Fine-resolution GRACE underground water reserve monitoring method and system
By combining the GRACE dataset with FLDAS and XGBoost models, resolution downscaling was performed, which solved the accuracy problem of groundwater storage monitoring at the local scale, and achieved fine-resolution groundwater storage monitoring, identifying drought types and detecting groundwater depletion.
Patent Information
- Application Number
- CN202511004553.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-21
- Publication Date
- 2025-10-28
AI Technical Summary
Existing technologies make it difficult to accurately monitor changes in groundwater storage at the local scale, especially when data resolution is low, which complicates the identification of drought types.
The GRACE dataset is combined with the FLDAS model and the XGBoost model. By building a model containing multiple climate variables, resolution downscaling is performed to calculate groundwater storage anomalies and improve monitoring accuracy.
It has achieved fine-grained monitoring of groundwater reserves at a resolution of 0.1°, effectively identified drought types, enhanced distributed estimation of groundwater reserve changes, and improved the ability to detect droughts and pinpoint severe groundwater depletion.
Smart Images
Figure CN120850106A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of water storage detection technology, and in particular relates to a high-resolution GRACE groundwater storage monitoring method and system. Background Technology
[0002] Groundwater aquifers, as the largest available freshwater resource, play a crucial role in meeting basic water needs. Drought is a severe natural disaster that negatively impacts water (resource) and groundwater supply, agriculture and livestock, the economy, and the sustainability of human and wildlife communities. Drought is defined as prolonged water scarcity and is generally classified into four types: meteorological drought (insufficient precipitation), hydrological drought (insufficient runoff), agricultural drought (insufficient soil moisture), and groundwater drought. The ambiguity in the definition and characteristics of drought, as well as the challenges in assessing its precursors, has led to the formation and use of various drought indices. The development of drought from the meteorological stage to the hydrological and agricultural stages affects groundwater reserves, and continuous water extraction from aquifers during drought periods disrupts the hydrological balance, leading to groundwater drought.
[0003] Accurate identification of groundwater drought is crucial, especially in water-scarce regions where regional security is at stake. The intensity and duration of droughts are expected to increase due to the worsening effects of climate change. Therefore, monitoring and detecting drought dynamics is essential. However, the hidden nature of groundwater reserves limits available information about their fluctuations and dynamics, complicating the accurate identification of groundwater drought, which can ultimately be affected by climate, agricultural, and hydrological droughts. Accurately assessing groundwater reserves and their dynamics remains more challenging than assessing surface hydrological components because groundwater reserves cannot be directly observed. Therefore, there is an urgent need to provide a high-resolution GRACE groundwater reserve monitoring method and system to address the aforementioned technical challenges. Summary of the Invention
[0004] In view of this, the present invention provides a high-resolution GRACE groundwater storage monitoring method and system, which improves the accuracy of GRACE groundwater storage detection, effectively identifies drought types, and enhances the distributed estimation of groundwater storage changes using high-resolution FLDAS data. The specific technical solution adopted is as follows.
[0005] In a first aspect, the present invention provides a high-resolution GRACE groundwater storage monitoring method, comprising the following steps:
[0006] Obtain the GRACE dataset for the study area and estimate the initial land water storage anomalies based on the GRACE dataset;
[0007] Obtain the field observation dataset of the study area, and construct an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff.
[0008] An XGBoost model is constructed based on the initial land water storage anomaly and the multiple climate variables, and the predicted land water storage anomaly is determined by resolution downscaling based on the XGBoost model.
[0009] Calculate the residual between the predicted land water storage anomaly and the initial land water storage anomaly, and determine the groundwater storage anomaly in the study area based on the residual and the predicted land water storage anomaly.
[0010] As a preferred embodiment of the above technical solution, the GRACE dataset of the study area is obtained, and the initial land water storage anomaly is estimated based on the GRACE dataset, including:
[0011] Obtain the watershed surface model CLSM of the global hydrological model GLDAS. The spatial resolution of the watershed surface model CLSM is 0.25°×0.25°. The watershed surface model CLSM is a sub-model of the global hydrological model GLDAS.
[0012] The watershed surface model CLSM uses numerical simulations of terrestrial water storage and energy processes to derive hydrological drought indices to assess the wet and dry conditions of soil and groundwater. The hydrological drought indices are expressed as percentiles representing the probability of drought occurrence based on wet or dry conditions, which correspond to the groundwater storage percentage (GWSP).
[0013] As a preferred embodiment of the above technical solution, an in-situ observation dataset of the study area is obtained, and an FLDAS model containing multiple climate variables is constructed based on the GRACE dataset and the in-situ observation dataset, including:
[0014] Multiple well data were acquired in the study area. The groundwater depth (DTG) values obtained from these well data were converted into groundwater GWL using topographic elevations related to mean sea level. The corresponding expression is as follows:
[0015] GWL = H - DTG (1)
[0016] Using average water production S y The seasonal variation of groundwater storage is calculated by multiplying the groundwater level anomaly value (GWIA) and the groundwater level anomaly value (GWSA). The GWSA of each well is calculated using formula (2).
