Underground water reserve inversion method and system based on multi-source data

Abstract

The invention discloses an underground water reserve inversion method and system based on multi-source data, relates to the technical field of hydrological data processing, and solves the technical problem that underground water reserve changes are difficult to accurately and finely describe due to multi-aspect limitations in the prior art. The method comprises the following steps: acquiring and preprocessing underground water reserve related data; the land water reserves are from a plurality of mechanisms, uncertainty evaluation is carried out on the land water reserves of data of each mechanism, a synthesis coefficient is calculated, and then synthesis is carried out according to the synthesis coefficient; carrying out interpolation on the missing data of the land water reserves; carrying out downscaling processing on the land water reserve data and other groundwater reserve related data after interpolation; performing underground water reserve inversion based on a water balance formula by using the downscaled underground water reserve related data; according to the method, the time-space consistent land water reserves data are obtained through data synthesis, and the accuracy of the data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological data processing, and in particular to a groundwater storage inversion method and system based on multi-source data. Background Art

[0002] Groundwater is an important part of the hydrological cycle and plays a role in ecological environmental protection and economic and social development. In recent years, under the background of climate change and the deepening impact of human activities, groundwater reserves have changed dramatically, accompanied by a series of resource, ecological and environmental problems, including groundwater level decline, water source depletion, ground subsidence, ecological degradation, land desertification, etc., which seriously affect the sustainable development of social economy. Therefore, it is particularly urgent to accurately invert groundwater reserves and study the changes in groundwater reserves.

[0003] Previous studies on groundwater storage inversion mainly relied on observation data from actual measurement sites, which had problems such as scattered and uneven monitoring sites, limited monitoring range, and high construction costs. Since March 2002, the GRACE satellite launched by the United States has realized the observation of terrestrial water storage changes using a time-varying gravity field. Compared with monitoring sites, it has a wider coverage, is continuous in time and space, and is suitable for large-scale monitoring, with obvious advantages.

[0004] However, the terrestrial water storage change data inverted by GRACE include changes in surface water storage and groundwater storage. Therefore, it is necessary to separate groundwater storage changes from terrestrial water storage changes. At present, the relevant methods for groundwater storage inversion have limitations such as insufficient consistency of GRACE data, missing data for some time periods, and low spatial resolution of data, making it difficult to accurately and finely describe groundwater storage changes. Summary of the invention

[0005] In order to solve the above-mentioned problems existing in the prior art, the present invention provides a groundwater reserve inversion method and system based on multi-source data, which solves the technical problem that the prior art has many limitations and is difficult to accurately and finely characterize the changes in groundwater reserves.

[0006] A groundwater storage inversion method based on multi-source data includes the following steps:

[0007] S1: Obtaining and preprocessing groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data;

[0008] S2: The terrestrial water reserves are from multiple institutions, and the uncertainty of the terrestrial water reserves of each institution's data is evaluated to calculate the synthesis coefficient, and then the terrestrial water reserves of multiple institutions are synthesized according to the synthesis coefficient;

[0009] S3: interpolate the missing data of the synthesized terrestrial water storage; S4: downscale the interpolated terrestrial water storage data and other groundwater storage related data;

[0010] S5: Using the downscaled groundwater storage data, the groundwater storage inversion is performed based on the water balance formula.

[0011] Furthermore, the terrestrial water reserves of each institution's data include the terrestrial water reserves of Level 2 and Level 3 of GRACE FO, and S2 includes: synthesizing the Level 2 data of multiple institutions into the Level 2 terrestrial water reserves, and synthesizing the Level 3 data of multiple institutions into the Level 3 terrestrial water reserves. The specific synthesis formula is as follows:

[0012] X i =X true +ε i ,i=1,2,3,…,N (1)

[0013] Where, X i is the time series of terrestrial water storage, X true is the true value of the terrestrial water storage time series, ε i is the time series error of terrestrial water storage;

[0014] Y i,M =X i -X R =ε i -ε R ,i=1,2,3,…,N-1 (2)

