A method for spatiotemporal reconstruction of land surface temperature

Abstract

The present invention relates to the technical field of surface temperature model reconstruction, and in particular to a surface temperature spatiotemporal reconstruction method. A surface temperature spatiotemporal reconstruction method of the present invention solves the problem that the existing surface temperature data downscaling statistical scheme lacks physical mechanism ambiguity, and the problem that multi-source remote sensing observation is limited in cloud coverage areas. At the same time, it solves the problem of large-area missing data recovery and spatial downscaling on a global scale, as well as the reconstruction of surface temperature data on a continuous time scale, and finally realizes the reconstruction of global surface temperature data with high spatial resolution, continuous time (hourly level), and not affected by clouds.

Description

Technical Field

[0001] The present invention relates to the technical field of land surface temperature model reconstruction, and in particular to a land surface temperature spatiotemporal reconstruction method. Background Art

[0002] Land surface temperature has important practical value in agricultural drought monitoring, climate change, agricultural meteorology, and ecological environment monitoring. Since the 1970s, the use of satellite thermal infrared remote sensing to monitor land surface temperature has become relatively mature. However, due to the limitations of sensor monitoring capabilities, the land surface temperature data directly obtained by satellite remote sensing cannot have the characteristics of spatial high resolution and temporal continuity at the same time. In addition, satellite cloud images show that there is always a large area of ​​cloud coverage on the global surface. Due to the shielding effect of clouds on thermal infrared radiation, there is a phenomenon of missing surface temperature data in cloud-covered areas. Therefore, providing a land surface temperature reconstruction method that can achieve spatial high resolution, temporal continuity, and is not affected by clouds plays an important role in the study of agricultural production activities and climate change.

[0003] The reconstruction of missing data in the spatial dimension is divided into two aspects. On the one hand, the pixels covered by clouds are restored, and on the other hand, the resolution of pixels is reduced from a large spatial scale to a small scale. Traditional methods use the spatial variability, similarity and dependence of adjacent pixels, and there are also methods that combine ground station measurements for interpolation, but they cannot effectively restore missing values ​​over large areas, and it is difficult to restore extreme temperatures and drastic temperature changes. There are also methods that combine driving factors provided by other satellite remote sensing (for example, longitude, latitude, altitude, normalized difference vegetation index, land cover type, solar radiation, etc.), and establish statistical models of these predictors and surface temperature to reconstruct surface temperature data. Although geostationary satellites and polar-orbiting satellites can provide remote sensing images with temporal continuity and spatial high resolution, respectively, the above methods still cannot provide spatially high-resolution, temporally continuous, and cloud-free surface temperature data.

[0004] The following is a method for recovering and reconstructing missing surface temperature data:

[0005] Firstly, a suitable study area is selected, and a nonlinear model is established based on local geographic autocorrelation to downscale the surface temperature. Then, the normalized difference vegetation index, land cover type and digital elevation model are selected as auxiliary parameters to spatially downscale the surface temperature data. Finally, the parameters and errors are spatially interpolated to improve the spatial resolution and accuracy of the surface temperature data provided by satellite remote sensing products.

[0006] The above methods still have the following shortcomings:

[0007] (1) When surface temperature satellite data have large areas of missing values, existing methods are unable to effectively restore extreme temperatures and drastic temperature changes.

[0008] (2) Given the characteristics and limitations of satellite data, multi-source satellite remote sensing data alone cannot provide global surface temperature data that are spatially high-resolution, temporally continuous (hourly level), and not affected by clouds. Summary of the invention

[0009] In order to solve the problems of the prior art, the present invention provides a method for spatiotemporal reconstruction of surface temperature, which solves the problem that the physical mechanism of the existing statistical scheme for downscaling surface temperature data is ambiguous, and the problem that multi-source remote sensing observations are limited in cloud coverage areas. At the same time, it solves the problems of satellite data recovery and spatial downscaling of large-area missing data on a global scale, and reconstruction of surface temperature data on a continuous time scale, and ultimately realizes the reconstruction of global surface temperature data with high spatial resolution, continuous time (hourly level), and no influence of clouds.

[0010] The technical solution adopted by the present invention is as follows:

[0011] A method for spatiotemporal reconstruction of land surface temperature comprises the following steps:

[0012] A. preprocessing satellite and reanalysis data to obtain global temperature data, wherein the satellite and reanalysis data include land surface temperature data and sea surface temperature data;

[0013] B. Spatial scale reconstruction of satellite and reanalysis data, including: spatial scale interpolation of satellite sea surface temperature data, spatial interpolation of reanalysis land surface temperature data, and splicing of satellite land and sea surface temperature data;

[0014] C. Reconstruction of spatiotemporal scales of surface temperature data.

