Hydrological physics constraint and integrated machine learning coupling driven groundwater storage anomaly downscaling method

Abstract

The present application relates to the technical field of geophysical remote sensing and hydrological information processing, and in particular to a groundwater storage anomaly downscaling method based on hydrological physical constraint and integrated machine learning coupling driving, comprising: step S1: obtaining multi-source gravity satellite Mascon grid product and hydrological model data, determining several water storage components of the region to be studied, inversely obtaining groundwater storage anomaly data, and preprocessing; step S2: based on the preprocessed groundwater storage anomaly data, selecting environmental variables, constructing a groundwater storage anomaly downscaling fusion model under a coarse spatial resolution, introducing hydrological physical constraints, and completing the training of the groundwater storage anomaly downscaling fusion model; step S3: using the trained groundwater storage anomaly downscaling fusion model to obtain corresponding groundwater storage anomaly prediction values for the high-resolution environmental variables fed back by the region to be studied. The present application effectively generates a groundwater storage anomaly distribution with high spatial details and strong physical rationality.

Description

Technical Field

[0001] This invention relates to the field of geophysical remote sensing and hydrological information processing technology, specifically to a method for downscaling groundwater storage anomalies driven by a coupling of hydrophysical constraints and integrated machine learning. Background Technology

[0002] Groundwater storage anomalies (GWSA) are important indicators of regional groundwater changes and play a significant role in water resource assessment and ecological monitoring. Although the GRACE (Gravity Recovery and Climate Experiment) / GRACE-FO (Gravity Recovery and Climate Experiment Follow-On) satellites provide effective means for large-scale monitoring, their spatial resolution is relatively low, and downscaling is usually required to obtain more refined information.

[0003] Existing downscaling methods mainly include dynamic downscaling and statistical downscaling. The former has a physical basis but requires high data and computing resources. The latter has advantages in efficiency and spatial representation, but it is mostly data-driven, lacks systematic consideration of the uncertainty of GRACE products, and is difficult to introduce physical constraints. In addition, the construction of target data often uses a single product or simple averaging, which fails to fully reflect the error differences between different data sources. Summary of the Invention

[0004] To address the shortcomings of existing technologies in uncertainty handling, physical constraints, and model stability, this invention aims to provide a groundwater storage anomaly downscaling method based on a coupling of hydrophysical constraints and integrated machine learning. The specific technical solution adopted is as follows:

[0005] Step S1: Obtain multi-source gravity satellite Mascon grid products and hydrological model data, determine several water storage components in the area to be studied, invert groundwater storage anomaly data, and perform preprocessing;

[0006] Step S2: Based on the preprocessed groundwater storage anomaly data, select environmental variables, construct a groundwater storage anomaly downscaling fusion model at coarse spatial resolution, introduce hydrophysical constraints, and complete the training of the groundwater storage anomaly downscaling fusion model.

[0007] Step S3: Use the trained groundwater storage anomaly downscaling fusion model to predict the high-resolution environmental variables fed back to the study area, and obtain the corresponding predicted values ​​of groundwater storage anomalies.

[0008] Preferably, step S1 includes:

[0009] Step S11: Obtain the terrestrial water storage anomaly values ​​of the area under study using the Mascon grid product from the multi-source gravity satellite;

[0010] Step S12: Obtain hydrological model data through the global land data assimilation system, and based on the hydrological model data, obtain the vegetation canopy water storage, soil water storage and snow water equivalent of the area to be studied, and determine the corresponding outliers;

[0011] Step S13: Combine the anomalies of terrestrial water storage and the anomalies corresponding to the water storage components to obtain groundwater storage anomaly data and fill in the missing values.

[0012] Preferably, step S12 specifically includes:

[0013] A long-term time series is established based on the current study period of the area under study. The average values ​​of vegetation canopy water storage, soil water storage and snow water equivalent are obtained through the long-term time series. Combined with the vegetation canopy water storage, soil water storage and snow water equivalent of the area under study, the corresponding outliers are obtained.

