Day-to-day reconstruction method and device for land surface temperature based on MODIS remote sensing data

Abstract

This invention provides a method and apparatus for daily land surface temperature reconstruction based on MODIS remote sensing data. This method can independently reconstruct daily land surface temperature for any region under assumed clear-sky conditions on a personal terminal or cloud platform (such as Google Earth Engine). The model first uses nearest-neighbor linear interpolation to initially fill in missing values, primarily due to cloud pollution. Then, it narrows the missing range by weighted fusion of data from the same period of the preceding year. For existing noise points and overfitting issues, SG filtering from signal processing is introduced for smoothing and correction, ultimately generating a seamless daily 1km land surface temperature product. Verification results show that the product obtained by this invention can better reflect the true state of land surface temperature under assumed clear-sky conditions, and more effectively reflects the seasonal characteristics of land surface temperature after excluding weather interference, demonstrating greater reliability in long-term time-series analysis studies.

Description

Technical Field

[0001] This invention relates to the field of remote sensing technology, and provides a method and apparatus for reconstructing daily land surface temperature based on MODIS remote sensing data. Background Technology

[0002] Land surface temperature (LST) refers to the temperature measured on the Earth's surface after solar radiation reaches it and is reflected and absorbed. Unlike air temperature, which is commonly referred to as ambient temperature, LST represents the temperature at the interface between the land surface and the air, while LST represents the temperature of air currents at a height of approximately 1.5 meters above the surface. Despite these differences, both atmospheric and LST are barometers reflecting the condition of the atmosphere, lithosphere, biosphere, and other ubiquitous Earth systems. In addition to global temperature change, LST, as a direct driver of energy exchange between the land surface and the atmosphere, is widely used in research fields such as ecological environment monitoring, crop evapotranspiration estimation, and urban thermal environment distribution. As the most reliable method for obtaining regional or global LST products, thermal infrared remote sensing inversion possesses technical advantages that traditional methods, such as ground-based fixed-point observations, cannot match, and has long been the preferred method for obtaining high spatiotemporal resolution LST products. Thermal infrared remote sensing relies on thermal infrared sensors mounted on satellites to obtain LST through non-contact measurement.

[0003] Since 1960, approximately 80 types of thermal infrared radiometers have been carried on more than 240 satellites to acquire a range of Earth parameters, including surface temperature, cloud top temperature, atmospheric water vapor, and precipitation. Various algorithms based on the radiative transfer equation are relatively mature, such as single-channel, multi-channel, multi-angle, multi-temporal, and hyperspectral inversion algorithms. Although the surface temperature retrieved from thermal infrared radiometers has high spatial resolution and accuracy, it is easily affected by cloud contamination, resulting in an average data utilization rate of only one-third. Full-coverage surface temperature reconstruction products are mainly obtained through three methods: spatiotemporal data interpolation, multi-source data fusion, and physical model extrapolation. However, these methods all have some shortcomings. For example, the accuracy of data assimilation products is low in areas with sparse meteorological stations, such as western regions, leading to unsatisfactory local fusion results; spatiotemporal interpolation methods do not calculate the true surface temperature, and the results can be severely distorted when large areas or long-term data are missing. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and apparatus for reconstructing daily land surface temperature based on MODIS remote sensing data, achieving a seamless spatiotemporal effect for MODIS land surface temperature data.

[0005] According to one aspect of the present invention, a method for reconstructing daily land surface temperature based on MODIS remote sensing data is provided, comprising the following steps:

[0006] S1: Obtain MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels;

[0007] S2: Perform global phase compensation on the missing daily global imagery of MOD11A1 land surface temperature products to obtain continuous MOD11A1 land surface temperature products.

[0008] S3: Using the nearest neighbor method, linear interpolation is performed on the previous and next image sets of the missing images in the MOD11A1 surface temperature product, on a daily basis.