[0015] Preferably, the step A specifically comprises:

[0016] A1. The original land surface temperature data is converted into land surface temperature data with physical meaning by using the following formula: T = α·DN + b, where DN is the value recorded in the original land surface temperature data; a is the scale factor, whose value is 0.02; b is the offset, whose value is 0; the unit of the converted land surface temperature data T is Kelvin;

[0017] A2. The original sea surface temperature data is converted into a numerical value by a formula to obtain sea surface temperature data with physical meaning. The conversion formula is the same as the formula in step A1. The scale factor a of the sea surface temperature data is 0.000717185; the offset b is -2; and the unit of the converted sea surface temperature data T is Celsius.

[0018] A3. Use the following formula to unify the dimensions: T c =T k -273.15, of which T C Indicates degrees Celsius, T K It represents degrees Kelvin, and finally all the global temperature data are obtained in degrees Celsius.

[0019] Preferably, step B specifically includes:

[0020] B1. Improve the spatial resolution of sea surface temperature data based on bilinear interpolation to make it consistent with the spatial resolution of land surface temperature;

[0021] B2. We selected ERA5 surface temperature data with a high temporal resolution of 1 h to improve the temporal resolution of the MODIS temperature product from the Moderate-resolution Imaging Spectroradiometer.

[0022] Preferably, step B1 specifically includes: using the values ​​of four adjacent pixels, assigning different weights according to their distances from the interpolation point for linear interpolation, taking four pixels around the point (x, y), interpolating twice in the y direction or the x direction, and then interpolating once in the x direction or the y direction, thus completing an interpolation.

[0023] Preferably, step C specifically includes: spatially matching the satellite global surface temperature data with the reanalysis data surface temperature data, and fusing the satellite global surface temperature data with the reanalysis data surface temperature data.

[0024] Preferably, step C specifically comprises:

[0025] C1. Resample the vegetation data and digital elevation model data with a spatial resolution of 30m to 1km respectively, so that they have the same spatial scale as the global temperature data;

[0026] C2. Spatial matching of the surface temperature reanalysis data at the corresponding time with the vegetation data and digital elevation model data provided by satellites;

[0027] C3. Use geographically weighted regression method to downscale the surface temperature data of reanalysis data;

[0028] C4. The data fusion method used is the universal kriging method, which is solved by Lagrange multipliers to fuse the daytime and nighttime global surface temperature data at a spatial resolution of 1 km with the surface temperature data reconstructed at a spatiotemporal scale of 1 km spatial resolution and 1 h time interval.

[0029] The beneficial effects brought by the technical solution provided by the present invention are:

[0030] The introduction of globally complete and consistent reanalysis data based on physical laws ensures data recovery in areas with low satellite data coverage, simplifies the nonlinear perturbations of statistical models to a certain extent, and improves prediction accuracy. At the same time, the temporal and spatial distribution of the temperature diurnal variation law lays a scientific foundation for providing a global surface temperature database that is spatially high-resolution, temporally continuous (hourly level), and not affected by clouds. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0032] Figure 1 The present invention is a flow chart of a method for spatiotemporal reconstruction of land surface temperature. DETAILED DESCRIPTION

[0033] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0034] Embodiment 1

[0035] The present embodiment provides a method for spatiotemporal reconstruction of land surface temperature, which mainly includes three steps, namely, satellite and reanalysis data preprocessing, satellite and reanalysis data spatial downscaling, and spatiotemporal reconstruction of land surface temperature data. In the present embodiment of China, the main input data sources include: 1km spatial resolution land surface temperature data of medium resolution imaging spectrometer, 4km spatial resolution MODIS sea surface temperature data, 0.1° spatial resolution reanalysis data ERA5 land surface temperature data (temporal resolution is 1h), 250m spatial resolution MODIS normalized difference vegetation index (NDVI) data and 30m spatial resolution ground digital elevation DEM data.

[0036] The specific steps of the method include:

[0037] Step 1: Data preprocessing of satellite and reanalysis data.

[0038] This step requires restoring the DN value of the satellite MODIS temperature data and unifying the dimension of the satellite MODIS temperature data.

[0039] In order to generate global temperature data, it is necessary to establish a unified data standard for the data attributes of the two temperature products, and then on this basis, mosaic synthesis is carried out based on geographical location relationships to obtain global temperature data.

[0040] According to the product description, in order to save data storage space, the original surface temperature product is converted to a numerical value. Here, it is necessary to convert the numerical value through the correlation coefficient and formula provided in the attribute to obtain the surface temperature data with physical meaning. The conversion formula is as follows:

[0041] T=α·DN+b

[0042] Among them, DN is the value recorded in the product; a is the scale factor, whose value is 0.02; b is the offset, whose value is 0. The unit of the converted surface temperature data is Kelvin.

