Groundwater model method for constructing snow melting supply influence based on CMA-RA / Land data

Abstract

The invention belongs to the technical field of mine engineering, and particularly discloses a method for constructing a groundwater model influenced by snow melting supply based on CMA-RA / Land data. The method comprises the following steps: acquiring a CMA-RA / Land data set of a research area and creating data extraction points; a meteorological variable time sequence corresponding to each data extraction point is extracted from the data set, and spatial interpolation and unit conversion are carried out; determining a source sink item based on the meteorological variable time sequence and in combination with an underground water burial depth condition; establishing a groundwater model, setting initial and boundary conditions, taking the source sink item as an input parameter and setting condition combination, and simulating a groundwater dynamic process by adopting the groundwater model; and based on the simulation result, an SHAP method is adopted to analyze and output characteristics of the underground water model, and the underground water model influenced by snow melting supply is obtained according to analysis and optimization of the underground water model. The method has the characteristics of easiness in data acquisition, clear physical significance of parameters and strong certainty of simulation results.

Description

Technical Field

[0001] This invention belongs to the field of mining engineering technology, and specifically relates to a method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, which features readily available data, clear physical meaning of parameters, and strong deterministic simulation results. Background Technology

[0002] In cold and arid mountainous-basin regions, snowmelt recharge (including glacial meltwater and seasonal snowmelt) is one of the most important and distinctive sources of groundwater recharge, exhibiting significant seasonality and spatial heterogeneity. For open-pit coal mines and their surrounding aquifers, spring snowmelt infiltration determines the recovery process of groundwater levels during the dry season and directly affects the dynamic changes in mine water inflow. Therefore, accurately characterizing the snowmelt recharge process and its response mechanism to the groundwater system is a crucial prerequisite for constructing high-precision and high-reliability groundwater models.

[0003] In current engineering practice, the estimation of equivalent snowmelt recharge largely relies on the degree-day method or the energy balance method. The former is widely used due to its fewer required parameters and simpler calculations, but its core parameters (such as the degree-day factor and snowmelt threshold temperature) are significantly affected by regional climate, topography, and underlying surface conditions, making it difficult to maintain spatiotemporal stability. The latter, while possessing a more complete physical mechanism, is highly dependent on high spatiotemporal resolution meteorological elements such as solar radiation, wind speed, and humidity. At the mining scale, it is often limited by a lack of observational data, making it difficult to generalize to actual groundwater modeling scenarios. Furthermore, existing methods typically simplify snowmelt recharge to an empirical infiltration coefficient multiplied by the total snowmelt volume, neglecting the complex regulatory effects of vadose zone thickness, soil freezing-thawing processes, vegetation cover, and human engineering activities (such as mine drainage and seepage barriers) on infiltration paths and efficiency. This results in highly subjective source-sink term settings for groundwater models and a weak physical foundation.

[0004] In recent years, with the development of reanalysis techniques, the China Meteorological Administration's Land Surface Reanalysis Dataset (CMA-RA / Land) has provided continuous meteorological and hydrological variables with high spatiotemporal resolution (approximately 9 km, hourly) by integrating multi-source observations and land surface process models. These variables include key elements such as snow cover water equivalent (SWE), precipitation, evapotranspiration, and snowmelt equivalent water volume. Because this dataset has been systematically validated in China and possesses good accuracy and consistency, it has been widely applied in fields such as climate change and watershed hydrological simulation. However, current work primarily focuses on climate change analysis and watershed-scale surface hydrological simulation, lacking a technical approach to directly and efficiently convert snowmelt information from CMA-RA / Land into input parameters for groundwater models. On the one hand, the netCDF format raster data is difficult to precisely interface with the spatial discrete structure of groundwater models (such as MODFLOW grids) or monitoring points; on the other hand, the units of the original variables (such as kg / (m³)) are often insufficient. 2 The units used in groundwater models (such as mm / d) are inconsistent with those used in groundwater models, and the time scale needs to be converted from instantaneous / average to daily scale, so a standardized processing procedure is urgently needed.

[0005] More importantly, existing groundwater modeling methods often employ empirical constants or simplified rainfall infiltration parameterization when constructing source and sink terms, failing to fully utilize the comprehensive "rainfall-evaporation-snowmelt" coordinated information provided by CMA-RA / Land. This results in unclear physical meanings for source and sink terms. Particularly in arid and semi-arid mining areas, due to the deep groundwater depth and thick vadose zone, rainfall infiltration efficiency is extremely low, while snowmelt, with its slow and prolonged process, is more likely to generate effective recharge. If the traditional assumption of rainfall-dominated recharge is still used, the actual contribution of snowmelt will be severely underestimated, leading to biased predictions of mine inflow. Furthermore, mine excavation, large-scale pumping, and seepage prevention projects significantly alter the natural groundwater flow field, making the transport path, residence time, and discharge methods after snowmelt infiltration highly nonlinear. Traditional constant-parameter models struggle to capture such complex responses, resulting in significant uncertainty in mine inflow simulation results. Furthermore, current groundwater model parameter optimization largely relies on trial and error or local sensitivity analysis, lacking a global, interpretable quantitative assessment of the impact of multiple input sources (such as different meteorological elements, geological parameters, and boundary conditions) on outputs (such as mine water inflow). This makes the source of model uncertainty unclear, hindering targeted improvements.

