Efficient prediction method for spatio-temporal distribution of groundwater level at grid scale in large-scale agricultural irrigation area

Abstract

The application provides a large-scale agricultural irrigation area grid scale groundwater level space-time distribution efficient prediction method, belongs to the technical field of groundwater prediction, and steps include: firstly, a SWAT model of a basin where an irrigation area is located and a MODFLOW model covering the irrigation area are respectively constructed, the two are coupled to obtain a SWAT-MODFLOW model, and parameter correction is carried out; secondly, an irrigation area groundwater level data set is constructed through scenario combination and model simulation, is used for training and screening the highest-precision LSTM model, and the optimal control range of each LSTM model is determined through space compression and reconstruction, and multiple LSTM models are obtained; finally, the multiple LSTM models are parallelly calculated to predict the space-time distribution of the grid scale groundwater level of the irrigation area. The optimal control range of each LSTM model is determined through space compression-reconstruction, finally, the prediction of the space-time distribution of the grid scale groundwater level of the large-scale agricultural irrigation area is realized through the parallel calculation of the multiple LSTM models, and the prediction efficiency is improved on the premise of ensuring the prediction accuracy of the irrigation area groundwater level.

Description

Technical Field

[0001] This invention belongs to the field of groundwater prediction technology, and relates to an efficient prediction method for the spatiotemporal distribution of groundwater level at the grid scale in large-scale agricultural irrigation areas. In particular, it relates to an efficient prediction method for the spatiotemporal distribution of groundwater level at the grid scale in large-scale agricultural irrigation areas based on the coupling of spatial compression, reconstruction and artificial intelligence technologies. Background Technology

[0002] As a populous and agriculturally significant country, China's agricultural production is highly dependent on irrigation. In some agricultural irrigation areas located in arid and semi-arid regions of northern China, limited surface water resources are insufficient to sustain agricultural production, making groundwater the primary source of irrigation, accounting for over 70% of total water consumption. Long-term, large-scale groundwater extraction has led to a continuous decline in groundwater levels within irrigation areas, resulting in a series of ecological and environmental problems such as groundwater depletion, reduced river flow, and ecological degradation. Against this backdrop, it is particularly necessary to determine the response of regional groundwater levels to meteorological conditions and various groundwater resource management measures, and then to determine targeted groundwater resource management measures through optimization calculations.

[0003] Currently, there are two main types of methods for predicting the response of regional groundwater levels to meteorological conditions and groundwater resource management measures. One type is the physical model represented by MODFLOW (Modular Three-Dimensional Finite-Difference Groundwater Flow Model), which mainly solves partial differential equations using methods such as the finite difference method or the finite element method, and then predicts the spatiotemporal distribution of regional groundwater levels on a grid basis. For example, Chinese invention patent (201911057334.1) provides a method for simulating hydrogeological characteristics and regional groundwater circulation for quantitative evaluation of groundwater flow. In this patent, the MODFLOW model is used to simulate regional groundwater dynamics. The MODFLOW model can achieve accurate calculation of regional groundwater levels at the grid scale, but the computation time cost is high. This disadvantage is further amplified when coupled with optimization algorithms to optimize groundwater resource management measures.

[0004] Another type is machine learning models such as LSTM (Long Short-Term Memory) and RNN (Recurrent Neural Network), which mainly determine the response of regional groundwater levels to meteorological conditions and groundwater resource management measures through nonlinear mapping. In this approach, the mainstream modeling scheme is to train a machine learning model for all sample points (observation wells or grids) within the region and use the trained model to predict the groundwater level at all sample points in the entire region. For example, Chinese invention patent (202510300451.5) provides a training method and related device for a groundwater level prediction model. This patent uses monitoring point data within the region to train a machine learning model and uses the trained model to predict the groundwater level at all monitoring points within the region. Compared to the MODFLOW model, the higher computational efficiency is an advantage of machine learning models. However, due to the heterogeneity of hydrological conditions, geological conditions, and human activity intensity within large-scale agricultural irrigation areas, groundwater level fluctuation patterns in different regions are often inconsistent. This modeling scheme, relying solely on a single model, cannot guarantee the prediction accuracy at each grid point, and therefore does not achieve satisfactory performance in large-scale agricultural irrigation areas.

[0005] Existing technologies cannot achieve a balance between accuracy and efficiency in groundwater level prediction. Improving the prediction efficiency of models while maintaining accuracy in groundwater level prediction for large-scale agricultural irrigation areas is a problem that needs to be solved. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides an efficient method for predicting the spatiotemporal distribution of groundwater levels at the grid scale in large-scale agricultural irrigation districts. Starting from the influencing factors and response processes of groundwater levels in each grid, and guided by the Nash-Sutcliffe Efficiency (NSE) LSTM prediction accuracy evaluation, grids with similar physical and hydrological characteristics are clustered. Through parallel computation of LSTM models established for the control grids at the centroid locations of each cluster, efficient and accurate prediction of monthly groundwater levels in all grids within the irrigation district is achieved.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A highly efficient method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas is proposed. Specifically, it is a method based on the coupling of spatial compression, reconstruction, and artificial intelligence technologies, and includes the following steps:

[0009] Step 1: Construct a SWAT (Soil and Water Assessment Tool) model of the watershed where the irrigation district is located. Specifically:

[0010] Step 1.1: Combining relevant data including water system data, DEM (Digital Elevation Model) data, land use data, soil data, and irrigation water intake data of the watershed where the irrigation district is located, use ArcGIS software as the platform and ArcSWAT toolkit to construct a SWAT model of the watershed where the irrigation district is located.