[0014] Preferably, step S2 includes:

[0015] Step S21: Based on the preprocessed groundwater storage anomaly data, select environmental variables for spatiotemporal matching and standardization preprocessing to establish a coarse spatial resolution feature set;

[0016] Step S22: Construct a downscaling fusion model of groundwater storage anomaly using a coarse spatial resolution feature set, introduce hydrophysical constraints, optimize the model parameters, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the downscaling fusion model of groundwater storage anomaly.

[0017] Preferably, step S22 includes:

[0018] Step S221: Based on the coarse spatial resolution feature set, construct a groundwater storage anomaly downscaling fusion model using a gradient boosting tree model, and output the fitting error;

[0019] Step S222: Introduce hydrophysical constraints through the principle of water balance, and establish a coupling loss function based on the fitting error;

[0020] Step S223: Optimize the model parameters by constraining them according to the coupling loss function, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the groundwater storage anomaly downscaling fusion model.

[0021] Preferably, step S221 includes:

[0022] Step S2211: Based on the coarse spatial resolution feature set, establish a gradient boosting tree model by sequentially integrating multiple decision trees;

[0023] Step S2212: Iteratively optimize the gradient boosting tree model, gradually merge decision trees, and continuously introduce new decision trees to correct the errors of the previous model until the training of the preset number of iterations is completed, and a downscaled fusion model of groundwater storage anomaly is obtained.

[0024] Step S2213: Output the predicted value through the downscaling fusion model of groundwater storage anomaly, and combine it with the observed value in the feature set to obtain the fitting error.

[0025] Preferably, step S222 includes:

[0026] Step S2221: Obtain precipitation, evapotranspiration and runoff in the study area during the corresponding study period. Based on the predicted groundwater storage anomaly data output by the groundwater storage anomaly downscaling fusion model during the study period and the observed groundwater storage anomaly data in the coarse spatial resolution feature set, obtain the change in groundwater storage anomaly data. Combine precipitation, evapotranspiration and runoff to construct hydrophysical constraints.

[0027] Step S2222: Weight and fuse the fitting error and hydrophysical constraints to establish a coupling loss function.

[0028] Preferably, step S3 specifically includes:

[0029] The high-resolution environmental variables of the area under study during the target period are input into the trained downscaling fusion model of groundwater storage anomaly for forward calculation, and the corresponding predicted values ​​of groundwater storage anomaly are output.

[0030] To address the aforementioned technical problems, the present invention also provides an electronic device comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus, and the processor calls logical instructions in the memory to execute the groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning described in any of the preceding claims.

[0031] The present invention has the following beneficial effects:

[0032] 1. In this method, firstly, multi-level hydrological data provided by relevant products and systems are comprehensively utilized to invert GWSA data for the area under study. Then, a downscaling fusion model under an integrated learning framework is constructed, i.e., a groundwater storage anomaly downscaling fusion model is built at coarse spatial resolution. Hydrophysical constraints are introduced and coupled into the model's loss function to achieve deep constraint and fusion training of the data-driven process by physical mechanisms. Finally, the trained groundwater storage anomaly downscaling fusion model outputs the corresponding predicted groundwater storage anomaly values. While significantly improving the spatial resolution of GWSA, the method ensures the hydrophysical consistency of the prediction results, outputting a GWSA spatiotemporal distribution product with high spatial resolution and conforming to natural laws. This addresses the challenge of traditional methods failing to guarantee physical consistency when improving resolution. Through a training mechanism driven by physical constraints, the model follows both data patterns and physical laws during optimization. In practical applications, this method effectively generates groundwater storage anomaly distributions with both high spatial detail and strong physical rationality, significantly improving the accuracy, reliability, and scientific value of monitoring results. It helps to reveal the spatiotemporal evolution mechanism of groundwater storage and its response characteristics to climate and human activities in greater detail.