[0043] The sea surface temperature product also needs to be converted into a physically meaningful value, using the same conversion formula as above. The scale factor a of the sea surface temperature data is 0.000717185; the offset b is -2. The converted sea surface temperature data is in degrees Celsius, and then the dimensions are unified using the following formula.

[0044] T C =T K -273.15

[0045] Among them, TC stands for Celsius and TK stands for Kelvin. Here, the dimensions of both temperature products are converted to Celsius to achieve the unification of data dimensions.

[0046] Step 2: Spatial downscaling of satellite and reanalysis data.

[0047] This step requires the spatial scale interpolation of satellite MODIS sea surface temperature data and reanalysis data, and the pan-Krieg integration of the spliced ​​satellite MODIS land surface temperature data and reanalysis data.

[0048] The spatial resolution of satellite land surface temperature data is 1km, while the spatial resolution of satellite sea surface temperature data is 4km. Therefore, the spatial resolution of the two needs to be unified. Considering that the surface of the ocean area is relatively simple and the difference between adjacent pixels is small, in order to obtain a global temperature product with a higher spatial resolution, the method used here is to improve the spatial resolution of the sea surface temperature data based on the bilinear interpolation method, so that it is consistent with the spatial resolution of the land surface temperature, both of which are 1km. Similarly, in order to obtain the downscaled product of the surface temperature reanalysis data, the original reanalysis data is also subjected to bilinear spatial interpolation.

[0049] First, the spatial resolution of MODIS data products needs to be unified.

[0050] MODIS land surface temperature and sea surface temperature data products have different spatial resolutions. The spatial resolution of land surface temperature data is 1km, while the spatial resolution of sea surface temperature data is 4km. Therefore, the spatial resolutions of the two need to be unified. Considering that the surface of the ocean area is relatively simple and the difference between adjacent pixels is small, the bilinear interpolation method is used here to reconstruct the spatial scale of the sea surface temperature data. The values ​​of the four adjacent pixels are used to perform linear interpolation according to their distance from the interpolation point and different weights are assigned. Take the four pixels around the point (x, y), interpolate twice in the y direction (or x direction), and then interpolate once in the x direction (or y direction), that is, complete an interpolation. The spatial resolution of the MODIS sea surface temperature data product is improved to be consistent with the spatial resolution of the land surface temperature data of 1km. Finally, the global daytime and nighttime surface temperature distribution is obtained.

[0051] Then, in order to obtain more detailed surface temperature information on a time scale, the ERA5 surface temperature data with a high time resolution of 1h is selected here to improve the time resolution of the MODIS temperature product. However, since the spatial resolution of the ERA5 reanalysis data is 0.1°, which is lower than the expected spatial resolution of 1km, it cannot be directly used for data fusion. Here, the nearest neighbor interpolation method is first used to obtain surface temperature reanalysis data with the same spatial scale as MODIS. The spatial variability of surface temperature is large due to the complex underlying surface, which is often related to the surface type. The surface temperature data obtained by a single interpolation method cannot truly and effectively represent the temperature differences on a small scale. In order to further obtain spatial downscaling products, data products consistent with the spatial resolution of MODIS are provided here.

[0052] Step 3: Reconstruct the spatiotemporal scale of surface temperature data.

[0053] This step includes spatial matching of satellite MODIS global surface temperature data with reanalysis data surface temperature, and fusion of satellite MODIS global surface temperature data with reanalysis data surface temperature data. Spatial matching is performed based on geographical location relationships, and global daytime and nighttime surface temperature products with improved temporal and spatial scale resolution are obtained.

[0054] First, the NDVI data with a spatial resolution of 250m and the DEM data with a spatial resolution of 30m were resampled to 1km, which has the same spatial scale as MODIS. The surface temperature reanalysis data at the corresponding time were spatially matched with the NDVI and DEM data provided by the satellite.

[0055] Then, the geographically weighted regression method is used to achieve downscaling of the surface temperature data of the reanalysis data. Surface parameters that affect the spatial variability of surface temperature are introduced here, including the normalized difference vegetation index NDVI and the ground elevation DEM. In areas without vegetation coverage, there are also complex surface types (for example, bare soil, desert, concrete ground, etc.), and their surface albedo is different. Here, the surface temperature reanalysis data is used as an auxiliary factor of the statistical model, which simplifies the nonlinear perturbation of the statistical model to a certain extent and improves the prediction accuracy. The geographically weighted method is actually an extension of the traditional standard regression method. Compared with the global scale parameter estimation, this method adopts a local parameter estimation method. By constructing the regression coefficient of the surface spatial change, the geographically weighted method can be used to study the dynamic regression relationship between the independent variable and the dependent variable. Therefore, the downscaling based on geographical weighting can obtain the spatial dynamic relationship between the surface temperature and related parameters.

