CN114049570B

CN114049570B - Satellite-borne remote sensing water vapor space inversion method and system based on neural network

Info

Publication number: CN114049570B
Application number: CN202210034648.5A
Authority: CN
Inventors: 许艳; 高兆瑞; 江楠; 徐天河
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2022-01-13
Filing date: 2022-01-13
Publication date: 2022-04-22
Anticipated expiration: 2042-01-13
Also published as: CN114049570A

Abstract

The invention provides a satellite-borne remote sensing water vapor space inversion method and system based on a neural network, which belong to the technical field of water vapor inversion prediction, and the scheme comprises the following steps: acquiring space-based remote sensing observation data, land coverage type data and ground-based GNSS water vapor observation data; performing space-time matching on the acquired three types of data in the world based on a linear interpolation algorithm to obtain matching data in the global range; determining matching data corresponding to the current position information based on the position information of the region to be subjected to water vapor inversion and the matching data in the global range; and inputting the position information and the corresponding matching data into a pre-trained deep learning model, and outputting a water vapor inversion result of a region to be subjected to water vapor inversion.

Description

Satellite-borne remote sensing water vapor space inversion method and system based on neural network

Technical Field

The invention belongs to the technical field of water vapor space inversion prediction, and particularly relates to a satellite-borne remote sensing water vapor space inversion method and system based on a neural network.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

Although a number of neural network algorithms for pwv (probable Water vapor) inversion have been proposed and have achieved good results, most of these algorithms are built by re-analyzing data such as the NCEP (national environmental prediction center) and the ECMEF (european mesoscale weather prediction center).

Ground-based GNSS (global navigation satellite system) observations can provide higher accuracy and time resolution PWV values on fixed stations than reanalyzed data. However, the inventor finds that the scattered distribution of the GNSS stations reduces the spatial universality of the algorithm, the existing combined satellite-borne remote sensing data and ground-based GNSS data, and the neural network algorithm established based on the ground-based GNSS stations is usually only time domain verification on a fixed GNSS station, and for areas without GNSS stations, because corresponding GNSS station historical data does not exist, accurate PWV inversion cannot be performed by using the neural network algorithm; meanwhile, due to the complex land cover type in the space, particularly in the microwave inversion, the complexity of the land cover type causes the variability of microwave emissivity parameters, so that the algorithm established based on the fixed station is still a point domain algorithm and is not expanded to a space domain.

Disclosure of Invention

In order to solve the problems, the invention provides a satellite-borne remote sensing water vapor space inversion method and a system based on a neural network, wherein the method is expanded from a point domain to a space domain by introducing longitude and latitude data of a GNSS survey station as space parameters in the construction of a neural network model; meanwhile, the accurate inversion of PWV of all regions including the region without the GNSS survey station is realized by combining the influence of the land coverage type on the inversion accuracy.

According to a first aspect of the embodiments of the present invention, there is provided a method for inverting a space-borne remote sensing water vapor space based on a neural network, including:

acquiring space-based remote sensing observation data, land coverage type data and ground-based GNSS water vapor observation data;

performing space-time matching on the acquired three types of data in the world based on a linear interpolation algorithm to obtain matching data in the global range;

determining matching data corresponding to the current position information based on the position information of the region to be subjected to water vapor inversion and the matching data in the global range;

and inputting the position information and the corresponding matching data into a pre-trained deep learning model, and outputting a water vapor inversion result of a region to be subjected to water vapor inversion.

Further, the training process of the deep learning model specifically includes:

acquiring space-based remote sensing observation historical data, land coverage type historical data and foundation GNSS water vapor observation historical data;

based on a linear interpolation algorithm, performing space-time matching on the three types of acquired historical data in the world to acquire historical matching data in the global range;

and training the constructed deep learning model by taking the historical matching data as a training set to obtain the trained deep learning model.

Further, the space-time matching is performed on the acquired three types of data, specifically comprising space matching and time matching, wherein the space matching is to respectively query 4 points around the survey station in AMSR2 global grid data and MODIS global grid data based on the longitude and latitude of the GNSS survey station, and determine the brightness temperature and the land coverage type of the survey station by using a bilinear interpolation method; the time matching is based on the atmospheric precipitation variable values corresponding to the two GNSS observation times with the AMSR2 scanning time closest to each other, and the PWV value of the AMSR2 scanning time is determined according to linear interpolation.

Furthermore, the position information adopts longitude and latitude information.

Furthermore, in the training process of the deep learning model, the input layer comprises brightness temperature polarization difference, surface elevation, surface temperature and longitude and latitude values under various frequencies in the historical matching data, and the output layer is the atmospheric precipitation magnitude value in the historical matching data.