[0017] GWSA=S y ×GWLA (2)
[0018] GWSA indicates anomaly in groundwater reserves, S y GWIA represents average water production and groundwater level anomaly.
[0019] As a preferred embodiment of the above technical solution, an XGBoost model is constructed based on the initial land water storage anomaly and the multiple climate variables, and the predicted land water storage anomaly is determined by resolution downscaling based on the XGBoost model, including:
[0020] The XGBoost model constructs multiple interconnected decision trees, where the latest residual of one tree guides the development of the next tree. The sample prediction value of the XGBoost model is determined by the sum of the leaf node values on different trees.
[0021] The VIMP test, a measure of variable importance, is used to assess the predictive significance of each independent variable. A higher VIMP value indicates a stronger predictive power for the independent variable.
[0022] As a preferred embodiment of the above technical solution, the XGBoost model is used to reduce the GRACETWSA value from 0.5° resolution to 0.1° resolution, including:
[0023] An XGBoost model was constructed, and TWSA at a resolution of 0.5° was predicted using initial TWSA data and multiple climate variables. TWSA represents the anomaly value of land water storage.
[0024] At a resolution of 0.5°, the residual value is calculated by subtracting the initial TWSA value from the predicted TWSA value;
[0025] Input data at 0.1° resolution was integrated into the XGBoost model to predict TWSA values at 0.1° resolution;
[0026] The residual value is added to the predicted TWSA value at 0.1° resolution to obtain the TWSA value at 0.1° resolution, wherein residual correction is used to correct the prediction bias caused by omission;
[0027] The XGBoost model was used to subtract the outliers of runoff and soil water storage from the TWSA values to obtain the downscaled GWSA values, where the GWSA values are groundwater storage outliers.
[0028] As a preferred embodiment of the above technical solution, the formula (3) corresponding to the correlation between runoff anomaly QsA, soil water storage anomaly SMSA, and groundwater storage anomaly GWSA and the study area is determined according to TWSA. Among them, the formula (4) corresponding to subtracting the SMSA and QsA values from TWSA and subtracting the SMSAzhi1 and QsA values after downscaling by the XGBoost model from TWSA is used. The SMSA and QsA components are from the Noah model.
[0029] TWSA=GWSA+SMSA+QsA (3)
[0030] GWSA=TWSA-[SMSA+QsA] (4).
[0031] As a preferred embodiment of the above technical solution, the calculation process of the GRACE Groundwater Drought Index (GGDI) includes:
[0032] The climatological values for each month are obtained using formula (5):
[0033]
[0034] Where Ci represents the long-term monthly average, i = 1, 2, ... 12; parameter n represents the duration of the value from 2003 to 2020, in months, n = 1, 2, ... 18; and GWSA represents the groundwater storage in the i-th month.
[0035] The groundwater storage deviation (GSD) is calculated by subtracting the monthly cumulative value from the GWSA. The GSD serves as an indicator of the net seasonal fluctuation of the GWSA and is standardized using the mean and standard deviation of the GSD.
[0036]
[0037] Where GGDI represents the standardized net change in GWSA, t is the duration in months, and t ranges from 1 to 216; variables This represents the overall average value of GSD; It represents the overall standard deviation of GSD and serves as an indicator of drought conditions.
[0038] As a preferred embodiment of the above technical solution, four statistics are used to evaluate the performance of the XGBoost model. These four statistics include correlation (R), percentage of bias (PBIAS), root mean square error (RMSE), and Nash-Sutcliffe efficiency (NSE), with the corresponding expressions as follows:
[0039]
[0040] Among them, X i and Yi These are the observed value and the predicted value. and These are the values that reflect the average of the observed and predicted values, respectively, and the symbol N represents the total number of samples collected.
[0041] As a preferred embodiment of the above technical solution, the GRACE dataset is processed using spherical harmonic functions and Mascons functions. The TWSA data is derived from the GRACEJPLMasconsRL-06 dataset, with a resolution of 0.5°×0.5°.
[0042] Secondly, the present invention also provides a high-resolution GRACE groundwater storage monitoring system, applied to the aforementioned high-resolution GRACE groundwater storage monitoring method, comprising:
[0043] Anomaly estimation module is used to acquire the GRACE dataset of the study area and estimate the initial land water storage anomaly value based on the GRACE dataset;
[0044] The model building module is used to acquire the field observation dataset of the study area and build an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff.
