A method for filling orbital gaps in spaceborne passive microwave brightness temperature imaging

Abstract

This invention discloses a method for filling orbital gaps in spaceborne passive microwave brightness temperature imagery. This method belongs to the field of passive microwave remote sensing technology and can be used to fill orbital gaps in spaceborne passive microwave brightness temperature imagery in areas with special underlying surfaces such as snow and frozen soil. The method first uses indicator factors calculated from brightness temperatures across different channels to characterize special underlying surfaces such as snow and frozen soil. Secondly, it leverages the powerful one-dimensional nonlinear fitting capabilities of a CNN-LSTM model to capture the temporal characteristics of brightness temperatures. Finally, the method fills orbital gaps in spaceborne passive microwave brightness temperature imagery using only the brightness temperature data to be filled. This method fully considers the impact of special underlying surfaces such as snow and frozen soil on brightness temperature when filling missing brightness temperatures. Therefore, the method is applicable not only to low-altitude areas with stable annual brightness temperature variations, but also to high-altitude areas with drastic annual brightness temperature variations. Furthermore, the method utilizes only brightness temperature data to fill orbital gaps in brightness temperature imagery, leveraging the temporal characteristics of the data itself. This method is highly scalable and practical, and can be used for various types of spaceborne passive microwave brightness temperature imagery.

Description

Technical Field

[0001] The present invention belongs to the field of passive microwave remote sensing, and in particular relates to a method for filling orbital gaps in spaceborne passive microwave brightness temperature images based on physical factors and time series characteristics. Background Art

[0002] Passive microwave brightness temperature images generated by spaceborne passive microwave imagers are widely used in surface parameter inversion and surface condition monitoring. Passive microwave brightness temperature data can be used to generate important surface parameter products such as soil moisture, land surface temperature, sea surface temperature, and snow depth. Brightness temperature data are also widely used in vegetation monitoring and soil freeze-thaw transformation research. However, currently, all satellites carrying passive microwave imagers are polar-orbiting satellites, such as the DMSP SSM / I, Aqua AMSR-E, and GCOM-W1AMSR-2. Due to the limited bandwidth of passive microwave imagers, brightness temperature images obtained by polar-orbiting passive microwave imagers have observation gaps between adjacent orbits, known as orbital gaps. These gaps gradually widen with decreasing latitude. Furthermore, these orbital gaps in brightness temperature images are propagated to secondary products derived from them, resulting in spatiotemporal discontinuities in the secondary products and limiting their further application. Therefore, filling the orbital gaps in spaceborne passive microwave brightness temperature images is of great practical value.

[0003] Currently, research on filling missing values ​​in remote sensing data primarily focuses on surface parameter products, while relatively little research has been conducted on filling orbital gaps in spaceborne passive microwave brightness temperature imagery. Methods for filling missing values ​​in remote sensing data can generally be divided into two categories, depending on whether or not auxiliary data are used. Methods that use auxiliary data primarily aim to fill missing values ​​in the source data directly or indirectly through the auxiliary data. However, these methods are limited by the quality of the auxiliary data and have poor scalability. Methods that use the source data alone to fill missing values ​​through interpolation are the other way around. Common interpolation methods include kriging, empirical orthogonal function interpolation, and inverse distance weighted method (IDWM). These methods offer good data independence and a wide range of applications. However, their effectiveness and accuracy decline when there are a large number of missing data. For spaceborne passive microwave brightness temperature imagery, data gaps due to orbital gaps are regional, making them difficult to fill using spatial interpolation methods. One currently feasible approach is to use FY MWRI passive microwave brightness temperature data to fill orbital gaps in AMSR-2 brightness temperature imagery. However, due to the varying observation frequencies and transit times of different passive microwave imagers, this method cannot be extended to other spaceborne passive microwave brightness temperature data. Furthermore, passive microwave brightness temperature is significantly affected by snow cover and frozen ground, factors that must also be considered when filling orbital gaps. Therefore, it is necessary to develop a spaceborne passive microwave brightness temperature imagery orbital gap filling method that is suitable for specific underlying surfaces (snow cover and frozen ground) and uses only brightness temperature data. Summary of the Invention

[0004] The present invention discloses a method for filling orbital gaps in spaceborne passive microwave brightness temperature images, which is used to obtain spatially seamless passive brightness temperature images so as to generate spatially seamless secondary products such as soil moisture content and surface temperature.