[0006] Therefore, existing technologies face three major bottlenecks in modeling groundwater in mining areas driven by snowmelt recharge: (1) lack of automated extraction, interpolation and unit standardization processes for CMA-RA / Land data for groundwater models; (2) the source-sink term construction does not systematically integrate the dynamic coupling relationship between snowmelt, precipitation and evaporation, and the physical mechanism is unclear; (3) lack of parameter sensitivity quantification and model optimization mechanisms based on machine learning and interpretability analysis (such as SHAP), resulting in insufficient reliability of mine water inflow prediction.

[0007] Therefore, there is an urgent need for a new groundwater modeling method that integrates high-resolution reanalysis data, clearly defines physical processes, and supports automated processing and intelligent optimization, in order to accurately characterize the impact of snowmelt replenishment on mine water inflow and provide scientific support for water resource management and disaster prevention in cold and arid mining areas. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge. This method offers advantages such as readily available data, clear physical meaning of parameters, and strong deterministic simulation results.

[0009] The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in this invention, includes the following steps: data acquisition, creation of data extraction points, data processing, determination of source and sink terms, dynamic simulation, and model optimization. The specific details of each step are as follows: A. Data Acquisition: Obtain borehole data and field monitoring data on snowmelt infiltration and groundwater level for the study area, and obtain the corresponding CMA-RA / Land dataset for the study area from the National Meteorological Science Data Center; B. Creating data extraction points: Using geographic information system software, data extraction points are created for the study area; C. Data Processing: Extract the time series of meteorological variables corresponding to each data extraction point created in step B from the CMA-RA / Land dataset in batches, and perform spatial interpolation and unit conversion. D. Determine source and sink terms: Based on the meteorological variable time series extracted in step C, and combined with the groundwater depth conditions in the study area, determine the relationship between groundwater recharge, runoff and discharge, and determine the source and sink terms including rainfall, evaporation and snowmelt equivalent water volume. E. Dynamic Simulation: A three-dimensional geological structure model is constructed based on borehole data of the study area, and a groundwater model is established based on the three-dimensional geological structure model. The source and sink terms determined in step D are used as input parameters of the groundwater model, and the initial conditions and boundary conditions of the groundwater model are set. Multiple sets of parameter combinations are obtained by freely combining the aforementioned input parameters and set conditions. The groundwater model executes batch calculations based on multiple sets of parameter combinations to obtain dynamic groundwater data. Then, the snowmelt infiltration recharge coefficient of the groundwater model is adjusted according to the field monitoring measured values ​​obtained in step A. F. Model Optimization: Based on the groundwater dynamic data obtained in step E and the adjusted groundwater model, the SHAP method is used to optimize the groundwater model. n The output feature y= of each feature f ( p 1, p 2, p 3…,p n An importance and impact analysis was conducted. Based on the analysis, the influence of each parameter in the groundwater model on the prediction results of mine water inflow was quantified. The groundwater model was then optimized based on the quantified impact, resulting in a groundwater model influenced by snowmelt recharge, which was used to predict mine water inflow.

[0010] Furthermore, in step A, the CMA-RA / Land dataset corresponding to the study area is retrieved and downloaded from the National Meteorological Science Data Center website, and then Earth science data visualization software is used to identify the variable names and units contained in the dataset.

[0011] Furthermore, in step B, spatially distributed scattered points are deployed within the study area using geographic information system software, and the distance between any two scattered points is not less than the spatial resolution of the CMA-RA / Land dataset, and then a shapefile-formatted extraction point file extrac_point.shp is generated.

[0012] Furthermore, the specific process of step C is as follows: C10. Read the extrac_point.shp file and assign a unique identifier to each extraction point; then read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset. C20. Calculate the extraction points using the nearest neighbor interpolation method. P s ( l s , f s ) and each grid point in the CMA-RA / Land dataset ( l i , f j The Euclidean distance is used to select the grid point index with the smallest distance. i \* ,j \* ), index the aforementioned nearest grid point ( i \* ,j \* The physical quantity values ​​of ) V i\*,j\* ( t k Directly assign to the extraction point V s ( t k); in,( l s , f s ) is the extraction point P s latitude and longitude, ( l i , f j ) is the grid number in the CMA-RA / Land dataset. i Line number j The latitude and longitude of the grid points V s ( t k The physical quantity corresponding to the nearest grid point at that grid point and time is... t k The value, subscript " i \* ,j \* "" indicates the nearest grid point in the CMA-RA / Land dataset to that extraction point; C30. The units for both the instantaneous value of snowmelt equivalent water volume SWE_inst and the average value SWE_tavg are changed from kg / m³. 2 Convert to mm, and change the units of rainfall (TotalPrecip_tavg) and evaporation (Evap_tavg) from kg / (m²) 2 ·s) is converted to mm / d.