[0009] S4: By combining data from the same period in adjacent years, the reconstructed pixels are weighted and fused using the sinusoidal harmonic function variation law of surface temperature, thereby reducing the error in the linear interpolation process;

[0010] S5: Obtain MCD12Q1 land surface classification products, calculate the average land surface temperature of similar land features, and further fill in the missing land surface temperature pixels of similar land features.

[0011] S6: Use SG filtering to remove noise during the reconstruction process, and finally obtain the daily surface temperature reconstruction results.

[0012] Furthermore, in step S1, the preprocessing includes: conditional filtering, image mosaicking, study area cropping, pixel quality review, and other preprocessing.

[0013] Furthermore, step S1 specifically includes:

[0014] S1.1: Establish the region of interest, acquire the MOD11A1 land surface temperature product, and crop all images to the same region of interest;

[0015] S1.2: The quality of the original MOD11A1 land surface temperature product is further screened based on four indicators: cloud impact, data quality, emissivity error, and inversion error, to obtain qualified land surface temperature pixels.

[0016] Preferably, in step S1.2, the two thresholds of average surface temperature inversion error being less than or equal to 2K and average emissivity error being less than or equal to 0.04 are selected as the basis for screening qualified surface temperature pixels.

[0017] Furthermore, in step S2, the specific method for compensating for the global missing phase is as follows:

[0018] For missing MOD11A1 daily data, a near-term replica is performed, that is, the image closest to the missing time phase in the time series product is searched as the initial value for that period.

[0019] Further, in step S3, the linear interpolation process using the nearest neighbor method specifically includes:

[0020] By using the observed values ​​of clear-sky surface temperature pixels from adjacent time phases, a linear function is used to interpolate the missing pixels, thereby quickly filling in the missing surface temperature values.

[0021] Combining the randomness of short-term surface temperature changes with the pixel-level unit operation method of GEE, the linear interpolation process is performed pixel by pixel, and the introduced clear-sky surface temperature pixels are the closest to the missing pixels in time.

[0022] Predicting time As the dividing line, the interval from that time is... Images within a day are grouped into a set of images from adjacent time phases and divided into two groups, each containing a different time period. and ;

[0023] For the image set before the prediction time, the last bit compression method is used to integrate the latest clear sky surface temperature pixels at each location in that time period into one image.

[0024] For the image set after the predicted time, the first-to-last compression method is used to integrate the latest clear sky surface temperature pixels at each location during that time period into one image.

[0025] For the two compressed images, each pixel is associated with different temporal information. Based on the dependent variable of land surface temperature and the independent variable of time represented by the pixel, univariate linear interpolation is performed to calculate the land surface temperature of the pixel at the predicted time. The expression is as follows:

[0026]

[0027] in, For the predicted time Pixel values, and These represent the compressed pixel values ​​before and after the surface temperature band, respectively. and The time information contained in a pixel. The intercept of the univariate linear equation is given.

[0028] Furthermore, in step S4, the step of combining data from the same period of adjacent years and using the sinusoidal harmonic function variation law of surface temperature to perform weighted fusion of reconstructed pixels specifically includes:

[0029] For LST data from different years, an inverse temporal distance weighting method is used to weight and fuse multiple contemporaneous data before and after the prediction time. The expression is as follows:

[0030]

[0031] in, The weighted and merged surface temperature. and Represents the predicted time Surface temperature and weight, For the year, and Represents the predicted time Surface temperature and weight, This represents the maximum annual interval from the predicted year;

[0032] The weighting constraints are as follows:

[0033]

[0034] Further, in step S5, obtaining the MCD12Q1 land surface classification product, calculating the average land surface temperature of similar land features, and further filling in the missing land surface temperature pixels of similar land features specifically includes:

[0035] According to the IGBP classification standard of MCD12Q1 land cover use products, the land cover in the study area was divided into 17 categories. The average surface temperature of each category was calculated as follows:

[0036]

[0037] in, It is the average surface temperature of a certain type. This represents the total number of pixels for that land type. For this land category The surface temperature value of each pixel;

[0038] use Further fill in the missing LST pixels of similar land features.