[0033] 2. The electronic device provided by this invention has the same beneficial effects as the groundwater storage anomaly downscaling method based on hydrophysical constraints and integrated machine learning coupling provided by this invention, and will not be described in detail here. Attached Figure Description

[0034] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0035] Figure 1 The following is a flowchart illustrating the implementation of a groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning, as provided in one embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the verification results of an example of a groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning, provided in an embodiment of the present invention. Detailed Implementation

[0037] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning proposed in accordance with the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0039] The following describes in detail, with reference to the accompanying drawings, a specific scheme for a groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning provided by the present invention.

[0040] To better illustrate this point, groundwater, as a vital drinking water source, has a storage change that directly impacts water resource security. Groundwater storage and analysis (GWSA) is an important indicator of regional groundwater changes. By predicting GWSA, we can identify areas at risk of groundwater over-extraction. Furthermore, predicting GWSA helps assess the impact of groundwater changes on ecosystem services, revealing more precisely the spatiotemporal evolution mechanism of groundwater storage and its response characteristics to climate and human activities.

[0041] Please see Figure 1 The diagram illustrates a flowchart of a groundwater storage anomaly downscaling method based on a coupling of hydrophysical constraints and integrated machine learning, according to an embodiment of the present invention. The method includes:

[0042] Step S1: Obtain multi-source gravity satellite Mascon grid products and hydrological model data, determine several water storage components in the area to be studied, invert groundwater storage anomaly data, and perform preprocessing;

[0043] Step S2: Based on the preprocessed groundwater storage anomaly data, select environmental variables, construct a groundwater storage anomaly downscaling fusion model at coarse spatial resolution, introduce hydrophysical constraints, and complete the training of the groundwater storage anomaly downscaling fusion model.

[0044] Step S3: Use the trained groundwater storage anomaly downscaling fusion model to predict the high-resolution environmental variables fed back to the study area, and obtain the corresponding predicted values ​​of groundwater storage anomalies.

[0045] It is explained that, given the important value of GRACE / GRACE-FO groundwater products for regional water resource management and social sustainable development, this study on a groundwater storage anomaly downscaling method driven by a coupling of hydrophysical constraints and integrated machine learning has significant theoretical and practical implications for improving the spatial resolution of GRACE / GRACE-FO groundwater products.

[0046] As an alternative implementation method, the area to be studied usually refers to an area with a relatively independent groundwater system, representative hydrogeological conditions, and potential needs for water resource management or environmental change monitoring. It is generally an agricultural irrigation area, a densely populated area, an arid or semi-arid area, or a key monitoring area where excessive groundwater extraction has led to land subsidence or water quality deterioration.

[0047] Further, step S1 includes:

[0048] Step S11: Obtain the terrestrial water storage anomaly values ​​of the area to be studied using the Mascon grid product from the multi-source gravity satellite.

[0049] To clarify, in this embodiment, the multi-source gravity satellite Mascon grid product refers to the Mascon (mass concentration) products from three institutions: CSR (Center for Space Research, University of Texas at Austin), JPL (Jet Propulsion Laboratory), and GSFC (Goddard Space Flight Center, NASA).

[0050] Specifically, we obtained the terrestrial water storage anomaly data (TWSA) from the Mascon products of CSR, JPL, and GSFC, and determined the corresponding error variance, denoted as [missing information]. , and The inverse variance weighting method is used to adaptively determine the weights of each data source in the fusion process, which is used to quantify the uncertainty of each data source. The corresponding calculation formula is as follows:

[0051]

[0052] in, This indicates the weight of the data source corresponding to the land water storage anomaly data of Mascon products from CSR, JPL, and GSFC. Indicates the first Error variance of each data source; , Both represent data source indexes. , representing the data sources of feedback from Mascon products from CSR, JPL, and GSFC, respectively.

[0053] The formula for calculating the terrestrial water storage anomaly value of the study area using the adaptive weighted fusion method is as follows:

[0054]

[0055] in, This indicates anomalies in terrestrial water storage in the area under study; , , These represent anomalous terrestrial water storage data from Mascon products of CSR, JPL, and GSFC, respectively.

[0056] Step S12: Obtain hydrological model data through the global land data assimilation system, and based on the hydrological model data, obtain the vegetation canopy water storage, soil water storage and snow water equivalent of the area to be studied, and determine the corresponding outliers.