According to a second aspect of the embodiments of the present invention, there is provided a neural network-based spaceborne remote sensing water vapor space inversion system, including:

the data acquisition unit is used for acquiring space-based remote sensing observation data, land coverage type data and ground-based GNSS water vapor observation data;

the space-time matching unit is used for carrying out space-time matching on the acquired three types of data in the world based on a linear interpolation algorithm to obtain matching data in the global range;

the water vapor inversion unit is used for determining matching data corresponding to the current position information based on the position information of the region to be subjected to water vapor inversion and the matching data in the global range; and inputting the position information and the corresponding matching data into a pre-trained deep learning model, and outputting a water vapor inversion result of a region to be subjected to water vapor inversion.

Compared with the prior art, the invention has the beneficial effects that:

(1) the invention provides a satellite-borne remote sensing water vapor space inversion method and a system based on a neural network, wherein the method is expanded from a point domain to a space domain by introducing longitude and latitude data of a GNSS survey station as space parameters in the construction of a neural network model; meanwhile, the accurate inversion of PWV of all regions including the region without the GNSS survey station is realized by combining the influence of the land coverage type on the inversion accuracy.

(2) The space inversion algorithm fusing the advantages of the satellite-borne microwave remote sensing data and the high-precision and time-resolution ground-based GNSS data is constructed by utilizing a BP neural network algorithm based on the high-space-continuity satellite-borne microwave remote sensing data and the high-precision and time-resolution ground-based GNSS data; in the construction of the neural network, not only Tbs (Brightness temperature values) data from an AMSR2 microwave sensor is taken as input, but also earth surface temperature (Ts), earth surface elevation (H) and longitude and latitude data from a ground-based GNSS measuring station are taken as input layers to further improve the precision, and PWV data (GNSS-PWV) derived from the GNSS measuring station is taken as an output layer; wherein, the longitude and latitude data are used as the space parameters in the neural network to extend the model from the point domain to the space domain.

(3) The method analyzes the relation between the space distribution of inversion errors and the land cover type by combining an error space distribution diagram and a land cover type diagram of a survey station according to the complexity of the land cover type, and plays a certain guiding role in the region PWV inversion of different land cover types.

Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.

FIG. 1 is a model training flow chart of a neural network-based spaceborne remote sensing water vapor space inversion method according to a first embodiment of the disclosure;

fig. 2(a) is a diagram illustrating the distribution of 207 GNSS stations (without edge black dots) in the western united states and 20 stations (with edge black dots) used in the test in the rail ascending state according to the first embodiment of the disclosure;

fig. 2(b) is a distribution of 207 GNSS stations in the western united states (without edge black dots) and 20 stations used in the test (with edge black dots) in the down-track state according to the first embodiment of the disclosure;

FIG. 3(a) is a scatter plot of the elevated orbit states AMSR2-PWV and GNSS-PWV in the training set according to the first embodiment of the disclosure (the number of data samples NUM is 82137, the correlation coefficient is 0.74, and the RMSE is 3.80 mm);

FIG. 3(b) is a scatter diagram of the elevated orbit states AMSR2-PWV and GNSS-PWV in the test set according to the first embodiment of the disclosure (the number of data samples NUM is 7527, the correlation coefficient is 0.71, and the RMSE is 3.93 mm);

FIG. 3(c) is a scatter plot of the orbit reduction states AMSR2-PWV and GNSS-PWV in the training set according to the first embodiment of the disclosure (the number of data samples NUM is 94985, the correlation coefficient is 0.79, and the RMSE is 3.74 mm);

FIG. 3(d) is a scatter plot of the orbit reduction states AMSR2-PWV and GNSS-PWV in the test set according to the first embodiment of the disclosure (the number of data samples NUM is 11936, the correlation coefficient is 0.74, and the RMSE is 3.80 mm);

FIG. 4(a) is a schematic diagram illustrating a spatial distribution of a single GNSS station RMSE in an elevated state according to a first embodiment of the disclosure;

FIG. 4(b) is a schematic diagram illustrating a spatial distribution of a single GNSS station RMSE in a down-track state according to a first embodiment of the disclosure;

FIG. 5(a) is a schematic diagram illustrating a spatial distribution of GNSS stations RMSE with grass coverage in an elevated state according to a first embodiment of the disclosure;

fig. 5(b) shows the spatial distribution of GNSS stations RMSE covered by shrubs in the rail-lifting state according to the first embodiment of the disclosure;

FIG. 5(c) is a schematic diagram illustrating a spatial distribution of GNSS stations RMSE with grass coverage in a down-track state according to a first embodiment of the disclosure;

fig. 5(d) shows the spatial distribution of GNSS stations RMSE covered by shrubs in the down-track state according to the first embodiment of the disclosure.

