High-resolution soil moisture inversion method and device, electronic equipment and storage medium

Abstract

The invention provides a high-resolution soil moisture inversion method and device, electronic equipment and a storage medium, and relates to the technical field of remote sensing monitoring. Target soil moisture inversion data under a first resolution and TVDI data under a second resolution corresponding to a target area can be acquired; wherein the target soil moisture inversion data comprises MWRI data, NDVI data and albedo data; the second resolution is greater than the first resolution; performing soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data under the first resolution; and downscaling the initial soil moisture data according to the TVDI data to obtain target soil moisture data under the second resolution. In this way, the initial soil moisture data is downscaled according to the high-resolution TVDI data, the drought characterization capability of TVDI can be fully utilized, and the resolution and stability of soil moisture inversion are improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing monitoring technology, and in particular to a high-resolution soil moisture inversion method, device, electronic equipment and storage medium. Background Art

[0002] Soil moisture (SM) is an important factor in the land energy balance and a key parameter controlling the water and heat energy exchange between the land and the atmosphere. It is directly related to atmospheric humidity and affects the heat flux from the land to the atmosphere. At the same time, SM is also a link between terrestrial surface water and groundwater and an important part of the water cycle of terrestrial ecosystems. It is widely used in agriculture, such as sowing planning, irrigation planning, etc.

[0003] The inversion methods for soil moisture can be roughly divided into three categories: microwave remote sensing method, vegetation index inversion method, thermal inertia model method, etc. Among them, the microwave remote sensing method uses the principle that the dielectric constant of soils with different water contents is different, which can effectively monitor soil moisture content and its changes; and it can be used all day, becoming the most flexible, applicable and effective method for monitoring soil moisture. Microwave remote sensing is divided into active microwave remote sensing and passive microwave remote sensing. In addition, the method of microwave soil moisture inversion using machine learning has been widely studied by scholars in recent years. The main idea is that the highly nonlinear relationship between soil moisture products and microwave channel brightness temperature can be statistically established based on auxiliary data and a large amount of training data.

[0004] Soil moisture downscaling refers to the process of converting low spatial resolution soil moisture remote sensing products to high spatial resolution to improve the spatial detail and accuracy of soil moisture data. At present, scholars have conducted a lot of research on soil moisture downscaling methods, forming several types of soil moisture downscaling methods such as statistical model method, physical model method, and machine learning model method.

[0005] Although there are many microwave soil moisture inversion and downscaling methods in academia, there is still a lack of relatively stable and spatiotemporally continuous soil moisture datasets with high spatiotemporal resolution. Summary of the invention

[0006] The object of the present invention is to provide a high-resolution soil moisture inversion method, device, electronic equipment and storage medium to improve the resolution and stability of soil moisture inversion.

[0007] In a first aspect, an embodiment of the present invention provides a high-resolution soil moisture inversion method, comprising:

[0008] Obtain target soil moisture inversion data at a first resolution and temperature vegetation drought index TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes microwave imager MWRI data, normalized difference vegetation index NDVI data and albedo data; the second resolution is greater than the first resolution;

[0009] Perform soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at a first resolution;

[0010] The initial soil moisture data are downscaled according to the TVDI data to obtain the target soil moisture data at the second resolution.

[0011] Further, obtaining target soil moisture inversion data at a first resolution and temperature vegetation drought index TVDI data at a second resolution corresponding to the target area includes:

[0012] Obtain the original soil moisture inversion data and original surface temperature data of the target area within the target time range. The original soil moisture inversion data includes original MWRI data, original NDVI data and original albedo data;

[0013] Preprocessing the original soil moisture inversion data and the original surface temperature data to obtain target soil moisture inversion data at a first resolution, and second NDVI data and target surface temperature data at a second resolution; wherein the preprocessing includes one or more of data parsing, projection conversion, spatial scale conversion, and time scale aggregation;

[0014] The TVDI data is generated according to the second NDVI data and the target land surface temperature data.

[0015] Further, TVDI data is generated according to the second NDVI data and the target surface temperature data, including:

[0016] TVDI data is calculated by the following formula:

[0017]

[0018] T Smax =a1+b1×NDVI;

[0019] T Smin =a2+b2×NDVI;

[0020] Among them, TVDI is the TVDI value of the pixel in the TVDI data, T s is the surface temperature value of the pixel in the target surface temperature data, T sminis the surface temperature of the wet edge, a2 and b2 are the linear fitting coefficients of the wet edge model, NDVI is the NDVI value of the pixel in the second NDVI data, T smax is the surface temperature of the dry side, a1 and b1 are the linear fitting coefficients of the dry side model.

[0021] Further, soil moisture inversion is performed on the target soil moisture inversion data to obtain initial soil moisture data at a first resolution, including:

[0022] Normalizing the target soil moisture inversion data to obtain normalized target soil moisture inversion data;

[0023] Inputting the normalized target soil moisture inversion data into the trained deep neural network model to obtain the first soil moisture data output by the deep neural network model;

[0024] The first soil moisture data is denormalized to obtain initial soil moisture data.

[0025] Furthermore, the initial soil moisture data is downscaled according to the TVDI data to obtain the target soil moisture data at the second resolution, including:

[0026] Perform bilinear interpolation on the initial soil moisture data to obtain simulated soil moisture data at a second resolution, where the simulated soil moisture data includes a simulated soil moisture value for each pixel;

[0027] According to the TVDI data, the statistical deviation data is calculated, and the statistical deviation data includes the statistical deviation coefficient of the TVDI value of each pixel within its preset neighborhood;

[0028] The simulated soil moisture data are corrected using statistical deviation data to obtain the target soil moisture data.