[0013] Further, in step C10, Python code is used to write the program code, and then the GeoPandas package is used to read the extrac_point.shp file containing geographic coordinates and assign a unique identifier to each extracted point; then the netCDF4 package is used to read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset; In step C20, the extraction point is first calculated. P s ( l s , f s ) and each grid point in the CMA-RA / Land dataset ( l i , f j Euclidean distance: Then find and extract points.P s ( l s , f s Recent grid point index: Then index the nearest grid point ( i \* ,j \* The corresponding physical quantity value V i\*,j\* ( t k Assign it directly to the extraction point: V s ( t k )= V i\*,j\* ( t k ); where the superscript (s) represents the time series at the extraction point Ps.

[0014] Furthermore, in step D, the equivalent water volume of rainfall, evaporation and snowmelt in the extracted meteorological variable time series is compared, and the relationship between groundwater recharge, runoff and discharge is determined in combination with the groundwater depth conditions of the study area to determine the composition of source and sink terms; wherein, if the study area is an arid to semi-arid region and the groundwater depth is greater than 5 m, the composition of source and sink terms may not include rainfall.

[0015] Furthermore, in step E, the groundwater flow in the groundwater model is expressed as a three-dimensional transient Darcy flow equation: , In the formula, ▽ represents the Nabra operator; S s The water storage rate is expressed in units of 1 / m. H Water head, in meters (m); K This is the permeability tensor, in m / d. W The unit time source and sink terms include infiltration recharge, river recharge, mine boundary drainage, and well pumping, in units of 1 / d; t is time, in units of days. Among them, the source and sink terms W per unit time are represented by the grid cells (i,j, in the groundwater model) k The part marked ) is represented as: , In the formula, i , j , k Number the grid points; Q p ( i , j ,k , t This represents the amount of rainfall / snowmelt infiltration replenishment. Q h ( i , j , k ) represents the boundary recharge flow rate for a given head. Q w ( i , j , k () represents the flow rate of the pumping well; Q d ( i , j , k ) represents the drainage volume at the mine boundary; Δ V i,j,k This represents the volume of the grid cell.

[0016] Furthermore, in step E, setting the initial and boundary conditions of the groundwater model includes: Initial conditions: , Given head boundary: , Waterproof boundary: , In the formula: H 0 represents the initial water level of the model, in meters. For a steady-state model, the initial water level is set to the ground elevation, while for a non-steady-state model that is simulated step by step, the initial water level is the result of the previous step. H river Γ1 represents the river level in meters (m); Γ2 represents the Dirichlet and Newman boundaries, respectively; and n represents the normal unit vector of the boundary.

[0017] Furthermore, in step E, the source and sink terms determined in step D are first standardized by Z-score to eliminate the influence of dimensions, and then used as input parameters for the groundwater model; subsequently, based on the groundwater dynamic simulation results of multiple parameter combinations, a numerical model to replace the groundwater model is constructed using the random forest machine learning method.

[0018] Furthermore, in step F, the SHAP method is used to process the output features y= of the aforementioned numerical model that have n features. f ( p 1, p 2, p 3…, p n The importance and impact of each parameter in the numerical model are analyzed to determine its influence on the predicted mine water inflow. Based on this analysis, the numerical model is optimized to obtain a groundwater model that considers the impact of snowmelt recharge. The specific process is as follows: F10. First, calculate the output feature y= f ( p 1, p 2, p 3…, p n Each feature in ) i The SHAP value is: , In the formula: M = {1, 2, 3, ..., n}, representing the set of all features; Not including the first i A subset of features; f S ( p S (Based solely on feature subsets) S The numerical model predicts the output; f S∪{i} ( p S∪{i} ) to add features i Post-numerical model prediction increment, i.e. i The marginal contribution; f and p These are the importance index function and its corresponding characteristic terms; F20, then, for each feature i The global importance index is obtained by taking the average absolute value of all samples. I i : , F30, subsequently, the global importance indicators I i Normalization processing : , In the formula: I i Features i Global importance; Features i The global importance normalized value; N The total number of samples; j For sample index; f i,j For the first i The feature in the first j SHAP values ​​of each sample; n The total number of features; F40. Finally, utilize the normalized global importance index. The average marginal contribution of each feature to the prediction results of the groundwater model is measured in different feature subset combinations. The importance of each parameter in the groundwater model is quantitatively ranked, and the groundwater model considering the influence of snowmelt recharge is obtained based on the ranking.