[0011] Step 1.2: Based on the measured historical runoff data of the hydrological station in the watershed where the irrigation area is located, and using SWAT-CUP software as the platform, the SUFI-2 (Sequential Uncertainty Fitting, version 2) global search algorithm is used to complete the parameter calibration of the SWAT model in Step 1.1, so that the output of the SWAT model can fit the surface runoff process.

[0012] Step 1.3: Organize the riverbed seepage, precipitation infiltration, and groundwater evaporation data output by the SWAT model in Step 1.2 for use in the coupling of the SWAT and MODFLOW models in the subsequent Step 3.

[0013] Step 2, construct the MODFLOW model covering the irrigation area, specifically:

[0014] Combining borehole data, hydrogeological maps, hydrogeological parameters, and other relevant data, a MODFLOW model covering the irrigation area was constructed using VisualMODFLOW software as a platform, with a grid scale of 1km×1km, to output groundwater level simulation results at the grid scale of the irrigation area.

[0015] Step 3: Couple the SWAT model with the MODFLOW model to obtain the SWAT-MODFLOW model, and perform parameter correction so that the surface runoff and groundwater level output by the SWAT-MODFLOW model can fit historical measured data; specifically:

[0016] Step 3.1: The riverbed seepage, precipitation infiltration and groundwater evaporation output from the SWAT model in Step 1.3 are used as input data and input into the MODFLOW model constructed in Step 2.1 to drive the simulation of the groundwater level in the irrigation area, complete the coupling of the SWAT model and the MODFLOW model, and obtain the SWAT-MODFLOW model.

[0017] Step 3.2: Using the historical measured groundwater level of the observation well as a reference, adjust the hydrogeological parameters of each grid in the MODFLOW model constructed in Step 3.1 so that the groundwater level of the irrigation area simulated by the SWAT-MODFLOW model fits the historical measured groundwater level process.

[0018] Step 3.3, if the simulation accuracy index R of the groundwater level in step 3.2 is...2 If the Coefficient of Determination is less than 0.99, the range of values ​​for parameters related to riverbed seepage, precipitation infiltration, and groundwater evaporation in the SWAT model constructed in step 1 is re-estimated, and the output of the SWAT model is again transferred to the MODFLOW model constructed in step 2. This process is iterated until the surface runoff and groundwater level output by the SWAT-MODFLOW model in step 3.1 can fit the historical measured data, where the surface runoff simulation accuracy requirement is R0. 2 ≥0.7, groundwater level simulation accuracy requirement R 2 ≥0.99.

[0019] Step 4: Construct various meteorological and groundwater extraction scenarios. Through scenario combinations and model simulations, build a groundwater level dataset for the irrigation area, which will be used to train and select LSTM models. Specifically:

[0020] Step 4.1: Combine historical and future meteorological data of the watershed to construct a dataset of multiple meteorological scenarios, which will be used as input to the SWAT model in the SWAT-MODFLOW model in Step 3.

[0021] Step 4.2: Combine historical groundwater extraction data of the irrigation area to construct a dataset of multiple groundwater extraction scenarios, which will be used as input to the MODFLOW model in the SWAT-MODFLOW model in Step 3.

[0022] Step 4.3: Combine the multiple meteorological datasets constructed in Step 4.1 and the multiple groundwater extraction datasets constructed in Step 4.2 in pairs to construct combined scenarios where multiple meteorological scenarios and groundwater extraction scenarios intersect; the pairwise combination method specifically refers to the combination of each meteorological scenario and each groundwater extraction scenario to form a groundwater extraction scenario.

[0023] Step 4.4: Using the various combined scenarios obtained in Step 4.3 as input to the SWAT-MODFLOW model in Step 3, the groundwater level response process of each grid in the irrigation area under various combined scenarios is simulated and determined through the SWAT-MODFLOW model.

[0024] Step 4.5: Based on the time span, the groundwater level response processes of each grid within the irrigation area under various combined scenarios determined in Step 4.4 are divided into training, validation, and test sets in proportions of 70%, 15%, and 15%, respectively. The training set is used for parameter training of the LSTM model, the validation set is used for hyperparameter tuning of the LSTM model, and the test set is used to evaluate the prediction accuracy of the LSTM model. Finally, the LSTM model with the highest accuracy on the test set that meets the accuracy requirements (Nash efficiency coefficient NSE ≥ 0.5) is selected.

