A method and system for long-term reconstruction of groundwater storage (GWS) at a watershed scale

Abstract

The application provides a method and system for long-term reconstruction of groundwater storage GWS at a watershed scale, comprising: obtaining and fusing long-term meteorological and hydrological data calibrated in 1960-2023 and short-term total land water storage TWS after 2002; calculating a short-term groundwater GWS sequence according to a water balance equation ΔTWS = ΔSWS + ΔSMS + ΔGWS at a watershed scale; using the short-term groundwater GWS sequence and the part after 2002 in the fused long-term meteorological and hydrological data as target data and driving data to train a random forest RF sub-model of an RF-KF coupled model; inputting the fused long-term meteorological and hydrological data into the trained random forest RF sub-model to obtain a long-term prediction result of the groundwater storage GWS; replacing the output of a prediction module of a Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS, using an updating module of the Kalman filter KF sub-model to simulate time dynamics and correct accumulated errors in long-term prediction, and obtaining a final long-term reconstruction result of the groundwater storage GWS. The application can perform long-term reconstruction of GWS.

Description

Technical Field

[0001] This invention relates to the field of long-term reconstruction technology of groundwater storage GWS, and in particular to a method and system for long-term reconstruction of groundwater storage GWS at the watershed scale. Background Technology

[0002] Groundwater storage (GWS) is the collective term for unconfined and confined water buried in the voids of underground rock and soil. It is an important component of the water cycle in a watershed and a vital source of replenishment for rivers and lakes in arid and semi-arid regions. Its replenishment accounts for approximately 50% of the total water resource replenishment and is a significant factor affecting the sustainability of regional ecosystems.

[0003] Traditional monitoring methods rely on groundwater wells to record groundwater levels, providing the highest data confidence. However, with the current trend of environmental governance shifting towards watershed management, traditional monitoring methods are limited by spatial distribution and face difficulties in identifying groundwater conditions within watersheds. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a method and system for long-term reconstruction of groundwater storage at the watershed scale (GWS), the technical solution of which is as follows:

[0005] On the one hand, a long-term reconstruction method for watershed-scale groundwater storage using the groundwater storage sphere (GWS) is provided, which includes:

[0006] S1. Acquire and fuse long-term meteorological and hydrological data from 1960 to 2023, including: precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale.

[0007] S2. Acquire and fuse the short-term total terrestrial water storage (TWS) data after 2002;

[0008] S3. Based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS, the short-term groundwater GWS sequence is calculated.

[0009] S4. Using the short-term groundwater GWS sequence and the portion of the fused long-term meteorological and hydrological data after 2002 as target data and driving data, train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model. The long-term groundwater storage reconstruction RF-KF coupled model includes the random forest RF sub-model and the Kalman filter KF sub-model.

[0010] S5. Input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS).

[0011] S6. Replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS, and use the update module of the Kalman filter KF sub-model to simulate the time dynamics and correct the accumulated error of the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

[0012] Optionally, S1 specifically includes:

[0013] S11. Obtain the long-term meteorological and hydrological data through meteorological reanalysis data ERA5 and land data assimilation model GLDAS. The data of ERA5 and GLDAS correspond to each other and both provide precipitation, evapotranspiration, air temperature, surface water storage SWS, and soil water storage SMS at the watershed scale.

[0014] S12. Obtain meteorological station data from the China Meteorological Administration (CMA), including precipitation, evapotranspiration, and temperature. ERA5 and GLDAS data often differ from the meteorological station data. Use the meteorological station data to fuse and calibrate the ERA5 and GLDAS data. This fusion calibration uses a Bayesian linear fusion algorithm, targeting a specific meteorological station, and fuses and calibrates the ERA5 and GLDAS data for its corresponding location. The formula is as follows:

[0015] y fused-1 =α·y ERA5 +(1-α)·y GLDAS +∈1

[0016] Where y fused-1 To fuse the calibrated data, and y ERA5 and y GLDAS For the corresponding ERA5 and GLDAS data, both are unified to a spatial resolution of 0.1°, consistent with the ERA5 raster size. Through optimization of coefficient α and residual ∈1, the root mean square error (RMSE) between the fused and calibrated data and the meteorological station data is minimized.

