A high spatio-temporal resolution land surface temperature reconstruction method in cloud and fog environment

Abstract

The application belongs to the field of land surface temperature reconstruction, and relates to a high spatiotemporal resolution land surface temperature reconstruction method in a cloud and fog environment, comprising the following steps: calculating a first reference land surface temperature and a first land surface temperature residual of a clear sky pixel under low spatial resolution according to a first land surface temperature annual variation model; obtaining seamless land surface reflectivity data under high spatial resolution; defining a relative solar radiation index based on high spatial resolution digital elevation model data; screening clear sky pixels participating in XGBoost model modeling; adopting an XGBoost model to respectively construct a land surface temperature annual variation model coefficient under low spatial resolution and a plurality of model relationships between daily land surface temperature residuals and land surface attributes; obtaining a high spatial resolution land surface temperature annual variation model coefficient and daily land surface temperature residual based on the plurality of model relationships constructed, and obtaining a high spatial resolution daily land surface temperature; and improving the accuracy and stability of the land surface temperature reconstruction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of land surface temperature reconstruction, and specifically discloses a high spatiotemporal resolution land surface temperature reconstruction method in a cloud and fog environment. BACKGROUND

[0002] Land surface temperature (LST) is a key parameter in the process of dynamic balance of land surface energy, and has important scientific research and application value in meteorology, hydrology, ecology and geology. At present, satellite observation has become the only way to obtain land surface temperature at the pixel scale due to its wide range of observation and regular revisit. However, it is difficult to meet the needs of high temporal resolution and high spatial resolution at the same time in the current satellite remote sensing of land surface temperature. Satellites with high temporal resolution usually have low spatial resolution, and satellites with high spatial resolution have long revisit periods, which makes it difficult to meet the needs of continuous and high-precision monitoring. In addition, the current remote sensing inversion of land surface temperature relies on the thermal infrared band, and thermal infrared remote sensing is sensitive to atmospheric conditions and is easily disturbed by clouds and fog. Especially in mountainous areas with complex terrain, the available land surface temperature observation data is extremely limited due to the influence of clouds and fog all year round, which further exacerbates the difficulty of land surface temperature application in mountainous areas.

[0003] At present, some studies have proposed cloud-free reconstruction and spatial downscaling methods for land surface temperature to alleviate the problem of data missing caused by cloud and fog obstruction or insufficient spatial resolution in remote sensing observation. However, most of these methods focus on solving a single problem and are difficult to fully cope with the complex situation of temporal and spatial information missing in areas severely disturbed by clouds and fog. In addition, the step-by-step processing strategy of "cloud-free reconstruction first, then spatial downscaling" or "downscaling first, then temporal interpolation" is prone to error accumulation during processing, which ultimately reduces the accuracy and stability of the land surface temperature reconstruction results.

[0004] Therefore, the application provides a high spatio-temporal resolution land surface temperature reconstruction method in a cloud and fog environment. SUMMARY

[0005] The application aims to provide a high spatio-temporal resolution land surface temperature reconstruction method in a cloud and fog environment, solve the problem, improve the accuracy and stability of the land surface temperature reconstruction result, and specifically provide the following technical scheme.

[0006] Further, step 2 includes: constructing a first annual surface temperature variation model, and calculating a first reference surface temperature using the first annual surface temperature variation model; calculating the daily first surface temperature residual based on the difference between the daily first surface temperature and the first reference surface temperature; the first surface temperature refers to the actual surface temperature obtained from low spatial resolution data.

[0007] Further, step 3 includes: filling in the missing parts of the low spatial resolution reflectance data to be fused to obtain seamless low spatial resolution reflectance filled data; fusing the seamless low spatial resolution filled data with high spatial resolution data using spatiotemporal fusion technology, and filling in the missing parts of the high spatial resolution reflectance data using the fused reflectance data to obtain high spatial resolution reflectance filled data; and using the high spatial resolution reflectance filled data as one of the surface attribute data under high spatial resolution.

[0008] Furthermore, the calculation formula for imputing missing data in the low spatial resolution data to be fused is as follows:

[0009] ;

[0010] In the formula, The surface reflectance representing the missing pixels. The surface reflectance representing the missing pixel in the preceding clear-sky pixel. The surface reflectance of the missing pixel in the subsequent clear-sky pixel. Represents the accumulated days of the year for missing pixels. and These represent the accumulated days of the year corresponding to the preceding and following clear sky pixels, respectively.

[0011] Further, step 4 includes: calculating the solar radiation received at the top of the atmosphere based on the solar constant;

[0012] Based on the hour angle, the times of sunrise and sunset are determined; based on the solar radiation received at the top of the atmosphere, the times of sunrise and sunset, and the solar incidence angle under mountainous terrain, the solar radiation received on rugged mountain surfaces is determined; based on the solar radiation received at the top of the atmosphere, the times of sunrise and sunset, and the solar zenith angle, the solar radiation received on flat surfaces is determined; based on the ratio of solar radiation received on rugged mountain surfaces to that received on flat surfaces, the relative solar radiation index is determined.