[0039] Further, in step S6, the step of using SG filtering to remove noise during the reconstruction process specifically includes:

[0040] First, determine a radius as... The moving window incorporates all data within it into a set, with the time nodes of the time-series data contained within the window being... ,use The polynomial of degree applies to the window. Fitting is performed at each time point, and its polynomial expression is as follows:

[0041]

[0042] in, These are the filtered pixel values. For polynomial parameters;

[0043] Each time point within the window corresponds to a polynomial, totaling [number missing]. The number of pixels in the window Greater than polynomial parameters When the system of equations is solved, the parameters are obtained, and their expression is:

[0044]

[0045] Represented in matrix form as follows:

[0046]

[0047] in, This is the filtered LST value. The annual LST value used for filtering. For parameter matrices, It is a random error vector;

[0048] Fitting parameters Least square solution for:

[0049]

[0050] Model prediction or filtered value for:

[0051]

[0052] For the MOD11A1 time-series surface temperature product The window represents the time before and after the prediction. Time of day This represents the land surface temperature value after weighted merging and filling with land use averages. This represents the LST value after noise removal using SG filtering.

[0053] According to another aspect of the present invention, the present invention provides a daily surface temperature reconstruction device based on MODIS remote sensing data, comprising the following modules:

[0054] The preprocessing module is used to acquire MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels.

[0055] The global missing phase compensation module is used to perform global missing phase compensation on the single-day global missing image of MOD11A1 land surface temperature product to obtain continuous MOD11A1 land surface temperature product.

[0056] The nearest neighbor interpolation module is used to perform linear interpolation on a daily basis on the image sets before and after the missing images in the MOD11A1 surface temperature product using the nearest neighbor method.

[0057] The weighted fusion module is used to combine data from the same period of adjacent years and utilize the sinusoidal harmonic function variation law of surface temperature to perform weighted fusion of reconstructed pixels, thereby reducing errors in the linear interpolation process;

[0058] The missing pixel filling module is used to obtain MCD12Q1 land surface classification products, calculate the average land surface temperature of similar land features, and further fill in the missing land surface temperature pixels of similar land features.

[0059] The SG filtering module is used to remove noise during the reconstruction process, ultimately obtaining the daily surface temperature reconstruction results.

[0060] The technical solution provided by this invention has the following beneficial effects:

[0061] This invention enables independent reconstruction of land surface temperature under assumed clear-sky conditions for any time phase and region. The method first uses nearest-neighbor linear interpolation to initially fill in missing values, then further narrows the missing range through weighted fusion of data from the same period of the preceding year. For noise points and overfitting issues, SG filtering is introduced for smoothing and correction, ultimately generating a seamless land surface temperature product for 1km daily. Surface station observation data validates the proposed method's overall good performance. Cross-validation with similar products like GF (Gap-Filled) shows high similarity, but the proposed method is more robust for predicting large-area missing regions. Attached Figure Description

[0062] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0063] Figure 1 This is a flowchart of a method for reconstructing daily land surface temperature based on MODIS remote sensing data, provided by an embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram of the nearest neighbor linear interpolation in an embodiment of the present invention;

[0065] Figure 3 This is a time series diagram of MODIS clear sky LST at DM sites from 2018 to 2020 in an embodiment of the present invention;

[0066] Figure 4 This is a partial process diagram of time-domain reconstruction on July 15, 2020, in an embodiment of the present invention;

[0067] Figure 5This is a structural diagram of a daily surface temperature reconstruction device based on MODIS remote sensing data provided in an embodiment of the present invention. Detailed Implementation

[0068] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0069] refer to Figure 1 This invention provides a method for reconstructing daily land surface temperature based on MODIS remote sensing data, the method comprising the following steps:

[0070] S1: Obtain MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels;

[0071] Specifically, in this embodiment, the MOD11A1 land surface temperature product retrieved by the MODIS sensor carried by the Terra satellite is selected to establish the region of interest, and all images are cropped to the same range as the region of interest. The quality of the original data is further screened based on four indicators: cloud impact, data quality, emissivity error, and retrieval error. Preferably, the average retrieval error of land surface temperature is less than or equal to 2K and the average emissivity error is less than or equal to 0.04 as the basis for screening qualified land surface temperature (LST) pixels. In other embodiments, other threshold conditions can also be selected according to actual conditions and needs.