Detailed Description

The invention is further described with reference to the following figures and examples.

It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

The embodiments and features of the embodiments of the present invention may be combined with each other without conflict.

The first embodiment is as follows:

the embodiment aims to provide a satellite-borne remote sensing water vapor space inversion method based on a neural network.

A satellite-borne remote sensing water vapor space inversion method based on a neural network comprises the following steps:

Further, the training process of the deep learning model specifically includes:

Further, the plurality of frequencies includes 18GHz, 23GHz, 36GHz and 89 GHz.

Further, the space-based remote sensing observation data adopt AMSR2 global grid data, the land coverage type data adopt MODIS global grid data, and the foundation GNSS water vapor observation data adopt Suominet GNSS observation station data.

Further, the MODIS global grid data specifically includes: each grid point represents the land cover type of the area by an integer, and the value range of the grid point is an integer which is greater than or equal to 0 and less than or equal to 16. In particular, the MODIS global grid data used in the present embodiment is based on the global land cover type product MCD12C1, which describes land cover characteristics observed from Terra and Aqua MODIS data and provides a single image of the global land cover type per year. For each product file, the product file is a global grid product with the resolution of 0.05 degrees and comprises 17 land cover categories defined by the international life circle program (IGBP), and each grid point of the product file represents the land cover category by a number with the value range of more than or equal to 0 and less than or equal to 16.

Further, the deep learning model adopts a BP neural network type.

Specifically, for the convenience of understanding, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings and actual data:

the invention provides a water vapor space inversion method for observing high spatial continuity and high-precision data by combining remote sensing and ground-based GNSS based on corresponding data of 207 Suominet GNSS observation stations and AMSR2 microwave sensors in the west of North America and based on a BP neural network; meanwhile, in order to prove the feasibility of the scheme of the invention, in the embodiment, the GNSS-PWV data is used for precision verification, and the relation between the space model precision and the land cover type is analyzed by combining with the MODIS land cover type product. The scheme of the present disclosure is described in detail below with reference to fig. 1:

as shown in FIG. 1, the invention provides a neural network-based spaceborne remote sensing water vapor space inversion method, which comprises the following steps:

the method comprises the following steps: and (4) preprocessing data. The data preprocessing process comprises the space-time matching of AMSR2, MODIS and ground-based GNSS data, and the specific implementation process is as follows:

the Tbs global grid data spatial resolution of AMSR2 is 0.1 degrees, providing only one value per day at each grid point on either the ascending or descending trajectory. Similarly, the land cover type product MCD12C1 is a global grid product with a resolution of 0.05 degrees, and there is only one value at each grid point (each grid point represents the land cover type of the area by an integer, and the range of the value is an integer greater than or equal to 0 and less than or equal to 16). However, the GNSS stations of the SuomiNet network are distributed discretely, and they can provide a PWV value every 15 minutes at a fixed station, as well as provide latitude and longitude, surface elevation (H), surface temperature (Ts), etc. parameters of the stations.

For space matching of the three types of data, firstly, finding 4 points around a GNSS survey station in a grid product of a satellite-borne sensor according to the longitude and latitude of the GNSS survey station, and then determining Tbs and a land coverage type of the survey station according to a bilinear interpolation method; for time matching, two GNSS observation times closest to the AMSR2 scanning time are firstly found, and then the GNSS-PWV value of the AMSR2 scanning time is determined according to linear interpolation, so that space-time matching of three types of data is realized.

Step two: constructing a neural network and finding out the optimal design of the neural network, wherein the optimal design comprises the input layer, the output layer, the number of hidden layers and the number of nodes in each layer; and finding the optimal neural network structure design. The specific implementation process is as follows:

in the neural network construction, the tool used in the embodiment is MATLAB2020a, the function used is "feedback forward net", and the built-in parameter is a default value. We set the learning step size to 0.3, the maximum training number to 1000, and the minimum target error to 10^-8. For the number of implicit layer numbers and the number of nodes per layer, we set to 1-5 cycles and 1-10 cycles, respectively. We randomly selected 20 stations from the 207 stations in north america for use as test samples and the remaining stations for use as training samples. Thus, the construction of the BP neural network is completed.