[0029] Furthermore, the simulated soil moisture data is corrected using statistical deviation data to obtain target soil moisture data, including:

[0030] The target soil moisture data is calculated by the following formula:

[0031] SM i =SM j +α×δ TVDI ;

[0032] Among them, SM i is the soil moisture value of pixel i in the target soil moisture data, SM j is the simulated soil moisture value of pixel i in the simulated soil moisture data, α is the preset deviation influence coefficient, δ TVDIis the statistical deviation coefficient corresponding to pixel i in the statistical deviation data.

[0033] Furthermore, the above method also includes:

[0034] Acquire a training sample library, the training sample library includes a plurality of sample data at a first resolution obtained through preprocessing, data screening and normalization, each sample data includes sample soil moisture inversion data and corresponding SMAP soil moisture product data;

[0035] The initial deep neural network model is trained according to the training sample library to obtain a trained deep neural network model; wherein the sample soil moisture inversion data in each sample data is input, and the corresponding SMAP soil moisture product data is the true value.

[0036] In a second aspect, an embodiment of the present invention further provides a high-resolution soil moisture inversion device, comprising:

[0037] An acquisition module is used to acquire target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data and albedo data; and the second resolution is greater than the first resolution;

[0038] An inversion module, used to perform soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at a first resolution;

[0039] The downscaling module is used to downscale the initial soil moisture data according to the TVDI data to obtain the target soil moisture data at the second resolution.

[0040] In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the high-resolution soil moisture inversion method of the first aspect is implemented.

[0041] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the high-resolution soil moisture inversion method of the first aspect is executed.

[0042] The high-resolution soil moisture inversion method, device, electronic device and storage medium provided by the embodiment of the present invention can obtain target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data and albedo data; the second resolution is greater than the first resolution; soil moisture inversion is performed on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution; the initial soil moisture data is downscaled according to the TVDI data to obtain target soil moisture data at the second resolution. In this way, soil moisture inversion is performed on MWRI data, NDVI data, albedo data, etc. to obtain low-resolution initial soil moisture data, and then the initial soil moisture data is downscaled according to the high-resolution TVDI data, which can make full use of the ability of TVDI to characterize drought and improve the resolution and stability of soil moisture inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A schematic diagram of a process flow of a high-resolution soil moisture inversion method provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of the structure of a deep neural network model provided by an embodiment of the present invention;

[0046] Figure 3 A schematic flow chart of another high-resolution soil moisture inversion method provided by an embodiment of the present invention;

[0047] Figure 4 A schematic structural diagram of a high-resolution soil moisture inversion device provided by an embodiment of the present invention;

[0048] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] There is currently a lack of mature soil moisture inversion methods with high temporal and spatial resolution for the observation data of the FY-3 satellite microwave imager (MWRI). Based on this, the embodiments of the present invention provide a high-resolution soil moisture inversion method, device, electronic device and storage medium, which can utilize the FY-3 satellite MWRI observation data and other multi-source satellite data, and fuse the Temperature Vegetation Dryness Index (TVDI) to obtain a daily-scale microwave soil moisture inversion result with a resolution of 1 km.

[0051] It should be noted that the embodiments of the present invention can also be applied to microwave sensors of various other satellites to generate high-resolution soil moisture inversion data.

[0052] To facilitate understanding of this embodiment, a high-resolution soil moisture inversion method disclosed in an embodiment of the present invention is first introduced in detail.

[0053] The embodiment of the present invention provides a high-resolution soil moisture inversion method, which can be executed by an electronic device with data processing capability. Figure 1 The flowchart of a high-resolution soil moisture inversion method shown in FIG. 1 mainly includes the following steps S110 to S130:

[0054] Step S110, obtaining target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to the target area.

[0055] Among them, the target soil moisture inversion data includes MWRI data, Normalized Difference Vegetation Index (NDVI) data and albedo data. The target soil moisture inversion data can also include other auxiliary data describing pixel characteristic information, such as Digital Elevation Model (DEM), land cover type (LC), latitude / longitude (Lat / lon) and day of year (DOY), etc.; the second resolution is greater than the first resolution.

[0056] The target area is the required research area, which can be a certain region or a global region. The first resolution is a lower spatial resolution, and the second resolution is a higher spatial resolution. For example, the first resolution is 10km and the second resolution is 1km. TVDI is a temperature vegetation drought index, which has a very significant negative correlation with soil moisture and can well estimate the soil surface moisture conditions.

[0057] In some possible embodiments, the above step S110 may include the following sub-steps 111 to 113:

[0058] Sub-step 111, obtaining original soil moisture inversion data and original surface temperature data of the target area within a target time range, wherein the original soil moisture inversion data includes original MWRI data, original NDVI data and original albedo data.

[0059] The above target time range is the required time range, which can be one day or multiple days. The original MWRI data can be the MWRI L1 channel data of FY3D. The L1 channel data refers to the first-level data product obtained by the microwave radiometer carried by the Fengyun-3D satellite, which can include radiation brightness temperature, geographic location information and timestamp, etc. The spatial resolution of the original MWRI data can be 10km. The original NDVI data can be the NDVI remote sensing data of MODIS (Moderate Resolution Imaging Spectroradiometer). The original albedo data can be the Albedo remote sensing data of MODIS. The original surface temperature data can be the LST (Land Surface Temperature) data of MODIS, with a spatial resolution of 1km.