[0057] To clarify, the Global Land Data Assimilation System (GLDAS) is a comprehensive system that integrates multi-source observation data with numerical model simulation results, and dynamically updates and optimizes land hydrological state parameters through data assimilation technology. Its core objective is to provide hydrological model data, namely multi-level hydrological data formed by vegetation canopy water storage, soil water storage, and snow cover water equivalent.

[0058] It can be noted that canopy water storage (CWS) is usually characterized by vegetation water content or biomass water; soil water storage (SMS) is the water storage of soil profiles at different depths such as 0-10cm, 10-30cm, and 30-100cm. In this embodiment, it is the sum of soil moisture content in each layer within a depth of 0-200cm; snow water equivalent (SWE) is the amount of water melted per unit area of ​​snow.

[0059] Further, in step S12, specifically:

[0060] A long-term time series is established based on the current study period of the area under study. The average values ​​of vegetation canopy water storage, soil water storage and snow water equivalent are obtained through the long-term time series. Combined with the vegetation canopy water storage, soil water storage and snow water equivalent of the area under study, the corresponding outliers are obtained.

[0061] Specifically, a long-term time series refers to a sequence formed by continuously and systematically collecting and organizing observational data of a specific region or phenomenon over a relatively long time span, i.e., the study period. In this embodiment, the study period refers to 2002-2024, and the average values ​​of vegetation canopy water storage, soil water storage, and snow cover water equivalent are obtained, respectively denoted as... , and To identify outliers, the corresponding calculation formula is:

[0062]

[0063]

[0064]

[0065] in, , , These represent outliers corresponding to vegetation canopy water storage, soil water storage, and snow water equivalent, respectively. , , These represent the water storage capacity of the vegetation canopy, the soil water storage capacity, and the snow water equivalent, respectively.

[0066] Step S13: Combine the anomalies of terrestrial water storage and the anomalies corresponding to the water storage components to obtain groundwater storage anomaly data and fill in the missing values.

[0067] Specifically, based on the principle of water balance—that is, the change in water input minus water output equals the change in water storage—groundwater storage anomaly data, namely GWSA (Groundwater Storage Anomaly), is calculated. This data reflects the degree of deviation of groundwater storage from the baseline period during the current analysis period. The baseline period is typically the multi-year average or groundwater storage for a specific historical period. The corresponding calculation formula is:

[0068]

[0069] in, This indicates abnormal data regarding groundwater reserves.

[0070] Singular Spectrum Analysis (SSA) is used to impute missing values ​​in groundwater storage anomaly data. The core of this method is to decompose the original time series corresponding to the groundwater storage anomaly data into subsequences with different characteristics such as trend, periodic and noise terms, and to reconstruct these subsequences to achieve a reasonable estimate of the missing data.

[0071] Further, step S2 includes:

[0072] Step S21: Based on the preprocessed groundwater storage anomaly data, select environmental variables for spatiotemporal matching and standardized preprocessing to establish a coarse spatial resolution feature set.

[0073] In this embodiment, environmental variables refer to data such as NDVI (Normalized Difference Vegetation Index), slope, and soil moisture. NDVI is an important indicator reflecting vegetation cover and growth status obtained through remote sensing technology. Slope is a key parameter that directly affects surface runoff, soil erosion, and the layout of human activities. Soil moisture is an indicator that measures the water content in the soil.

[0074] Specifically, environmental variables are selected to perform spatiotemporal matching and standardization preprocessing on the preprocessed groundwater storage anomaly data, establishing a feature set characterizing the driving force of the groundwater system, i.e., establishing a coarse spatial resolution feature set, denoted as . High-precision spatiotemporal interpolation techniques, such as Kriging interpolation, are used to achieve spatiotemporal alignment between environmental variables and groundwater storage anomaly data, ensuring mutual matching between them. Z-score standardization is employed to dimensionlessly process each environmental variable, eliminating dimensional differences and magnitude effects to ensure data comparability. Finally, the processed data are integrated to form a feature set. .