[0013] Furthermore, the formula for calculating the solar radiation received at the top of the atmosphere is as follows:

[0014] ;

[0015] The formula for calculating the time of sunrise and sunset is:

[0016] ;

[0017] ;

[0018] ;

[0019] The formula for calculating the solar radiation received by the rugged mountainous surface is:

[0020] ;

[0021] The formula for calculating the solar radiation received by the flat surface is:

[0022] ;

[0023] The formula for calculating the relative solar radiation index is:

[0024] ;

[0025] In the formula, represents the relative solar radiation index on the dth day of the year, and respectively represent the solar radiation received by the rugged mountainous surface and the flat surface; is the solar radiation received by the top of the atmosphere on the dth day of the year, cos represents the cosine function, dt represents the derivative with respect to time t, and t represents the time variable, is the solar zenith angle, is the solar incident angle under the mountainous terrain; and respectively represent the sunset and sunrise times of the sun; is the time angle corresponding to the sunrise and sunset of the sun; is the latitude; is the solar declination; d represents the date variable; represents the circular constant, is the solar constant, and cos represents the cosine function.

[0026] Further, step 5 comprises: determining the number of pixel thresholds of the clear-sky pixels participating in the XGBoost model training; taking the geometric center of the study area as the center, searching for the clear-sky pixels in the order of distance from short to long, obtaining a set of clear-sky pixels for participating in the XGBoost model training; and taking the distance between the center and the pixel farthest from the set of clear-sky pixels as the radius of the dynamic window.

[0027] Further, step 6 comprises: establishing a first model relationship between the first land surface attribute and the first land surface temperature residual at a low spatial resolution by an XGBoost model; the first land surface attribute comprises latitude, elevation, slope, sky view factor, small-scale self-heating factor, solar incidence angle, normalized vegetation index, land surface reflectivity, land use type and cross-polarization ratio; establishing a second model relationship between the second land surface attribute and the annual mean temperature parameter in the first land surface temperature annual variation model at a low spatial resolution by an XGBoost model; the second land surface attribute comprises latitude, elevation, slope, sky view factor, small-scale self-heating factor, relative solar index, annual mean normalized vegetation index, annual mean reflectivity and annual mean cross-polarization ratio; establishing a third model relationship between the third land surface attribute and the annual amplitude parameter in the first land surface temperature annual variation model at a low spatial resolution by an XGBoost model; the third land surface attribute comprises latitude, elevation, slope, relative solar index, annual mean normalized vegetation index, normalized vegetation index variation, annual mean reflectivity, annual mean cross-polarization ratio and cross-polarization ratio variation.

[0028] Further, the calculation formula of the sky view factor is:

[0029] ;

[0030] In the formula, SVF represents the sky view factor, Di represents the total number of divided directions, k represents the direction variable, represents the horizon occlusion angle in the kth direction; sin represents the sine function;

[0031] The calculation formula of the small-scale self-heating factor is:

[0032] ;

[0033] In the formula, SSP represents the small-scale self-heating factor, i represents the pixel variable, m and n are the pixel sizes of the coarse resolution and fine resolution DEM data respectively, and the size of the coarse resolution m is larger than that of the fine resolution n; represents the number of fine resolution pixels contained in the coarse resolution pixel, and respectively represent the slope of the coarse resolution pixel and the slope of the fine resolution pixel, and j represents the fine resolution pixel variable in the coarse resolution pixel;

[0034] The calculation formula of the cross-polarization ratio is:

[0035] ;

[0036] In the formula, q is the cross-polarization ratio, and respectively represent the backscattering coefficients of vv polarization and vh polarization in the SAR data.

[0037] Further, step 7 comprises: determining a plurality of high spatial resolution land surface property data based on the cloud-free land surface reflectance data at high spatial resolution and the high resolution land surface property resampling data; processing the plurality of high spatial resolution land surface property data through a plurality of model relationships to obtain a high spatial resolution land surface temperature annual variation model coefficient and a second land surface temperature residual; constructing a second land surface temperature annual variation model based on the high spatial resolution land surface temperature annual variation model coefficient, and calculating a second reference land surface temperature; and determining a high spatial resolution daily land surface temperature based on the sum of the second reference land surface temperature and the second land surface temperature residual.

[0038] The present application has the following advantages and beneficial effects:

[0039] The present application directly applies the relationship between the parameters in the LST residual and the ATC model and the land surface properties at low resolution to the high resolution land surface property data, thereby obtaining high resolution, cloud-free land surface temperature data at one time. Compared with the traditional "first cloud filling and then downscaling" method, the present application has a more concise process and better operability, while effectively reducing the error and improving the reconstruction accuracy and efficiency.