[0060] Sub-step 112, pre-processing the original soil moisture inversion data and the original surface temperature data to obtain target soil moisture inversion data at a first resolution, and second NDVI data and target surface temperature data at a second resolution; wherein the pre-processing includes one or more of data parsing, projection conversion, spatial scale conversion and time scale aggregation.

[0061] Computer grid processing tools can be used to perform one or more of data analysis, projection conversion, spatial scale conversion, and time scale aggregation on raw MWRI data, raw NDVI data, and raw albedo data, and obtain daily-scale grid data with a resolution of 0.25° in longitude and latitude coordinate systems such as WGS84, which is the target soil moisture inversion data; wherein, a resolution of 0.25° means that the longitude and latitude range covered by each grid unit are both 0.25°. Data analysis, projection conversion, spatial scale conversion, filtering, and daily-scale aggregation can be performed on raw NDVI data and raw albedo data; data analysis, projection conversion, spatial scale conversion, and daily-scale aggregation can be performed on raw MWRI data; wherein, projection conversion and spatial scale conversion can be achieved by one algorithm or by two algorithms respectively. In specific implementation, the Savitzky-Golay filtering method can be used to filter the original NDVI data and the original albedo data and interpolate missing values, and the linear averaging method can be used to obtain the target NDVI data and target albedo data on a daily scale. The pixel averaging method can be used to aggregate the original MWRI data on a daily scale, and the bilinear interpolation or mean aggregation method can be used for spatial scale conversion.

[0062] Similarly, computer raster processing tools can be used to perform one or more of data analysis, projection conversion, spatial scale conversion, filtering, and daily scale aggregation on the original NDVI data and the original surface temperature data to obtain daily scale raster data with a spatial resolution of 1 km in a longitude and latitude coordinate system such as WGS84. This data is the second NDVI data and the target surface temperature data.

[0063] Sub-step 113, generating TVDI data according to the second NDVI data and the target surface temperature data.

[0064] For the above sub-step 113, the TVDI data can be calculated by the following formula:

[0065]

[0066] T Smax =a1+b1×NDVI;

[0067] T Smin =a2+b2×NDVI;

[0068] Among them, TVDI is the TVDI value of the pixel in the TVDI data, T s is the surface temperature value of the pixel in the target surface temperature data, T sminis the surface temperature of the wet edge, a2 and b2 are the linear fitting coefficients of the wet edge model, NDVI is the NDVI value of the pixel in the second NDVI data, T smax is the surface temperature of the dry side, a1 and b1 are the linear fitting coefficients of the dry side model.

[0069] Step S120: performing soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at a first resolution.

[0070] The neural network model can be trained through machine learning, and the trained neural network model can be used to realize soil moisture inversion. Deep Neural Network (DNN) is a technology in the field of machine learning. It is a neural network model for deep learning developed on the basis of perceptron. The trained DNN can be used to perform soil moisture inversion on the target soil moisture inversion data.

[0071] Based on this, in some possible embodiments, the above step S120 can be implemented through the following process: normalizing the target soil moisture inversion data to obtain normalized target soil moisture inversion data; inputting the normalized target soil moisture inversion data into the trained deep neural network model to obtain the first soil moisture data output by the deep neural network model; and denormalizing the first soil moisture data to obtain initial soil moisture data.

[0072] The specific method of normalization can be selected according to actual needs. For example, the maximum and minimum normalization method can be used. It should be noted that the normalization method used before applying the deep neural network model is the same as the normalization method used when training the deep neural network model. For example, for the maximum and minimum normalization method, the maximum and minimum values ​​used when normalizing the target soil moisture inversion data are the same as the maximum and minimum values ​​used when normalizing the sample data used to train the deep neural network model.

[0073] Step S130, downscaling the initial soil moisture data according to the TVDI data to obtain target soil moisture data at a second resolution.

[0074] Considering that both soil moisture and remote sensing spectral drought index have the ability to characterize drought, medium and high resolution spectral drought index can be used to participate in soil moisture downscaling. Among many drought indices, TVDI has a very significant negative correlation with soil moisture. TVDI is a temperature vegetation drought index that can well estimate the soil surface moisture conditions.

[0075] In some possible embodiments, the above step S130 may include the following sub-steps 131 to 133:

[0076] Sub-step 131 , performing bilinear interpolation on the initial soil moisture data to obtain simulated soil moisture data at a second resolution, where the simulated soil moisture data includes a simulated soil moisture value for each pixel.

[0077] The spatial resolution of the initial soil moisture data can be increased to the second resolution through bilinear interpolation to obtain simulated soil moisture data. The simulated soil moisture data obtained in this way will produce block effects and lack natural transitions, so it is necessary to use TVDI data for correction.

[0078] Sub-step 132, calculating statistical deviation data based on the TVDI data, the statistical deviation data including the statistical deviation coefficient of the TVDI value of each pixel within its preset neighborhood.

[0079] Each TVDI value within the preset neighborhood of the pixel can constitute a TVDI matrix, and the statistical deviation coefficient corresponding to the pixel can be the statistical deviation coefficient between the TVDI value of the pixel and its TVDI matrix. The preset neighborhood of the pixel can be set according to actual needs, and is not limited here. Optionally, the preset neighborhood of the pixel is a pixel area with the pixel as the center and the first resolution as the radius. For example, the preset neighborhood of the pixel is a 10km×10km pixel area with the pixel as the center. The statistical deviation coefficient can be calculated by the following formula:

[0080] δ TVDI =(TVDI i -TVDI μ ) / TVDI σ .