[0075] Step S22: Construct a downscaling fusion model of groundwater storage anomaly using a coarse spatial resolution feature set, introduce hydrophysical constraints, optimize the model parameters, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the downscaling fusion model of groundwater storage anomaly.

[0076] Understandably, constructing an ensemble learning downscaling model involves building a groundwater storage anomaly downscaling fusion model based on a coarse spatial resolution feature set. Here, coarse spatial resolution represents the actual ground area represented by each pixel in the environmental variable data feedback. Furthermore, hydrophysical laws such as water balance are coupled into the model's loss function as constraint terms to achieve the fusion training of physical mechanisms and data-driven approaches, thereby improving the model's universality and reliability and providing high-resolution groundwater dynamic information.

[0077] Further, step S22 includes:

[0078] Step S221: Based on the coarse spatial resolution feature set, a gradient boosting tree model is used to construct a downscaling fusion model of groundwater storage anomalies, and the fitting error is output.

[0079] The groundwater storage anomaly downscaling fusion model, also known as the integrated machine learning fusion model, includes a multi-scale feature engineering module, a physical consistency mask module, a regression mapping module, and a global reconstruction and attribution module. It integrates the input layer, filter layer, core learning layer, and output layer to obtain the fusion model.

[0080] Further, step S221 includes:

[0081] Step S2211: Based on the coarse spatial resolution feature set, establish a gradient boosting tree model by sequentially integrating multiple decision trees.

[0082] Specifically, the coarse spatial resolution feature set Input is fed into a multi-scale feature engineering module to facilitate the extraction of information at different granularities and levels; a physically consistent mask module is used for coarse spatial resolution feature sets. Filtering is performed to exclude abnormal data that does not conform to the corresponding domain knowledge or physical laws, effectively reducing noise interference and improving the model's generalization ability; by sequentially integrating multiple decision trees, i.e., base learners, in this embodiment, a total of A decision tree is used to effectively alleviate the overfitting problem of a single decision tree; a gradient boosting tree model is established, and the corresponding calculation formula is:

[0083]

[0084] in, This represents a gradient boosting tree model; Represents the initial constant; Indicates the learning rate; Represents the decision tree index; Indicates the number of decision trees; Indicates the first A decision tree, i.e., a base learner; This represents the internal structure and parameters of the decision tree. .

[0085] Step S2212: Iteratively optimize the gradient boosting tree model, gradually merge decision trees, and continuously introduce new decision trees to correct the errors of the preceding model until the training of the preset number of iterations is completed, and a downscaling fusion model of groundwater storage anomaly is obtained.

[0086] Specifically, the number of iteration rounds is preset, i.e. In the iterative optimization process, each iteration corresponds one-to-one with the addition of a base learner. That is, in the regression mapping module, at the current iteration... In the round of iteration, add the first... Each base learner, based on the current model The negative gradient of the prediction error is calculated as the residual, denoted as . The residual As a learning objective, train a new base learner. Then the base learner By learning rate Weighted superposition to the current model In this process, model updates are implemented, and the corresponding calculation formula is:

[0087]

[0088] Through the above iterative process, new base learners are continuously introduced to correct the errors of the preceding model until the preset number of rounds is completed. Through training, a nonlinear mapping model with high-precision fitting ability for groundwater storage anomalies is finally formed, namely the groundwater storage anomaly downscaling fusion model, denoted as... .

[0089] Step S2213: Output the predicted value through the downscaling fusion model of groundwater storage anomaly, and combine it with the observed value in the feature set to obtain the fitting error.

[0090] Specifically, a downscaling fusion model of groundwater storage anomalies is used to simulate and predict groundwater storage anomalies in the study area, generating a high-resolution grid of predicted values. This grid, after global reconstruction and attribution, outputs predicted values ​​containing predicted groundwater storage anomaly data and predicted environmental variables. Then, the predicted values ​​are compared with the observed values ​​in the feature set at a region-by-region level to obtain the fitting error. The root mean square error (RMSE) or mean absolute error (MAE) is typically used to determine the fitting error, denoted as . .