[0017] Optionally, S2 specifically includes:

[0018] S21. The Total Land Storage (TWS) data for 2002-2023 was obtained using the GRACE satellite mission launched in 2002. GRACE Level 3 remote sensing product data from JPL, GSFC, and CSR were used, with a spatial resolution of 3°.

[0019] S22. Obtain at least one year's worth of TWS measured values ​​through sampling or by obtaining them from officially published water resources statistical yearbooks. Use the TWS measured values ​​to perform fusion calibration on the Level 3 remote sensing product data. The fusion calibration uses a Bayesian linear fusion algorithm, and its formula is as follows:

[0020] y fused-tws =α1y GSFC +α2y CSR +α3y JPL +∈2

[0021] Where y fused-tws To fuse the calibrated data, α i The coefficient is y, the sum is 1, and y GSFC ,y CSR ,y JPL All data are remote sensing product data, ∈2 represents the residuals, and the coefficient α is used to calculate the residuals. i The optimization of residuals ∈2 minimizes the root mean square error (RMSE) between the fused calibrated data and the TWS measured values.

[0022] Optionally, S3 specifically includes:

[0023] The short-term groundwater GWS sequence of the watershed is estimated by subtracting the known changes in SWS and SMS from the changes in TWS using the watershed-scale water balance equation: ΔGWS = ΔTWS - ΔSWS - ΔSMS. The GWS data for a certain year is obtained by calibration using local water resources bulletins for that year.

[0024] Optionally, the Random Forest (RF) sub-model aggregates multiple decision trees to capture the complex nonlinear relationship between meteorological and hydrological elements and GWS fluctuations, as shown in the following formula:

[0025]

[0026] Where f(R) is the long-term prediction result of the groundwater storage GWS, and T i (x) is the result predicted by the decision tree, and N is the number of decision trees.

[0027] Optionally, the Kalman filter KF sub-model uses an improved Kalman filter algorithm for data reconstruction;

[0028] The original Kalman filter algorithm operates within a state-space framework, continuously improving predictions by merging new observations and dynamically adjusting weights through the state covariance matrix. It estimates the system state by combining model predictions with observed data. The prediction steps of the prediction module are as follows:

[0029] X k =F k x k-1 +B k u k +w k

[0030] Where X kIt is the Kalman filter prediction at time k; F k It is the state transition matrix, representing the relationship between the predicted value at the previous time point and the predicted value at the current time point; B k It is the control input matrix, representing the importance of the control factors; u k It is a meteorological and hydrological driving factor; w k It is process noise;

[0031] The update module optimizes the prediction results based on the observed data:

[0032] f(K) = X k +K k (z k -X k )

[0033]

[0034] K k =P k (P k +R k ) -1

[0035] Where z k It is the observed value of groundwater storage; K k It is the Kalman gain; R k and Q k All of these are model noise;

[0036] The improved Kalman filter algorithm uses the prediction results of the random forest (RF) sub-model to replace the output of the prediction module in the original Kalman filter, and uses an update module to simulate time dynamics and correct for long-term prediction accumulated errors. The equations are as follows:

[0037] f'(R,K) after2002 =f(R)+K k (z k -f(R))

[0038] f'(R,K) before2002 =f(R)+K k总 (z k总 -f(R) 总 )

[0039] Where f'(R,K) after2002 This is the result of GWS reconstruction of groundwater storage after 2002, f'(R,K) before2002 This is a long-term reconstruction of groundwater storage using the GWS model prior to 2002. f(R) is the prediction result from the Random Forest (RF) sub-model, replacing the original Kalman filter prediction result X. k ;z kThis refers to the actual observed groundwater storage for a specific year (k) after 2002 that needs to be predicted; z k总 This represents the total observed values ​​from one year after the year k that needs to be predicted, up to the final year, before 2002. For each year after 2002, the actual observed groundwater storage for that year is used. For each year before 2002, the reconstructed values ​​from the Kalman filter KF sub-model are used as the observed values ​​for that year. These observed values ​​are summed to form z. k总 ;K k It is the Kalman gain, and its value is related to the prediction error F of the random forest. k It shows a negative correlation; for data after 2002, F... k F represents the error between the random forest prediction and the actual short-term groundwater observation for year k. For data prior to 2002, F k总 K represents the total error from year k to the final year; k总 With F k总 Showing a negative correlation, through and K k =P k (P k +R k ) -1 Calculated.