[0025] Step 5: Determine the optimal control range for each LSTM model through spatial compression and reconstruction. Specifically:

[0026] Step 5.1: Based on the spatial distribution of six features of each grid in the irrigation area, namely permeability coefficient, water supply, land use type, elevation, aquifer thickness and standard deviation of groundwater level fluctuation, the entire irrigation area is clustered into multiple partitions using the K-means clustering algorithm and the silhouette coefficient optimization method, and the grid where the centroid of each partition is located is used as the control grid of that partition.

[0027] Step 5.2: Build and train an LSTM model for each control grid of each partition separately, which will be used to predict the groundwater level at the control grid under various combined scenarios in Step 4.3.

[0028] Step 5.3 involves generalizing the parameters of the LSTM model corresponding to each control grid to all grids within the irrigation area through a traversal approach. This determines the accuracy distribution of each LSTM model parameter across all grids predicting groundwater levels, and preliminarily constructs a multi-LSTM model accuracy evaluation matrix. ;

[0029] The accuracy evaluation matrix of the multi-LSTM model is shown in Equation (1):

[0030] (1)

[0031] Among them, the multi-model accuracy evaluation matrix line number of each line Represents a grid, with column labels for each column. Represents a control grid. Okay, number The value of the column Representing the The grid is using the first The average accuracy obtained on test sets for various scenarios when the parameters of a control grid model are set.

[0032] Step 5.4: Extract the multi-LSTM model accuracy evaluation matrix obtained in Step 5.3. The maximum value of each row determines the optimal accuracy that can be achieved by multiple LSTM models jointly predicting the groundwater level of all grids in the irrigation area, and the spatial distribution of the optimal accuracy is determined according to the spatial location of each grid.

[0033] Step 5.5: Add control grids in areas with weak accuracy and repeat steps 5.2 to 5.4 to complete the correction of the accuracy evaluation matrix of the multi-LSTM model;

[0034] Step 5.6: Determine the optimal control range for each control grid using a precision clustering algorithm. Specifically:

[0035] If the matrix in step 5.5 The Walking in its first If the column reaches its maximum value, then the [number]th [column]... The grid was summarized in the first... The optimal control range of an LSTM model.

[0036] Step 6: Parallel computation of multiple LSTM models to predict the spatiotemporal distribution of groundwater levels at the grid scale in the irrigation area. Specifically:

[0037] In the process of predicting the spatiotemporal distribution of groundwater level at the grid scale in the irrigation area, the LSTM model built and trained for each control grid is computed in parallel to obtain the groundwater level of all grids within the optimal control range of each LSTM model. Combined with the spatial location of each grid, the spatiotemporal distribution of groundwater level at the grid scale of the entire irrigation area is obtained.

[0038] The advantages of this invention compared to the prior art are as follows:

[0039] (1) Traditional physical models such as MODFLOW and FEFLOW can predict groundwater levels at the grid scale of large-scale agricultural irrigation areas. The calculation results are accurate and the spatial resolution is high. However, their main disadvantage is the high time cost required for calculation. This makes it difficult to couple the simulation of groundwater level with optimization algorithms to optimize the calculation of groundwater resource management measures. In this invention, the SWAT-MODFLOW model constructed in steps 1-3 is used to determine the response process of groundwater level in each grid of the irrigation area under various meteorological conditions and groundwater extraction conditions. Then, the groundwater level response process (output) of the irrigation area under various meteorological conditions and groundwater extraction conditions (input) is used to train the LSTM model. Finally, the groundwater level process of each grid of the irrigation area is predicted by parallel computing technology (step 6). While ensuring accuracy and resolution, the calculation efficiency is greatly improved.

[0040] (2) Current artificial intelligence technologies mostly use a single model to cover the entire research area. However, due to the heterogeneity of groundwater level fluctuation patterns caused by the differences in physical characteristics and hydrological properties between grids, the model training accuracy is poor. Furthermore, in the actual prediction process, it is impossible to take into account the accuracy of each grid, ultimately resulting in poor overall model accuracy or generalization ability. In contrast, this invention first uses spatial compression (steps 5.1~5.2) to determine the control grid, and then performs one-to-one modeling for each control grid, thereby avoiding the interference of heterogeneous groundwater level fluctuation patterns on parameter training and significantly improving training accuracy. Then, spatial reconstruction (steps 5.3~5.6) is used to find the spatial range with the most similar physical characteristics, hydrological properties and control grids, thereby significantly improving the prediction accuracy of groundwater levels in all grids of the irrigation area under study. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas according to the present invention.

[0042] Figure 2 This is a structural diagram of the LSTM model constructed in this invention;

[0043] Figure 3 This is a schematic diagram of irrigation district grid division according to an embodiment of the present invention;

[0044] Figure 4 This is a graph showing the simulation results of combined scenarios within the irrigation area and the division results of the groundwater level dataset in an embodiment of the present invention. Each broken line in the graph represents the response process of the groundwater level of a single grid under a certain type of combined scenario, with a time span from January 2001 to December 2064. According to the proportions of 70%, 15%, and 15%, the data from January 2001 to October 2045 is divided into the training set, the data from October 2045 to May 2055 is divided into the validation set, and the data from May 2055 to December 2064 is divided into the test set.