[0072] S2: Perform global phase compensation on the missing daily global imagery of MOD11A1 land surface temperature products to obtain continuous MOD11A1 land surface temperature products.

[0073] Specifically, the missing daily global imagery of the MOD11A1 land surface temperature product is replicated in advance. This is done by searching for the image closest to the missing phase in the time series product as the initial value for that period, in order to meet the requirement of daily data availability.

[0074] S3: Using the nearest neighbor method, linear interpolation is performed on the previous and next image sets of the missing images in the MOD11A1 surface temperature product, on a daily basis.

[0075] The specific process for linear interpolation using the nearest neighbor method is as follows: Figure 2 As shown, the specific steps are as follows:

[0076] The first step in missing pixel recovery is to use the observation values ​​of clear-sky surface temperature pixels from adjacent time phases to interpolate the missing pixels using a linear function, thereby quickly compensating for the LST values ​​of the missing pixels.

[0077] Although surface temperature exhibits a sinusoidal harmonic function variation pattern over long time, its short-term fluctuations are difficult to fit with a suitable function. Linear interpolation often uses multiple linear regression of known data from multiple periods to solve for unknowns, but this method may cause overfitting and lead to significant deviations in local predictions.

[0078] Combining the randomness of short-term surface temperature changes with the pixel-level unit operation method of GEE, the linear interpolation process of this invention is performed pixel by pixel, and the introduced clear-sky surface temperature pixels are the closest to the missing pixels in time.

[0079] Predicting time As the dividing line, the interval from that time is... Images within a day are grouped into a set of images from adjacent time phases and divided into two groups, each containing a different time period. and For the image set before the prediction time, a last-order compression method is used, integrating the latest clear-sky surface temperature pixels at each location within that time period into a single image. The reverse is applied to the image set after the prediction time. For the two compressed images, each pixel is associated with different temporal information. Based on the dependent variable of surface temperature and the independent variable of time represented by each pixel, univariate linear interpolation is performed to calculate the surface temperature of the pixel at the prediction time. The expression is as follows:

[0080]

[0081] in, For the predicted time Pixel values, and These represent the compressed pixel values ​​before and after the surface temperature band, respectively. and The time information contained in a pixel. The intercept of the univariate linear equation is given.

[0082] S4: By combining data from the same period in adjacent years, the reconstructed pixels are weighted and fused using the sinusoidal harmonic function variation law of surface temperature, thereby reducing the error in the linear interpolation process;

[0083] Due to the periodic variation in the intensity of solar radiation received by ground features, the surface temperature they represent generally follows a sinusoidal curve pattern. For example... Figure 3 As shown, taking the DM station of the Heihe River Basin Surface Processes Integrated Observation Network as an example, the LST moving average fitting curve of the underlying cornfields from 2018 to 2020 not only shows good linearity in the short term, but also highly matches the LST variation range of the same period over many years. The specific steps of S4 are as follows:

[0084] For LST data from different years, an inverse temporal distance weighting method is used to weight and fuse multiple contemporaneous data before and after the prediction time. The expression is as follows:

[0085]

[0086] in, The weighted and merged surface temperature. and Represents the predicted time Surface temperature and weight, For the year, and Represents the predicted time Surface temperature and weight, This represents the maximum annual interval from the predicted year;

[0087] The weighting constraints are as follows:

[0088]

[0089] S5: Obtain MCD12Q1 land surface classification products, calculate the average land surface temperature of similar land features, and further fill in the missing land surface temperature pixels of similar land features.