[0081] Among them, δ TVDI is the statistical deviation coefficient corresponding to pixel i in TVDI data, TVDI i is the TVDI value of pixel i in the TVDI data, TVDI μ is the average TVDI value of each pixel in the preset neighborhood of pixel i in the TVDI data, TVDI σ It is the standard deviation of the TVDI values ​​of each pixel in the preset neighborhood of pixel i in the TVDI data.

[0082] Sub-step 133, using the statistical deviation data to correct the simulated soil moisture data to obtain the target soil moisture data.

[0083] In a possible implementation, the target soil moisture data can be calculated using the following formula:

[0084] SMi =SM j +α×δ TVDI ;

[0085] Among them, SM i is the soil moisture value of pixel i in the target soil moisture data, SM j is the simulated soil moisture value of pixel i in the simulated soil moisture data, α is the preset deviation influence coefficient, δ TVDI is the statistical deviation coefficient corresponding to pixel i in the statistical deviation data.

[0086] The above α is a preset empirical value, for example, α=-0.4.

[0087] The high-resolution soil moisture inversion method provided by the embodiment of the present invention can obtain target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data, and albedo data; the second resolution is greater than the first resolution; soil moisture inversion is performed on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution; the initial soil moisture data is downscaled according to the TVDI data to obtain target soil moisture data at the second resolution. In this way, soil moisture inversion is performed on MWRI data, NDVI data, albedo data, etc. to obtain low-resolution initial soil moisture data, and then the initial soil moisture data is downscaled according to the high-resolution TVDI data, which can make full use of the ability of TVDI to characterize drought and improve the resolution and stability of soil moisture inversion.

[0088] The embodiment of the present invention also provides a training method for the above-mentioned deep neural network model, including:

[0089] A training sample library is obtained, wherein the training sample library includes a plurality of sample data at a first resolution obtained through preprocessing, data screening and normalization, wherein each sample data includes sample soil moisture inversion data and corresponding SMAP (Soil Moisture Active Passive, soil moisture active passive detection satellite) soil moisture product data; an initial deep neural network model is trained according to the training sample library to obtain a trained deep neural network model; wherein the sample soil moisture inversion data in each sample data is input, and the corresponding SMAP soil moisture product data is true value.

[0090] The above-mentioned preprocessing includes one or more of data analysis, projection conversion, spatial scale conversion and time scale aggregation. Among them, the SMAP soil moisture product data can be obtained through data analysis, projection conversion and spatial scale conversion. Data screening can ensure the validity of the data, prevent deviations caused by data anomalies, and ensure the representativeness and accuracy of the DNN training samples, so as to avoid the influence of large extreme values ​​in the preprocessed training data set on the DNN training accuracy. Optionally, the screening conditions for data screening may include: NDVI<0.9, and Albedo<0.4, and the consistency of the surface type label reaches the preset consistency threshold pixel, and meets the following brightness temperature screening conditions Cond:

[0091] Cond1=TB 3V ≤264K&TB 3V ≤(175K+0.49×TB 5V );

[0092] Cond2=(TB 2V -TB 2H )≥18K&(TB 2V -TB 4V )≤10K&(TB 4V -TBs V )≤10K;

[0093] Cond3=(TB 2V -TB 2H )≥20K;

[0094] Cond4=TB 5V ≥254K&(TB 2V -TB 2H )≥7K;

[0095] Cond=(Cond1&Cond2)|(~Cond1)&(Cond3|Cond4).

[0096] Among them, TB xV is the observed brightness temperature of the xth channel of FY3D with polarization V (i.e. vertical polarization), for example, TB 3V is the observed brightness temperature of the third channel V polarization of FY3D, TB xH is the observed brightness temperature of the xth channel of FY3D with H polarization (i.e. horizontal polarization), for example, TB 2H is the observed brightness temperature of the second channel H polarization of FY3D; Cond means that when Cond1 is satisfied, those that satisfy Cond2 are retained; when Cond1 is not satisfied, those that satisfy Cond3 or Cond4 are retained; in other cases, they are not retained.

[0097] The above consistency threshold can be set according to actual needs and is not limited here. For example, the consistency threshold is 0.8, which means that the surface types of 20 grids in the 5*5 grid at the 5km scale of the pixel are the same.

[0098] For ease of understanding, the above-mentioned high-resolution soil moisture inversion method is introduced in detail below.

[0099] In view of the defects and problems of the existing microwave soil moisture inversion methods, an embodiment of the present invention proposes a method for obtaining 1km soil moisture data using the MWRI remote sensing image (i.e., MWRI data) of the Fengyun-3D satellite (FY3D). The method can generate a 1km resolution VSM (VolumetricSoilMoisture) dataset from the FY3D MWRI remote sensing image and the corresponding 1km TVDI dataset. The VSM dataset of this resolution can provide support for research in ecology, forestry, agriculture, etc. At the same time, the framework of this method can also be applied to microwave imagers of other satellites to generate soil moisture data of corresponding resolution.