[0015] Where Y i,M is an M×(N-1) matrix, where M is the number of time series periods, X i is the time series of terrestrial water storage, X R is the terrestrial water storage time series of the reference field. In this embodiment, the CSR-Mascon terrestrial water storage time series is selected as the reference field; ε i is the time series error of terrestrial water storage, ε R is the time series error of terrestrial water storage in the reference field;

[0016] S=cov(Y i,M )=J·R·J T (3)

[0017] In the formula, cov(Y i,M ) is Y i,M The covariance matrix of , R is the noise matrix,

[0018] The uncertainty of each set of land water storage change time series is expressed by the standard deviation, i.e. where r i,j are the diagonal elements of the R matrix;

[0019]

[0020] In the formula, X is the best estimate of the time series, X i is the time series of terrestrial water storage, ω i and σ i are the weighted coefficient and standard deviation of the time series of the ith model, and n is the number of reconstructed models.

[0021] Furthermore, S3 includes: combining singular spectrum analysis to interpolate the missing data of the synthesized terrestrial water storage, firstly setting the missing values ​​of the original data to 0 for interpolation, and then performing singular spectrum analysis on the interpolated sequence, including:

[0022] For the terrestrial water storage time series [x1, x2, …, x N ], where N represents the length of the time series;

[0023]

[0024] Where X is the trajectory matrix, M and N are the window length and the number of matrix columns after time series embedding, respectively, and M is between 1 and <M<N / 2,L=N-M+1;

[0025]

[0026] Where U is the left singular vector, V is the right singular vector, A is a diagonal matrix, and the singular values ​​are located on the main diagonal of A;

[0027]

[0028] In the formula, PC i is the result of principal component analysis of the right singular vector, EOF i is the result of empirical orthogonal function analysis of right singular vectors, U *,i is the i-th column of U, V *,i is the i-th column of V, σ i is the i-th diagonal element of A;

[0029]

[0030] In the formula, λ i is the eigenvalue, Λ=diag[λ1,λ2,··λ d ] is a diagonal matrix consisting of the first d largest non-zero eigenvalues ​​of the matrix X, and d = min{M,L};

[0031] X=X1+X2+…+X d (9)

[0032] In the formula, each X i is the reconstruction matrix of a certain group;

[0033]

[0034] Where M t is the weight. When 1≤t≤M-1, Mt=t; when M≤t≤N-M+1, Mt=M; when M+1≤t≤N, Mt=N-t+1.

[0035] Further, the S4 includes:

[0036] S4.1. Use the interpolated terrestrial water storage data as the dependent variable, and then use the preprocessed soil water storage, snow water equivalent and vegetation canopy water storage as well as precipitation, evaporation and DEM data as dependent variables to train and verify the random forest model:

[0037]

[0038] In the formula, D is the training set, X and Y are low-resolution feature variables and target variables, respectively, and N represents the length of the time series;

[0039]

[0040] Where B is the number of decision trees in the random forest, T b (X) and are the prediction results of the b-th decision tree with low resolution for the input features and the prediction results of the target variable;

[0041] S4.2. Use bilinear interpolation downscaling method to downscale soil water storage, snow water equivalent and vegetation canopy water storage as well as precipitation, evaporation and DEM data to obtain downscaled independent variables;

[0042] S4.3. Input the downscaled independent variables into the random forest model trained and verified in S4.1 to obtain the downscaled terrestrial water storage data:

[0043]

[0044] Where B is the number of decision trees in the random forest, T b (X h )and They are the prediction results of the b-th decision tree with high resolution for the input features and the target variable, respectively.

[0045] Further, S5 includes: using the downscaled groundwater storage related data to perform groundwater storage inversion based on the water balance formula, the formula is as follows:

[0046] GWS=TWS-SM-SWE-CAN (14)

[0047] Where GWS is groundwater storage; TWS is terrestrial water storage; SM is soil water storage; SWEA is snow water equivalent; CAN is vegetation canopy water storage.