[0019] The present invention has the following beneficial effects: 1. This invention, by creating standardized geographic extraction points, employing nearest neighbor interpolation to achieve spatial matching of data, and uniformly completing unit conversion and spatiotemporal scale transformation for multiple types of meteorological and hydrological variables, not only opens up the conversion path from netCDF format raster data to usable parameters for groundwater models and adapts to the spatial discrete structure of MODFLOW grids and mining area monitoring points, but also fully leverages the advantages of high spatiotemporal resolution, high precision, and easy data acquisition of CMA-RA / Land datasets. This establishes a standardized processing flow for CMA-RA / Land datasets for groundwater models, eliminating the problem of low reanalysis data utilization due to mismatches in data format, units, and scales in existing technologies. It fills the technical gap in applying this type of dataset to groundwater modeling in mining areas and effectively solves the technical challenge of accurately connecting high-resolution reanalysis raster data with groundwater models.

[0020] 2. This invention, based on the extracted complete meteorological variable time series and combined with the actual hydrogeological conditions of the groundwater depth in the study area, determines the groundwater recharge-drainage relationship, clarifies the source and sink components including rainfall, evaporation, and snowmelt equivalent water volume, and can specifically eliminate invalid rainfall infiltration recharge when the groundwater depth in arid-semi-arid mining areas is greater than 5m, thus accurately highlighting the core contribution of snowmelt infiltration recharge in cold and arid mining areas. This method abandons the crude model of simplifying snowmelt infiltration recharge to empirical coefficient conversion in existing technologies, fully integrates the dynamic coupling relationship of snowmelt-rainfall-evaporation, and constructs a groundwater model source and sink component system with clear physical meaning and strong targeting. It completely solves the core defects of existing technologies in the setting of source and sink components, which are highly subjective and have weak physical basis, greatly reduces the human subjectivity in the setting of source and sink components, and significantly improves the physical rationality and scientificity of the groundwater model.

[0021] 3. This invention utilizes a three-dimensional geological structure model constructed based on borehole data, coupled with the MODFLOW three-dimensional transient Darcy flow equation, to accurately depict the dynamic process of groundwater driven by snowmelt infiltration recharge. This model is adapted to the complex hydrogeological conditions of mining areas and the impact of human engineering activities. Furthermore, by adjusting the snowmelt infiltration recharge coefficient of the groundwater model using measured values ​​from on-site monitoring, and using the constructed source and sink terms as core input parameters, this invention can accurately capture the nonlinear responses of groundwater transmission paths, residence times, and discharge methods caused by human engineering activities such as mine excavation, large-scale pumping, and seepage prevention curtains. This overcomes the shortcomings of traditional constant parameter models in existing technologies that cannot adapt to such nonlinear changes, effectively restoring the real process of groundwater level recovery and dynamic changes in mine inflow during the dry season in cold and arid mining areas. The accuracy and fit of the simulation results are far superior to existing technologies.

[0022] 4. This invention first eliminates the influence of dimensions through Z-score standardization, constructs an alternative numerical model using random forest, and then uses the SHAP method to quantify the marginal contribution of each parameter to the prediction result of mine water inflow. After normalization, the global importance ranking of each parameter is obtained, and the core influencing parameters are optimized accordingly. This innovatively integrates machine learning and interpretability analysis to complete model optimization, achieving global quantification, scientific explanation, and precise optimization of the influence of groundwater model parameters. Compared with existing trial-and-error methods and local sensitivity analysis, this invention achieves intelligent optimization of model parameters, clarifies the core sources of model uncertainty, and possesses strong interpretability. The optimized groundwater model outputs with strong determinism, small prediction bias, and high reliability, completely solving the technical bottlenecks of trial-and-error parameter optimization and unclear sources of uncertainty in existing technologies, as well as the problem of high uncertainty in existing mine water inflow predictions.

[0023] 5. The overall process of this invention is standardized, modular, and easy to operate. The required basic data can be easily obtained from the authoritative National Meteorological Science Data Center. It does not rely on scarce high spatiotemporal resolution field meteorological observation elements. Therefore, it can balance the simplicity of calculation and the high accuracy of the model. It overcomes the defects of unstable parameters in the degree-day factor method and high data dependence in the energy balance method. It is easy to promote and apply in the engineering practice of open-pit coal mines.