[0040] On the other hand, a watershed-scale groundwater storage GWS long-term reconstruction system is provided, the system comprising:

[0041] The first acquisition and fusion calibration module is used to acquire and fuse long-term meteorological and hydrological data from 1960 to 2023. The long-term meteorological and hydrological data includes precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale.

[0042] The second acquisition and fusion calibration module is used to acquire and fuse the short-term total terrestrial water storage (TWS) data after 2002.

[0043] The calculation module is used to calculate the short-term groundwater GWS sequence based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS.

[0044] The training module is used to train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model using the short-term groundwater GWS sequence and the portion of the fused long-term meteorological and hydrological data after 2002 as target data and driving data. The long-term groundwater storage reconstruction RF-KF coupled model includes a random forest RF sub-model and a Kalman filter KF sub-model.

[0045] The prediction module is used to input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS).

[0046] The reconstruction module is used to replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS. The update module of the Kalman filter KF sub-model is used to simulate the time dynamics and correct the error accumulated in the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

[0047] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method for long-term reconstruction of groundwater storage GWS at the watershed scale.

[0048] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement the above-described method for long-term reconstruction of groundwater storage GWS at the watershed scale.

[0049] The beneficial effects of the technical solution provided by this invention include at least the following:

[0050] This invention performs long-term GWS reconstruction of watershed-scale groundwater storage based on multi-source data. It uses random forest and Kalman filter fusion to reconstruct watershed groundwater data, thereby improving the long-term reconstruction capability and accuracy of watershed-scale groundwater storage GWS. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a method for long-term reconstruction of groundwater storage at the watershed scale using the Geographic Water System (GWS) according to an embodiment of the present invention.

[0053] Figure 2 This is a general block diagram of a method for long-term reconstruction of groundwater storage at the watershed scale using the groundwater storage sphere (GWS) according to an embodiment of the present invention.

[0054] Figure 3 This is a detailed block diagram of a method for long-term reconstruction of groundwater storage at the watershed scale using the Geographic Water System (GWS) according to an embodiment of the present invention.

[0055] Figure 4This is a schematic diagram comparing the prediction effects of the long-term reconstruction RF-KF coupled model and the single model for groundwater storage provided in this embodiment of the invention.

[0056] Figure 5 This is a schematic diagram of the GWS distribution in the Hulun Lake basin provided in an embodiment of the present invention;

[0057] Figure 6 This is a block diagram of a watershed-scale groundwater storage GWS long-term reconstruction system provided by an embodiment of the present invention;

[0058] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0059] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0060] This invention provides a method for long-term reconstruction of groundwater storage at the watershed scale (GWS). This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart of this method is shown below. Figure 2 and Figure 3 The diagram shows the overall block diagram and detailed block diagram of the method. The processing flow may include the following steps:

[0061] S1. Acquire and fuse long-term meteorological and hydrological data from 1960 to 2023, including: precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale.

[0062] Optionally, S1 specifically includes:

[0063] S11. Obtain the long-term meteorological and hydrological data through meteorological reanalysis data ERA5 and land data assimilation model GLDAS. The data of ERA5 and GLDAS correspond to each other and both provide precipitation, evapotranspiration, temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale.

[0064] S12. Obtain meteorological station data from the China Meteorological Administration (CMA), including precipitation, evapotranspiration, and temperature. ERA5 and GLDAS data often differ from the meteorological station data. Use the meteorological station data to fuse and calibrate the ERA5 and GLDAS data. This fusion calibration uses a Bayesian linear fusion algorithm, targeting a specific meteorological station, and fuses and calibrates the ERA5 and GLDAS data for its corresponding location. The formula is as follows:

[0065] y fused-1 =α·y ERA5 +(1-α)·y GLDAS +∈1

[0066] Where y fused-1 To fuse the calibrated data, and y ERA5 and y GLDAS For the corresponding ERA5 and GLDAS data, both are unified to a spatial resolution of 0.1°, consistent with the ERA5 raster size. Through optimization of coefficient α and residual ∈1, the root mean square error (RMSE) between the fused and calibrated data and the meteorological station data is minimized.