[0045] The data processing module is used to construct an XGBoost model based on the initial land water storage anomaly and the multiple climate variables, and to determine the predicted land water storage anomaly by downscaling the XGBoost model.
[0046] An anomaly determination module is used to calculate the residual value between the predicted land water storage anomaly value and the initial land water storage anomaly value, and to determine the groundwater storage anomaly value of the study area based on the residual value and the predicted land water storage anomaly value.
[0047] This invention provides a high-resolution GRACE groundwater storage monitoring method and system. It acquires a GRACE dataset of the study area and estimates initial land water storage anomalies based on the dataset. It also acquires a field observation dataset of the study area and constructs an FLDAS model containing multiple climate variables based on the GRACE and field observation datasets. Furthermore, it constructs an XGBoost model based on the initial land water storage anomalies and multiple climate variables, and uses resolution downscaling based on the XGBoost model to determine predicted land water storage anomalies. The residuals between the predicted and initial land water storage anomalies are calculated, and groundwater storage anomalies in the study area are determined based on these residuals and the predicted land water storage anomalies. This improves the accuracy of GRACE groundwater storage detection, effectively identifies drought types, and the high-resolution FLDAS data enhances the distributed estimation of groundwater storage changes. The downscaled GRACE dataset is highly advantageous for detecting drought and identifying severe groundwater depletion. Attached Figure Description
[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.
[0049] Figure 1 A flowchart of the high-resolution GRACE groundwater storage monitoring method provided by the present invention;
[0050] Figure 2 This invention provides a schematic diagram illustrating the temporal correlation between the GGDI and GWSP drought indices.
[0051] Figure 3 The structural block diagram of the GRACE groundwater storage monitoring system with high resolution provided by the present invention. Detailed Implementation
[0052] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0053] The launch of the state-of-the-art Gravity Recovery and Climate Experiment (GRACE) satellite mission in 2002 marked a significant advancement in assessing regional water storage changes. Groundwater storage data estimated using GRACE can be used to analyze drought impacts, overcoming the limitations of traditional field monitoring methods. To date, two main approaches have been adopted: the first involves using GRACE observational data in conjunction with supplemental data to distinguish groundwater storage changes from other components of the water cycle to assess drought characteristics; the second approach involves researchers developing various GRACE-based drought indices for drought analysis, including the Total Storage Deficit Index (TSDI), Water Storage Deficit (WSD), the GRACE-based Hydrological Drought Index (GHDI), the GRACE Drought Severity Index (GDSI), the Water Storage Deficit Index (WSDI), the Modified Total Storage Deficit Index (MTSDI), the Combined Climate Bias Index (CCDI), and the Enhanced Water Storage Deficit Index (EWSDI). While these methods have proven effective for large-scale drought dynamics surveys, their applicability for assessing drought at local scales remains challenging, primarily due to low data resolution.
[0054] See Figure 1 This invention provides a high-resolution GRACE groundwater storage monitoring method, comprising the following steps:
[0055] S1: Obtain the GRACE dataset for the study area and estimate the initial land water storage anomaly based on the GRACE dataset;
[0056] S2: Obtain the field observation dataset of the study area, and construct an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff.
[0057] S3: Construct an XGBoost model based on the initial land water storage anomaly and the multiple climate variables, and determine the predicted land water storage anomaly by downscaling the XGBoost model using resolution.
[0058] S4: Calculate the residual between the predicted land water storage anomaly and the initial land water storage anomaly, and determine the groundwater storage anomaly in the study area based on the residual and the predicted land water storage anomaly.
[0059] In this embodiment, the study area is predominantly arid and highly susceptible to drought, which significantly impacts its agricultural economy. Agricultural land accounts for a large portion of the country's total land use, making drought monitoring crucial for the study area, particularly for local-scale analysis. Downscaled GRACE observations were used to monitor local-scale changes in water and groundwater storage. However, studies using high-resolution indices for mesoscale drought monitoring are limited. This study enhances the distributed estimation of groundwater storage changes using high-resolution data from the Famine Warning System Network Land Data Assimilation System (FLDAS). By applying machine learning methods to downscaled GRACE observations, the feasibility of developing and evaluating the GGDI index for assessing local-scale groundwater characteristics and groundwater drought dynamics at a finer resolution (0.1°) is demonstrated. The GRACE dataset was processed using spherical harmonic functions and Mascons functions. The TWSA data was derived from the GRACE JPL Mascons RL-06 dataset with a resolution of 0.5° × 0.5°. The TWSA estimates are derived from the GRACE record and use two basic functions: the spherical harmonic function (SH) and the Mascons function. The Mascons function can be utilized at smaller (smaller scales) levels through rich processing techniques and higher spatial resolution.