[0024] In summary, this invention innovatively employs dynamic simulation of snowmelt infiltration recharge, automated data processing, and machine learning parameter optimization. This allows for the precise quantification of the impact mechanism of snowmelt infiltration recharge on the groundwater system and mine inflow in mining areas. It solves the problems of "lack of physical mechanisms, difficulty in data fusion, and inefficient parameter calibration" in the application of traditional groundwater models in cold and arid regions. This significantly improves the accuracy and reliability of mine inflow prediction. The groundwater model it constructs can provide scientific and reliable technical support for the rational planning of groundwater resources in mines in cold and arid regions, the precise prevention and control of mine inflow disasters, and the ecological protection of aquifers. Attached Figure Description

[0025] Figure 1 The flowchart of the groundwater model based on CMA-RA / Land data for constructing a groundwater model of snowmelt recharge is shown in the present invention. Figure 2 This refers to the equivalent amount of snowmelt water replenishment per unit area in the open-pit mine area according to the embodiments of the present invention. Figure 3 The equivalent water volume of rainfall, evaporation, and snowmelt in the open-pit mine area in this embodiment of the invention; Figure 4 This invention provides a three-dimensional hydrogeological model of the open-pit mine area in an embodiment of the invention. In the picture: Figure 4 'a' represents the elevation of the study area and the model boundary. Figure 4 b represents a three-dimensional geological model constructed based on geological boreholes. Figure 4 c represents the B−B′ stratigraphic profile. Figure 4 d represents the continuous distribution of Quaternary aquifers in the study area. Figure 4 e represents the mudstone aquitard and coal seam structure, as well as the confined water recharge path; Figure 5 This invention presents an example of SHAP-based model parameter importance ranking and impact analysis. In the picture: Figure 5 'a' represents the ranking of the importance of model parameters based on SHAP. Figure 5 b represents the degree of influence of the SHAP-based model parameters on the model output. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this does not limit the present invention in any way. Any changes or improvements made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0027] like Figure 1 As shown, the method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge according to the present invention includes the following steps: data acquisition, creation of data extraction points, data processing, determination of source and sink terms, dynamic simulation, and model optimization. The specific details of each step are as follows: A. Data Acquisition: Obtain borehole data and field monitoring data on snowmelt infiltration and groundwater level for the study area, and obtain the corresponding CMA-RA / Land dataset for the study area from the National Meteorological Science Data Center; B. Creating data extraction points: Using geographic information system software, data extraction points are created for the study area; C. Data Processing: Extract the time series of meteorological variables corresponding to each data extraction point created in step B from the CMA-RA / Land dataset in batches, and perform spatial interpolation and unit conversion. D. Determine source and sink terms: Based on the meteorological variable time series extracted in step C, and combined with the groundwater depth conditions in the study area, determine the relationship between groundwater recharge, runoff and discharge, and determine the source and sink terms including rainfall, evaporation and snowmelt equivalent water volume. E. Dynamic Simulation: A three-dimensional geological structure model is constructed based on borehole data of the study area, and a groundwater model is established based on the three-dimensional geological structure model. The source and sink terms determined in step D are used as input parameters of the groundwater model, and the initial conditions and boundary conditions of the groundwater model are set. Multiple sets of parameter combinations are obtained by freely combining the aforementioned input parameters and set conditions. The groundwater model executes batch calculations based on multiple sets of parameter combinations to obtain dynamic groundwater data. Then, the snowmelt infiltration recharge coefficient of the groundwater model is adjusted according to the field monitoring measured values ​​obtained in step A. F. Model Optimization: Based on the groundwater dynamic data obtained in step E and the adjusted groundwater model, the SHAP method is used to optimize the groundwater model. n The output feature y= of each feature f ( p 1, p 2, p 3…, p n An importance and impact analysis was conducted. Based on the analysis, the influence of each parameter in the groundwater model on the prediction results of mine water inflow was quantified. The groundwater model was then optimized based on the quantified impact, resulting in a groundwater model influenced by snowmelt recharge, which was used to predict mine water inflow.

[0028] In step A, based on the latitude and longitude of the study area and the data time, elements, and spatiotemporal resolution, the corresponding CMA-RA / Land dataset for the study area is retrieved through the National Meteorological Science Data Center website (https: / / data.cma.cn / ). A txt file containing an FTP link is obtained, and then the corresponding netCDF format raw data files are downloaded in batches using a program such as aria2c. Subsequently, earth science data visualization software (such as Panoply, GrADS, MeteoInfo, IDV, NCL) is used to identify the variable names and units of meteorological / land surface data contained in the dataset.

[0029] In step B, geographic information system software (such as ArcGIS, SuperMap, MapGIS, MapInfo, QGIS) is used to deploy spatially distributed scattered points within the study area, and the distance between any two scattered points is not less than the spatial resolution of the CMA-RA / Land dataset. Then, the extracted point file extrac_point.shp in shapefile format is generated.

[0030] The CMA-RA / Land dataset has a minimum spatial resolution of 0.5° or approximately 34 km.