[0091] Step S222: Introduce hydrophysical constraints through the principle of water balance, and establish a coupling loss function based on the fitting error.

[0092] The explanation is that, based on the principle of water balance, physical constraints on groundwater changes are constructed to ensure that the model prediction results satisfy the conservation relationship between precipitation, evapotranspiration, runoff and groundwater changes; that is, in the process of building the model, the law of conservation of mass is strictly followed to ensure that the dynamic correlation between various hydrological elements in the model simulation process conforms to natural laws.

[0093] Further, step S222 includes:

[0094] Step S2221: Obtain precipitation, evapotranspiration and runoff in the study area during the corresponding study period. Based on the predicted groundwater storage anomaly data output by the groundwater storage anomaly downscaling fusion model during the study period and the observed groundwater storage anomaly data in the coarse spatial resolution feature set, obtain the change in groundwater storage anomaly data. Combine precipitation, evapotranspiration and runoff to construct hydrophysical constraints.

[0095] Specifically, precipitation, evapotranspiration, and runoff, i.e., hydrological components, are analyzed. Precipitation and evapotranspiration data are obtained from publicly available datasets published on relevant platforms, such as the daily precipitation dataset for China provided by the National Climate Center of the China Meteorological Administration, or the MODIS evapotranspiration product from NASA. Runoff data is obtained from the aforementioned hydrological model data. Based on the study period, hydrological components and groundwater storage anomaly data are differentiated to form multiple data samples, the number of which is denoted as [missing information]. Then, by combining the observed groundwater storage anomaly data fed back at the time corresponding to each data sample in the feature set, the change in groundwater storage anomaly data is obtained, denoted as . , This represents the data sample index; the deviation between each hydrological component is defined as a penalty term, establishing hydrophysical constraints, and the corresponding calculation formula is:

[0096]

[0097] in, Indicates hydrophysical constraints; Indicates the data sample index; Indicates the number of data samples; , , , They represent the first The changes in anomalies in precipitation, evapotranspiration, runoff, and groundwater storage corresponding to each data sample.

[0098] Step S2222: Weight and fuse the fitting error and hydrophysical constraints to establish a coupling loss function.

[0099] Specifically, the coupling loss function is established, and the corresponding calculation formula is as follows:

[0100]

[0101] in, Represents the coupling loss function; Indicates the time during the training process The changing weighting coefficients.

[0102] It can be explained that the weighting coefficients It can adaptively adjust based on the training status or error changes of the downscaling fusion model of groundwater storage anomalies, so as to dynamically balance the relationship between data fitting accuracy and physical consistency.

[0103] Step S223: Optimize the model parameters by constraining them according to the coupling loss function, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the groundwater storage anomaly downscaling fusion model.

[0104] It can be explained that the coupling loss function is used. The model parameters of the groundwater storage anomaly downscaling fusion model are constrained and optimized. During the training process, the influence of data error and physical constraints are considered simultaneously, so that the two losses jointly dominate the training process. This effectively avoids the problem of the model overfitting local data noise or violating physical laws, and allows the model to gradually converge to the optimal state of prediction accuracy and physical consistency. This enables the model to learn the nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, realize the fusion and downscaling prediction of groundwater storage anomaly information from macro scale to micro scale, and complete the training of the groundwater storage anomaly downscaling fusion model.

[0105] Furthermore, in step S3, specifically:

[0106] The high-resolution environmental variables of the area under study during the target period are input into the trained downscaling fusion model of groundwater storage anomaly for forward calculation, and the corresponding predicted values ​​of groundwater storage anomaly are output.

[0107] Specifically, the target time period, i.e. the period studied according to the current task requirements, involves acquiring environmental variables of the area under study during that time period, i.e., standardized high spatial resolution multi-source environmental factor data, which are then input into a groundwater storage anomaly downscaling fusion model with fixed model parameters after training. , Represents high-resolution environmental variables of the area under study during the target time period; This represents the optimal parameters obtained through physical constraint coupling-driven optimization, which are then used for forward calculation to directly output the corresponding groundwater storage anomaly prediction value. Among them, high resolution and coarse spatial resolution are a set of relative descriptions. In this embodiment, high resolution means that each pixel in the environmental variable represents a smaller actual surface area, which provides rich detailed information, while coarse spatial resolution has a wider coverage and lower data storage and processing costs.