[0060] It should be noted that FLDAS is a global data assimilation system that integrates field-based data, remote sensing observations, and modeling outputs to simulate multiple parameters. The VIC model (Variable Permeability Model) and the Noah model provide the basis for FLDAS products with spatial resolutions of 0.25° and 0.1°, respectively. The Noah model has gained significant recognition in the hydrological community due to its reliable predictions of water and energy fluxes. The FLDAS-Noah model was used to calculate variables such as soil evapotranspiration (ET), soil water storage (SMS), precipitation (P), temperature (T), and runoff (Qs). The GRACE dataset for the study area was acquired, and initial land storage anomalies were estimated based on the GRACE dataset. This included: acquiring the watershed surface model CLSM of the global hydrological model GLDAS, with a spatial resolution of 0.25°×0.25°. The watershed surface model CLSM is a sub-model of the global hydrological model GLDAS. The watershed surface model CLSM uses numerical simulations of land storage and energy processes to derive hydrological drought indices to assess the wet and dry conditions of soil and groundwater. The hydrological drought indices are expressed as percentiles representing the probability of drought occurrence based on wet or dry conditions, corresponding to the groundwater storage percentage (GWSP).
[0061] It should be understood that by acquiring the GRACE dataset of the study area and estimating the initial land storage anomaly based on the GRACE dataset, acquiring the field observation dataset of the study area, and constructing an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset, an XGBoost model is constructed based on the initial land storage anomaly and multiple climate variables, and the predicted land storage anomaly is determined by downscaling the XGBoost model using resolution. The residual values between the predicted land storage anomaly and the initial land storage anomaly are calculated, and the groundwater storage anomaly values of the study area are determined based on the residual values and the predicted land storage anomaly values. This improves the accuracy of groundwater storage detection by GRACE and effectively identifies drought types. The high-resolution data of FLDAS enhances the distributed estimation of groundwater storage changes, and the downscaled GWSA dataset is very useful in detecting drought and identifying severe groundwater depletion.
[0062] Optionally, an in-situ observation dataset of the study area is obtained, and an FLDAS model containing multiple climate variables is constructed based on the GRACE dataset and the in-situ observation dataset, including:
[0063] Multiple well data were acquired in the study area. The groundwater depth (DTG) values obtained from these well data were converted into groundwater GWL using topographic elevations related to mean sea level. The corresponding expression is as follows:
[0064] GWL = H - DTG (1)
[0065] Using average water production S y The seasonal variation of groundwater storage is calculated by multiplying the groundwater level anomaly value (GWIA) and the groundwater level anomaly value (GWSA). The GWSA of each well is calculated using formula (2).
[0066] GWSA=S y ×GWLA (2)
[0067] GWSA indicates anomaly in groundwater reserves, S y GWIA represents average water production and groundwater level anomaly.
[0068] In this embodiment, data from 3368 wells (representing June and October, i.e., the period before and after the monsoon) were used in the in-situ groundwater level observation dataset for the study area. Existing outliers, irregular fluctuations, and discontinuities in groundwater level observations were excluded from the main dataset; therefore, 398 wells with records from June 2003 to June 2014 were selected for validation. The groundwater depth (DTG) values obtained from these wells were initially converted to the corresponding baseline with a groundwater level range of 0.01 to 0.40 using topographic elevations associated with mean sea level (H), as used in the GRACE mission. The seasonal variation of groundwater storage was calculated using the product of area-average water production (Sy) and GWLA, ranging from 0.01 to 0.40. GWLA was determined using an average Sy value of 0.14.
[0069] Optionally, an XGBoost model is constructed based on the initial land water storage anomaly and the multiple climate variables, and the predicted land water storage anomaly is determined by resolution downscaling based on the XGBoost model, including:
[0070] The XGBoost model constructs multiple interconnected decision trees, where the latest residual of one tree guides the development of the next tree. The sample prediction value of the XGBoost model is determined by the sum of the leaf node values on different trees.
[0071] The VIMP test, a measure of variable importance, is used to assess the predictive significance of each independent variable. A higher VIMP value indicates a stronger predictive power for the independent variable.
[0072] In this embodiment, the XGBoost model is used to reduce the GRACE TWSA value from 0.5° resolution to 0.1° resolution. This includes: constructing an XGBoost model; using initial TWSA data and multiple climate variables to predict the TWSA at 0.5° resolution, where TWSA represents anomalies in land water storage; calculating residual values at 0.5° resolution by subtracting the initial TWSA value from the predicted TWSA value; integrating the 0.1° resolution input data into the XGBoost model to predict the TWSA value at 0.1° resolution; adding the residual values to the predicted TWSA value at 0.1° resolution to obtain the TWSA value at 0.1° resolution, where residual correction is used to correct prediction bias caused by omissions; and using the XGBoost model to subtract the anomalies in runoff and soil water storage from the TWSA value to obtain the downscaled GWSA value, where GWSA represents anomalies in groundwater storage. The XGBoost model is built based on the gradient boosting tree algorithm. As the gradient increases, other models are incorporated into the overall prediction to correct residuals or errors generated by previous models. XGBoost constructs multiple interconnected decision trees, where the latest residual or error in one tree guides the development of the next tree.