[0048] Furthermore, it also includes S6: using measured groundwater level data to verify the inverted groundwater storage results. First, the groundwater depth data is converted into monthly water level change values, and the water level change values ​​are multiplied by the regional water supply degree to obtain the measured groundwater reserves, which are then compared and verified with the inverted groundwater reserves.

[0049] Further, the plurality of institutions include CSR, JPL and GFZ.

[0050] Furthermore, the preprocessing of terrestrial water storage includes replacing low-order terms, de-banding and Gaussian filtering of Level2 data.

[0051] A groundwater storage inversion system based on multi-source data is used to implement a groundwater storage inversion method based on multi-source data, including: a data acquisition and processing module, a land water storage synthesis module, a land water storage interpolation module, a downscaling processing module, and a groundwater storage inversion module;

[0052] The data acquisition and processing module is used to obtain and pre-process the groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data;

[0053] The terrestrial water storage synthesis module is used for the terrestrial water storage from multiple institutions, and performs uncertainty assessment on the terrestrial water storage of each institution's data and calculates the synthesis coefficient and then synthesizes according to the synthesis coefficient;

[0054] The terrestrial water storage interpolation module is used to interpolate missing data of terrestrial water storage;

[0055] The downscaling processing module is used to downscale the interpolated terrestrial water storage data and other groundwater storage related data;

[0056] The groundwater storage inversion module is used to use the downscaled groundwater storage related data to perform groundwater storage inversion based on the water balance formula.

[0057] The beneficial effects of the present invention include:

[0058] (1) By integrating various GRACE data from three institutions, namely CSR, JPL, and GFZ, we can obtain temporally and spatially consistent terrestrial water storage data and improve the accuracy of the data.

[0059] (2) The continuity and integrity of groundwater storage data can be improved by interpolating terrestrial water storage data.

[0060] (3) The terrestrial water storage is downscaled based on multi-source data, which improves the spatial resolution of groundwater storage and provides a more accurate reference for regional groundwater storage changes.

[0061] (4) Combining GRACE and GLDAS data, the groundwater storage data is inverted using the water balance formula.

[0062] (5) Using the measured groundwater depth data, the measured groundwater reserves are compared and verified with the inverted groundwater reserves data. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a groundwater storage inversion method based on multi-source data involved in an embodiment of the present application.

[0064] Figure 2 This is the GRACE missing data involved in the embodiments of the present application.

[0065] Figure 3 It is the spatial change of terrestrial water storage in a certain area after random forest downscaling involved in the embodiment of the present application. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0067] A groundwater storage inversion method based on multi-source data, such as Figure 1 As shown, the following steps are included:

[0068] S1: Obtaining and preprocessing groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data;

[0069] S2: The terrestrial water reserves are from multiple institutions, and the uncertainty of the terrestrial water reserves of each institution's data is evaluated to calculate the synthesis coefficient, and then the terrestrial water reserves of multiple institutions are synthesized according to the synthesis coefficient;

[0070] S3: interpolate the missing data of the synthesized terrestrial water storage; S4: downscale the interpolated terrestrial water storage data and other groundwater storage related data;

[0071] S5: Using the downscaled groundwater storage data, the groundwater storage inversion is performed based on the water balance formula.

[0072] In another embodiment, obtaining groundwater storage related data and performing preprocessing includes:

[0073] The original data include terrestrial water storage of Level 2 and Level 3 of GRACE and GRACE FO, soil water storage, snow water equivalent and vegetation canopy water storage of GLDAS, precipitation of CN05.1, evaporation of GLEAM, GEBCO DEM and other data. The above data within the study area are obtained by clipping using the boundary range layer of the study area.

[0074] GRACE's Level 2 and Level 3 come from three institutions, CSR, JPL and GFZ, with a total of six types of data. Level 2 data needs to replace low-order terms, remove bands and Gaussian filters. The spatial resolution of terrestrial water storage, soil water storage, snow water equivalent and vegetation canopy water storage data is 1°. In addition, precipitation, evaporation and DEM are uniformly converted to 1°.