[0005] The technical solution adopted in the present invention is:

[0006] A method for filling orbital gaps in spaceborne passive microwave brightness temperature imaging, the method comprising the following steps:

[0007] Step 1: Collect the annual pixel brightness temperature sequence of the passive microwave brightness temperature image pixels;

[0008] Step 2: Due to the orbital gap, the brightness temperature series of pixels within the year is incomplete. The missing values ​​of the brightness temperature series of pixels within the year are filled using the time interpolation method so that they can be input into the constructed filling model later.

[0009] Step 3: Use the passive microwave brightness temperature of different channels to calculate the frozen soil indicator factor (PR), snow indicator factor (SD), soil water content indicator factor (SM), time series indicator factor (t d );

[0010] Step 4: Build a multi-factor passive microwave brightness temperature image track gap filling model based on convolutional neural network (CNN) and long short-term memory network (LSTM);

[0011] The input of the filling model is the historical time series of a specified time step of a single pixel, which includes: pixel brightness temperature series, frozen soil indicator factor series, snow cover indicator factor series, soil water content indicator factor series and time series indicator factor; the output of the filling model is the brightness temperature filling estimate of a single pixel;

[0012] Step 5: Using the pixel brightness temperature sequence, frozen soil indicator factor sequence, snow cover indicator factor sequence, soil water content indicator factor sequence, and time series indicator factor sequence within the specified time step as input data, the filling model is trained;

[0013] Use the simulated missing area to verify the currently trained filling model. If the verification passes, proceed to step 7; otherwise, continue training the current filling model until it passes the verification.

[0014] Step 7: Fill the brightness temperature of the pixels in the target area located in the track gap based on the filling model;

[0015] Extract the historical time series of the specified time step for each target pixel located in the track gap in the target area. If there is missing data in the pixel brightness temperature series in the historical time series, interpolation method is used to fill it.

[0016] The indicator factor and brightness temperature historical time series of the target pixel are input into the verified filling model, and the brightness temperature filling estimate of the target pixel is obtained based on the output of the filling model, and the brightness temperature filling estimate is used to replace the missing value of the target pixel.

[0017] Furthermore, in step 1, the selected acquisition area is sampled at intervals, a specified number of pixels are evenly selected, and the annual pixel brightness temperature sequence of the selected pixels is obtained based on the pixel brightness temperature.

[0018] Furthermore, in step 2, a time window of size 4 is used to fill the missing values ​​of the pixel brightness temperature series within the year based on the window mean.

[0019] Furthermore, in step 4, the constructed filling model sequentially includes multiple convolutional network layers, multiple long short-term memory network layers, and at least one fully connected layer. The convolutional network layers sequentially include one-dimensional convolutional layers and activation functions, and the final fully connected layer uses a sigmoid function for linear transformation to obtain the output of the filling model. Optimally, the number of convolutional network layers is 2, the number of multi-layer long short-term memory network layers is 3, and the number of fully connected layers is 1.

[0020] Furthermore, when training the filling model, the loss function used is the mean square error.

[0021] Furthermore, in step 7, if there are missing data in the pixel brightness temperature sequence in the historical time series, the missing data are filled using the average of the observations of the previous day and the next day.

[0022] The technical solution provided by the present invention brings at least the following beneficial effects:

[0023] This paper proposes a method for filling orbital gaps in passive microwave brightness temperature imagery that uses only passive microwave brightness temperature data and is applicable to areas with unique underlying surfaces, such as frozen ground and snow. This method comprehensively considers the effects of frozen ground state transitions, snow scattering, and soil moisture content changes on passive microwave brightness temperature time series. Therefore, the method is applicable not only to low-altitude areas with stable brightness temperature time series, but also to high-altitude areas with snow and frozen ground, such as the Qinghai-Tibet Plateau. Furthermore, the method uses only the passive microwave brightness temperature data to be filled, without any other data. Therefore, the method is highly data-independent and applicable to different types of spaceborne passive microwave data. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 A schematic flow chart of a method for filling orbital gaps in spaceborne passive microwave brightness temperature imaging provided by an embodiment of the present invention;

[0026] Figure 2 is an example regional elevation distribution map in an embodiment of the present invention;

[0027] Figure 3 A diagram of a network structure used in an embodiment of the present invention;

[0028] Figure 4 The simulation results of the missing region verification in the embodiment of the present invention are shown in Figure 1, where (a)-(j) are scatter plots of the original and reconstructed brightness temperatures of different channels, and (k)-(l) are error distribution diagrams. In the figure, V represents the vertical polarization channel, and H represents the horizontal polarization channel.