[0040] The present application sets a variable window to gradually filter out available clear-sky pixels as the input of the XGBoost model from the center of the study area outward, dynamically balances the number and spatial correlation of training data while ensuring that the model has sufficient training samples, and both ensures the sufficiency of the model input data and suppresses the interference of low correlation samples far from the study area on the model accuracy.

[0041] The present application defines a relative solar radiation index and uses it as an important variable for constructing the relationship between the land surface properties and the ATC model parameters, effectively considers the regulation of the terrain on the annual solar radiation reception, significantly enhances the representation ability between the land surface properties and the ATC model parameters, and further effectively improves the reconstruction accuracy of the high resolution LST. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The present application provides a flowchart of a high spatiotemporal resolution land surface temperature reconstruction method suitable for cloud and fog environment for the embodiments of the present application;

[0043] Figure 2 The present application provides a study area elevation map and a ground station position diagram for verification;

[0044] Figure 3a The present application provides an image of the parameter a in the ATC model in the study area at high spatial resolution;

[0045] Figure 3bA map of the parameter β in the ATC model in the study area under high spatial resolution is obtained;

[0046] Figure 4a A comparison map of the reconstructed land surface temperature and the ground station temperature is obtained for all cases;

[0047] Figure 4b A comparison map of the reconstructed land surface temperature and the ground station temperature is obtained for the case where the cloud cover is greater than 0.99;

[0048] Figure 5 A map of the average RMSE of XGBoost training under different dynamic window radii is obtained;

[0049] Figure 6 The low-resolution land surface temperature map of the study area before reconstruction is provided by the present application;

[0050] Figure 7 The high-resolution all-weather land surface temperature map of the study area after reconstruction is provided by the present application;

[0051] Figure 8 The RSRI map of the relative solar index of the study area is provided by the present application. DETAILED DESCRIPTION

[0052] To make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings of the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0053] Figure 1 The flowchart of the high-temporal and high-spatial resolution land surface temperature reconstruction method suitable for cloud and fog environments provided by the embodiments of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the high-temporal and high-spatial resolution land surface temperature reconstruction method suitable for cloud and fog environments disclosed by the present application includes the following contents:

[0054] Step 1: Collect remote sensing image data; the remote sensing image data includes low spatial resolution data and high spatial resolution data.

[0055] Step 2: Calculate the first reference land surface temperature and the first land surface temperature residual of the clear-sky pixel in each day at low spatial resolution according to the first land surface temperature annual variation model. The first land surface temperature annual variation model refers to the land surface temperature annual variation model constructed by low spatial resolution data. The first reference land surface temperature refers to the land surface temperature output by the first land surface temperature annual variation model at low spatial resolution. The first land surface temperature residual refers to the land surface temperature residual at low spatial resolution. For example, the difference between the land surface temperature fitted by the first land surface temperature annual variation model and the MOD11A1 land surface temperature.

[0056] In some embodiments, step 2 includes: constructing the first land surface temperature annual variation model, and calculating the first reference land surface temperature by the first land surface temperature annual variation model. The first land surface temperature residual is calculated based on the difference between the first land surface temperature and the first reference land surface temperature. The first land surface temperature refers to the actual land surface temperature obtained according to the low spatial resolution data.

[0057] Step 3: Fuse the low spatial resolution reflectance data and the high spatial resolution reflectance data to obtain seamless land surface reflectance data at high spatial resolution based on the ESTARFM (Enhanced STARFM, Enhanced STARFM) spatio-temporal fusion model.

[0058] In some embodiments, step 3 includes: filling the missing data in the low spatial resolution data to be fused to obtain low spatial resolution filling data. The low spatial resolution filling data and the high spatial resolution data are fused by the spatio-temporal fusion technology, and the missing part of the high spatial resolution data is filled by the reflectance data obtained by the fusion to obtain high spatial resolution filling data. The low spatial resolution filling data refers to the complete low spatial resolution data obtained by filling the missing data in the low spatial resolution data. The high spatial resolution filling data refers to the seamless high spatial resolution data obtained by filling the original high spatial resolution data using the synthesized high resolution data obtained by the ESTARFM. The high spatial resolution reflectance filling data is used as one of the land surface attribute data at high spatial resolution.

[0059] Step 4: Define the relative solar radiation index (RSRI) based on the high spatial resolution digital elevation model (DEM) data. The relative solar radiation index is used to evaluate the difference between the solar radiation received by the mountainous land surface and the solar radiation received by the flat land surface.