[0075] In another embodiment, the S2 includes:

[0076] The generalized tricornut method is used to carry out uncertainty assessment of GRACE single-institution data and synthesis of multiple models.

[0077] Among them, the three types of Level2 data are used to synthesize Level2 terrestrial water reserves, and the three types of Level3 data are used to synthesize Level3 terrestrial water reserves.

[0078] X i =X true +ε i ,i=1,2,3,…,N (1)

[0079] Where, X i is the time series of terrestrial water storage, X true is the true value of the terrestrial water storage time series, ε i is the time series error of terrestrial water storage.

[0080] Y i,M =X i -X R =ε i -ε R ,i=1,2,3,…,N-1 (2)

[0081] Where Y i,M is an M×(N-1) matrix, where M is the number of time series periods, X i is the time series of terrestrial water storage, X R is the terrestrial water storage time series of the reference field. In this embodiment, the CSR-Mascon terrestrial water storage time series is selected as the reference field. i is the time series error of terrestrial water storage, ε R is the time series error of terrestrial water storage in the reference field.

[0082] S=cov(Y i,M )=J·R·J T (3)

[0083] In the formula, cov(Y i,M ) is Y i,M The covariance matrix of , R is the noise matrix,

[0084] The uncertainty of each set of land water storage change time series is expressed by the standard deviation, i.e. where r i,j are the diagonal elements of the R matrix.

[0085]

[0086] In the formula, X is the best estimate of the time series, X i is the time series of terrestrial water storage, ω i and σ i are the weighted coefficient and standard deviation of the time series of the ith model, and n is the number of reconstructed models.

[0087] Specifically, the synthesis coefficient of this embodiment is as follows:

[0088] Table 1. Synthesis coefficients of the generalized triangular hat method

[0089]

[0090]

[0091] In the table, CSR, JPL and GFZ are the three institutions that obtain Level 2 and Level 3 data.

[0092] In another embodiment, the S3 includes:

[0093] Since the GRACE satellite is missing 33 months of data from April 2002 to March 2023, Figure 2 As shown, it specifically involves missing data and the corresponding years and months, where red rectangles represent missing data and blue rectangles represent existing data.

[0094] The present invention combines singular spectrum analysis to interpolate the missing data of terrestrial water reserves. First, the missing values ​​of the original data are set to 0 for filling. Then, singular spectrum analysis is performed on the filled sequence to create a trajectory matrix, perform singular value decomposition, group the trajectory matrix, and reconstruct the missing values.

[0095] For the terrestrial water storage time series [x1, x2, …, x N ], where N represents the length of the time series.

[0096]

[0097] Where X is the trajectory matrix, M and N are the window length and the number of matrix columns after time series embedding, respectively, and M is between 1 and <M<N / 2,L=N-M+1。

[0098]

[0099] Where U is the left singular vector, V is the right singular vector, A is a diagonal matrix, and the singular values ​​are located on the main diagonal of A.

[0100]

[0101] In the formula, PC i is the result of principal component analysis of the right singular vector, EOF i is the result of empirical orthogonal function analysis of right singular vectors, U *,i is the i-th column of U, V *,i is the i-th column of V, σ i is the i-th diagonal element of A.

[0102]

[0103] In the formula, λ i is the eigenvalue, Λ=diag[λ1,λ2,··λ d ] is a diagonal matrix consisting of the first d largest non-zero eigenvalues ​​of the matrix X, and d = min{M,L}.

[0104] X=X1+X2+…+X d (9)

[0105] In the formula, each X i is the reconstruction matrix of a group.

[0106]

[0107] Where M t is the weight. When 1≤t≤M-1, Mt=t; when M≤t≤N-M+1, Mt=M; when M+1≤t≤N, Mt=N-t+1.

[0108] In another embodiment, the S4 includes:

[0109] The interpolated terrestrial water storage data and the preprocessed soil water storage, snow water equivalent and vegetation canopy water storage with a spatial resolution of 1° as well as precipitation, evaporation and DEM data were downscaled to 0.25° by bilinear interpolation.