[0056] T modis =β0+β1NDVI+β2DEM+β3T era

[0057] Next, a multivariate linear regression model between the surface temperature and the above regression parameters is established, and the corresponding fitting coefficients and residuals are obtained by using the least squares estimation of the weights. Then, the statistically obtained fitting coefficients are used to calculate the surface temperature values ​​at other times (except the MODIS transit time). In this way, the spatial downscaling products of the surface temperature reanalysis data are obtained, and the reconstruction of the surface temperature data on a temporal scale can also be achieved.

[0058] Finally, the data fusion method used is the universal kriging method, which is solved by Lagrange multipliers. The MODIS global surface temperature data at day and night with a spatial resolution of 1km is fused with the surface temperature data reconstructed at a spatial and temporal scale of 1km spatial resolution and 1h time interval. Given the surface temperature values ​​of n stations (times), the estimated surface temperature distribution s at m stations (times) can be expressed as the sum of an unknown term Xsβ and a random term v with a mean of 0:

[0059] s=X s β+v

[0060] Among them, X s (m×2) is defined as the trend model, and β(2×1) is defined as the drift coefficient, which can express the weights of the two variables in the trend model. Here, a linear model is used as the trend model:

[0061]

[0062] Similar to the regression model, the constant term is used to obtain the net offset not affected by the MODIS results. As for the random term v, it represents the residual of the two temperature data used for fusion at (n), and at the estimation site (m) it represents the estimated residual between the true low temperature value and the weighted temperature value.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for spatiotemporal reconstruction of land surface temperature, comprising the following steps: A. preprocessing satellite and reanalysis data to obtain global temperature data, wherein the satellite and reanalysis data include land surface temperature data and sea surface temperature data; B. Spatial scale reconstruction of satellite and reanalysis data, including: spatial scale interpolation of satellite sea surface temperature data, spatial interpolation of reanalysis land surface temperature data, and splicing of satellite land and sea surface temperature data; The step B specifically comprises: B1. Improve the spatial resolution of sea surface temperature data based on bilinear interpolation to make it consistent with the spatial resolution of land surface temperature; B2. We selected the surface temperature data of ERA5, a reanalysis data with a high temporal resolution of 1 h, and improved the temporal resolution of the temperature product of the Moderate Resolution Imaging Spectroradiometer MODIS. C. Reconstruction of the spatiotemporal scale of land surface temperature data, spatial matching of satellite global land surface temperature data and reanalysis data land surface temperature data, and fusion of satellite global land surface temperature data and reanalysis data land surface temperature data; Specifically include: C1. Resample the vegetation data with 250m spatial resolution and the digital elevation model data with 30m spatial resolution to 1km respectively, so that they have the same spatial scale as the global temperature data; C2. Spatial matching of the surface temperature reanalysis data at the corresponding time with the vegetation normalized index data and digital elevation model data provided by satellites; C3. Use geographically weighted regression method to downscale the surface temperature data of reanalysis data; C4. The data fusion method used is the universal kriging method, which is solved by Lagrange multipliers to fuse the daytime and nighttime global surface temperature data at a spatial resolution of 1 km with the surface temperature data reconstructed at a spatiotemporal scale of 1 km spatial resolution and 1 h time interval.

2. A method for spatiotemporal reconstruction of land surface temperature according to claim 1, characterized in that: The step A specifically comprises: A1. The original land surface temperature data is converted into numerical values ​​through the formula to obtain land surface temperature data with physical meaning. The conversion formula is as follows: , where DN is the value recorded in the original land surface temperature data; a is the scale factor, whose value is 0.02; b is the offset, whose value is 0; the unit of the converted land surface temperature data T is Kelvin; A2. The original sea surface temperature data is converted into a numerical value by a formula to obtain sea surface temperature data with physical meaning. The conversion formula is the same as the formula in step A1. The scale factor a of the sea surface temperature data is 0.000717185; the offset b is -2; and the unit of the converted sea surface temperature data T is Celsius. A3. Use the following formula to unify the dimensions: T c =T k -273.15, of which T C Indicates degrees Celsius, T K It represents degrees Kelvin, and finally all the global temperature data are obtained in degrees Celsius.

3. A method for spatiotemporal reconstruction of land surface temperature according to claim 1, characterized in that: The step B1 is specifically as follows: use the values ​​of the four neighboring pixels, assign different weights according to their distance from the interpolation point for linear interpolation, take the four pixels around the point (x, y), interpolate twice in the y direction or the x direction, and then interpolate once in the x direction or the y direction, thus completing an interpolation.

Images