[0060] In some embodiments, step 4 comprises: calculating the solar radiation received by the top of the atmosphere based on the solar constant; determining the time of sunset and sunrise based on the hour angle; determining the solar radiation received by the rugged terrain of the mountainous area based on the solar radiation received by the top of the atmosphere, the time of sunset and sunrise, and the solar incident angle under the terrain of the mountainous area; determining the solar radiation received by the flat terrain based on the solar radiation received by the top of the atmosphere, the time of sunset and sunrise, and the solar zenith angle; and determining the relative solar radiation index based on the ratio of the solar radiation received by the rugged terrain of the mountainous area to the solar radiation received by the flat terrain.

[0061] Step 5: Using a dynamic window, screen the clear-sky pixels participating in the XGBoost model modeling, with the study area as the center. The dynamic window can be sorted according to the distance of each pixel from the center of the study area. The distance can refer to the distance between the geometric center of the study area and the geometric center of the pixel. For example, based on the MOD11A1 1km daily land surface temperature product, reconstruct the land surface temperature, and screen the clear-sky pixels according to the quality control identifier of MOD11A1. XGBoost models the two parameters related to the annual average temperature and the annual amplitude in the ATC model and the residual in the ATC fitting process, respectively. Because the cloud cover is different every day, the spatial distribution of the selected clear-sky pixels is also different, so the radius of the dynamic window is set according to the cloud cover of each day.

[0062] In some embodiments, step 5 comprises: determining the pixel threshold number of clear-sky pixels participating in the XGBoost model training; taking the geometric center of the study area as the center, searching for the training clear-sky pixels step by step from short to long distance, and obtaining a set of clear-sky pixels; and taking the distance between the center and the farthest pixel in the set of clear-sky pixels as the radius of the dynamic window. The pixel threshold number refers to the total number of clear-sky pixels participating in modeling. The training clear-sky pixels refer to the clear-sky pixels participating in the XGBoost model training. For example, search for available clear-sky pixels step by step outward according to the distance of the pixel from the geometric center of the study area, and take the distance between the farthest pixel from the center as the radius of the dynamic window.

[0063] Step 6: Use the XGBoost model (extreme gradient boosting model) to build a low spatial resolution land surface temperature annual variation model coefficient and a plurality of model relationships between daily land surface temperature residuals and land surface attributes. Use XGBoost as a regression model to establish the nonlinear relationship between the two parameters of the ATC model and the land surface attributes and the LST residuals and the land surface attributes. The two parameters of the ATC model include a and β, which represent the parameters related to the annual average LST and the LST amplitude, respectively.

[0064] In some embodiments, step 6 comprises: establishing, by the XGBoost model, a first model relationship between first land surface properties and the first land surface temperature residual at a low spatial resolution; the first land surface properties include latitude (Lat), elevation, slope (Slope), sky view factor (SVF), small-scale self-heating factor (SSP), solar elevation index (SEI), normalized difference vegetation index (NDVI), surface reflectance (SR, including red, green, blue, and near-infrared bands), land use type, and cross-polarization ratio (q);

[0065] establishing, by the XGBoost model, a second model relationship between second land surface properties and the annual mean temperature parameter in the first land surface temperature annual variation model at a low spatial resolution; the second land surface properties include latitude (Lat), elevation (Elv), slope (Slope), sky view factor (SVF), small-scale self-heating factor (SSP), relative solar index (RSRI), annual mean normalized difference vegetation index (aNDVI), annual mean reflectance (aSR, including red, green, blue, and near-infrared bands), and annual mean cross-polarization ratio (aq).

[0066] establishing, by the XGBoost model, a third model relationship between third land surface properties and the annual amplitude parameter in the first land surface temperature annual variation model at a low spatial resolution; the third land surface properties include latitude (Lat), elevation (Elv), slope (Slope), relative solar index (RSRI), annual mean normalized difference vegetation index (aNDVI), normalized difference vegetation index variation (vNDVI), annual mean reflectance (aSR, including red, green, blue, and near-infrared bands), annual mean cross-polarization ratio (aq), and cross-polarization ratio variation (vq).

[0067] Step 7: applying the plurality of model relationships to the high spatial resolution land surface property data, respectively, to obtain high spatial resolution land surface temperature annual variation model coefficients and daily high spatial resolution land surface temperature residuals, and thereby obtaining daily high spatial resolution land surface temperatures in the study area. For example, by obtaining the high spatial resolution ATC model coefficients, a second land surface temperature annual variation model is constructed, and the daily high spatial resolution LST is obtained by adding the daily high spatial resolution LST residual to the reference value of the daily high spatial resolution LST calculated by the second land surface temperature annual variation model. The second land surface temperature annual variation model is used to obtain the reference land surface temperature at a high spatial resolution (i.e., the second reference land surface temperature).