[0110] The random forest model was trained and evaluated based on the 1° data, using 90% of the data for training and 10% for validation. The 0.25° independent variable was input into the random forest regression model to obtain the terrestrial water storage at 0.25°.

[0111]

[0112] Where D is the training set, X and Y are the low-resolution feature variables and target variables, respectively, and N represents the length of the time series.

[0113]

[0114] Where B is the number of decision trees in the random forest, T b (X) and They are the prediction results of the b-th decision tree with low resolution for the input features and the prediction results of the target variable, respectively.

[0115]

[0116] Where B is the number of decision trees in the random forest, T b (X h )and They are the prediction results of the b-th decision tree with high resolution for the input features and the target variable, respectively.

[0117] Specifically, the spatial change of terrestrial water storage in a certain region after random forest downscaling is as follows: Figure 3 shown.

[0118] In another embodiment, S5 includes: using the downscaled terrestrial water storage to invert the groundwater storage based on the water balance formula, the formula is as follows:

[0119] GWS=TWS-SM-SWE-CAN (14)

[0120] Where GWS is groundwater storage; TWS is terrestrial water storage; SM is soil water storage; SWEA is snow water equivalent; CAN is vegetation canopy water storage.

[0121] In another embodiment, S6 includes: using measured groundwater level data to verify the inverted groundwater reserves. First, the groundwater depth data is converted into monthly water level change values. The water level change value is multiplied by the regional water supply degree to obtain the measured groundwater reserves, which are then compared and verified with the groundwater reserves inverted by GRACE. The measured water levels of three stations, Youfang Village, Baojiadian Town, Tougong Village, Ergong Township, and the former Wuxing Sixth Team (now Datuguli Fourth Team) in Dafeng Town, Changji Prefecture, are selected and converted into measured groundwater reserves. The groundwater reserves inverted by level2 and level3 under the longitude and latitude of the station are extracted, and the specific longitude and latitude information is shown in Table 2. The correlation coefficient between the measured groundwater reserves of the station and the groundwater reserves inverted by evel2 and level3 is calculated.

[0122] GWSA 实测 =ΔH·S

[0123] Where, GWSA 实测 is the measured change in groundwater reserves, ΔH is the measured change in groundwater level, and S is the average groundwater supply degree, which is taken as 0.01.

[0124] Table 2. Site longitude and latitude

[0125]

[0126] Table 3. Groundwater storage verification

[0127]

[0128] The verification results of groundwater reserves obtained by S6 are shown in Table 3. The correlation coefficients between the groundwater reserves inverted by level 2 and the measured groundwater reserves of Youfang Village, Baojiadian Town, Tougong Village, Ergong Township, and the former Wuxing Sixth Team (now Datuguli Fourth Team) in Dafeng Town in Changji Prefecture are -0.069, 0.061, and 0.074, respectively.

[0129] The correlation coefficients between the groundwater reserves inverted by Level3 and the measured groundwater reserves at the three stations in Youfang Village, Baojiadian Town, Changji Prefecture, Tougong Village, Ergong Township, and the former Wuxing Sixth Team (now Datuguli Fourth Team) in Dafeng Town are 0.015, -0.110, and -0.029, respectively.

[0130] In general, after the groundwater reserves inverted by level 2 and level 3 were verified with the measured groundwater reserves, the correlation coefficients of the groundwater reserves inverted by level 2 and level 3 were positively correlated at the two stations of Tougong Village, Ergong Township and the former Wuxing Sixth Team (now Datuguli Fourth Team) in Dafeng Town, but the inverted and measured groundwater reserves were generally correlated, which indicates that the inverted groundwater reserves meet the requirements.

[0131] Specifically, the data processing process of using a groundwater storage inversion method based on multi-source data is as follows:

[0132] Among them, Table 4 shows the synthetic terrestrial water storage results after S2 treatment;

[0133] Table 5 shows the interpolation of terrestrial water storage after S3 processing;

[0134] Table 6 shows the groundwater storage inversion results obtained through S5.