[0073] It should be noted that the predicted values of the samples are determined by the sum of the leaf values from different trees. Both Random Forest and XGBoost can effectively manage multicollinearity and complex nonlinear interactions without sacrificing accuracy. The Variable Importance Metric Prediction (VIMP) test is used to evaluate the predictive significance of each independent variable, where a higher VIMP value indicates a stronger independent variable. Using the XGBoost model, the GRACETWSA value is reduced from 0.5° to 0.1° resolution through the following specific steps: 1) Construct an XGBoost model using the original (initial) TWSA data and other input data to predict TWSA at 0.5° resolution; 2) Calculate the residual value at 0.5° resolution by subtracting the original TWSA value from the predicted TWSA value; 3) Integrate the high-resolution (0.1°) input data into the trained model to predict TWSA at 0.1° resolution; 4) Obtain the fine-resolution (0.1°) TWSA by incorporating the interpolation residuals from the 0.5° data (step 2) into the existing predicted TWSA value at 0.1°. This method uses residual correction to correct the prediction bias caused by ignoring factors, thereby maintaining consistency with the original 0.1° dataset; 5) Finally, the outliers of runoff and soil water storage are subtracted from the TWSA values using the XGBoost model to obtain the reduced GWSA values.
[0074] Optionally, the formula (3) corresponding to the correlation between runoff anomaly QsA, soil water storage anomaly SMSA, and groundwater storage anomaly GWSA and the study area is determined based on TWSA. In this formula (4), the SMSA and QsA values are subtracted from TWSA, and the SMSAzhi1 and QsA values after downscaling by the XGBoost model are subtracted from TWSA. The SMSA and QsA components are derived from the Noah model.
[0075] TWSA=GWSA+SMSA+QsA (3)
[0076] GWSA=TWSA-[SMSA+QsA] (4).
[0077] In this embodiment, the calculation process of the GRACE Groundwater Drought Index (GGDI) includes:
[0078] The climatological values for each month are obtained using formula (5):
[0079]
[0080] Where Ci represents the long-term monthly average, i = 1, 2, ... 12; parameter n represents the duration of the value from 2003 to 2020, in months, n = 1, 2, ... 18; and GWSA represents the groundwater storage in the i-th month.
[0081] The groundwater storage deviation (GSD) is calculated by subtracting the monthly cumulative value from the GWSA. The GSD serves as an indicator of the net seasonal fluctuation of the GWSA and is standardized using the mean and standard deviation of the GSD.
[0082]
[0083] Where GGDI represents the standardized net change in GWSA, t is the duration in months, and t ranges from 1 to 216; variables This represents the overall average value of GSD; It represents the overall standard deviation of GSD and serves as an indicator of drought conditions.
[0084] It should be noted that four statistics are used to evaluate the performance of the XGBoost model. These four statistics include correlation (R), percentage of bias (PBIAS), root mean square error (RMSE), and Nash-Sutcliffe efficiency (NSE), and their corresponding expressions are as follows:
[0085]
[0086] Among them, X i and Y i These are the observed value and the predicted value. and These represent the average values of the observed and predicted values, respectively, with N indicating the total number of samples collected. The downscaled TWSA generated by XGBoost and the TWSA from GRACE reflect a parallel decreasing trend across the entire study area, exhibiting almost identical variability. The temporal variations of GRACE TWSA (0.5°) and downscaled TWSA (0.1°) show a significant correlation of 0.99 between the two time series. The monthly volatility of the GRACE TWSA data is 9.78 mm / month, while the monthly volatility of the downscaled TWSA data is 11.05 mm / month. The downscaling accuracy was evaluated by comparing the TWSA time series obtained from the GRACE model and the XGBoost model over time.
[0087] Specifically, see Figure 2 The downscaled TWSA generated by XGBoost and the TWSA from GRACE reflect a parallel decreasing trend across the entire study area, exhibiting almost identical variability. Comparing the temporal variations of GRACE TWSA (0.5°) and the downscaled TWSA (0.1°), the significant correlation between the two time series is 0.99. The monthly volatility of GRACE TWSA data is 9.78 mm / month, while the monthly variation of the downscaled TWSA data is 11.05 mm / month. The downscaling accuracy was evaluated by comparing the TWSA time series obtained from the GRACE model and the XGBoost model, with correlation coefficients, NSE, and RMSE of 0.95, 0.90, and 23.98 mm, respectively. The spatial consistency of TWSA volatility in August 2020 before and after downscaling was analyzed, along with GWSA estimation. The observed consistency indicates that the downscaling model performs well in effectively capturing the spatial trends of the TWSA derived from GRACE. Both GRACE and the downscaled TWSA show similar geographic variations, with values increasing in the northern and western regions, while values significantly decreasing within the basin and directly north of the basin. By subtracting SMSA and QsA from TWSA to extract GWSA, it was found that the downscaled TWSA and GWSA results provide better regional texture information extraction than the original GRACE data.