[0029] Figure 5 The original AMSR-2 images and reconstructed brightness temperature images of different channels in the simulated missing area in the embodiment of the present invention are shown. In the figure, V represents the vertical polarization channel, and H represents the horizontal polarization channel.

[0030] Figure 6 1 is a comparison chart of the original AMSR-2 brightness temperature image and the padded spatial seamless brightness temperature image on different dates in an embodiment of the present invention.

[0031] Figure 7 The figures show the original AMSR-2 36GHz brightness temperature images and the brightness temperature images after orbital gap filling on different dates in the embodiments of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in detail and completely in conjunction with the drawings in the implementation of the present invention. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings can be arranged and designed using different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present invention.

[0033] The present invention provides a method for filling orbital gaps in spaceborne passive microwave brightness temperature imagery. First, the method utilizes indicator factors calculated from brightness temperatures across different channels to characterize special underlying surfaces, such as snow and frozen ground. Second, the powerful one-dimensional nonlinear fitting capabilities of the CNN-LSTM model (convolutional neural network-long short-term memory network) are combined to capture the temporal characteristics of brightness temperature. Finally, orbital gaps in spaceborne passive microwave brightness temperature imagery are filled using only the brightness temperature data to be filled. The method proposed in this embodiment of the present invention fully considers the impact of special underlying surfaces, such as snow and frozen ground, on brightness temperature when filling missing brightness temperatures. Therefore, the method proposed in this embodiment of the present invention is applicable not only to low-altitude areas with stable annual brightness temperature variations, but also to high-altitude areas with drastic annual brightness temperature variations. Furthermore, the present invention utilizes only brightness temperature data to fill orbital gaps in brightness temperature imagery, taking into account the temporal characteristics of the data itself. Therefore, the present invention exhibits excellent scalability and practicality, and can serve various spaceborne passive microwave brightness temperature imagery applications.

[0034] like Figure 1 As shown, the embodiment of the present invention provides a method for filling orbital gaps in satellite-borne passive microwave brightness temperature images, which includes nine steps: S1, extracting the intra-year brightness temperature sequence of the passive microwave brightness temperature image pixels in the training and verification data; S2, performing temporal interpolation on the missing data; S3, calculating the permafrost, snow cover, soil water content and time indicator factors; S4, constructing a passive microwave brightness temperature orbital gap filling model; S5, using the training data to train the constructed model; S6, constructing a simulated missing area verification model, i.e., using the simulated missing area for verification; S7, extracting the time series data of the pixels in the brightness temperature image to be filled and interpolating the missing values ​​in the extracted pixel time series; S8, inputting the processed data into the model to obtain the filling value; S9, traversing all pixels in the brightness temperature image to be filled that are located in the orbital gap until a spatially seamless passive microwave brightness temperature image is generated. The specific processing process of each step is as follows:

[0035] Step S1: Extract the annual brightness temperature sequence of the passive microwave brightness temperature image pixels within the sample area.

[0036] This example uses the AMSR-2L3 passive microwave brightness temperature data from 2020, 2021, and 2022 as an example (the 2020 data is used for training, the 2021 data is used for verification, and the 2022 data is used for testing). Figure 2 As shown, the example region ranges from 73°E to 103°E and 26°N to 40°N. The AMSR-2L3 passive microwave brightness temperature used in this example includes five channels: 7, 10, 18.7, 36, and 89 GHz, each of which includes both vertical and horizontal polarization modes.

[0037] In this step, we first sample the sample area at intervals, evenly selecting 500 pixels to ensure that all land cover types are covered. Then, we extract the annual brightness temperature series for the sample points.

[0038] Step S2: interpolate the missing values ​​of the brightness temperature series extracted in step S1.

[0039] Due to missing observations caused by orbital gaps, the extracted brightness temperature time series is not continuous. Model training requires a complete brightness temperature time series as input, so missing values ​​need to be filled. By analyzing the AMSR-2 orbital gap pattern, it is found that pixels are not missing for more than three consecutive days. Therefore, temporal interpolation is used in this step to fill in missing values. The time window size is 4, and the window mean is used to fill in missing values.