[0031] The specific process of step C is as follows: C10. Read the extrac_point.shp file and assign a unique identifier to each extraction point; then read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset. C20. Calculate the extraction points using the nearest neighbor interpolation method. P s ( l s , f s ) and each grid point in the CMA-RA / Land dataset ( l i , f j The Euclidean distance is used to select the grid point index with the smallest distance. i \* ,j \* ), index the aforementioned nearest grid point ( i \* ,j \* The physical quantities (such as daily average snowmelt equivalent water volume SWE_tavg, total precipitation TotalPrecip_tavg, or actual evapotranspiration Evap_tavg) V i\*,j\* ( t k Directly assign to the extraction point V s ( t k This converts the netCDF grid data into hourly or daily sequences at each extraction point, providing input for the subsequent construction of groundwater source and sink terms; in,( l s , f s ) is the extraction point P s latitude and longitude, ( l i , f j ) is the grid number in the CMA-RA / Land dataset. i Line number j The latitude and longitude of the grid pointsV s ( t k The physical quantity corresponding to the nearest grid point at that grid point and time is... t k The value, subscript " i \* ,j \* "" indicates the nearest grid point in the CMA-RA / Land dataset to that extraction point; C30. The units for both the instantaneous value of snowmelt equivalent water volume SWE_inst and the average value SWE_tavg are changed from kg / m³. 2 Convert to mm, and change the units of rainfall (TotalPrecip_tavg) and evaporation (Evap_tavg) from kg / (m²) 2 ·s) is converted to mm / d.

[0032] In step C10, Python code is used to read the extrac_point.shp file containing geographic coordinates using the GeoPandas package, and a unique identifier (Point_1, Point_2,...) is assigned to each extracted point. Subsequently, the netCDF4 package is used to read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset (netCDF file). In step C20, the extraction point is first calculated. P s ( l s , f s ) and each grid point in the CMA-RA / Land dataset ( l i , f j Euclidean distance: Then find and extract points. P s ( l s , f s Recent grid point index: Then index the nearest grid point ( i \* ,j \* The corresponding physical quantity value V i\*,j\*( t k Assign it directly to the extraction point: V s ( t k )= V i\*,j\* ( t k ); where the superscript (s) represents the time series at the extraction point Ps.

[0033] The various elements involved in this invention include: Daily average equivalent snowmelt water volume: , Total daily precipitation: , Actual evaporation rate: .

[0034] In step D, the equivalent water volume of rainfall, evaporation and snowmelt in the extracted meteorological variable time series is compared, and the relationship between groundwater recharge, runoff and discharge is determined in combination with the groundwater depth conditions of the study area to determine the composition of source and sink terms. If the study area is an arid to semi-arid region and the groundwater depth is greater than 5 m, the composition of source and sink terms may not include rainfall.

[0035] In step E, the groundwater flow in the groundwater model is represented by the three-dimensional transient Darcy flow equation: , In the formula, ▽ represents the Nabra operator; S s The water storage rate is expressed in units of 1 / m. H Water head, in meters (m); K This is the permeability tensor, in m / d. W The unit time source and sink terms include infiltration recharge, river recharge, mine boundary drainage, and well pumping, in units of 1 / d; t is time, in units of days. Among them, the source and sink terms W per unit time (including infiltration recharge, river recharge, mine boundary drainage, well pumping, etc.) are represented by the grid cells (i,j,) in the groundwater model. k The part marked ) is represented as: , In the formula, i , j , k Number the grid points; Q p ( i , j , k , t This represents the amount of rainfall / snowmelt infiltration replenishment. Qh ( i , j , k ) represents the boundary recharge flow rate for a given head. Q w ( i , j , k () represents the flow rate of the pumping well; Q d ( i , j , k ) represents the drainage volume at the mine boundary; Δ V i,j,k This represents the volume of the grid cell.

[0036] In step E, the initial and boundary conditions of the groundwater model are set as follows: Initial conditions: , Given head boundary: , Waterproof boundary: , In the formula: H 0 represents the initial water level of the model, in meters. For a steady-state model, the initial water level is set to the ground elevation, while for a non-steady-state model that is simulated step by step, the initial water level is the result of the previous step. H river Γ1 represents the river level in meters (m); Γ2 represents the Dirichlet and Newman boundaries, respectively; and n represents the normal unit vector of the boundary.

[0037] In step E, data including stratum thickness and hydrogeological parameters are obtained from boreholes in the study area to construct a three-dimensional geological structure model; then, the three-dimensional geological structure model is imported into MODFLOW, MODFLOW-NWT, MODFLOW-6 or Visual MODFLOW software to build a groundwater model.

[0038] In step E, the infiltration rate determined in step D is used to determine the amount of groundwater replenished by snowmelt infiltration based on the empirical infiltration coefficient (not greater than 0.25); at the same time, other hydrogeological parameters of the groundwater model are determined based on the hydrogeological data of the study area.

[0039] In step E, the source and sink terms determined in step D are first standardized by Z-score to eliminate the influence of dimensions, and then used as input parameters for the groundwater model. Subsequently, based on the groundwater dynamic simulation results of multiple parameter combinations, a numerical model to replace the groundwater model is constructed using the random forest machine learning method. Then, the snowmelt infiltration recharge coefficient of the numerical model is adjusted according to the field monitoring measured values ​​obtained in step A.