[0088] Specifically, through cross-validation of GRACE and downscaled GWSA with groundwater level observations, the downscaled GWSA showed a stronger seasonal correlation with the original (initial) GWSA, quantified as 0.77, compared to the original GRACE GWSA correlation coefficient of 0.75. This improved correlation underscores the effectiveness of the XGBoost model in accurately representing downscaled GWSA data at a resolution of 0.1°.
[0089] Specifically, the temporal variation of the GGDI across the entire watershed was compared with the model-based groundwater drought index GWSP. Since the CLSM-based GWSP has a lower resolution, the comparison was made across the entire watershed rather than its sub-watersheds. The overall correlation between the GGDI and GWSP across the entire watershed was 0.66, indicating a strong correlation between the two indices. Both indices declined sharply as the hydrological conditions in the study area continued to deteriorate.
[0090] Specifically, drought severity was assessed using the scales from previous GGDI analyses (Table 1). From 2016 to 2020, GGDI values showed a trend of worsening drought, with diverse characteristics. Drought events have become increasingly common in recent decades, with severe droughts occurring in 2010, 2019, and 2020, consistent with the significant groundwater depletion observed in previous analyses. Severe drought events were observed at different time points from 2003 to 2020. The average GGDI value for the study area was -1.52 in August 2010 (severe drought), -2.24 in January 2020 (extreme drought), and -2.25 in April 2020 (extreme drought). The drought in April 2020 affected almost the entire study area, necessitating the implementation of robust drought mitigation strategies to reduce the impact of drought and enhance drought resilience.
[0091] Table 1 Classification of GGDI
[0092]
[0093] Specifically, the water storage in the study area showed a downward trend over time, with two distinct downward trends between 2003 and 2010 and between 2016 and 2020, while an upward trend occurred between 2011 and 2015. The analysis of drought periods in the river basins of the study area matched the drought periods recorded in GWSA. The reduction in groundwater storage after 2016 can be attributed to the minimal growth of well installations and the phasing out of older wells, leading to their eventual abandonment. Another contributing factor was the widespread use of wells. Nearly 90% of the groundwater extraction in the study area was supplied by Province A, which owns approximately 85% of the wells in the area. Between 1950 and 2017, planting intensity increased significantly from 80% to 189%, a change primarily attributed to the proliferation of wells.
[0094] See Figure 3 The present invention also provides a high-resolution GRACE groundwater storage monitoring system, applied to the above-mentioned high-resolution GRACE groundwater storage monitoring method, comprising:
[0095] Anomaly estimation module is used to acquire the GRACE dataset of the study area and estimate the initial land water storage anomaly value based on the GRACE dataset;
[0096] The model building module is used to acquire the field observation dataset of the study area and build an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff.
[0097] The data processing module is used to construct an XGBoost model based on the initial land water storage anomaly and the multiple climate variables, and to determine the predicted land water storage anomaly by downscaling the XGBoost model.
[0098] An anomaly determination module is used to calculate the residual value between the predicted land water storage anomaly value and the initial land water storage anomaly value, and to determine the groundwater storage anomaly value of the study area based on the residual value and the predicted land water storage anomaly value.
[0099] In this embodiment, human activities, along with natural factors, play a crucial role in drought conditions, where water resources are constrained due to uneven spatial and temporal distribution. During the agricultural season, excessive groundwater extraction for irrigation creates a supply-demand imbalance, often leading to drought crises. Groundwater depletion not only hinders sustainable social and economic development but also increases the risk of drought-related disasters. Rapid population growth further exacerbates water demand, leading to over-exploitation and intensified drought. The severe drought of 2010-2011 can likely be attributed to the significant groundwater depletion in 2010, forcing people to rely more heavily on groundwater to compensate for surface water shortages. While climate factors are the primary drivers of drought, human activities also significantly contribute to groundwater drought in the study area; therefore, improving water management systems is crucial, including developing water-saving irrigation systems, promoting groundwater recharge, and carefully regulating groundwater extraction. The spatiotemporal irregularities of rainfall distribution throughout the study area lead to shortages of both surface and groundwater resources. Insufficient rainfall forces farmers to heavily rely on groundwater extraction, resulting in aquifer depletion and exacerbating global drought conditions.