[0067] S2. Acquire and fuse calibrated short-term total terrestrial water storage (TWS) since 2002;

[0068] Optionally, S2 specifically includes:

[0069] S21. The Total Land Storage (TWS) data for 2002-2023 was obtained using the GRACE satellite mission launched in 2002. GRACE Level 3 remote sensing product data from JPL, GSFC, and CSR were used, with a spatial resolution of 3°.

[0070] S22. Obtain at least one year's worth of TWS measured values ​​through sampling or by obtaining them from officially published water resources statistical yearbooks. Use the TWS measured values ​​to perform fusion calibration on the Level 3 remote sensing product data. The fusion calibration uses a Bayesian linear fusion algorithm, and its formula is as follows:

[0071] y fused-tws =α1y GSFC +α2y CSR +α3y JPL +∈2

[0072] Where y fused-tws To fuse the calibrated data, α i The coefficient is y, the sum is 1, and y GSFC ,y CSR ,yJPL All data are remote sensing product data, ∈2 represents the residuals, and the coefficient α is used to calculate the residuals. i The optimization of residuals ∈2 minimizes the root mean square error (RMSE) between the fused calibrated data and the TWS measured values.

[0073] S3. Based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS, the short-term groundwater GWS sequence is calculated.

[0074] The groundwater reconstruction method of this invention is based on the principle of mass conservation, which states that in a closed hydrological system, changes in water input, output, and storage must be balanced. By quantifying the various components of the watershed hydrological cycle, the temporal dynamics of water resources can be assessed. Some components, such as snowmelt and hillside runoff, contribute negligibly to the regional water balance and are therefore considered insignificant, and can be represented by the watershed-scale water balance equation.

[0075] Optionally, S3 specifically includes:

[0076] The short-term groundwater GWS sequence is estimated by subtracting the known changes in SWS and SMS from the change in TWS (ΔGWS) using the watershed-scale water balance equation: ΔGWS = ΔTWS - ΔSWS - ΔSMS. The GWS data for a specific year is used to calibrate the data and obtain the short-term groundwater GWS sequence. This GWS data is obtained from the local water resources bulletin for that year (the water resources bulletin can be GWS data from any year after 2002; ΔGWS can be used to calculate the data from that year to 2002 and from that year to 2023. For example, the 2010 water resources bulletin can be used to push back to 2009 up to 2002, or forward to 2011 up to 2023; the short-term groundwater GWS sequence includes data from 2002 to 2023).

[0077] S4. Using the short-term groundwater GWS sequence and the portion of the fused long-term meteorological and hydrological data after 2002 as target data and driving data, train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model. The long-term groundwater storage reconstruction RF-KF coupled model includes the random forest RF sub-model and the Kalman filter KF sub-model.

[0078] S5. Input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS).

[0079] Optionally, the Random Forest (RF) sub-model aggregates multiple decision trees to capture the complex nonlinear relationship between meteorological and hydrological elements and GWS fluctuations, as shown in the following formula:

[0080]

[0081] Where f(R) is the long-term prediction result of the groundwater storage GWS, and T i (x) is the result predicted by the decision tree, and N is the number of decision trees.

[0082] However, due to changes in the distribution of historical data, extrapolation based on static data after 2002 may accumulate errors over a long period of time. Therefore, embodiments of the present invention use a Kalman filter KF sub-model to correct these errors.

[0083] S6. Replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS, and use the update module of the Kalman filter KF sub-model to simulate the time dynamics and correct the accumulated error of the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

[0084] The Kalman filter KF sub-model uses an improved Kalman filter algorithm for data reconstruction.