[0040] In step F, the SHAP method is used to process the output feature y= of the aforementioned numerical model, which has n features. f ( p 1, p 2, p 3…, p n The importance and impact of each parameter in the numerical model are analyzed to determine its influence on the predicted mine water inflow. Based on this analysis, the numerical model is optimized to obtain a groundwater model that considers the impact of snowmelt recharge. The specific process is as follows: F10. First, calculate the output feature y= f ( p 1, p 2, p 3…, p n Each feature in ) i The SHAP value is: , In the formula: M = {1, 2, 3, ..., n}, representing the set of all features; Not including the first i A subset of features; f S ( p S (Based solely on feature subsets) S The numerical model predicts the output (other features are masked or averaged). f S∪{i} ( p S∪{i} ) to add features i Post-numerical model prediction increment, i.e. i The marginal contribution; f and p These are the importance index function and its corresponding characteristic terms; F20, then, for each feature i The global importance index is obtained by taking the average absolute value of all samples. I i : , F30, subsequently, the global importance indicators I i Normalization processing : , In the formula: I i Features i Global importance; Features i The global importance normalized value; N The total number of samples; j For sample index; f i,j For the first i The feature in the first j SHAP values ​​of each sample; n The total number of features; F40. Finally, utilize the normalized global importance index. The average marginal contribution of each feature to the prediction results of the groundwater model is measured in different feature subset combinations. The importance of each parameter in the groundwater model is quantitatively ranked, and the groundwater model considering the influence of snowmelt recharge is obtained based on the ranking.

[0041] Example This paper studies and applies a model case study of an open-pit coal mine in a semi-arid grassland region of Northeast China, considering snowmelt replenishing groundwater.

[0042] S100: This open-pit coal mine is located on an alluvial plain, in a transitional zone between erosion, deposition, and gentle hills. The southeastern part of the mining area consists of low hills covered with grassland vegetation; the northern part is low-lying swampy land. Climatically, it belongs to the mid-temperate semi-arid continental climate zone, with an average annual rainfall of 345 mm and an average annual potential evaporation of 1314.7 mm.

[0043] Before the excavation of the mine pit, the area was primarily fed by lateral river recharge and drainage. After excavation, pumping wells and drainage at the mine pit boundaries became the main sources of groundwater discharge. On-site monitoring was conducted within the study area (e.g., after excavation). Figure 2 The snowmelt infiltration and groundwater level are shown in the figure. Pumping wells, mining operations, and curtain walls are the main factors that alter the natural groundwater flow field.

[0044] Based on the latitude and longitude of the aforementioned mining area and the data time, elements, and spatiotemporal resolution, the corresponding CMA-RA / Land dataset for the mining area was retrieved through the National Meteorological Science Data Center website (https: / / data.cma.cn / ). A txt file containing an FTP link was obtained, and then the corresponding netCDF format raw data files were downloaded in batches using a program such as aria2c. Subsequently, Panoply software was used to identify the variable names and units of meteorological / land surface data contained in the CMA-RA / Land dataset.

[0045] S200: Using ArcGIS, spatially distributed scattered points are laid out within the above-mentioned mining area, and the distance between any two scattered points is not less than the spatial resolution of the CMA-RA / Land dataset (such as 0.5° or about 34 km). Then, the extracted point file extrac_point.shp in shapefile format is generated.

[0046] S300: Extracts the time series of meteorological variables corresponding to each data extraction point created in S200 from the CMA-RA / Land dataset in batches, and performs spatial interpolation and unit conversion processing; the specific process is as follows: S310: Write Python code and then use the GeoPandas package to read the extrac_point.shp file containing geographic coordinates and assign a unique identifier (Point_1, Point_2, ...) to each extracted point; then use the netCDF4 package to read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset (netCDF file).

[0047] S320: Using the nearest neighbor interpolation method, first calculate the extraction points. P s ( l s , f s ) and each grid point in the CMA-RA / Land dataset ( l i , f j Euclidean distance: Then find and extract points. P s ( l s , f s Recent grid point index: Then index the nearest grid point ( i \* ,j \* The corresponding physical quantity value V i\*,j\* ( t k Assign it directly to the extraction point: V s ( t k )= Vi\*,j\* ( t k This converts the netCDF grid data into hourly or daily sequences at each extraction point, providing input for the subsequent construction of groundwater source and sink terms.

[0048] Wherein, the superscript (s) denotes the time series at extraction point Ps, ( l s , f s ) is the extraction point P s latitude and longitude, ( l i , f j ) is the grid number in the CMA-RA / Land dataset. i Line number j The latitude and longitude of the grid points V s ( t k The physical quantity corresponding to the nearest grid point at that grid point and time is... t k The value, subscript " i \* ,j \* "" indicates the nearest grid point in the CMA-RA / Land dataset to the extraction point.