[0090] Specifically, S5 includes:

[0091] According to the IGBP classification standard of MCD12Q1 land cover use products, the land cover in the study area was divided into 17 categories. The average surface temperature of each category was calculated as follows:

[0092]

[0093] in, It is the average surface temperature of a certain type. This represents the total number of pixels for that land type. For this land category The surface temperature value of each pixel, using Missing values ​​are further filled.

[0094] S6: Use SG filtering to remove noise during the reconstruction process to ensure the stability of the time series, and finally obtain the daily surface temperature reconstruction results;

[0095] Step S6 specifically includes:

[0096] First, determine a radius as... The moving window incorporates all data within it into a set, with the time nodes of the time-series data contained within the window being... ,use The polynomial of degree applies to the window. Fitting is performed at each time point, and its polynomial expression is as follows:

[0097]

[0098] in, These are the filtered pixel values. For polynomial parameters;

[0099] Each time point within the window corresponds to a polynomial, totaling [number missing]. The number of pixels in the window Greater than polynomial parameters When the system of equations is solved, the parameters are obtained, and their expression is:

[0100]

[0101] Represented in matrix form as follows:

[0102]

[0103] in, This is the filtered LST value. The annual LST value used for filtering. For parameter matrices, It is a random error vector;

[0104] Fitting parameters Least square solution for:

[0105]

[0106] Model prediction or filtered value for:

[0107]

[0108] For the MOD11A1 time-series surface temperature product The window represents the time before and after the prediction. Time of day This represents the land surface temperature value after weighted merging and filling with land use averages. This represents the LST value after noise removal using SG filtering.

[0109] The reconstruction results of the daily surface temperature reconstruction method based on MODIS remote sensing data provided by this invention are as follows: Figure 4As shown, a) is the reconstruction result of some areas within the administrative region of China on July 15, 2020; b) is the surface temperature before rectangular fusion, where most areas have missing surface temperature values ​​due to cloud cover and other reasons; c) is the surface temperature image after processing with adjacent temporal linear interpolation proposed by S3, showing that most missing pixels have been filled; d) is the surface temperature image after weighted fusion of adjacent year data proposed by S4, where all missing pixels have been filled after processing; e) is the surface temperature after SG filtering, which removes noise during the reconstruction process to ensure temporal stability.

[0110] The following describes a daily land surface temperature reconstruction device based on MODIS remote sensing data provided by the present invention. The daily land surface temperature reconstruction device based on MODIS remote sensing data described below can be referred to in correspondence with the daily land surface temperature reconstruction method based on MODIS remote sensing data described above.

[0111] refer to Figure 5 The present invention provides a daily surface temperature reconstruction device based on MODIS remote sensing data, which specifically includes the following modules:

[0112] Preprocessing module 01 is used to acquire MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels.

[0113] The global missing phase compensation module 02 is used to perform global missing phase compensation on the single-day global missing image of MOD11A1 land surface temperature product to obtain continuous MOD11A1 land surface temperature product.

[0114] The nearest neighbor interpolation module 03 is used to perform linear interpolation on a daily basis on the image sets before and after the missing images of the MOD11A1 surface temperature product using the nearest neighbor method.

[0115] The weighted fusion module 04 is used to combine data from the same period of adjacent years and utilize the sinusoidal harmonic function variation law of surface temperature to perform weighted fusion of reconstructed pixels, thereby reducing errors in the linear interpolation process;

[0116] Missing pixel filling module 05 is used to obtain MCD12Q1 land surface classification products, calculate the LST mean of similar land features, and further fill the missing land surface temperature pixels of similar land features.

[0117] SG filter module 06 is used to remove noise during the reconstruction process using SG filtering, and finally obtain the daily surface temperature reconstruction results.