[0085] The original Kalman filter algorithm operates within a state-space framework, continuously improving predictions by merging new observations and dynamically adjusting weights through the state covariance matrix. It estimates the system state by combining model predictions with observed data. The prediction steps of the prediction module are as follows:

[0086] X k =F k x k-1 +B k u k +w k

[0087] Where X k It is the Kalman filter prediction at time k; F k It is the state transition matrix, representing the relationship between the predicted value at the previous time point and the predicted value at the current time point; B k It is the control input matrix, representing the importance of the control factors; u k It is a meteorological and hydrological driving factor; w k It is process noise;

[0088] The update module optimizes the prediction results based on the observed data:

[0089] f(K) = X k +K k (z k -X k )

[0090]

[0091] K k =P k (P k +R k ) -1

[0092] Where z k It is the observed value of groundwater storage; K k It is the Kalman gain; R k and Q k All of these are model noise;

[0093] The improved Kalman filter algorithm uses the prediction results of the random forest (RF) sub-model to replace the output of the prediction module in the original Kalman filter, and uses an update module to simulate time dynamics and correct for long-term prediction accumulated errors. The equations are as follows:

[0094] f'(R,K) after2002 =f(R)+K k (z k -f(R))

[0095] f'(R,K) before2002 =f(R)+K k总 (z k总 -f(R) 总 )

[0096] Where f'(R,K) after2002 This is the result of GWS reconstruction of groundwater storage after 2002, f'(R,K) before2002 This is a long-term reconstruction of groundwater storage using the GWS model prior to 2002. f(R) is the prediction result from the Random Forest (RF) sub-model, replacing the original Kalman filter prediction result X. k ;z k This refers to the actual observed groundwater storage for a specific year (k) after 2002 that needs to be predicted; z k总 This represents the total observed values ​​from one year after the year k that needs to be predicted, up to the final year, before 2002. For each year after 2002, the actual observed groundwater storage for that year is used. For each year before 2002, the reconstructed values ​​from the Kalman filter KF sub-model are used as the observed values ​​for that year. These observed values ​​are summed to form z. k总 ;K k It is the Kalman gain, and its value is related to the prediction error F of the random forest. k It shows a negative correlation; for data after 2002, F... k F represents the error between the random forest prediction and the actual short-term groundwater observation for year k. For data prior to 2002, F k总 K represents the total error from year k to the final year;k总 With F k总 Showing a negative correlation, through and K k =P k (P k +R k ) -1 Calculated.

[0097] like Figure 4 As shown, the results of single random forest prediction and Kalman filter simulation are compared with those of the RF-KF coupled model simulation for long-term reconstruction of groundwater storage in this embodiment of the invention. The correlation coefficient R2 between the simulated and measured groundwater values, also known as the coefficient of determination, is an important indicator of the goodness of fit of the statistical model. The results show that the R2 of the coupled model simulation increased from 0.88 to 0.96, and the root mean square error (RMSE) decreased from 18.59 to 10.86, indicating that the embodiment of the invention has a better reconstruction effect for groundwater simulation. Figure 5 As shown, it can achieve 150,000 km. 2 The reconstruction of groundwater reserves in the Hulun Lake basin demonstrates its capacity for large-scale groundwater reconstruction.

[0098] like Figure 6 As shown, this embodiment of the invention also provides a watershed-scale groundwater storage GWS long-term reconstruction system, the system comprising:

[0099] The first acquisition and fusion calibration module 610 is used to acquire and fuse long-term meteorological and hydrological data from 1960 to 2023, including precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale.

[0100] The second acquisition and fusion calibration module 620 is used to acquire and fuse calibration the short-term total terrestrial water storage (TWS) since 2002.

[0101] Calculation module 630 is used to calculate the short-term groundwater GWS sequence based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS.

[0102] Training module 640 is used to train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model using the short-term groundwater GWS sequence and the portion after 2002 in the fused long-term meteorological and hydrological data as target data and driving data. The long-term groundwater storage reconstruction RF-KF coupled model includes a random forest RF sub-model and a Kalman filter KF sub-model.

[0103] Prediction module 650 is used to input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS).

[0104] The reconstruction module 660 is used to replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS, and use the update module of the Kalman filter KF sub-model to simulate the time dynamics and correct the error accumulated in the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

[0105] The watershed-scale groundwater storage GWS long-term reconstruction system provided in this embodiment of the invention has a functional structure that corresponds to the watershed-scale groundwater storage GWS long-term reconstruction method provided in this embodiment of the invention, and will not be described again here.

[0106] Figure 7 This is a schematic diagram of the structure of an electronic device 700 provided in an embodiment of the present invention. The electronic device 700 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 701 and one or more memories 702. The memory 702 stores at least one instruction, which is loaded and executed by the processor 701 to implement the steps of the above-mentioned watershed-scale groundwater storage GWS long-term reconstruction method.