[0068] In some embodiments, step 7 includes: determining multiple high spatial resolution surface attribute data based on cloudless surface reflectance data at high spatial resolution and high-resolution surface attribute resampling data; processing the multiple high spatial resolution surface attribute data through multiple model relationships to obtain high spatial resolution annual surface temperature variation model coefficients and a second surface temperature residual; constructing a second surface temperature annual variation model based on the high spatial resolution annual surface temperature variation model coefficients and calculating a second reference surface temperature; determining the high spatial resolution daily surface temperature based on the sum of the second reference surface temperature and the second surface temperature residual. The second surface temperature residual refers to the high spatial resolution surface temperature residual. The second surface temperature annual variation model is used to calculate the high spatial resolution reference surface temperature. The second reference surface temperature refers to the high spatial resolution reference surface temperature.

[0069] Example 1

[0070] As an example, see the attached document. Figure 1 As shown, taking the Wanglang Nature Reserve in Sichuan Province as an example, this embodiment provides a method for high spatiotemporal resolution land surface temperature reconstruction under cloud and fog conditions. This includes constructing an annual land surface temperature variation model, acquiring high spatiotemporal resolution cloudless land surface reflectance data, defining the relative solar radiation index, dynamically filtering XGBoost model input pixels using a window, model training, and acquiring high spatial resolution cloudless land surface temperature. The study area and ground stations used for verification are shown in the attached figure. Figure 2 As shown.

[0071] Step 1: Acquire remote sensing image data, including high-resolution and low-resolution images. Low-spatial-resolution data includes thermal infrared data (MOD11A1), surface reflectance data (MOD09A1), and normalized difference vegetation index (NDVI) data (MOD13A2), which are resampled to 1 km. High-spatial-resolution data includes surface reflectance data (Landsat-8 / 9, Sentinel-2), DEM data (Alos), SAR data (Sentinel-1), and land classification data (CLCD), which are resampled to 30 m. Subsequently, the DEM, SAR, and land classification data are resampled to 1 km to meet the requirements for model building at low spatial resolution.

[0072] Step 2: Based on the Annual Temperature Cycle (ATC) model, calculate the reference LST and residuals for clear-sky pixels at low spatial resolution for each day. The daily surface temperature can be expressed as:

[0073] ;

[0074] In the formula, LST d is the LST of the d th day of the year, LST ref is the LST reference value calculated by the ATC model, LST res is the LST residual. The ATC model can be expressed as:

[0075] ;

[0076] In the formula, LST d is the LST of the d th day of the year, LST d is the LST of the d th day of the year, is the phase shift relative to the vernal equinox, N is the length of the annual cycle, usually set to 365 days, ignoring leap day, e represents the exponential function; NDVI(d) represents the normalized vegetation index of the d th day of the year; sin represents the sine function; d represents the date variable; LST d is the LST of the d th day of the year,

[0077] For this embodiment, the MOD11A1 land surface temperature product is used, the clear sky pixel observation data is screened through the quality control layer, and the MOD13A2 NDVI product is used to obtain the NDVI corresponding to the clear sky observation LST.

[0078] Step 3: Considering that the study area is affected by clouds and fog, there are serious missing data in the high-resolution land surface reflectivity data, the ESTARFM spatio-temporal fusion technology is used to fill the missing part of the high-resolution land surface reflectivity data. The high-resolution land surface reflectivity data uses Landsat-8 / 9 and Sentinel-2 data, and the low-resolution land surface reflectivity data used for fusion uses the MOD09A1 8-day composite reflectivity product. For the area seriously affected by cloud and fog, the missing data of the 8-day composite reflectivity product is still missing, and the missing data is filled before spatio-temporal fusion:

[0079] ;

[0080] In the formula, R miss represents the land surface reflectivity of the missing pixel, R miss represents the land surface reflectivity of the missing pixel, R miss represents the land surface reflectivity of the missing pixel, R miss represents the land surface reflectivity of the missing pixel, R miss represents the land surface reflectivity of the missing pixel, R miss represents the land surface reflectivity of the missing pixel,

[0081] For the ESTARFM spatio-temporal fusion method, low spatial resolution reflectance data corresponding to the target time and each group of high spatial resolution and low spatial resolution land surface reflectance images before and after the target time are needed. In view of the missing part of the Landsat-8 / 9 reflectance data, the ground surface reflectance data meeting the requirements are selected respectively for spatio-temporal fusion, and the fused reflectance data are used to fill the missing part of the original data, so as to obtain high spatial resolution cloud-free reflectance data every 8 days.

[0082] Step 4: In order to measure the influence of mountainous terrain on the ground surface temperature, the present application defines a relative solar radiation index (RSRI) to evaluate the difference between the solar radiation received by the mountainous ground surface and the solar radiation received by the flat ground surface:

[0083] ;

[0084] In the formula, represents the relative solar radiation index on the dth day of the year, and respectively represent the solar radiation received by the mountainous rugged ground surface and the flat ground surface, and the influence of the atmospheric transmittance is ignored, wherein:

[0085] ;

[0086] ;

[0087] In the formula, is the solar radiation received by the top of the atmosphere on the dth day of the year, cos represents the cosine function, dt represents the derivative with respect to time t, and t represents the time variable, is the solar zenith angle, is the solar incident angle under the complex terrain of the mountainous area; and respectively represent the sunset and sunrise times.