[0045] Figure 5 This is a location distribution diagram of the irrigation area control grid according to an embodiment of the present invention. Each black circle in the diagram represents a control grid.

[0046] Figure 6 This is a map showing the accuracy distribution of the LSTM model of each control grid in the irrigation area in an embodiment of the present invention in predicting the groundwater level of all grids in the irrigation area. The circles in the map represent control grids, the green circles represent the current control grids, and the colors of the other grids correspond to the NSE prediction accuracy. Figure 6 (a) in the figure is the accuracy distribution of the LSTM model of control grid 1 in predicting the groundwater level of all grids in the irrigation area; Figure 6 (b) in the figure shows the accuracy distribution of the LSTM model of control grid 2 in predicting the groundwater level of all grids in the irrigation area; Figure 6 (c) in the figure is the accuracy distribution of the LSTM model of control grid 3 in predicting the groundwater level of all grids in the irrigation area; Figure 6 (d) in the figure is the accuracy distribution of the LSTM model of control grid 4 in predicting the groundwater level of all grids in the irrigation area; Figure 6 (e) in the figure is the accuracy distribution of the LSTM model of control grid 5 on the groundwater level prediction of all grids in the irrigation area; Figure 6 (f) in the figure is the accuracy distribution of the LSTM model of control grid 6 on the groundwater level prediction of all grids in the irrigation area; Figure 6 (g) in the figure is the accuracy distribution of the LSTM model of control grid 7 on the groundwater level prediction of all grids in the irrigation area; Figure 6(h) in the figure is the accuracy distribution of the LSTM model of control grid 8 on the groundwater level prediction of all grids in the irrigation area; Figure 6 (i) in the figure is the accuracy distribution of the LSTM model of control grid 9 in predicting the groundwater level of all grids in the irrigation area;

[0047] Figure 7 This is a distribution map of the optimal control range of each control grid in the irrigation area according to an embodiment of the present invention. The circles in the map represent control grids, the green circles represent the current control grids, and the blue grids represent the optimal control range of the current control grids. Figure 7 (a) in the diagram is the optimal control range distribution map of control grid 1; Figure 7 (b) in the diagram is the optimal control range distribution map of control grid 2; Figure 7 (c) in the diagram is the optimal control range distribution map of control grid 3; Figure 7 (d) in the diagram represents the optimal control range distribution of control grid 4. Figure 7 (e) in the diagram represents the optimal control range distribution of control grid 5. Figure 7 (f) in the diagram represents the optimal control range distribution of control grid 6. Figure 7 (g) in the diagram represents the optimal control range distribution of control grid 7. Figure 7 (h) in the diagram represents the optimal control range distribution of control grid 8. Figure 7 (i) in the diagram represents the optimal control range distribution of control grid 9.

[0048] Figure 8 This is a distribution map showing the accuracy of the groundwater level prediction results in the irrigation area according to an embodiment of the present invention. Figure 8 (a) in the figure is the precision distribution of the NSE index; Figure 8 (b) in the figure is the accuracy distribution of the Mean Absolute Error (MAE) index; Figure 8 (c) in the figure is the precision distribution of the Root Mean Square Error (RMSE) index;

[0049] Figure 9 This is a comparison chart of the cumulative frequency distribution results of the prediction accuracy of groundwater level in irrigation areas in the embodiments of the present invention and the prediction accuracy of the control examples. The blue line represents the results of the embodiments, and the orange line represents the results of the control examples. Figure 9 (a) in the figure is a comparison chart of the cumulative frequency distribution results of the NSE index; Figure 9 (b) in the figure is a comparison chart of the cumulative frequency distribution results of the MAE index; Figure 9 (c) in the figure is a comparison of the cumulative frequency distribution results of the RMSE index. Detailed Implementation

[0050] As mentioned above, this invention provides an efficient method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas based on the coupling of spatial compression, reconstruction, and artificial intelligence technologies. The specific process is as follows: Figure 1 As shown, the method includes the following steps:

[0051] Step 1: Construct a SWAT model of the watershed where the irrigation district is located. Specifically:

[0052] Step 1.1: Combining relevant data including water system data, DEM data, land use data, soil data, and irrigation water intake data of the watershed where the irrigation district is located, a SWAT model of the watershed where the irrigation district is located is constructed using ArcGIS software and the ArcSWAT toolkit to characterize surface hydrological processes. The expression is as follows:

[0053] (2)

[0054] In the formula, Represents the final soil moisture content, in mm; Represents the initial soil moisture content, in mm; Represents the time step. Representing the Rainfall, mm; Representing the Surface runoff, mm; Representing the Evapotranspiration per day, mm; Representing the Soil infiltration and flow rate measured daily, mm; Representing the The groundwater runoff per day, in mm.