[0107] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned watershed-scale groundwater storage GWS long-term reconstruction method. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0108] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for long-term reconstruction of groundwater storage at the watershed scale using the Groundwater Water Storage System (GWS), characterized in that, The method includes: S1. Acquire and fuse long-term meteorological and hydrological data from 1960 to 2023, including: precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale. S2. Acquire and fuse the short-term total terrestrial water storage (TWS) data after 2002; S3. Based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS, the short-term groundwater GWS sequence is calculated. S4. Using the short-term groundwater GWS sequence and the portion of the fused long-term meteorological and hydrological data after 2002 as target data and driving data, train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model. The long-term groundwater storage reconstruction RF-KF coupled model includes the random forest RF sub-model and the Kalman filter KF sub-model. S5. Input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS). S6. Replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS, and use the update module of the Kalman filter KF sub-model to simulate the time dynamics and correct the accumulated error of the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

2. The method according to claim 1, characterized in that, S1 specifically includes: S11. Obtain the long-term meteorological and hydrological data through meteorological reanalysis data ERA5 and land data assimilation model GLDAS. The data of ERA5 and GLDAS correspond to each other and both provide precipitation, evapotranspiration, air temperature, surface water storage SWS, and soil water storage SMS at the watershed scale. S12. Obtain meteorological station data from the China Meteorological Administration (CMA), including precipitation, evapotranspiration, and temperature. ERA5 and GLDAS data often differ from the meteorological station data. Use the meteorological station data to fuse and calibrate the ERA5 and GLDAS data. This fusion calibration uses a Bayesian linear fusion algorithm, targeting a specific meteorological station, and fuses and calibrates the ERA5 and GLDAS data for its corresponding location. The formula is as follows: and fused-1 =α·y ERA5 +(1-α)·y GLDAS +∈1 Where y fused-1 To fuse the calibrated data, and y ERA5 and y GLDAS For the corresponding ERA5 and GLDAS data, both are unified to a spatial resolution of 0.1°, consistent with the ERA5 raster size. Through optimization of coefficient α and residual ∈1, the root mean square error (RMSE) between the fused and calibrated data and the meteorological station data is minimized.

3. The method according to claim 1, characterized in that, S2 specifically includes: S21. The Total Land Storage (TWS) data for 2002-2023 was obtained using the GRACE satellite mission launched in 2002. GRACE Level 3 remote sensing product data from JPL, GSFC, and CSR were used, with a spatial resolution of 3°. S22. Obtain at least one year's worth of TWS measured values ​​through sampling or by obtaining them from officially published water resources statistical yearbooks. Use the TWS measured values ​​to perform fusion calibration on the Level 3 remote sensing product data. The fusion calibration uses a Bayesian linear fusion algorithm, and its formula is as follows: y fused-tws =α1y GSFC +α2y CSR +α3y JPL +∈2 Where y fused-tws To fuse the calibrated data, α i The coefficient is y, the sum is 1, and y GSFC ,y CSR ,y JPL All data are remote sensing product data, ∈2 represents the residuals, and the coefficient α is used to calculate the residuals. i The optimization of residuals ∈2 minimizes the root mean square error (RMSE) between the fused calibrated data and the TWS measured values.

4. The method according to claim 1, characterized in that, S3 specifically includes: The short-term groundwater GWS sequence of the watershed is estimated by subtracting the known changes in SWS and SMS from the changes in TWS using the watershed-scale water balance equation: ΔGWS = ΔTWS - ΔSWS - ΔSMS. The GWS data for a certain year is obtained by calibration using local water resources bulletins for that year.

5. The method according to claim 1, characterized in that, The Random Forest (RF) sub-model aggregates multiple decision trees to capture the complex nonlinear relationship between meteorological and hydrological elements and GWS fluctuations. Its formula is as follows: Where f(R) is the long-term prediction result of the groundwater storage GWS, and T i (x) is the result predicted by the decision tree, and N is the number of decision trees.