In the PWV inversion, the Tbs polarization difference of AMSR2 is determined

，

，

，

And from GNSS stationsH，TsAnd longitude and latitude parameters are used as an input layer, and the GNSS-PWV value is used as an output layer. The neural network is trained and the error of the test sample is calculated. And finding the combination of the hidden layer number and the node array under the condition of the minimum error of the test sample through the continuous circulation of the hidden layer number and the node number of the neural network, namely the optimal design of the neural network.

Step three: and (6) evaluating the precision. The method comprises the steps of calculating the precision of a training sample and a testing sample by utilizing GNSS-PWV, and analyzing the relation between the RMSE size of the measuring station and the land cover type by combining an error space distribution diagram and a MODIS land cover type product diagram.

To demonstrate the feasibility of the protocol of the invention, the following is illustrated by specific experiments:

(1) test area.

The experimental area ranged from 30 ° to 50 ° N and 100 ° to 130 ° W, which is located on the western land of the united states. Fig. 2(a) and 2(b) show the distribution of 207 GNSS stations in the experimental area. 187 sites (no edge black dots) were used as training set and 20 sites (edge black dots) were used as test set. All sites 2015 + 2017 data from three years were used to build models.

(2) Verification of spatial model accuracy using GNSS-PWV data

After space-time matching, 82137 pairs of training set matching points and 7527 pairs of test set matching points are reserved on an ascending rail, and 94985 pairs of matching points and 11936 pairs of matching points are reserved on a descending rail respectively. The overall correlation coefficient and RMSE (root mean square error) are shown in fig. 3(a) to 3 (d). It is clear that the PWV values are distributed approximately between 0-60 mm, with points within 30 mm being very dense and few points above 50 mm. In the ascending orbit, the RMSE for the training and test sets were 3.80 and 3.93mm, R²The values are 0.74 and 0.71, respectively. Under the circumstancesIn the downtrack, the RMSE for the training and test sets were 3.74 and 3.80mm, R²The values are 0.79 and 0.74, respectively. The results show that the falling track can obtain more data than the rising track, and the accuracy of the falling track is better than that of the rising track. According to the microwave radiation transmission theory, the radiation received by the microwave sensor is microwave radiation emitted from the atmosphere and the earth's surface. Since the up-track data is collected during the day, microwave radiation from sunlight, though becoming very weak after two atmospheric attenuations and surface reflections, can still interfere with microwave measurements. The rail descending data is collected at night, and the microwave radiation interference from sunlight cannot occur, so the rail descending precision is slightly better than that of rail ascending.

We have calculated the RMSE for a single GNSS station, as shown in fig. 4(a) to 4(b), showing the spatial distribution of the RMSE, with the stations represented by the black dots with the edges as the test set and the remaining stations as the training set. The results show the RMSE spatial distribution of individual GNSS stations in the selected area. In the area with better training set precision, the precision of the nearby test set is also better. Comparing the results of the up-and down-tracking, the down-tracking results have fewer stations with particularly poor accuracy (stations with lower grey values) than the up-tracking, which means that the down-tracking results have better accuracy, which also corresponds to the results shown in fig. 3. The distribution characteristics of the RMSE show that stations with good accuracy (stations with higher gray values) in the ascending and descending tracks are mostly distributed in the north and middle regions, while stations with poor accuracy (stations with lower gray values) are mostly distributed in the southwest and southeast regions.

(3) Research on relation between RMSE spatial distribution and land coverage type

In microwave inversion, different types of ground cover have different microwave emissivity, and therefore, the complexity of ground cover types results in variability in microwave emissivity. Variations in emissivity parameters can affect the accuracy of PWV inversion.

The land cover type on land is always complex and various, and GNSS stations are divided into 8 types according to the land cover type of the stations. As shown in table 1, there were the highest number of stations on the green, 94 stations, 27 stations next to shrubbery, and fewer stations on the remaining land cover types, all less than 20. Grassland and shrub distributions are relatively concentrated, the number of sites used to model is abundant, while other types of land cover distributions are relatively scattered. Therefore, we take grass and shrubs as examples to build a spatial model to study the RMSE distribution over a single land cover type. Fig. 5(a) to 5(d) show the results.

As can be seen in fig. 5(a) to 5(d), stations located on a single centrally distributed grass and shrub land cover type have better accuracy, while less accurate stations are located in areas where grass and shrubs are mixed with other land cover types. Taking fig. 5(a) and 5(c) as an example, although the southwest and southeast sites are located on grasslands, these grasslands are scattered and surrounded by arable land and other land cover types. Therefore, mixed ground cover types result in variability of emissivity, which in turn results in poor PWV inversion accuracy. Also, most sites located in areas like the central area where grass distribution is concentrated and not mixed with other land cover types have good inversion accuracy because these areas have stable emissivity. Fig. 5(b) and 5(d) show that the accuracy of most stations on the brush is better because the brush is relatively concentrated and not sporadically distributed. However, there are still a few stations in a mixed land-covering type area such as shrubs and grasslands, which are poor in accuracy. It can be concluded that the spatial model established by the method has better precision on a single large-range centralized distribution land cover type, but has poorer precision on a mixed land cover type, and the method also conforms to the theoretical characteristics of microwave inversion.