[0040] Step S3: Calculate frozen soil, snow cover, soil water content and time series indicator factors.

[0041] The example area has a seasonally frozen underlying surface. When the soil freezes, the emissivity increases, causing a sudden increase in brightness temperature. When the soil is frozen, the emissivity is stable and changes little, and the brightness temperature changes are relatively stable during this stage. When the soil thaws, the emissivity will drop rapidly, causing the brightness temperature to drop sharply. In addition, during the thawing period, the soil moisture changes greatly, causing the emissivity to fluctuate greatly, which in turn causes the brightness temperature to fluctuate greatly. Because the polarization ratio (PR) is sensitive to the transformation of the frozen soil state, the polarization ratio is selected as the frozen soil indicator factor, and the calculation method is as follows:

[0042]

[0043] Where, and They represent the brightness temperature of the vertical polarization channel and the brightness temperature of the horizontal polarization channel at frequency f, respectively.

[0044] The soil moisture indicator factor (SM) is calculated as follows:

[0045]

[0046] Where, Tb 10H and Tb 36H Represent the brightness temperatures of the 10 GHz and 36 GHz horizontal polarization channels, respectively.

[0047] Some pixels in the example area are covered with snow. Snow accumulation can cause strong volume scattering of passive microwave radiation, resulting in a sharp drop in the brightness temperature on the satellite. In this embodiment of the present invention, the snow depth indicator factor (SM) is used to characterize snow accumulation. The calculation method is as follows:

[0048] SD=Tb 18H -Tb 36H (3)

[0049] Where, Tb 18H and Tb 36H Represent the brightness temperatures of the 18 GHz and 36 GHz horizontal channels, respectively.

[0050] In addition, brightness temperature varies with the seasons within a year, with certain patterns. Therefore, the present embodiment also uses the accumulated days per year (DOY) as an input factor. However, since the accumulated days per year cannot reflect the regularity of brightness temperature changes with the seasons, it needs to be converted. The conversion method is as follows:

[0051]

[0052] Where T is 365 or 366 (leap year), which represents the number of days in a year; DOY i is the i-th day of the year; t d is the converted timing indicator factor.

[0053] Step S4: Construct a multi-factor spaceborne passive microwave brightness temperature image orbit gap filling model.

[0054] In this embodiment, a spaceborne passive microwave brightness temperature image orbit gap filling model (CNN-LSTM model for short) is constructed based on convolutional neural network and long short-term memory network. The model consists of an input layer, two CNN layers, three LSTM layers, and a fully connected layer (i.e., output layer). Figure 3 As shown in the figure, the CNN layer in the model primarily consists of convolutional layers and activation functions. The convolutional layers extract high-dimensional features from the input data, while the activation functions use ReLU for nonlinear transformations. The LSTM layers primarily construct long-term dependencies from the features extracted by the CNN, while the fully connected layers use the Sigmoid function for linear transformations. The fully connected layers are placed after the LSTM layers to improve the model's ability to capture nonlinear changes while reducing overfitting. The combination of these two models leverages the strengths of each to bridge the gap in passive microwave brightness temperature imagery.

[0055] In this embodiment, the specific parameters involved in the constructed model are as follows: the convolution kernel of the CNN layer is one-dimensional, with a total of 64 kernels, each with a size of 3, and the stride and padding are both set to 1; each LSTM layer has 64 LSTM units.

[0056] Step S5: Train the model designed in step S4.

[0057] The input time step is 5, that is, the input of the model is the PR, SD, SM, BT, t of the previous 5 days. dTime series data, outputting brightness temperatures at the time to be filled. In practical applications, time series data with other time steps can also be input as needed. Generally, if the time step is too long, the model will have difficulty capturing sudden changes in the time series; if the time step is too short, the model will be difficult to apply due to missing data. During the training phase, the batch size is 370, the number of iterations is 50, and the loss function is the mean squared error (MSE).

[0058] Step S6: Verify using simulated missing regions.

[0059] Since there are no observed brightness temperatures at the track gaps, the model filling results cannot be verified at these gaps. In this embodiment, the model effectiveness is verified by simulating missing regions. This simulated missing region assumes that a region with no missing observations on a particular day is missing due to track gaps. This region is then reconstructed using the filling model constructed in this embodiment of the present invention. The model is verified by comparing the original brightness temperature of the simulated missing region with the reconstructed brightness temperature.