[0100] The overall idea of ​​the method provided by the embodiment of the present invention is: preprocessing FY3D MWRI data, SMAP soil moisture product data, and MODIS NDVI, Albedo and other data to prepare a training sample library for the soil moisture inversion model; constructing a deep neural network model, taking SMAP soil moisture product data as the true value, inputting FY3D MWRI data and variables such as NDVI and Aldebo to train the DNN model; using the preprocessed 10km resolution MWRI data and other model input data to obtain 10km resolution soil moisture results; using 1km LST and NDVI to calculate TVDI, establishing an empirical statistical model of TVDI and soil moisture, and obtaining the downscaled 1km soil moisture results.

[0101] The specific process is as follows:

[0102] 1. Build a training sample library

[0103] The FY3D MWRI data, SMAP soil moisture product data, MODIS NDVI, Albedo and other data were preprocessed to produce a training sample library for the soil moisture inversion model. The production process of the training sample library mainly includes satellite data preprocessing, quality control and normalization.

[0104] (1) Satellite data preprocessing

[0105] Computer raster processing tools are used to perform data analysis, projection conversion, spatial scale conversion, and time scale aggregation on FY3D MWRI L1 channel data, SMAP soil moisture product data, MODIS NDVI, Albedo, etc., to obtain a daily scale raster dataset with a resolution of 0.25° in the longitude and latitude coordinate system such as WGS84. Among them, the MODIS NDVI and Albedo satellite observation data are filtered and missing value interpolated using the Savitzky-Golay filter method, and the daily scale NDVI dataset and Albedo dataset are obtained using the linear averaging method; the MWRI orbit data are aggregated on a daily scale using the pixel averaging method; the FY3D MWRI L1 channel data and SMAP soil moisture product data are spatially scaled using bilinear interpolation or mean aggregation methods, so that they are converted to a global scale grid of 0.25° (global scale grid refers to a geographical range covering the world).

[0106] (2) Training sample library quality control

[0107] In order to ensure the validity of the data, prevent deviations caused by data anomalies, ensure the representativeness and accuracy of DNN training samples, and avoid the influence of large extreme values ​​in the preprocessed training data set on the DNN training accuracy as much as possible, this embodiment performs data screening on the input samples put into DNN training. When the vegetation is too dense, the microwave observation brightness temperature will produce uncertainty; when the homogeneity of low spatial resolution pixels is poor, the pixel surface type parameters cannot represent the actual surface coverage of the pixel; when the surface temperature is too low, the soil moisture will be affected by frozen soil, ice and snow, and can be approximately screened by microwave brightness temperature. Therefore, this embodiment sets the training set threshold to: NDVI <0.9, and Albedo <0.4, and the surface type label has good consistency of the pixel (such as the pixel has 20 grids with the same surface type in a 5×5 grid at a scale of 5km), and the following brightness temperature screening conditions Cond should be met:

[0108] Cond1=TB 3V ≤264K&TB 3V ≤(175K+0.49×TB 5V ); (1)

[0109] Cond2=(TB 2V -TB 2H )≥18K&(TB 2V -TB 4V )≤10K&(TB 4V -TB 5V )≤10K; (2)

[0110] Cond3=(TB2V -TB 2H )≥20K; (3)

[0111] Cond4=TB 5V ≥254K&(TB 2V -TB 2H )≥7K; (4)

[0112] Cond=(Cond1&Cond2)|(~Cond1)&(Cond3|Cond4); (5)

[0113] Among them, TB xV is the observed brightness temperature of the xth channel of FY3D with polarization V (i.e. vertical polarization), for example, TB 3V is the observed brightness temperature of the third channel V polarization of FY3D, TB xH is the observed brightness temperature of the xth channel of FY3D with H polarization (i.e. horizontal polarization), for example, TB 2H is the observed brightness temperature of the second channel H polarization of FY3D; Cond means that when Cond1 is satisfied, those that satisfy Cond2 are retained; when Cond1 is not satisfied, those that satisfy Cond3 or Cond4 are retained; in other cases, they are not retained.

[0114] (3) Data normalization

[0115] The main purpose of normalization is to normalize the units of all input data, which can unify the data into a certain range, achieve consistent iteration speed for gradient descent parameters, unify the range of each feature, make the iteration speed consistent, obtain the optimal solution at the same time, converge at the same time, reduce the number of iterations, and speed up the training of the model.

[0116] The maximum and minimum normalization method can be used, and the formula is:

[0117]

[0118] Among them, X nom is the normalized value, X is the original value, and X max and X min They are respectively the maximum and minimum values ​​of the original sequence (in the entire training sample library).

[0119] 2. DNN model training

[0120] DNN is a technology in the field of machine learning. It is a neural network model for deep learning developed on the basis of perceptron. DNN can be understood as a neural network with many hidden layers, also known as Deep Feedforward Network (DFN) and Multilayer Perceptron (MLP). Figure 2 As shown in the figure, according to the position of different layers, the neural network layers inside DNN can be divided into: input layer, hidden layer and output layer. Generally, the first layer is the input layer, the last layer is the output layer, and the layers in between are all hidden layers. The layers are fully connected, that is, any neuron in the i-th layer must be connected to any neuron in the i+1-th layer.