[0055] Step 1.2: Based on the measured historical runoff data of the hydrological station in the watershed where the irrigation area is located, the SWAT-CUP software is used as the platform, and the SUFI-2 global search algorithm is used to complete the parameter calibration of the SWAT model in Step 1.1, so that the model output can fit the surface runoff process.

[0056] Step 1.3: Organize the riverbed seepage, precipitation infiltration, and groundwater evaporation data output by the SWAT model in Step 1.2 for use in the coupling of the SWAT and MODFLOW models in the subsequent Step 3.

[0057] Step 2, construct the MODFLOW model covering the irrigation area, specifically:

[0058] Combining borehole data, hydrogeological maps, and hydrogeological parameters, a MODFLOW model covering the irrigation area was constructed using VisualMODFLOW software at a grid scale of 1 km × 1 km. This model was used to output groundwater level simulation results at the grid scale for the irrigation area. The grid division results are shown below. Figure 3 As shown. Since the groundwater source of the irrigation area under study is mainly the Quaternary unconfined aquifer, the mathematical model of groundwater flow in MODFLOW is generalized into a heterogeneous, isotropic two-dimensional unsteady groundwater flow, as expressed below:

[0059] (3)

[0060] In the formula, To simulate the seepage zone, m 2 (x,y) represents the spatial coordinates of the grid, in meters (m). The free surface of groundwater; It is a type of boundary; It is a type II boundary; For time, The flow rate per unit area at the flow boundary of the seepage zone, m 3 / d / m; The initial groundwater level is in meters (m). The river level is in meters (m). The direction of the outward normal to the boundary; Permeability coefficient, m / d; The water storage capacity of the aquifer is expressed in l / m. The gravity yield of the unconfined aquifer above the water surface; Water that replenishes or discharges to the water surface per unit time and per unit area, including precipitation infiltration and evaporation; For the source and sink terms of the aquifer, l / d.

[0061] Step 3, SWAT-MODFLOW model coupling and parameter correction, specifically:

[0062] Step 3.1: Transfer the riverbed seepage, precipitation infiltration and groundwater evaporation output from the SWAT model in Step 1.3 to the MODFLOW model constructed in Step 2 to drive the simulation of the groundwater level in the irrigation area and complete the coupling of the SWAT-MODFLOW model.

[0063] Step 3.2: Using the historical measured groundwater level of the observation well as a reference, adjust the hydrogeological parameters of each grid in the MODFLOW model constructed in Step 3.1 so that the groundwater level of the irrigation area simulated by the model fits the historical measured groundwater level process.

[0064] Step 3.3, if the simulation accuracy index R of the groundwater level in step 3.2 is...2 If the Coefficient of Determination is less than 0.99, the range of values ​​for parameters related to riverbed seepage, precipitation infiltration, and groundwater evaporation in the SWAT model constructed in step 1 is re-estimated, and the output of the SWAT model is again transferred to the MODFLOW model constructed in step 2. This process is repeated iteratively until the surface runoff and groundwater level output by the SWAT-MODFLOW model in step 3.1 can fit the historical measured data (where the surface runoff simulation accuracy requirement is R). 2 ≥0.7, groundwater level simulation accuracy requirement R 2 ≥0.99).

[0065] Step 4: Construct various meteorological and groundwater extraction scenarios. Through scenario combinations and model simulations, build a groundwater level dataset for the irrigation area. This dataset is then used to train and select the most accurate LSTM model. Specifically:

[0066] Step 4.1: Combining historical and future meteorological data from the watershed, four types of meteorological scenario datasets are constructed for input into the SWAT model in Step 3's SWAT-MODFLOW model. Specifically, the future meteorological scenarios include historical reenactments and three shared socioeconomic pathways (SSPs) under the BCC-CSM2-MR (Beijing Climate Center Climate System Model version 2 Medium Resolution) atmospheric circulation model from the CMIP6 (Coupled Model Intercomparison Project Phase 6) project: SSP126, SSP245, and SSP370. The relevant data for each shared socioeconomic pathway scenario are mapped from grid points to four meteorological stations within the watershed of the irrigation area under study after inverse distance weight interpolation and bias correction.

[0067] Step 4.2: Combining historical groundwater extraction data from the irrigation district, construct four types of groundwater extraction scenario datasets, which will be used as input to the MODFLOW model in the SWAT-MODFLOW model of Step 3. Specifically, the groundwater extraction scenarios are set according to the monthly historical groundwater extraction intensity of various land uses within the irrigation district, with four groundwater extraction scenarios set for each land use type: 25%, 50%, 75%, and 100% of the average historical groundwater extraction intensity.

[0068] Step 4.3: Combine the four types of meteorological datasets constructed in Step 4.1 with the four types of groundwater extraction datasets constructed in Step 4.2 in pairs to construct sixteen combined scenarios that intersect meteorological scenarios and groundwater extraction scenarios (the processing method specifically refers to the combination of each type of meteorological scenario and each type of groundwater extraction scenario to form a groundwater extraction scenario).