[0118] Each module of the daily land surface temperature reconstruction device based on MODIS remote sensing data corresponds one-to-one with each step of the daily land surface temperature reconstruction method based on MODIS remote sensing data, and is used to implement each step of the above method, which will not be described in detail here.

[0119] This invention provides a method and apparatus for daily land surface temperature reconstruction based on MODIS remote sensing data. This method can independently reconstruct the daily land surface temperature of any region under assumed clear-sky conditions on a personal terminal or cloud platform (such as Google Earth Engine). The model first uses nearest-neighbor linear interpolation to initially fill in missing values, primarily due to cloud pollution. Then, it narrows the missing range by weighted fusion of data from the same period of the preceding year. For existing noise points and overfitting issues, SG filtering from signal processing is introduced for smoothing and correction, ultimately generating a seamless 1km daily land surface temperature product. Verification results show that the product obtained by this invention can better reflect the true state of land surface temperature under assumed clear-sky conditions, and more effectively reflects the seasonal characteristics of land surface temperature after excluding weather interference, demonstrating greater reliability in long-term time-series analysis studies.

[0120] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0121] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. The use of the terms first, second, and third, etc., does not indicate any order and can be interpreted as identifiers.

[0122] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for reconstructing daily land surface temperature based on MODIS remote sensing data, characterized in that, Includes the following steps: S1: Obtain MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels; S2: Perform global phase compensation on the missing daily global imagery of MOD11A1 land surface temperature products to obtain continuous MOD11A1 land surface temperature products. S3: Using the nearest neighbor method, linear interpolation is performed on the previous and next image sets of the missing images in the MOD11A1 surface temperature product, on a daily basis. S4: Combining data from the same period in adjacent years, the reconstructed pixels are weighted and fused using the sinusoidal harmonic function variation pattern of surface temperature, thereby reducing errors in the linear interpolation process; specifically including: For surface temperature data from different years, an inverse temporal distance weighting method is used to weight and fuse multiple contemporaneous data before and after the prediction time. The expression is as follows: in, The weighted and merged surface temperature. and Represents the predicted time Surface temperature and weight, For the year, and Represents the predicted time Surface temperature and weight, This represents the maximum annual interval from the predicted year; The weighting constraints are as follows: ； S5: Obtain MCD12Q1 land surface classification products, calculate the average land surface temperature of similar land features, and further fill in the missing land surface temperature pixels of similar land features. S6: Use SG filtering to remove noise during the reconstruction process, and finally obtain the daily surface temperature reconstruction results.

2. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, In step S1, the preprocessing includes: conditional filtering, image mosaicking, study area cropping, and pixel quality review.

3. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, Step S1 specifically includes: S1.1: Establish the region of interest, acquire the MOD11A1 land surface temperature product, and crop all images to the same region of interest; S1.2: The quality of the original MOD11A1 land surface temperature product is further screened based on four indicators: cloud impact, data quality, emissivity error, and inversion error, to obtain qualified land surface temperature pixels.

4. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 3, characterized in that, In step S1.2, two thresholds are selected: the average inversion error of land surface temperature is less than or equal to 2K and the average emissivity error is less than or equal to 0.04 as the basis for screening qualified land surface temperature pixels, where K is the unit of temperature, Kelvin.

5. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, In step S2, the specific method for compensating for the global missing phase is as follows: For missing MOD11A1 daily data, a near-term replica is performed, that is, the image closest to the missing time phase in the time series product is searched as the initial value for that period.

6. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, In step S3, the linear interpolation process using the nearest neighbor method specifically includes: By using the observed values ​​of clear-sky surface temperature pixels from adjacent time phases, a linear function is used to interpolate the missing pixels, thereby quickly filling in the missing surface temperature values. Combining the randomness of short-term surface temperature changes with the pixel-level unit operation method of GEE, the linear interpolation process is carried out pixel by pixel, and the introduced clear sky surface temperature pixels are closest to the time corresponding to the missing pixels in terms of time phase. Predicting time As the dividing line, the interval from that time is... Images within a day are grouped into a set of images from adjacent time phases and divided into two groups, each containing a different time period. and ; For the image set before the prediction time, the last bit compression method is used to integrate the latest clear sky surface temperature pixels at each location in that time period into one image. For the image set after the predicted time, the first-to-last compression method is used to integrate the latest clear sky surface temperature pixels at each location during that time period into one image. For the two compressed images, each pixel is associated with different temporal information. Based on the dependent variable of land surface temperature and the independent variable of time represented by the pixel, univariate linear interpolation is performed to calculate the land surface temperature of the pixel at the predicted time. The expression is as follows: in, For the predicted time Pixel values, and These represent the compressed pixel values ​​before and after the surface temperature band, respectively. and This represents the time corresponding to the pixel before and after. The intercept of the univariate linear equation is given.

7. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, In step S5, the MCD12Q1 land cover classification product is obtained, the average land cover temperature of similar land cover is calculated, and the missing LST pixels of similar land cover are further filled, specifically including: According to the IGBP classification standard of MCD12Q1 land cover use products, the land cover in the study area was divided into 17 categories. The average surface temperature of each category was calculated as follows: in, It is the average surface temperature of a certain type. This represents the total number of pixels for that land type. For this land category The surface temperature value of each pixel; use Further fill in the missing surface temperature pixels for similar land features.

8. The method for reconstructing daily land surface temperature based on MODIS remote sensing data according to claim 1, characterized in that, In step S6, the step of using SG filtering to remove noise during the reconstruction process specifically includes: First, determine a radius as... The moving window incorporates all data within it into a set, with the time nodes of the time-series data contained within the window being... ,use The polynomial of degree applies to the window. Fitting is performed at each time point, and its polynomial expression is as follows: in, These are the filtered pixel values. For polynomial parameters, For the window The pixel value of the land surface temperature at each time point; Each time point within the window corresponds to a polynomial, totaling [number missing]. The number of pixels in the window Greater than polynomial parameters When the system of equations is solved, the parameters are obtained, and their expression is: Represented in matrix form as follows: in, These are the filtered surface temperature pixel values. The annual surface temperature pixel values ​​are used for filtering. For parameter matrices, It is a random error vector; Fitting parameters Least square solution for: Model prediction or filtered value for: For the MOD11A1 time-series surface temperature product The window represents the time before and after the prediction. Time of day This represents the land surface temperature value after weighted merging and filling with land use averages. This represents the surface temperature pixel value after noise removal using SG filtering.

9. A daily surface temperature reconstruction device based on MODIS remote sensing data, characterized in that, Includes the following modules: The preprocessing module is used to acquire MOD11A1 land surface temperature products, preprocess them, and then select qualified land surface temperature pixels. The global missing phase compensation module is used to perform global missing phase compensation on the single-day global missing image of MOD11A1 land surface temperature product to obtain continuous MOD11A1 land surface temperature product. The nearest neighbor interpolation module is used to perform nearest neighbor interpolation on a daily basis on the image sets before and after the missing images in the MOD11A1 surface temperature product. The weighted fusion module combines data from adjacent years and utilizes the sinusoidal harmonic function variation of land surface temperature to perform weighted fusion of reconstructed pixels, thereby reducing errors in the linear interpolation process; specifically, it includes: For surface temperature data from different years, an inverse temporal distance weighting method is used to weight and fuse multiple contemporaneous data before and after the prediction time. The expression is as follows: in, The weighted and merged surface temperature. and Represents the predicted time Surface temperature and weight, For the year, and Represents the predicted time Surface temperature and weight, This represents the maximum annual interval from the predicted year; The weighting constraints are as follows: ； The missing pixel filling module is used to obtain MCD12Q1 land surface classification products, calculate the average land surface temperature of similar land features, and further fill in the missing land surface temperature pixels of similar land features. The SG filtering module is used to remove noise during the reconstruction process, ultimately obtaining the daily surface temperature reconstruction results.

Images