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] wherein, is the hour angle, tan denotes the tangent function, denotes the latitude, represents the solar declination, sin denotes the sine function, and d denotes the date variable; denotes the circle constant, is the solar constant, cos denotes the cosine function, represents the solar azimuth, slope and Aspect represent the slope and aspect of the ground, respectively. Thus, the relative solar radiation index RSRI of each pixel in the study area for one year can be expressed as:

[0098] ;

[0099] The relative solar index is calculated by ALOS digital elevation data in this embodiment, Figure 8 is the RSRI spatial distribution image of the study area calculated in this embodiment.

[0100] Step 5: Taking the geometric center of the study area as the center, the available clear-sky pixels are searched step by step outward according to the distance of the pixel from the study area, and the clear-sky pixels available for participating in the XGBoost model training reach a pre-set threshold number. The distance of the pixel farthest from the center is the radius of the dynamic window.

[0101] Step 6: The XGBoost model is used to build the model relationship between the low spatial resolution ATC model coefficients (annual mean temperature and annual amplitude), the daily LST residual and the surface properties in step 1, respectively.

[0102] For the daily LST residual, first, the relationship between the latitude (Lat), the elevation (Elv), the slope (Slope), the sky view factor (SVF), the small-scale self-heating factor (SSP), the solar incident angle (SEI), the normalized vegetation index (NDVI), the surface reflectance (SR, including the red band, the green band, the blue band and the near-infrared band), the land use type, the cross-polarization ratio (q) and the residual is established at the resolution of 1 km.

[0103] ;

[0104] wherein, SVF represents the sky view factor, Di indicates that Di directions are divided in total, k represents the direction variable, denotes the ground blocking angle in the kth direction; sin denotes the sine function.

[0105] ;

[0106] In the formula, SSP represents the small-scale self-heating factor, i represents the pixel variable, and m and n are the pixel sizes of the coarse-resolution and fine-resolution DEM data, respectively. The size of m in the coarse-resolution data is larger than the size of n in the fine-resolution data. This represents the number of fine-resolution pixels contained within a coarse-resolution pixel, and is a positive integer with a value of [value missing]. This indicates the number of fine-resolution pixels contained in a single coarse-resolution pixel. and Let represent the slope of the coarse-resolution pixel and the slope of the fine-resolution pixel, respectively, and j represent the fine-resolution pixel variable within the coarse-resolution pixel. For the ALOS DEM data used in this embodiment, the 12.5m resolution DEM data is resampled to 500m and 1km respectively as fine-resolution pixels and coarse-resolution pixels in the SSP calculation process.

[0107] ;

[0108] In the formula, q is the cross-polarization ratio, which is one of the parameters containing soil moisture information. and These represent the backscattering coefficients of vv and vh polarizations in SAR data, respectively.

[0109] In this embodiment, the DEM data comes from ALOS 12.5 m DEM data, the surface reflectance data comes from MOD09A1 surface reflectance product, the NDVI comes from MOD13A2 NDVI product, the land use data comes from 30 m CLCD data, and the SAR data comes from Sentinel-1. The spatial resolution of the above data is unified to 1 km through resampling, and XGBoost is used to establish a model relationship between these surface attributes and LST residuals at a 1 km spatial resolution.

[0110] For the ATC model parameter α, the relationships between latitude (Lat), elevation (Elv), slope (Slope), sky visibility factor (SVF), small-scale self-heating factor (SSP), relative solar index (RSRI), annual normalized vegetation index (aNDVI), annual reflectance (aSR, including red, green, blue, and near-infrared bands), annual cross-polarization ratio (aq), and α were established at a 1 km resolution. The annual normalized vegetation index, annual reflectance, and annual cross-polarization ratio were directly obtained by averaging the cross-polarization ratio data calculated by MOD13A2, MOD09A1, and Sentinel-1.

[0111] For the ATC model parameter β, the relationship between latitude (Lat), elevation (Elv), slope (Slope), relative solar index (RSRI), annual mean normalized difference vegetation index (aNDVI), normalized difference vegetation index variation (vNDVI), annual mean reflectivity (aSR, including red, green, blue, and near-infrared bands), annual mean cross-polarization ratio (aq), and cross-polarization ratio variation (vq) and β was established at a 1 km resolution. The normalized difference vegetation index and mean cross-polarization ratio were directly obtained by calculating the difference between the maximum and minimum values of the cross-polarization ratio data calculated from MOD13A2 and Sentinel-1.