[0069] Step 4.4: Using the various combined scenarios obtained in Step 4.3 as input to the SWAT-MODFLOW model in Step 3, the groundwater level response process of each grid in the irrigation area under various combined scenarios is determined through model simulation. Figure 4 );

[0070] Step 4.5: Based on the time span, the groundwater level response processes of each grid within the irrigation district under various combined scenarios determined in Step 4.4 are divided into training, validation, and test sets according to a ratio of 70%, 15%, and 15%, respectively. Figure 4 The training set is used for training the parameters of the LSTM model, the validation set is used for tuning the hyperparameters of the LSTM model, and the test set is used to evaluate the prediction accuracy of the LSTM model and select the LSTM model with the highest accuracy that meets the accuracy requirement (NSE≥0.5).

[0071] Step 5: Spatial compression and reconstruction determine the optimal control range for each LSTM model. Specifically:

[0072] Step 5.1: Based on the spatial distribution of six features—permeability coefficient, specific yield, land use type, elevation, aquifer thickness, and standard deviation of groundwater level fluctuation—within each grid in the irrigation area, the entire irrigation area is clustered into multiple zones using the K-means clustering algorithm and silhouette coefficient optimization method. The grid containing the centroid of each zone is then used as the control grid for that zone. (See [link to control grid location]). Figure 5 ;

[0073] Step 5.2: Establish and train a separate LSTM model for each control grid partition. This model will be used to predict groundwater levels at the control grid under various combined scenarios in Step 4.3. The LSTM model takes monthly cumulative precipitation, monthly cumulative evaporation, monthly cumulative groundwater extraction depth, river flow, and the groundwater level at the beginning of the month as input, and the groundwater level at the end of the month as output. The Adam optimization algorithm is used for parameter training. Hyperparameters are manually adjusted based on their performance on the validation set. The model structure is as follows: Figure 2 As shown;

[0074] Step 5.3: By traversing through the grid, the parameters of the LSTM model corresponding to each control grid are extended to all grids in the irrigation area to determine the accuracy distribution of each LSTM model parameter on the groundwater level of all grids, and a preliminary multi-LSTM model accuracy evaluation matrix M is constructed.

[0075] The accuracy evaluation matrix of the multi-LSTM model is shown in Equation (4):

[0076] (4)

[0077] Among them, the multi-model accuracy evaluation matrix line number of each line Represents a grid, with column labels for each column. Represents a control grid. Okay, number The value of the column Representing the The grid is using the first The average accuracy obtained on test sets for various scenarios when the parameters of a control grid model are set.

[0078] Step 5.4: Extract the multi-LSTM model accuracy evaluation matrix obtained in Step 5.3. The maximum value of each row determines the optimal accuracy that can be achieved by multiple LSTM models jointly predicting the groundwater level of all grids in the irrigation area, and the spatial distribution of the optimal accuracy is determined according to the spatial location of each grid.

[0079] Step 5.5: Add control grids in areas with low accuracy and repeat steps 5.2 to 5.4 to complete the correction of the multi-LSTM model accuracy evaluation matrix. The spatial distribution of the optimal accuracy determined by this matrix is ​​as follows: Figure 6 ;

[0080] Step 5.6: Determine the optimal control range for each control grid using a precision clustering algorithm. The specific distribution of the optimal control range is as follows: Figure 7 Specifically:

[0081] If the matrix in step 5.5 The Walking in its first If the column reaches its maximum value, then the first... The grid was summarized in the first... The optimal control range of an LSTM model.

[0082] Step 6: Parallel computation of multiple LSTM models to predict the spatiotemporal distribution of groundwater levels at the grid scale in the irrigation area. Specifically:

[0083] In the prediction of the spatiotemporal distribution of groundwater level at the grid scale in the irrigation area, LSTM models constructed and trained for each control grid are computed in parallel to predict the groundwater level of all grids within the optimal control range of each LSTM model. The numerical distribution and cumulative frequency distribution of the groundwater level prediction accuracy are as follows: Figure 8 and Figure 9By combining the spatial location of each grid, the spatiotemporal distribution of groundwater level at the grid scale throughout the irrigation area is obtained.

[0084] This embodiment selects an irrigation district in the Songliao River Basin as a research example and uses the method described in this invention to efficiently predict the spatiotemporal distribution of groundwater level at the grid scale in the irrigation district.

[0085] The irrigation district studied is located in the Songliao River Basin in Northeast China. The district has a typical temperate semi-arid continental climate, with rainfall during the rainy season (June-August) accounting for over 80% of the total annual precipitation. The irrigation district is bounded by the main river channel, and its land use is primarily agricultural. In recent years, the large-scale expansion of arable land has led to a sharp increase in water demand. Long-term, large-scale groundwater extraction has caused the groundwater level to decline year after year, resulting in a large-scale depression cone that now covers almost the entire irrigation district.

[0086] The first step is to construct a SWAT model of the watershed where the irrigation district is located and complete the preliminary parameter calibration.