[0121] The core idea of ​​DNN is to establish a stable nonlinear function between input variables and outputs based on multi-source auxiliary data. In the DNN used in this embodiment, the output is SMAP soil moisture VSM, and the input is the channel brightness temperature data of the pixel position (including 10.65Ghz band (ie, the first channel) H polarization brightness temperature data, 10.65Ghz band V polarization brightness temperature data, 18.7Ghz band (ie, the second channel) H polarization brightness temperature data, 18.7Ghz band V polarization brightness temperature data) and auxiliary data describing pixel feature information (including normalized difference vegetation index NDVI, surface albedo Albedo, digital elevation model DEM, surface cover type LC, longitude and latitude Lat / lon and number of days in the year DOY, etc.). Therefore, DNN aims to establish the following relationship:

[0122] VSM=f DNN (TB 1H , TB 1V ,TB 2H , TB 2V ,NDVI,Albedo,DEM,LC,Lat,Lon,DOY); (7)

[0123] In formula (7), Albedo and LandCover represent the influence of surface characteristics, NDVI represents the influence of vegetation, DEM represents the influence of terrain, and Lat, Lon, and DOY represent the influence of three-dimensional data corresponding to geographic location and time.

[0124] After the model training is completed, there is no need to obtain SMAP soil moisture product data. It is only necessary to obtain real-time FY-3D or FY-3F MWRI channel brightness temperature data, as well as auxiliary data such as NDVI, Albedo, surface type and DEM. The trained machine learning model (i.e., DNN model) is used to predict soil moisture and obtain FY-3D or FY-3F MWRI 10km resolution soil moisture products with SMAP soil moisture characteristics.

[0125] 3. Soil moisture downscaling method based on TVDI

[0126] Considering that both soil moisture and remote sensing spectral drought index have the ability to characterize drought, medium and high resolution spectral drought index can be used to participate in soil moisture downscaling. Among many drought indices, the temperature vegetation drought index (TVDI) has a very significant negative correlation with soil moisture. Based on the negative correlation between soil moisture and TVDI, a spatial weight decomposition model is established using TVDI with a resolution of 1 km, and soil moisture is downscaled pixel by pixel, and finally a 1 km soil moisture product can be obtained.

[0127] TVDI is a temperature vegetation drought index that can be used to estimate the soil surface moisture conditions. The expression of TVDI is shown in the following formula:

[0128]

[0129] T Smax =a1+b1×NDVI; (9)

[0130] T Smin =a2+b2×NDVI; (10)

[0131] Among them, TVDI is the TVDI value of the pixel, T s is the surface temperature value of the pixel, T smin is the surface temperature of the wet edge (i.e. the temperature when the surface is completely wet), a2 and b2 are the linear fitting coefficients of the wet edge model, NDVI is the NDVI value of the pixel, T smax is the surface temperature of the dry side (i.e. the temperature when the surface is completely dry), and a1 and b1 are the linear fitting coefficients of the dry side model.

[0132] The wet edge model describes the temperature distribution when the surface is completely wet, and the surface temperature is lower than other surfaces with the same NDVI value. The linear fitting coefficients of the wet edge model can be updated in real time or periodically, or can be determined based on the values ​​of the same period in previous years. Taking real-time update as an example, the linear fitting coefficients of the wet edge model can be obtained as follows: divide the NDVI value range into multiple intervals, find all pixels whose NDVI values ​​belong to each interval in the current NDVI data, and use the minimum LST value in these pixels as the wet edge surface temperature corresponding to the interval; use NDVI as the horizontal coordinate and LST as the vertical coordinate to perform linear fitting on the wet edge surface temperatures corresponding to each interval to obtain the linear fitting coefficients of the wet edge model.

[0133] The dry edge model describes the temperature distribution when the surface is completely dry, and the surface temperature is higher than other surfaces with the same NDVI value. The linear fitting coefficients of the dry edge model can be updated in real time or periodically, or can be determined based on the values ​​of the same period in previous years. Taking real-time update as an example, the linear fitting coefficients of the dry edge model can be obtained as follows: divide the range of NDVI into multiple intervals, find all pixels whose NDVI values ​​belong to each interval in the current NDVI data, and use the maximum value of the LST value in these pixels as the dry edge surface temperature corresponding to the interval; use NDVI as the horizontal coordinate and LST as the vertical coordinate to perform linear fitting on the dry edge surface temperature corresponding to each interval, and you can get the linear fitting coefficients of the dry edge model.

[0134] The low-resolution soil moisture data is weighted pixel by pixel using high-resolution TVDI, and then the low spatial resolution soil moisture product is decomposed into a 1km soil moisture product using the weight. The spatial weight decomposition formula from the low-resolution pixel to the high-resolution pixel i contained in the pixel is as follows:

[0135] SM i =SM j +α×δ TVDI ; (11)

[0136] Among them, SM i is the soil moisture value of high-resolution pixel i, SM j is the simulated value of soil moisture at pixel i, α is the influence coefficient of the preset soil moisture on the deviation of TVDI (negative number), δ TVDIis the statistical deviation coefficient between the TVDI value of pixel i and its TVDI matrix. The TVDI matrix of pixel i is a matrix composed of the TVDI values ​​in the area with a radius of 10 km resolution and pixel i as the center. The simulated soil moisture value is obtained by bilinear interpolation of the original low-resolution soil moisture data based on the latitude and longitude coordinates of pixel i. Formula (11) shows that the difference in soil moisture relative to the baseline mean of adjacent pixels is mainly caused by the deviation of TVDI relative to the mean, and the more TVDI deviates from the mean, the more soil moisture deviates from the mean in the opposite direction.

