthy and shows a certain degree of reliability. The determination coefficient of the overall result is high, indicating that the model has a good fit at different stations and a strong prediction ability. The results of the scatter plot further verify the consistency of the S30 results with the L30 results in terms of prediction accuracy and spatial distribution. The determination coefficients of all three stations are over 0.92, indicating that the model has a high degree of fit. The predicted results are highly consistent with the surface temperature values ​​of the ground observation stations, which further verifies the reliability of the model. These results also prove the reliability of the multi-factor XGBoost model in predicting the surface temperature of the virtual constellation HLS.

Claims

1. A bidirectional multi-factor XGBoost surface temperature reconstruction method based on virtual constellations, characterized by: The following steps are involved: Step 1: Data preparation Step 2: Data preprocessing Obtain HLS remote sensing images and Landsat 8 surface temperature data in the study area and perform necessary preprocessing; preprocessing includes image registration and image cropping; first, all HLS and Landsat 8 images are mutually registered to ensure that they can be accurately superimposed in the same geographic coordinate system; second, in order to reduce the amount of calculation, the vector boundary data of the study area is used to uniformly crop the registered images so that they only cover the study area; Step 3: Surface temperature reconstruction In order to ensure full coverage of the study area and prevent the loss of important information, the method of manual selection of sample points was adopted, and the model was constructed by comprehensively considering multiple factors; terrain parameters included digital elevation model, aspect and slope; atmospheric parameters included atmospheric water vapor; radiation parameters involved shortwave downlink radiation; index parameters included normalized vegetation index, normalized desert index, normalized building index and normalized water index; by comprehensively considering these factors, the surface temperature can be predicted more accurately; on this basis, based on the HLS L30 data of two known periods, two independent multi-factor XGBoost models were constructed for different land cover types, as shown below: LST1=f1(factors F ) (1) Among them, factors F Including LC F NDVI F NDSI F NDWI F NDBI F 、DEM F 、Aspect F 、Slope F 、water Vapor F 、RSD F and land surface temperature LST1, which are the first period HLS L30 data used to construct the training set f1; LST2=f2(factors B ) (2) Among them, factors B Including LC B NDVI B NDSI B NDWI B NDBI B 、DEM B 、Aspect B 、Slope B 、water Vapor B 、RSD B and land surface temperature LST2, which are the HLS L30 data of the later period used to construct the training set f2; The two obtained training sets were applied to train the HLS S30 factor; LST F =f1(factors S ) (3) LST B =f2(factors S ) (4) Among them, factors S Including LC S NDVI S NDSI S NDWI S NDBI S 、DEM S 、Aspect S 、Slope S 、water Vapor S 、RSD S is a predictor of HLS S30; The final surface temperature is determined by the following formula: LST=LST F P F +LST B P B (5) Where P F and P B is the weight, and the known data farther from the prediction point in the prediction period are given a greater weight; P F and P B It can be calculated by formula (6)-(7): P F =d / T (6) P B =(T-d) / T (7) Where T is the repetition period of the HLS L30 satellite, which is 16 days; d represents the position of S30 within the L30 period; The shortwave downlink radiation is calculated by formula (8): R sd =(I0 / d 2 )cosθτ sw (8) Where I0 is the solar constant, d is the distance between the sun and the earth, θ is the solar incidence angle; τ sw is the atmospheric transmittance, which can be calculated by the following formula (9): Among them, K t is the correction factor (0<K t ≤1.0), in clear sky K t is 1.0, in polluted air K t is 0.5; θ hor is the horizontal incident angle of the sun, W is the atmospheric water vapor content (mm), which can be calculated using formula (10): W=0.14e0·P (10) Among them, q is the specific humidity, P is the near-surface air pressure; The normalized vegetation index is calculated by formula (12): NDVI=(r Nir -r Red ) / (ρ Nir +r Red ) (12) Among them, ρ Nir and ρ Red are the reflectances of the near-infrared and red bands of HLS, respectively; The normalized desert index is calculated by the following formula (13): NSDI=(ρ Red –r coastalAerosal ) / (ρ Red +r coastalAerosal ) (13) where ρ coastalAerosal is the reflectivity of band 1 of HLS; The normalized water index is calculated by the following formula (14): NDWI=(ρ Green -r Nir ) / ( p Green +r Nir ) (14) where ρ Green is the reflectance of the green band of HLS; The normalized building index can be calculated using the following formula (15): NDBI=(ρ SWIR1 -r Nir ) / (ρ SWIR1 +r Nir ) (15) where ρ SWIR1 is the reflectivity of the shortwave infrared band of HLS; Step 4: Accuracy Verification In order to evaluate the accuracy of the two-way multi-factor XGBoost model in predicting land surface temperature, ground station data were used to verify the land surface temperature results in the study area.

Images