[0112] The model relationships between the ATC model coefficients, daily LST residuals, and surface properties were established, respectively, and the established models were used for the corresponding high-resolution surface property data to obtain the ATC model coefficients and daily LST residuals at high resolution. In the surface property data, the reflectivity data and NDVI data were obtained from the results of ESTARFM data fusion, and the remaining surface property data was obtained by resampling the high-resolution surface property data. Figure 3a With the Figure 3b The spatial distribution of the high-resolution ATC model parameters a and β in the study area in 2022 is shown.

[0113] Step 6: Based on the high-resolution ATC model coefficients obtained in Step 5, the daily LST reference values in the study area were calculated, and combined with the daily LST residuals to obtain the high-spatial-resolution daily LST image in the study area.

[0114] From Figure 4a It can be seen that, by comparing with the measurement data of the three ground stations of Wanglang Station, the RMSE of the reconstructed high-resolution land surface temperature is 3.78 K. From Figure 4b It can be seen that, on the days with severe cloud cover (cloud cover greater than 0.99), by comparing with the measurement data of the three ground stations of Wanglang Station, the RMSE of the reconstructed high-resolution land surface temperature is 3.92 K, which is a small decrease in reconstruction accuracy compared to the annual reconstruction result. Figure 5 The average RMSE of using the XGBoost model to establish the relationship between the daily LST residual and the surface property under different dynamic window radii is shown. When the dynamic window radius is less than 100 km, the average RMSE of the XGBoost model and the RMSE between the measured temperature of the ground station and the reconstructed land surface temperature are both minimum values. As the cloud and fog interference gradually increases, the radius of the dynamic window also increases, and the average RMSE of the XGBoost model and the RMSE between the measured temperature of the ground station and the reconstructed land surface temperature also increase accordingly. Figure 6The 15th day of each month in 2022 MOD11A1 land surface temperature product image is displayed, before reconstruction, the research area is seriously interfered by cloud and fog, and there is a large amount of data loss. Figure 7 The land surface temperature after reconstruction on the 15th day of each month in 2022 is displayed, it can be found that the spatial resolution of the land surface temperature after reconstruction has been obviously improved, and the part caused by cloud and fog interference and leading to data loss has also been effectively filled.

[0115] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment, characterized in that, The method comprises the following steps: Step 1: collecting remote sensing image data; the remote sensing image data comprises low spatial resolution data and high spatial resolution data; Step 2: calculating a first reference land surface temperature and a first land surface temperature residual of a clear sky pixel under low spatial resolution according to a first land surface temperature annual variation model; the first land surface temperature annual variation model is a land surface temperature annual variation model constructed by using the low spatial resolution data; Step 3: fusing low spatial resolution land surface reflectivity data and high spatial resolution land surface reflectivity data based on a space-time fusion model to obtain seamless high spatial resolution land surface reflectivity data; Step 4: defining a relative solar radiation index based on high spatial resolution digital elevation model data; The relative solar radiation index is used to evaluate the difference between the solar radiation received by the mountainous land surface and the solar radiation received by the flat land surface; Step 5: taking the research area as the center, using a dynamic window to screen the clear sky pixels participating in the modeling of the XGBoost model; Step 6: using the XGBoost model to respectively construct a plurality of model relationships between the low spatial resolution land surface temperature annual variation model coefficients and the daily land surface temperature residual and the land surface attributes; Step 7: respectively applying the plurality of model relationships to the high spatial resolution land surface attribute data to obtain the high spatial resolution land surface temperature annual variation model coefficients and the daily land surface temperature residual, and obtaining the daily land surface temperature of high spatial resolution in the research area.

2. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 2 comprises: constructing a first land surface temperature annual variation model to obtain a first reference land surface temperature through the first land surface temperature annual variation model; Based on the difference between the daily first land surface temperature and the first reference land surface temperature, the first land surface temperature residual of each day is calculated; the first land surface temperature is the actual land surface temperature obtained according to the low spatial resolution data.

3. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 3 comprises: filling in the missing part of the low spatial resolution reflectivity data to be fused to obtain seamless low spatial resolution reflectivity filling data; The seamless low spatial resolution filling data and the high spatial resolution data are fused through the space-time fusion technology, and the missing part of the high spatial resolution reflectivity data is filled in through the reflectivity obtained by the fusion to obtain high spatial resolution reflectivity filling data; The high spatial resolution reflectivity filling data is used as one of the high spatial resolution land surface attribute data.

4. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 3, characterized in that, The calculation formula for filling in the missing data in the low spatial resolution data to be fused is: ； wherein, represents the surface reflectance of the missing pixel, represents the surface reflectance of the missing pixel at the previous clear-sky pixel, represents the surface reflectance of the missing pixel at the next clear-sky pixel, represents the day of the year of the missing pixel, and respectively represent the day of the year of the previous and next clear-sky pixels corresponding to the missing pixel.

5. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 4 comprises: calculating the solar radiation received by the top of the atmosphere based on the solar constant; determining the sunset and sunrise time based on the hour angle; determining the solar radiation received by the rugged mountainous land surface based on the solar radiation received by the top of the atmosphere, the sunset and sunrise time and the solar incident angle under the mountainous terrain; determining the solar radiation received by the flat land surface based on the solar radiation received by the top of the atmosphere, the sunset and sunrise time and the solar zenith angle; determining the relative solar radiation index based on the ratio of the solar radiation received by the rugged mountainous land surface to the solar radiation received by the flat land surface.

6. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 5, characterized in that, The calculation formula of the solar radiation received by the top of the atmosphere is: ； The calculation formula of the sunset and sunrise time is: ； ； ； The formula for calculating the solar radiation received by a mountainous rugged surface is: ； The formula for calculating the solar radiation received by a flat surface is: ； The formula for calculating the relative solar radiation index is: ; wherein Rsdrepresents the relative solar radiation index on the dth day of a year, and Rsdand Rsprespectively represent the solar radiation received by the mountainous rugged surface and the flat surface; Rsdrepresents the solar radiation received by the top of the atmosphere on the dth day of a year, cos represents the cosine function, dt represents the derivative with respect to time t, and t represents the time variable, Rsdrepresents the solar zenith angle, Rsdrepresents the solar incident angle under the mountainous terrain; and Rsdand Rsprespectively represent the solar radiation received by the mountainous rugged surface and the flat surface; Rsdrepresents the time angle corresponding to the sunrise and sunset of the sun; Rsdrepresents the latitude; Rsdrepresents the solar declination; d represents the date variable; Rsdrepresents the circular constant, Rsdrepresents the solar constant, and cos represents the cosine function.

7. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 5 includes: Determining the number of pixel thresholds for clear-sky pixels participating in XGBoost model training; Taking the geometric center of the study area as the center, searching for clear-sky pixels in the order of distance from short to long, obtaining a set of clear-sky pixels for participating in XGBoost model training; Taking the distance between the center and the farthest pixel in the set of clear-sky pixels as the radius of the dynamic window.

8. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 6 includes: Establishing a first model relationship between the first land surface attribute and the first land surface temperature residual at low spatial resolution through the XGBoost model; the first land surface attribute includes latitude, elevation, slope, sky visibility factor, small-scale self-heating factor, solar incident angle, normalized vegetation index, land surface reflectivity, land use type, and cross-polarization ratio; Establishing a second model relationship between the second land surface attribute and the annual average temperature parameter in the first land surface temperature annual variation model at low spatial resolution through the XGBoost model; the second land surface attribute includes latitude, elevation, slope, sky visibility factor, small-scale self-heating factor, relative solar index, annual average normalized vegetation index, annual average reflectivity, and annual average cross-polarization ratio; Establishing a third model relationship between the third land surface attribute and the annual amplitude parameter in the first land surface temperature annual variation model at low spatial resolution through the XGBoost model; the third land surface attribute includes latitude, elevation, slope, relative solar index, annual average normalized vegetation index, normalized vegetation index change, annual average reflectivity, annual average cross-polarization ratio, and cross-polarization ratio change.

9. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 8, characterized in that, The formula for calculating the sky visibility factor is: ； In the formula, SVF represents a sky view factor, Di represents the total number of divided directions, k represents a direction variable, represents the horizontal obscuration angle in the kth direction; and sin represents a sine function. The formula for calculating the small-scale self-heating factor is: ； In the formula, SSP represents the small-scale self-heating factor, i represents the pixel variable, and m and n are the pixel sizes of the coarse-resolution and fine-resolution DEM data, respectively. The size of m in the coarse-resolution data is larger than the size of n in the fine-resolution data. This indicates the number of fine-resolution pixels contained within a coarse-resolution pixel. and represents the slope of the coarse resolution pixel and the slope of the fine resolution pixel, respectively, and j represents the fine resolution pixel variable in the coarse resolution pixel; The formula for calculating the cross-polarization ratio is: ； where q is the cross-polarization ratio, and denote the backscatter coefficients for vv and vh polarization in the SAR data, respectively.

10. The method for high spatio-temporal resolution land surface temperature reconstruction in cloud and fog environment according to claim 1, characterized in that, Step 7 includes: Based on the cloud-free land surface reflectivity data at high spatial resolution and the high-resolution land surface attribute resampling data, determine multiple high spatial resolution land surface attribute data; Process the multiple high spatial resolution land surface attribute data through the multiple model relationships to obtain the high spatial resolution land surface temperature annual variation model coefficients and the second land surface temperature residual; Based on the high spatial resolution land surface temperature annual variation model coefficients, construct the second land surface temperature annual variation model and calculate the second reference land surface temperature; Based on the sum of the second reference land surface temperature and the second land surface temperature residual, determine the high spatial resolution daily land surface temperature.

Images