[0137] For ease of understanding, the present invention also provides Figure 3 The flowchart of another high-resolution soil moisture inversion method is shown in FIG. The overall implementation process of this method is introduced below: Figure 3 As shown in the figure, the model training is carried out through the following process: 1.1 data reading, the data read include FY3D brightness temperature data (i.e. FY3D MWRI data), SMAP soil moisture product data and auxiliary data, the auxiliary data include DEM, LC, NDVI and Albedo, etc.; 1.2 data preprocessing to obtain a training sample library; 1.3 machine learning model training, that is, using the training sample library to train the machine learning model to obtain the FY3D-SMAP machine learning model training result (i.e. the trained DNN model). Soil moisture inversion is performed through the following process: 2.1 Data reading, the read data includes FY3D-SMAP machine learning model training results, real-time FY3D brightness temperature data and auxiliary data, including DEM, LC, NDVI and Albedo, etc.; 2.2 Model prediction, that is, inputting real-time FY3D brightness temperature data and auxiliary data into the trained DNN model to obtain 10km soil moisture prediction results; 2.3 TVDI calculation, that is, using FY3D LST and FY3D NDVI to calculate TVDI; 2.4 Downscaling calculation, that is, using 10km soil moisture prediction results and calculated TVDI data for downscaling; 2.5 Data output, that is, outputting the 1km soil moisture results obtained by downscaling.

[0138] Corresponding to the above-mentioned high-resolution soil moisture inversion method, the embodiment of the present invention further provides a high-resolution soil moisture inversion device. Figure 4 A schematic diagram of a high-resolution soil moisture inversion device is shown, the device comprising:

[0139] An acquisition module 401 is used to acquire target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to a target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data and albedo data; and the second resolution is greater than the first resolution;

[0140] An inversion module 402 is used to perform soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at a first resolution;

[0141] The downscaling module 403 is used to downscale the initial soil moisture data according to the TVDI data to obtain target soil moisture data at a second resolution.

[0142] The high-resolution soil moisture inversion device provided by the embodiment of the present invention can obtain target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data, and albedo data; the second resolution is greater than the first resolution; soil moisture inversion is performed on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution; the initial soil moisture data is downscaled according to the TVDI data to obtain target soil moisture data at the second resolution. In this way, soil moisture inversion is performed on MWRI data, NDVI data, albedo data, etc. to obtain low-resolution initial soil moisture data, and then the initial soil moisture data is downscaled according to the high-resolution TVDI data, which can make full use of the ability of TVDI to characterize drought and improve the resolution and stability of soil moisture inversion.

[0143] Furthermore, the acquisition module 401 is specifically used to: obtain original soil moisture inversion data and original surface temperature data of the target area within a target time range, the original soil moisture inversion data including original MWRI data, original NDVI data and original albedo data; preprocess the original soil moisture inversion data and the original surface temperature data to obtain target soil moisture inversion data at a first resolution, and second NDVI data and target surface temperature data at a second resolution; wherein the preprocessing includes one or more of data analysis, projection conversion, spatial scale conversion and time scale aggregation; and generate TVDI data based on the second NDVI data and the target surface temperature data.

[0144] Furthermore, the acquisition module 401 is also used to calculate the TVDI data using the following formula:

[0145]

[0146] T Smax =a1+b1×NDVI;

[0147] T Smin =a2+b2×NDVI;

[0148] Among them, TVDI is the TVDI value of the pixel in the TVDI data, T sis the surface temperature value of the pixel in the target surface temperature data, T smin is the surface temperature of the wet edge, a2 and b2 are the linear fitting coefficients of the wet edge model, NDVI is the NDVI value of the pixel in the second NDVI data, T smax is the surface temperature of the dry side, a1 and b1 are the linear fitting coefficients of the dry side model.

[0149] Furthermore, the above-mentioned inversion module 402 is specifically used to: normalize the target soil moisture inversion data to obtain normalized target soil moisture inversion data; input the normalized target soil moisture inversion data into the trained deep neural network model to obtain the first soil moisture data output by the deep neural network model; and denormalize the first soil moisture data to obtain initial soil moisture data.

[0150] Furthermore, the downscaling module 403 is specifically used to: perform bilinear interpolation on the initial soil moisture data to obtain simulated soil moisture data at a second resolution, the simulated soil moisture data including a simulated soil moisture value for each pixel; calculate statistical deviation data based on the TVDI data, the statistical deviation data including a statistical deviation coefficient of the TVDI value of each pixel within its preset neighborhood; and use the statistical deviation data to correct the simulated soil moisture data to obtain target soil moisture data.

[0151] Furthermore, the downscaling module 403 is also used to calculate the target soil moisture data by the following formula:

[0152] SM i =SM j +α×δ TVDI ;

[0153] Among them, SM i is the soil moisture value of pixel i in the target soil moisture data, SM j is the simulated soil moisture value of pixel i in the simulated soil moisture data, α is the preset deviation influence coefficient, δ TVDI is the statistical deviation coefficient corresponding to pixel i in the statistical deviation data.

[0154] Furthermore, the above-mentioned device also includes a training module, which is used to: obtain a training sample library, the training sample library includes multiple sample data at a first resolution obtained through preprocessing, data screening and normalization, each sample data includes sample soil moisture inversion data and corresponding SMAP soil moisture product data; train the initial deep neural network model according to the training sample library to obtain a trained deep neural network model; wherein the sample soil moisture inversion data in each sample data is input, and the corresponding SMAP soil moisture product data is the true value.