[0100] It should be noted that the GRACE satellite plays a crucial role in detecting and monitoring drought, especially in data-scarce regions. Extensive data processing and filtering techniques can effectively improve the quality of GRACE datasets, specifically the error level related to spherical harmonic coefficients; however, the results may still contain weak signals. Therefore, a scaling factor is employed to address signal leakage during filtering. In data-scarce regions, GRACE data is a valuable resource for assessing and managing drought. This invention utilizes the GGDI drought index (a standardized index for objectively analyzing spatiotemporal drought) to assess groundwater surpluses and deficits. This analysis confirmed the drought that occurred from January 2010 to February 2011 due to groundwater depletion. While the GRACE dataset effectively identifies drought patterns (types), longer GRACE data records will improve accuracy, as longer records tend to produce more accurate results. GRACE technology advances terrestrial hydrology by providing precise, vertically integrated spatiotemporal data on water storage changes with millimeter-level accuracy in thickness. The TWSA data provided by GRACE Follow-up (GRACE-FO) will enhance the ability to assess long-term fluctuations in TWSA, ultimately improving insights into GRACE-related analyses. Human activities such as groundwater extraction, regional water allocation, and mining have significant impacts on the Earth's mass balance. However, these impacts are often overlooked due to a lack of observable data and difficulties in collecting relevant information. Despite these limitations, scaled-down global water resource assessment data can serve as a foundation for in-depth research into regional groundwater storage dynamics, ultimately contributing to future hydrological research and enhanced sustainable water resource management.
[0101] It should be understood that a scaled-down GWSA based on XGBoost was used to analyze and assess the dynamics of GWSA and related drought events from 2003 to 2020. The downscaled GWSA was evaluated based on seasonal GWSA observational data and monthly GWSP data, with correlation coefficients of 0.77 and 0.66, respectively. Spatial downscaling of GWSA improves the accuracy of identifying groundwater depletion areas with limited groundwater level data density. The scaled-down GWSA dataset is highly advantageous in detecting drought and identifying severe groundwater depletion, particularly in areas with limited hydrometeorological data and a lack of understanding of water storage changes. It reveals unique fluctuations in monthly GWSA in the study area, showing a significant downward trend from 2003 to 2020, an upward trend from 2011 to 2015, and a downward trend from 2016 to 2020. Severe global warming and drought trends were observed in 2010 and 2020, reflecting extreme drought conditions. The GGDI data show that drought events in the study area exhibit different temporal variations, with the most severe drought occurring between 2016 and 2020. Overall, the GGDI data indicate a downward trend in drought levels from 2003 to 2020, with significant temporal variations in the study area. The most severe drought occurred in April 2020, with a GGDI value of 0.25.
[0102] In all examples shown and described herein, any specific values should be interpreted as merely exemplary and not limiting, and thus other examples of the exemplary embodiments may have different values.
[0103] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.
[0104] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A high-resolution GRACE groundwater storage monitoring method, characterized in that, Includes the following steps: Obtain the GRACE dataset for the study area and estimate the initial land water storage anomalies based on the GRACE dataset; Obtain the field observation dataset of the study area, and construct an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff. An XGBoost model is constructed based on the initial land water storage anomaly and the multiple climate variables, and the predicted land water storage anomaly is determined by resolution downscaling based on the XGBoost model. Calculate the residual between the predicted land water storage anomaly and the initial land water storage anomaly, and determine the groundwater storage anomaly in the study area based on the residual and the predicted land water storage anomaly.
2. The high-resolution GRACE groundwater storage monitoring method according to claim 1, characterized in that, Obtain the GRACE dataset for the study area and estimate the initial land storage outliers based on the GRACE dataset, including: Obtain the watershed surface model CLSM of the global hydrological model GLDAS. The spatial resolution of the watershed surface model CLSM is 0.25°×0.25°. The watershed surface model CLSM is a sub-model of the global hydrological model GLDAS. The watershed surface model CLSM uses numerical simulations of terrestrial water storage and energy processes to derive hydrological drought indices to assess the wet and dry conditions of soil and groundwater. The hydrological drought indices are expressed as percentiles representing the probability of drought occurrence based on wet or dry conditions, which correspond to the groundwater storage percentage (GWSP).
3. The high-resolution GRACE groundwater storage monitoring method according to claim 1, characterized in that, Obtain the field observation dataset of the study area, and construct an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset, including: Multiple well data were acquired in the study area. The groundwater depth (DTG) values obtained from these well data were converted into groundwater GWL using topographic elevation correlated with mean sea level (H). The corresponding expression is as follows: GWL = H - DTG (1) Using average water production S y The seasonal variation of groundwater storage is calculated by multiplying the groundwater level anomaly value (GWIA) and the groundwater level anomaly value (GWSA). The GWSA of each well is calculated using formula (2). GWSA=S y ×GWLA (2) GWSA indicates anomaly in groundwater reserves, S y GWIA represents average water production and groundwater level anomaly.