6. The method according to claim 5, characterized in that, The Kalman filter KF sub-model uses an improved Kalman filter algorithm for data reconstruction. The original Kalman filter algorithm operates within a state-space framework, continuously improving predictions by merging new observations and dynamically adjusting weights through the state covariance matrix. It estimates the system state by combining model predictions with observed data. The prediction steps of the prediction module are as follows: X k =F k x k-1 +B k u k +w k Where X k It is the Kalman filter prediction at time k; F k It is the state transition matrix, representing the relationship between the predicted value at the previous time point and the predicted value at the current time point; B k It is a control input matrix, representing the importance of the control factors; u k It is a meteorological and hydrological driving factor; w k It is process noise; The update module optimizes the prediction results based on the observed data: f(K)=X k +K k (z k -X k ) K k =P k (P k +R k ) -1 Where zk is the observed value of groundwater storage; K k It is the Kalman gain; R k and Q k All of these are model noise; The improved Kalman filter algorithm uses the prediction results of the random forest (RF) sub-model to replace the output of the prediction module in the original Kalman filter, and uses an update module to simulate time dynamics and correct for long-term prediction accumulated errors. The equations are as follows: f’(R,K) after2002 =f(R)+K k (z k -f(R)) f’(R,K) before2002 =f(R)+K k总 (z k总 -f(R) 总 ) Where f'(R,K) after2002 This is the result of GWS reconstruction of groundwater storage after 2002, f'(R,K) before2002 This is a long-term reconstruction of groundwater storage using the GWS model prior to 2002. f(R) is the prediction result from the Random Forest (RF) sub-model, replacing the original Kalman filter prediction result X. k ;z k This refers to the actual observed groundwater storage for a specific year (k) after 2002 that needs to be predicted; z k总 This represents the total observed values ​​from one year after the year k that needs to be predicted, up to the final year, before 2002. For each year after 2002, the actual observed groundwater storage for that year is used. For each year before 2002, the reconstructed values ​​from the Kalman filter KF sub-model are used as the observed values ​​for that year. These observed values ​​are summed to form z. k总 ;K k It is the Kalman gain, and its value is related to the prediction error F of the random forest. k It shows a negative correlation; for data after 2002, F... k F represents the error between the random forest prediction and the actual short-term groundwater observation for year k. For data prior to 2002, F k总 K represents the total error from year k to the final year; k总 With F k总 Showing a negative correlation, through and K k =P k (P k +R k ) -1 Calculated.

7. A watershed-scale groundwater storage GWS long-term reconstruction system, characterized in that, The system includes: The first acquisition and fusion calibration module is used to acquire and fuse long-term meteorological and hydrological data from 1960 to 2023. The long-term meteorological and hydrological data includes precipitation, evapotranspiration, air temperature, surface water storage (SWS), and soil water storage (SMS) at the watershed scale. The second acquisition and fusion calibration module is used to acquire and fuse the short-term total terrestrial water storage (TWS) data after 2002. The calculation module is used to calculate the short-term groundwater GWS sequence based on the watershed-scale water balance equation ΔTWS=ΔSWS+ΔSMS+ΔGWS. The training module is used to train the random forest RF sub-model of the long-term groundwater storage reconstruction RF-KF coupled model using the short-term groundwater GWS sequence and the portion of the fused long-term meteorological and hydrological data after 2002 as target data and driving data. The long-term groundwater storage reconstruction RF-KF coupled model includes a random forest RF sub-model and a Kalman filter KF sub-model. The prediction module is used to input the fused long-term meteorological and hydrological data into the trained random forest (RF) sub-model to obtain the long-term prediction results of groundwater storage (GWS). The reconstruction module is used to replace the output of the prediction module of the Kalman filter KF sub-model with the long-term prediction result of the groundwater storage GWS. The update module of the Kalman filter KF sub-model is used to simulate the time dynamics and correct the error accumulated in the long-term prediction to obtain the final long-term reconstruction result of the groundwater storage GWS.

8. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, characterized in that, The processor loads and executes at least one instruction to implement the watershed-scale groundwater storage GWS long-term reconstruction method as described in any one of claims 1-6.

9. A computer-readable storage medium storing at least one instruction, characterized in that, The at least one instruction is loaded and executed by the processor to implement the watershed-scale groundwater storage GWS long-term reconstruction method as described in any one of claims 1-6.

Images