[0155] The high-resolution soil moisture inversion device provided in this embodiment has the same implementation principle and technical effects as those of the aforementioned high-resolution soil moisture inversion method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the high-resolution soil moisture inversion device, reference may be made to the corresponding contents in the aforementioned high-resolution soil moisture inversion method embodiment.

[0156] like Figure 5 As shown, an electronic device 500 provided by an embodiment of the present invention includes: a processor 501, a memory 502 and a bus, the memory 502 stores a computer program that can be run on the processor 501, when the electronic device 500 is running, the processor 501 and the memory 502 communicate through the bus, and the processor 501 executes the computer program to implement the above-mentioned high-resolution soil moisture inversion method.

[0157] Specifically, the memory 502 and the processor 501 can be general-purpose memories and processors, which are not specifically limited here.

[0158] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the high-resolution soil moisture inversion method in the previous method embodiment is executed. The computer-readable storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a RAM, a magnetic disk, or an optical disk, etc., various media that can store program codes.

[0159] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0160] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0161] The flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and a part of a module, a program segment or a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of the boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0162] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms.

[0163] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-resolution soil moisture inversion method, characterized in that: include: Obtain target soil moisture inversion data at a first resolution and temperature vegetation drought index TVDI data at a second resolution corresponding to the target area; wherein the target soil moisture inversion data includes microwave imager MWRI data, normalized difference vegetation index NDVI data and albedo data; the second resolution is greater than the first resolution; Performing soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution; The initial soil moisture data is downscaled according to the TVDI data to obtain target soil moisture data at the second resolution.

2. The method according to claim 1, characterized in that: The step of obtaining target soil moisture inversion data at a first resolution and temperature vegetation drought index TVDI data at a second resolution corresponding to the target area includes: Acquire original soil moisture inversion data and original surface temperature data of the target area within a target time range, wherein the original soil moisture inversion data includes original MWRI data, original NDVI data, and original albedo data; Preprocessing the original soil moisture inversion data and the original surface temperature data to obtain the target soil moisture inversion data at the first resolution, and the second NDVI data and target surface temperature data at the second resolution; wherein the preprocessing includes one or more of data parsing, projection conversion, spatial scale conversion, and time scale aggregation; The TVDI data is generated according to the second NDVI data and the target land surface temperature data.

3. The method according to claim 2, characterized in that The step of generating the TVDI data according to the second NDVI data and the target surface temperature data includes: The TVDI data is calculated by the following formula: Tsmax=a1+b1×NDVI Tsmin=a2+b2×NDVI Wherein, TVDI is the TVDI value of the pixel in the TVDI data, T s is the surface temperature value of the pixel in the target surface temperature data, Tsmi n is the surface temperature of the wet edge, a2 and b2 are the linear fitting coefficients of the wet edge model, NDVI is the NDVI value of the pixel in the second NDVI data, Tsma x is the surface temperature of the dry side, a1 and b1 are the linear fitting coefficients of the dry side model.

4. The method according to claim 1, characterized in that: The step of performing soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution includes: Normalizing the target soil moisture inversion data to obtain normalized target soil moisture inversion data; Inputting the normalized target soil moisture inversion data into the trained deep neural network model to obtain first soil moisture data output by the deep neural network model; The first soil moisture data is subjected to a denormalization process to obtain the initial soil moisture data.

5. The method according to claim 1, characterized in that The downscaling the initial soil moisture data according to the TVDI data to obtain target soil moisture data at the second resolution includes: Performing bilinear interpolation on the initial soil moisture data to obtain simulated soil moisture data at the second resolution, wherein the simulated soil moisture data includes a simulated soil moisture value of each pixel; Calculating statistical deviation data based on the TVDI data, wherein the statistical deviation data includes a statistical deviation coefficient of the TVDI value of each pixel within its preset neighborhood; The simulated soil moisture data is corrected using the statistical deviation data to obtain the target soil moisture data.

6. The method according to claim 5, characterized in that The method of correcting the simulated soil moisture data by using the statistical deviation data to obtain the target soil moisture data includes: The target soil moisture data is calculated by the following formula: SM=SMj+α×6 TVDI . Where SMi is the soil moisture value of pixel i in the target soil moisture data, SM j is the simulated soil moisture value of pixel i in the simulated soil moisture data, α is the preset deviation influence coefficient, δ TVDI is the statistical deviation coefficient corresponding to pixel i in the statistical deviation data.

7. The method according to claim 4, characterized in that The method further comprises: Acquire a training sample library, wherein the training sample library includes a plurality of sample data at the first resolution obtained through preprocessing, data screening and normalization, each of the sample data including sample soil moisture inversion data and corresponding SMAP soil moisture product data; The initial deep neural network model is trained according to the training sample library to obtain the trained deep neural network model; wherein the sample soil moisture inversion data in each of the sample data is input, and the corresponding SMAP soil moisture product data is the true value.

8. A high-resolution soil moisture inversion device, characterized in that: include: An acquisition module, used to acquire target soil moisture inversion data at a first resolution and TVDI data at a second resolution corresponding to a target area; wherein the target soil moisture inversion data includes MWRI data, NDVI data and albedo data; and the second resolution is greater than the first resolution; An inversion module, used to perform soil moisture inversion on the target soil moisture inversion data to obtain initial soil moisture data at the first resolution; A downscaling module is used to downscale the initial soil moisture data according to the TVDI data to obtain target soil moisture data at the second resolution.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the high-resolution soil moisture inversion method described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the high-resolution soil moisture inversion method according to any one of claims 1 to 7 is executed.

Images