[0108] Understandably, in this method, firstly, multi-level hydrological data provided by relevant products and systems are comprehensively utilized to invert and obtain GWSA data for the area under study; then, a downscaling fusion model under an ensemble learning framework is constructed, that is, a downscaling fusion model of groundwater storage anomalies is built at coarse spatial resolution; hydrophysical constraints are introduced and coupled into the loss function of the model to achieve deep constraint and fusion training of the data-driven process by the physical mechanism; finally, the trained downscaling fusion model of groundwater storage anomalies outputs the corresponding predicted values ​​of groundwater storage anomalies; while significantly improving the spatial resolution of GWSA, it ensures... To ensure the hydrophysical consistency of the predicted results, this method outputs a GWSA spatiotemporal distribution product with high spatial resolution and conformity to natural laws. Specifically, considering the difficulty of ensuring physical consistency when improving resolution using traditional methods, a training mechanism driven by physical constraint coupling is adopted. This allows the model to follow both data patterns and physical laws during the optimization process. In practical applications, this method effectively generates groundwater storage anomaly distributions with both high spatial detail and strong physical rationality, significantly improving the accuracy, reliability, and scientific value of monitoring results. It helps to reveal the spatiotemporal evolution mechanism of groundwater storage and its response characteristics to climate and human activities in greater detail.

[0109] Please see Figure 2 To better illustrate this point and to verify the reliability of the proposed method for spatial downscaling, during the verification process, the North China Plain Mascon products provided by CSR, GSFC, and JPL from April 2002 to December 2024, and the hydrological model products provided by GLDAS were used to invert and obtain the groundwater storage GWSA data for the North China Plain. The relevant implementation steps of this invention were then used to process the data sequentially to obtain the downscaling results.

[0110] It can be observed that the downscaling results improved the spatial resolution from 1° to 0.25°. Spatially, the downscaled GWSA exhibits a gradient variation characteristic consistent with the hydrogeological conditions of the North China Plain. The extremely water-deficient areas, which were previously smoothed in the coarse grid, show clearer spatial boundaries and core locations in the downscaled results. By introducing hydrophysical constraints, the downscaling model—the groundwater storage anomaly downscaling fusion model—refines the spatial structure while ensuring the physical rationality of the output results in terms of temporal evolution and spatial water distribution, providing a reliable data foundation for subsequent fine-scale groundwater dynamic monitoring and resource management.

[0111] In summary, a groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning has good effectiveness and reliability. It can effectively improve the spatial resolution of data and reveal potential local information in the study area, and can serve as a feasible method for groundwater storage downscaling.

[0112] The second embodiment of the present invention provides an electronic device, which includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The processor calls logical instructions in the memory to execute the groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning as described in any embodiment of the present invention.

[0113] When it is in operation, it needs to use a groundwater storage anomaly downscaling method based on hydrophysical constraints and integrated machine learning coupling. Therefore, whether the equipment and program data are integrated or different hardware is configured to produce functions with similar effects to those achieved by this invention, they all fall within the protection scope of this invention. This equipment has the same beneficial effects as the aforementioned groundwater storage anomaly downscaling method based on hydrophysical constraints and integrated machine learning coupling, and will not be elaborated here.

[0114] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for downscaling groundwater storage anomalies driven by a coupling of hydrophysical constraints and ensemble machine learning, characterized in that, The method includes: Step S1: Obtain multi-source gravity satellite Mascon grid products and hydrological model data, determine several water storage components in the area to be studied, invert groundwater storage anomaly data, and perform preprocessing; Step S2: Based on the preprocessed groundwater storage anomaly data, select environmental variables, construct a groundwater storage anomaly downscaling fusion model at coarse spatial resolution, introduce hydrophysical constraints, and complete the training of the groundwater storage anomaly downscaling fusion model. Step S3: Use the trained groundwater storage anomaly downscaling fusion model to predict the high-resolution environmental variables fed back to the study area, and obtain the corresponding predicted values ​​of groundwater storage anomalies.

2. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 1, characterized in that, Step S1 includes: Step S11: Obtain the terrestrial water storage anomaly values ​​of the area under study using the Mascon grid product from the multi-source gravity satellite; Step S12: Obtain hydrological model data through the global land data assimilation system, and based on the hydrological model data, obtain the vegetation canopy water storage, soil water storage and snow water equivalent of the area to be studied, and determine the corresponding outliers; Step S13: Combine the anomalies of terrestrial water storage and the anomalies corresponding to the water storage components to obtain groundwater storage anomaly data and fill in the missing values.

3. The groundwater storage anomaly downscaling method based on hydrophysical constraints and integrated machine learning driven by coupling, as described in claim 2, is characterized in that... In step S12, specifically: A long-term time series is established based on the current study period of the area under study. The average values ​​of vegetation canopy water storage, soil water storage and snow water equivalent are obtained through the long-term time series. Combined with the vegetation canopy water storage, soil water storage and snow water equivalent of the area under study, the corresponding outliers are obtained.

4. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 1, characterized in that, Step S2 includes: Step S21: Based on the preprocessed groundwater storage anomaly data, select environmental variables for spatiotemporal matching and standardization preprocessing to establish a coarse spatial resolution feature set; Step S22: Construct a downscaling fusion model of groundwater storage anomaly using a coarse spatial resolution feature set, introduce hydrophysical constraints, optimize the model parameters, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the downscaling fusion model of groundwater storage anomaly.

5. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 4, characterized in that, Step S22 includes: Step S221: Based on the coarse spatial resolution feature set, construct a groundwater storage anomaly downscaling fusion model using a gradient boosting tree model, and output the fitting error; Step S222: Introduce hydrophysical constraints through the principle of water balance, and establish a coupling loss function based on the fitting error; Step S223: Optimize the model parameters by constraining them according to the coupling loss function, establish a nonlinear mapping relationship between environmental variables and groundwater storage anomaly data, and complete the training of the groundwater storage anomaly downscaling fusion model.

6. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 5, characterized in that, Step S221 includes: Step S2211: Based on the coarse spatial resolution feature set, establish a gradient boosting tree model by sequentially integrating multiple decision trees; Step S2212: Iteratively optimize the gradient boosting tree model, gradually merge decision trees, and continuously introduce new decision trees to correct the errors of the previous model until the training of the preset number of iterations is completed, and a downscaled fusion model of groundwater storage anomaly is obtained. Step S2213: Output the predicted value through the downscaling fusion model of groundwater storage anomaly, and combine it with the observed value in the feature set to obtain the fitting error.

7. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 5, characterized in that, Step S222 includes: Step S2221: Obtain precipitation, evapotranspiration and runoff in the study area during the corresponding study period. Based on the predicted groundwater storage anomaly data output by the groundwater storage anomaly downscaling fusion model during the study period and the observed groundwater storage anomaly data in the coarse spatial resolution feature set, obtain the change in groundwater storage anomaly data. Combine precipitation, evapotranspiration and runoff to construct hydrophysical constraints. Step S2222: Weight and fuse the fitting error and hydrophysical constraints to establish a coupling loss function.

8. The method for downscaling groundwater storage anomalies based on the coupling of hydrophysical constraints and ensemble machine learning as described in claim 1, characterized in that, In step S3, specifically: The high-resolution environmental variables of the area under study during the target period are input into the trained downscaling fusion model of groundwater storage anomaly for forward calculation, and the corresponding predicted values ​​of groundwater storage anomaly are output.

9. An electronic device, characterized in that, The device includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The processor calls logical instructions in the memory to execute the groundwater storage anomaly downscaling method based on the coupling of hydrophysical constraints and integrated machine learning as described in any one of claims 1 to 8.

Images