[0060] In this example, the range of the simulated missing area is from 80°E to 83°E and from 32°N to 35°N. The observation time of the simulated missing area is March 26, 2022. In order to qualitatively evaluate the accuracy, the coefficient of determination (R 2 ), mean square error (MSE) and root mean square error (RMSE) are used as evaluation indicators, and the calculation method is as follows:

[0061]

[0062] Where n represents the number of samples; x and y represent the observed value and the estimated value (i.e., the output of the imputation model), respectively; and represent the means of x and y respectively.

[0063] Figure 4 The validation results of the simulated regional model are presented (MBE: -0.26K-0.15K; RMSE: 0.93K-2.59K; R 2 :0.98-0.99). From the verification results, the reconstructed brightness temperature is highly consistent with the original brightness temperature, indicating that the model has good accuracy. Figure 4 (k) and Figure 4 As shown in (l), as the frequency increases, the error distribution becomes wider and the accuracy decreases accordingly. In addition, due to its high sensitivity to the atmosphere, the reconstruction accuracy of the 89 GHz channel is lower than that of other channels (RMSE is 3.50K at night and 3.86K during the day).

[0064] Figure 5The original AMSR-2 brightness temperature image of the simulated area is shown alongside the reconstructed brightness temperature image. A comparison shows that the reconstructed brightness temperature is highly consistent with the original AMSR-2 brightness temperature in terms of spatial pattern and amplitude, with image detail becoming more pronounced as the frequency increases. Furthermore, at the same frequency, the brightness temperature of the vertically polarized channel is higher than that of the horizontally polarized channel, and the difference between the two decreases with increasing frequency.

[0065] Step S7: performing data preprocessing on the brightness temperature image to be filled.

[0066] For a passive microwave brightness temperature image, if there is an orbital gap, sequentially extract the five-day time series of the pixels in the orbital gap, interpolate the missing data in the brightness temperature series of the previous five days, and generate the corresponding indicator factor sequence according to step S3. If there is no orbital gap, exit.

[0067] In this step, the interpolation method for missing values ​​from the previous five days differs from that used during model training. Specifically, the missing values ​​are handled using the average of the observations from the previous and next day. This approach narrows the interpolation window and makes the model more flexible.

[0068] Step S8: Fill in the missing values ​​using the data extracted in step S7.

[0069] The data extracted in step S7 is input into the verified model to obtain the filling value, and the filling value is used to replace the missing value at the corresponding image position. The filling process can be expressed as:

[0070]

[0071] Where, is the brightness temperature estimate of the pixel to be filled; x and y are the pixel locations; t represents the observation date; PR, SM, T, SD, and BT represent the frozen soil indicator factor vector, soil water content indicator factor vector, time series indicator factor vector, snow cover indicator factor vector, and annual pixel brightness temperature vector, which are composed of the observation values ​​5 days before the observation date t.

[0072] Step S9: Generate a spatially seamless brightness temperature image after filling the track gaps corresponding to the original brightness temperature image.

[0073] Traverse all pixels at the track gaps of the brightness temperature image to be filled, and repeat steps S7-S9 to generate a spatially seamless passive microwave brightness temperature image after the track gaps are filled.

[0074] Figure 6 The original AMSR-2 7GHz brightness temperature images and the brightness temperature images after orbital gap filling on different dates are shown. Figure 7The original AMSR-2 36GHz brightness temperature images and the brightness temperature images after orbital gap filling on different dates are shown. Figure 6 and Figure 7 It can be seen that the data missing ratio of the original AMSR-2 brightness temperature image in the example area due to the orbital gap is between 28% and 38%. The method proposed in the embodiment of the present invention is used to fill all the missing values ​​in the original brightness temperature image, and a spatially seamless brightness temperature image is generated. The brightness temperature image after orbital gap filling is highly consistent with the original AMSR-2BT image in terms of spatial pattern and texture information, and no obvious boundary effect is produced at the edge of the orbital gap. In addition, due to the influence of the surface and the atmosphere, the 36GHz brightness temperature image shows richer details and more obvious spatial changes. In summary, the method for orbital gap filling for satellite-borne passive microwave brightness temperature images proposed in the embodiment of the present invention has good filling quality at different frequencies and polarization modes, and is applicable to different seasons.