TABLE 1 number of GNSS stations on different types of land cover

Type of land cover	Number of GNSS survey stations
		Grassland	94
Shrub and shrub	27
		Crops	19
Multi-tree grassland	18
		Forest (forest)	17
Urban and built-up area	12
		Water (W)	11

Example two:

the embodiment aims to provide a satellite-borne remote sensing water vapor space inversion system based on a neural network.

A satellite-borne remote sensing water vapor space inversion system based on a neural network comprises:

In further embodiments, there is also provided:

an electronic device comprising a memory and a processor, and computer instructions stored on the memory and executed on the processor, the computer instructions when executed by the processor performing the method of embodiment one. For brevity, no further description is provided herein.

It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and so on. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory may include both read-only memory and random access memory, and may provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.

A computer readable storage medium storing computer instructions which, when executed by a processor, perform the method of embodiment one.

The method in the first embodiment may be directly implemented by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor. The software modules may be located in ram, flash, rom, prom, or eprom, registers, among other storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor. To avoid repetition, it is not described in detail here.

Those of ordinary skill in the art will appreciate that the various illustrative elements, i.e., algorithm steps, described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The satellite-borne remote sensing water vapor space inversion method and system based on the neural network can be realized, and have wide application prospects.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A satellite-borne remote sensing water vapor space inversion method based on a neural network is characterized by comprising the following steps:

inputting the position information and the corresponding matching data into a pre-trained deep learning model, and outputting a water vapor inversion result of a region to be subjected to water vapor inversion; the position information adopts longitude and latitude information, and the corresponding matching data comprises brightness temperature polarization difference, earth surface elevation and earth surface temperature;

and carrying out accuracy evaluation on the obtained water vapor inversion result of the region to be subjected to water vapor inversion based on the land cover type in the matching data in the global range.

2. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 1, wherein the training process of the deep learning model specifically comprises the following steps:

3. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 1, wherein the space-time matching is performed on the obtained three types of data, specifically comprising space matching and time matching, the space matching is to respectively query 4 points around the station in AMSR2 global grid data and MODIS global grid data based on the longitude and latitude of the GNSS station, and determine the brightness temperature and the land coverage type of the station by respectively utilizing a bilinear interpolation method; the time matching is based on the atmospheric precipitation variable values corresponding to the two GNSS observation times with the AMSR2 scanning time closest to each other, and the PWV value of the AMSR2 scanning time is determined according to linear interpolation.

4. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 1, wherein in the training process of the deep learning model, input layers of the deep learning model comprise brightness temperature polarization difference, surface elevation, surface temperature and longitude and latitude values at multiple frequencies in historical matching data, and output layers are atmospheric precipitation quantity values in the historical matching data.

5. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 4, wherein the plurality of frequencies comprise 18GHz, 23GHz, 36GHz and 89 GHz.

6. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 1, wherein the space-based remote sensing observation data adopts AMSR2 global grid data, the land cover type data adopts MODIS global grid data, and the foundation GNSS water vapor observation data adopts Suominet GNSS survey station data.

7. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 6, wherein the MODIS global grid data specifically comprises: each grid point represents the land cover type of the area by an integer, and the value range of the grid point is an integer which is greater than or equal to 0 and less than or equal to 16.

8. The method for spaceborne remote sensing water vapor space inversion based on the neural network as claimed in claim 1, wherein the deep learning model is of a BP neural network type.

9. A satellite-borne remote sensing water vapor space inversion system based on a neural network is characterized by comprising the following components:

the water vapor inversion unit is used for determining matching data corresponding to the current position information based on the position information of the region to be subjected to water vapor inversion and the matching data in the global range; inputting the position information and the corresponding matching data into a pre-trained deep learning model, and outputting a water vapor inversion result of a region to be subjected to water vapor inversion; the position information adopts longitude and latitude information, and the corresponding matching data comprises brightness temperature polarization difference, earth surface elevation and earth surface temperature;

and the evaluation unit is used for evaluating the accuracy of the obtained water vapor inversion result of the region to be subjected to water vapor inversion based on the land cover type in the matching data in the global range.