[0087] The second step involves constructing a MODFLOW model covering the irrigation area using a 1km×1km grid scale. The grid subdivision results for the irrigation area are as follows: Figure 3 As shown, the figure contains 1904 grids, each with an area of ​​1 km².

[0088] The third step involves transmitting the data from the SWAT model, including riverbed seepage, precipitation infiltration, and groundwater evaporation, to the MODFLOW model to drive the simulation of groundwater level, thus completing the coupling of the SWAT-MODFLOW model. The parameters of the SWAT-MODFLOW model are then corrected based on historical measured runoff data and groundwater level data.

[0089] The fourth step involves multi-scenario simulation to determine the response process of groundwater levels in each grid within the irrigation district under various combined scenarios, and to construct a groundwater level dataset at the grid scale of the irrigation district. Here, we use grid number 125 as an example to illustrate the groundwater level dataset and its processing method. Figure 4 ).

[0090] The fifth step involves gradually determining the location of the control grid through spatial compression and reconstruction. Figure 5 The accuracy distribution of the LSTM model for each control grid in predicting groundwater levels across all grids in the irrigation area. Figure 6 ) and the optimal control range of each control grid ( Figure 7 ).

[0091] The sixth step involves parallel computation using multiple LSTM models to achieve efficient and accurate prediction of the spatiotemporal distribution of groundwater levels at the grid scale in the irrigation area. This computation is run on a computing server equipped with an Intel(R) Xeon(R) Gold5218R processor and an NVIDIA A100 GPU. Parallel computation of multiple LSTM models can complete a prediction of 116 months in approximately 13 seconds, representing a computational efficiency approximately 10 times higher than the physical model. The model's prediction accuracy is evaluated using three metrics: NSE, MAE, and RMSE. The numerical distribution of the prediction results, compared to the accuracy of the physical model, is shown in [Figure / link / reference]. Figure 8 The comparison of the cumulative frequency distribution of accuracy between the examples and the control examples is shown in the figure. Figure 9 .

[0092] Finally, it should be noted that the above is only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention (such as the application of various formulas, the order of steps, etc.) without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A highly efficient method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas, characterized in that, The efficient prediction method for the spatiotemporal distribution of groundwater levels at the grid scale in large-scale agricultural irrigation areas includes the following steps: Step 1: Construct a SWAT model of the watershed where the irrigation district is located; Step 2: Construct the MODFLOW model for the irrigated area; Step 3: Couple the SWAT model with the MODFLOW model to obtain the SWAT-MODFLOW model, and perform parameter correction so that the surface runoff and groundwater level output by the SWAT-MODFLOW model can fit the historical measured data. Step 4: Construct multiple meteorological and groundwater extraction scenarios. Construct a groundwater level dataset for irrigation area through scenario combination and model simulation, and divide it into training set, validation set and test set. The training set is used for parameter training of LSTM model, the validation set is used for hyperparameter tuning of LSTM model, and the test set is used to evaluate the prediction accuracy of LSTM model. Finally, select the LSTM model that meets the accuracy requirements and has the highest accuracy on the test set. Step 5: Determine the optimal control range of each LSTM model through spatial compression and reconstruction to obtain multiple LSTM models; Step 6: Parallel computation of multiple LSTM models to predict the spatiotemporal distribution of groundwater level at the grid scale in the irrigation area.

2. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Combine relevant data, including water system data, DEM data, land use data, soil data, and irrigation water intake data of the watershed where the irrigation district is located, to construct a SWAT model of the watershed where the irrigation district is located. Step 1.2: Based on the measured historical runoff data of the hydrological station in the watershed where the irrigation area is located, the SUFI-2 global search algorithm is used to complete the parameter calibration of the SWAT model in Step 1.1, so that the output of the SWAT model can fit the surface runoff process. Step 1.3: Organize the riverbed seepage, precipitation infiltration, and groundwater evaporation data output by the SWAT model in Step 1.2 for use in the coupling of the SWAT and MODFLOW models in the subsequent Step 3.

3. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 2, characterized in that, Step 2 specifically involves: combining borehole data, hydrogeological maps, and hydrogeological parameters to construct a MODFLOW model covering the irrigation area, which is used to output groundwater level simulation results at the grid scale of the irrigation area.

4. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 3, characterized in that, Step 3 specifically includes: Step 3.1: Input the riverbed seepage, precipitation infiltration and groundwater evaporation output by the SWAT model into the MODFLOW model to drive the simulation of the groundwater level in the irrigation area, and complete the coupling of the SWAT model and the MODFLOW model to obtain the SWAT-MODFLOW model. Step 3.2: Using the historical measured groundwater level of the observation well as a reference, adjust the hydrogeological parameters of each grid in the MODFLOW model of the SWAT-MODFLOW model so that the groundwater level of the irrigation area simulated by the SWAT-MODFLOW model fits the historical measured groundwater level process. Step 3.3, if the simulation accuracy index R of the groundwater level in step 3.2 is... 2 If the value is less than 0.99, the range of values ​​for parameters related to riverbed seepage, precipitation infiltration, and groundwater evaporation in the SWAT model constructed in step 1 is re-estimated, and the output of the SWAT model is transferred to the MODFLOW model constructed in step 2 again. This process is repeated until the surface runoff and groundwater level output by the SWAT-MODFLOW model can fit the historical measured data.

5. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 4, characterized in that, In step 3.3, the surface runoff and groundwater level output by the SWAT-MODFLOW model must be able to fit historical measured data. Specifically, the surface runoff simulation accuracy requirement R is: 2 ≥0.7, groundwater level simulation accuracy requirement R 2 ≥0.

99.

6. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 4, characterized in that, Step 4 specifically involves: Step 4.1: Combine historical meteorological data and future meteorological data of the watershed to construct a multi-class meteorological scenario dataset, which will be used as the input of the SWAT model in the SWAT-MODFLOW model in Step 3. Step 4.2: Combine historical groundwater extraction data of the irrigation area to construct a dataset of multiple groundwater extraction scenarios, which will be used as input to the MODFLOW model in the SWAT-MODFLOW model in Step 3. Step 4.3: Combine the multiple meteorological datasets constructed in Step 4.1 with the multiple groundwater extraction datasets constructed in Step 4.2 in pairs to construct a combined scenario that intersects multiple meteorological scenarios and groundwater extraction scenarios; Step 4.4: Using the various combined scenarios obtained in Step 4.3 as input to the SWAT-MODFLOW model in Step 3, the groundwater level response process of each grid in the irrigation area under various combined scenarios is simulated and determined through the SWAT-MODFLOW model. Step 4.5: Based on the time span, the groundwater level response process of each grid in the irrigation area under various combined scenarios determined in Step 4.4 is divided into training set, validation set and test set according to the proportion, so as to obtain the LSTM model with the highest accuracy on the test set that meets the accuracy requirements.

7. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 6, characterized in that, In step 4: In step 4.3, the pairwise combination processing method specifically refers to combining each type of meteorological scenario with each type of groundwater extraction scenario to form a groundwater extraction scenario. In step 4.5, the accuracy requirement means that the Nash efficiency coefficient NSE ≥ 0.

5.

8. The efficient method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas according to claim 6, characterized in that, Step 5 specifically involves: Step 5.1: Based on the spatial distribution of six features of each grid in the irrigation area, namely permeability coefficient, water supply, land use type, elevation, aquifer thickness and standard deviation of groundwater level fluctuation, the entire irrigation area is clustered into multiple partitions using the K-means clustering algorithm and the silhouette coefficient optimization method, and the grid where the centroid of each partition is located is used as the control grid of that partition. Step 5.2: Build and train an LSTM model for each control grid of each partition separately, which will be used to predict the groundwater level at the control grid under various combined scenarios in Step 4.

3. Step 5.3: By traversing through the grid, the parameters of the LSTM model corresponding to each control grid are extended to all grids in the irrigation area to determine the accuracy distribution of each LSTM model parameter on the groundwater level of all grids, and a preliminary multi-LSTM model accuracy evaluation matrix M is constructed. Step 5.4: Extract the multi-LSTM model accuracy evaluation matrix obtained in Step 5.

3. The maximum value of each row determines the optimal accuracy that can be achieved by multiple LSTM models jointly predicting the groundwater level of all grids in the irrigation area, and the spatial distribution of the optimal accuracy is determined according to the spatial location of each grid. Step 5.5: Add control grids in areas with weak accuracy and repeat steps 5.2 to 5.4 to complete the correction of the accuracy evaluation matrix of the multi-LSTM model; Step 5.6: Determine the optimal control range for each control grid using a precision clustering algorithm; specifically: If the matrix in step 5.5 The Walking in its first If the column reaches its maximum value, then the [number]th [column]... The grid was summarized in the first... The optimal control range of an LSTM model.

9. The efficient prediction method for the spatiotemporal distribution of groundwater level at a grid scale in large-scale agricultural irrigation areas according to claim 8, characterized in that, In step 5.3, the accuracy evaluation matrix of the multi-LSTM model As shown in formula (1): (1) Among them, the multi-model accuracy evaluation matrix line number of each line Represents a grid, with column labels for each column. Represents a control grid; the Okay, number The value of the column Representing the The grid is using the first The mean accuracy obtained on test sets for various scenarios when the parameters of a control grid model are set.

10. The efficient method for predicting the spatiotemporal distribution of groundwater levels at a grid scale in large-scale agricultural irrigation areas according to claim 8, characterized in that, Step 6 specifically involves: In the process of predicting the spatiotemporal distribution of groundwater level at the grid scale in the irrigation area, the LSTM model built and trained for each control grid is computed in parallel to obtain the groundwater level of all grids within the optimal control range of each LSTM model. Combined with the spatial location of each grid, the spatiotemporal distribution of groundwater level at the grid scale of the entire irrigation area is obtained.

Images