4. The high-resolution GRACE groundwater storage monitoring method according to claim 3, characterized in that, An XGBoost model is constructed based on the initial land water storage anomalies and the aforementioned multiple climate variables. The predicted land water storage anomalies are then determined using resolution downscaling based on the XGBoost model, including: The XGBoost model constructs multiple interconnected decision trees, where the latest residual of one tree guides the development of the next tree. The sample prediction value of the XGBoost model is determined by the sum of the leaf node values on different trees. The VIMP test, a measure of variable importance, is used to assess the predictive significance of each independent variable. A higher VIMP value indicates a stronger predictive power for the independent variable.
5. The high-resolution GRACE groundwater storage monitoring method according to claim 4, characterized in that, The XGBoost model was used to reduce the GRACE TWSA value from 0.5° resolution to 0.1° resolution, including: An XGBoost model was constructed, and TWSA at a resolution of 0.5° was predicted using initial TWSA data and multiple climate variables. TWSA represents the anomaly value of land water storage. At a resolution of 0.5°, the residual value is calculated by subtracting the initial TWSA value from the predicted TWSA value; Input data at 0.1° resolution was integrated into the XGBoost model to predict TWSA values at 0.1° resolution; The residual value is added to the predicted TWSA value at 0.1° resolution to obtain the TWSA value at 0.1° resolution, wherein residual correction is used to correct the prediction bias caused by omission; The XGBoost model was used to subtract the outliers of runoff and soil water storage from the TWSA values to obtain the downscaled GWSA values, where the GWSA values are groundwater storage outliers.
6. The high-resolution GRACE groundwater storage monitoring method according to claim 5, characterized in that, Formula (3) is used to determine the correlation between runoff anomaly QsA, soil water storage anomaly SMSA, and groundwater storage anomaly GWSA and the study area, where the formula (4) is used to subtract the SMSA and QsA values from the TWSA and subtract the SMSA and QsA values after downscaling by the XGBoost model from the TWSA. The SMSA and QsA components are from the Noah model. TWSA=GWSA+SMSA+QsA (3) GWSA=TWSA-[SMSA+QsA] (4).
7. The high-resolution GRACE groundwater storage monitoring method according to claim 2, characterized in that, The calculation process for the Grace Groundwater Drought Index (GGDI) includes: The climatological values for each month are obtained using formula (5): Where Ci represents the long-term monthly average, i = 1, 2, ... 12; parameter n represents the duration of the value from 2003 to 2020, in months, n = 1, 2, ... 18; and GWSA represents the groundwater storage in the i-th month. The groundwater storage deviation (GSD) is calculated by subtracting the monthly cumulative value from the GWSA. The GSD serves as an indicator of the net seasonal fluctuation of the GWSA and is standardized using the mean and standard deviation of the GSD. Where GGDI represents the standardized net change in GWSA, t is the duration in months, and t ranges from 1 to 216; variables This represents the overall average value of GSD; It represents the overall standard deviation of GSD and serves as an indicator of drought conditions.
8. The high-resolution GRACE groundwater storage monitoring method according to claim 7, characterized in that, The performance of the XGBoost model is evaluated using four statistics: correlation (R), percentage of bias (PBIAS), root mean square error (RMSE), and Nash-Sutcliffe efficiency (NSE). Their corresponding expressions are as follows: Among them, X i and Y i These are the observed value and the predicted value. and These are the values that reflect the average of the observed and predicted values, respectively, and the symbol N represents the total number of samples collected.
9. The high-resolution GRACE groundwater storage monitoring method according to claim 1, characterized in that, The GRACE dataset was processed using spherical harmonic functions and Mascons functions. The TWSA data was derived from the GRACE JPL Mascons RL-06 dataset with a resolution of 0.5° × 0.5°.
10. A high-resolution GRACE groundwater storage monitoring system, characterized in that, The GRACE groundwater storage monitoring method with fine resolution as described in any one of claims 1-9 includes: Anomaly estimation module is used to acquire the GRACE dataset of the study area and estimate the initial land water storage anomaly value based on the GRACE dataset; The model building module is used to acquire the field observation dataset of the study area and build an FLDAS model containing multiple climate variables based on the GRACE dataset and the field observation dataset. The multiple climate variables include soil evapotranspiration, soil water storage, precipitation, temperature and runoff. The data processing module is used to construct an XGBoost model based on the initial land water storage anomaly and the multiple climate variables, and to determine the predicted land water storage anomaly by downscaling the XGBoost model. An anomaly determination module is used to calculate the residual value between the predicted land water storage anomaly value and the initial land water storage anomaly value, and to determine the groundwater storage anomaly value of the study area based on the residual value and the predicted land water storage anomaly value.