[0135] Table 4. Synthetic terrestrial water storage results

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143] Table 5. Terrestrial water storage interpolation

[0144]

[0145]

[0146]

[0147]

[0148] Table 6. Groundwater storage inversion results

[0149]

[0150]

[0151]

[0152] A groundwater storage inversion system based on multi-source data is used to implement a groundwater storage inversion method based on multi-source data, including: a data acquisition and processing module, a land water storage synthesis module, a land water storage interpolation module, a downscaling processing module, and a groundwater storage inversion module;

[0153] The data acquisition and processing module is used to obtain and pre-process the groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data;

[0154] The terrestrial water storage synthesis module is used for the terrestrial water storage from multiple institutions, and performs uncertainty assessment on the terrestrial water storage of each institution's data and calculates the synthesis coefficient and then synthesizes according to the synthesis coefficient;

[0155] The terrestrial water storage interpolation module is used to interpolate missing data of terrestrial water storage;

[0156] The downscaling processing module is used to downscale the interpolated terrestrial water storage data and other groundwater storage related data;

[0157] The groundwater storage inversion module is used to use the downscaled groundwater storage related data to perform groundwater storage inversion based on the water balance formula.

[0158] The above-mentioned embodiments only express the specific implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A groundwater storage inversion method based on multi-source data, characterized in that: The following steps are involved: S1: Obtaining and preprocessing groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data; S2: The terrestrial water reserves are from multiple institutions, and the uncertainty of the terrestrial water reserves of each institution's data is evaluated to calculate the synthesis coefficient, and then the terrestrial water reserves of multiple institutions are synthesized according to the synthesis coefficient; S3: Interpolation of missing data for synthesized terrestrial water storage; S4: Downscaling the interpolated terrestrial water storage data and other groundwater storage related data; S5: Using the downscaled groundwater storage data, the groundwater storage inversion is performed based on the water balance formula.

2. The groundwater storage inversion method based on multi-source data according to claim 1 is characterized in that: The terrestrial water reserves of each institution's data include the terrestrial water reserves of Level 2 and Level 3 of GRACE FO. S2 includes: synthesizing the Level 2 data of multiple institutions into the Level 2 terrestrial water reserves, and synthesizing the Level 3 data of multiple institutions into the Level 3 terrestrial water reserves. The specific synthesis formula is as follows: X i =X true +ε i ,i=1,2,3,…,N (1) Where, X i is the time series of terrestrial water storage, X true is the true value of the terrestrial water storage time series, ε i is the time series error of terrestrial water storage; Y i,M =X i -X R =e i -e R ,i=1,2,3,…,N-1 (2) Where Y i,M is a (N-1)×M matrix, where M is the number of time series periods, X i is the time series of terrestrial water storage, X R is the terrestrial water storage time series of the reference field, ε R is the time series error of terrestrial water storage in the reference field; S=cov(Y i,M )=J·R·J T (3) In the formula, cov(Y i,M ) is Y i,M The covariance matrix of , R is the noise matrix, The uncertainty of each set of land water storage change time series is expressed by the standard deviation, i.e. where r i,j are the diagonal elements of the R matrix; In the formula, X is the best estimate of the time series, X i is the time series of terrestrial water storage, ω i and σ i are the weighted coefficient and standard deviation of the time series of the ith institution, and n is the number of reconstructed models.

3. The groundwater storage inversion method based on multi-source data according to claim 1 is characterized in that: S3 includes: combining singular spectrum analysis to interpolate the missing data of the synthesized terrestrial water storage, firstly setting the missing values ​​of the original data to 0 for interpolation, and then performing singular spectrum analysis on the interpolated sequence: For the terrestrial water storage time series [x1, x2, …, x N ], where N represents the length of the time series; Where X is the trajectory matrix, M and N are the window length and the number of matrix columns after time series embedding, respectively, and M is between 1 and <M<N / 2,L=N-M+1; Where U is the left singular vector, V is the right singular vector, A is a diagonal matrix, and the singular values ​​are located on the main diagonal of A; In the formula, PC i is the result of principal component analysis of the right singular vector, EOF i is the result of empirical orthogonal function analysis of right singular vectors, U *,i is the i-th column of U, V *,i is the i-th column of V, σ i is the i-th diagonal element of A; In the formula, λ i is the eigenvalue, Λ=diag[λ1,λ2,…λ d ] is a diagonal matrix consisting of the first d largest non-zero eigenvalues ​​of the matrix X, and d = min{M,L}; x=x1+X2+…+X d (9) In the formula, each X i is the reconstruction matrix of a certain group; Where M i is the weight. When 1≤t≤M-1, Mt=t; when M≤t≤N-M+1, Mt=M; when M+1≤t≤N, Mt=N-t+1.