[0075] 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

[0076] The above are only some embodiments of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the inventive concept of the present invention, which all fall within the scope of protection of the present invention.

Claims

1. A method for filling orbital gaps in spaceborne passive microwave brightness temperature imaging, characterized in that: The following steps are involved: Step 1: Collect the annual pixel brightness temperature sequence of the passive microwave brightness temperature image pixels; Step 2: Use the time interpolation method to fill the missing values ​​of the pixel brightness temperature series within the year; Step 3: Use the passive microwave brightness temperature of different channels to calculate the frozen soil indicator factor, snow cover indicator factor, soil water content indicator factor, and time series indicator factor; Step 4: Construct a multi-factor passive microwave brightness temperature image track gap filling model based on convolutional neural network and long short-term memory network; The input of the filling model is the historical time series of a specified time step of a single pixel, which includes: pixel brightness temperature series, frozen soil indicator factor series, snow cover indicator factor series, soil water content indicator factor series and time series indicator factor; the output of the filling model is the brightness temperature filling estimate of a single pixel; Step 5: Using the pixel brightness temperature sequence, frozen soil indicator factor sequence, snow cover indicator factor sequence, soil water content indicator factor sequence, and time series indicator factor sequence within the specified time step as input data, the filling model is trained; Step 6: Use the simulated missing area to verify the currently trained filling model. If the verification passes, proceed to step 7; otherwise, continue training the current filling model until it passes the verification. Step 7: Fill the brightness temperature of the pixels in the target area located in the track gap based on the filling model; Extract the historical time series of the specified time step for each target pixel located in the track gap in the target area. If there is missing data in the pixel brightness temperature series in the historical time series, interpolation method is used to fill it. The target pixel indicator factor and the brightness temperature historical time series are input into the verified filling model, and the brightness temperature filling estimate of the target pixel is obtained based on the output of the filling model, and the brightness temperature filling estimate is used to replace the missing value of the target pixel.

2. The method according to claim 1, wherein In step 1, the selected collection area is sampled at intervals, a specified number of pixels are evenly selected, and the annual pixel brightness temperature sequence of the selected pixels is obtained based on the pixel brightness temperature.

3. The method according to claim 1, wherein In step 2, a time window of size 4 is used to fill the missing values ​​of the pixel brightness temperature series within the year based on the window mean.

4. The method according to claim 1, wherein In step 3, the frozen soil indicator factor, snow cover indicator factor, soil water content indicator factor, and time series indicator factor are as follows: 1) The calculation method of frozen soil indicator factor RP is: in, and They represent the brightness temperature of the vertical polarization channel and the brightness temperature of the horizontal polarization channel at frequency f respectively; 2) The calculation method of soil moisture indicator factor SM is: Among them, Tb 10H and Tb 36H represent the brightness temperature of 10 GHz and 36 GHz horizontal polarization channels respectively; 3) The calculation method of snow cover indicator factor SM is: SD=Tb 18H -Tb 36H Among them, Tb 18H and Tb 36H Represent the brightness temperature of the 18 GHz and 36 GHz horizontal channels respectively; 4) Timing indicator factor t d The calculation method is: Where T represents the number of days in a year, DOY i is the i-th day of the year.

5. The method according to claim 1, wherein In step 4, the constructed filling model includes multiple convolutional network layers, multiple long short-term memory network layers and at least one fully connected layer in sequence, wherein the convolutional network layer includes a one-dimensional convolutional layer and an activation function in sequence, and the last fully connected layer uses a Sigmoid function for linear transformation to obtain the output of the filling model.

6. The method according to claim 5, wherein The number of convolutional network layers is 2, the number of multi-layer long short-term memory network layers is 3, and the number of fully connected layers is 1.

7. The method according to claim 5, wherein The convolutional kernel size of the convolutional network layer is 3, and the stride and padding are both set to 1.

8. The method according to claim 1, wherein When training the imputation model, the loss function used is mean squared error.

9. The method according to claim 1, wherein In step 7, if there are missing data in the pixel brightness temperature sequence in the historical time series, the missing data are filled using the average of the observations of the previous day and the next day.

10. The method according to claim 1, wherein Specify the time step to be set to 5 days.

Images