4. The method for groundwater storage inversion based on multi-source data according to claim 1 is characterized in that: The S4 includes: S4.

1. Use the interpolated terrestrial water storage data as the dependent variable, and then use the preprocessed soil water storage, snow water equivalent and vegetation canopy water storage as well as precipitation, evaporation and DEM data as dependent variables to train and verify the random forest model: In the formula, D is the training set, X and Y are low-resolution feature variables and target variables, respectively, and N represents the length of the time series; Where B is the number of decision trees in the random forest, T b (X) and are the prediction results of the b-th decision tree with low resolution for the input features and the prediction results of the target variable; S4.

2. Use bilinear interpolation downscaling method to downscale soil water storage, snow water equivalent and vegetation canopy water storage as well as precipitation, evaporation and DEM data to obtain downscaled independent variables; S4.

3. Input the downscaled independent variables into the random forest model trained and verified in S4.1 to obtain the downscaled terrestrial water storage data: Where B is the number of decision trees in the random forest, T b (X h )and They are the prediction results of the b-th decision tree with high resolution for the input features and the target variable, respectively.

5. The groundwater storage inversion method based on multi-source data according to claim 1 is characterized in that: S5 includes: using the downscaled groundwater storage related data to perform groundwater storage inversion based on the water balance formula, the formula is as follows: GWS=TWS-SM-SWE-CAN (14) Where GWS is groundwater storage; TWS is terrestrial water storage; SM is soil water storage; SWEA is snow water equivalent; CAN is vegetation canopy water storage.

6. The method for groundwater storage inversion based on multi-source data according to claim 1 is characterized in that: It also includes S6: using measured groundwater level data to verify the inverted groundwater storage results. First, the groundwater depth data is converted into monthly water level change values. The water level change values ​​are multiplied by the regional water supply degree to obtain the measured groundwater reserves, which are then compared and verified with the inverted groundwater reserves.

7. The groundwater storage inversion method based on multi-source data according to claim 1 is characterized in that: The multiple institutions include CSR, JPL and GFZ.

8. The method for groundwater storage inversion based on multi-source data according to claim 2 is characterized in that: The preprocessing of terrestrial water storage includes replacing low-order terms, debanding and Gaussian filtering of Level2 data.

9. A groundwater storage inversion system based on multi-source data, characterized in that: A method for implementing a groundwater storage inversion method based on multi-source data as described in any one of claims 1 to 8, comprising: a data acquisition and processing module, a land water storage synthesis module, a land water storage interpolation module, a downscaling processing module, and a groundwater storage inversion module; The data acquisition and processing module is used to obtain and pre-process the groundwater storage related data; the groundwater storage related data includes: land water storage, soil water storage, snow water equivalent and vegetation canopy water storage, precipitation, evaporation, DEM data; The terrestrial water storage synthesis module is used for the terrestrial water storage from multiple institutions, and performs uncertainty assessment on the terrestrial water storage of each institution's data and calculates the synthesis coefficient and then synthesizes according to the synthesis coefficient; The terrestrial water storage interpolation module is used to interpolate missing data of terrestrial water storage; The downscaling processing module is used to downscale the interpolated terrestrial water storage data and other groundwater storage related data; The groundwater storage inversion module is used to use the downscaled groundwater storage related data to perform groundwater storage inversion based on the water balance formula.

Images