[0049] The various elements involved in this mining area include: Daily average equivalent snowmelt water volume: , Total daily precipitation: , Actual evaporation rate: .

[0050] S330: The equivalent snowmelt water volume SWE_tavg is changed from kg / m³ 2 Convert to mm, and change the rainfall TotalPrecip_tavg and evaporation Evap_tavg from kg / (m²) 2 ·s) is converted to mm / d.

[0051] S400: Based on the comparison of equivalent water volume of rainfall, evaporation and snowmelt in the time series of meteorological variables extracted from S300, combined with the groundwater depth conditions in the mining area, the relationship between groundwater recharge, runoff and discharge is determined, and the composition of source and sink terms is identified.

[0052] Based on the above processing, the rainfall, snowmelt, and evaporation conditions of the mining area are obtained as follows: Figure 2 and Figure 3 As shown.

[0053] from Figure 2 It can obtain the snowmelt replenishment equivalent of the study area, from Figure 3 It can be seen that rainfall and actual evapotranspiration are basically synchronized and their values ​​are roughly equivalent. Therefore, rainfall recharge on groundwater can be ignored, and the source-sink term can exclude rainfall. Snowmelt infiltration recharge mainly occurs in spring. Since accurately obtaining the recharge coefficient is difficult, the snowmelt infiltration recharge coefficient is taken as approximately 0.25 based on existing research. Figure 3 The snowmelt equivalent water volume (SWE) for this period was determined to be approximately 0.75 mm / d, which translates to an infiltration recharge intensity of approximately 2 × 10⁻⁶ mm / d. -4 m / d. Considering the drawdown funnel formed by mine drainage and the drop in groundwater level around the mining area, snowmelt water will first replenish the moisture deficit in the vadose zone, and the final replenishment amount should be much less than 2 × 10 m / d. -4 m / d. Other hydrogeological parameters are shown in Table 1.

[0054] Table 1. Statistical table of hydrogeological parameters and infiltration recharge intensity of groundwater model in mining area Note: K1 and K3 are the permeability coefficients of the first and third aquifers in the model, respectively, and C d,1 and C d,3 The values ​​represent the equivalent hydraulic conductivity of the first and third layers of the mine pit, respectively. RCH represents the infiltration recharge intensity considering rainfall, snowmelt, and evaporation. The second layer of the hydrogeological model is an impermeable layer, and its parameter influence is ignored.

[0055] S500: A three-dimensional geological structure model is constructed based on data obtained from boreholes in the study area, including stratigraphic thickness data and hydrogeological parameters. Figure 4 Based on the three-dimensional geological structure model, a groundwater model was established using the MODFLOW program. The source and sink terms determined by S400 were standardized by Z-score to eliminate the influence of dimensions, and then used as input parameters for the groundwater model. Initial and boundary conditions of the groundwater model were set, and multiple sets of parameter combinations were obtained by freely combining the aforementioned input parameters and set conditions. The groundwater model was run in batches based on multiple sets of parameter combinations to simulate the dynamic process of groundwater, obtain groundwater dynamic data, and construct a numerical model to replace the groundwater model using the random forest machine learning method based on the groundwater dynamic data. Then, the snowmelt infiltration recharge coefficient of the numerical model was adjusted according to the field monitoring measured values ​​obtained by S100.

[0056] In the groundwater model, groundwater flow is represented by the three-dimensional transient Darcy flow equation: , In the formula, ▽ represents the Nabra operator; S sThe water storage rate is expressed in units of 1 / m. H Water head, in meters (m); K This is the permeability tensor, in m / d. W The unit time source and sink terms include infiltration recharge, river recharge, mine boundary drainage, and well pumping, in units of 1 / d; t is time, in units of d.

[0057] Among them, the source and sink terms W per unit time (including infiltration recharge, river recharge, mine boundary drainage, well pumping, etc.) are represented by the grid cells (i,j,) in the groundwater model. k The part marked ) is represented as: , In the formula, i , j , k Number the grid points; Q p ( i , j , k , t This represents the amount of rainfall / snowmelt infiltration replenishment. Q h ( i , j , k ) represents the boundary recharge flow rate for a given head. Q w ( i , j , k () represents the flow rate of the pumping well; Q d ( i , j , k ) represents the drainage volume at the mine boundary; Δ V i,j,k This represents the volume of the grid cell.

[0058] The initial and boundary conditions for setting the groundwater model include: Initial conditions: , Given head boundary: , Waterproof boundary: , In the formula: H 0 represents the initial water level of the model, in meters. For a steady-state model, the initial water level is set to the ground elevation, while for a non-steady-state model that is simulated step by step, the initial water level is the result of the previous step. H river Γ1 represents the river level in meters (m); Γ2 represents the Dirichlet and Newman boundaries, respectively; and n represents the normal unit vector of the boundary.

[0059] This embodiment mainly studies the stable water inflow in the mine before and after October 2018, during which the water inflow... Q d Stable at 9.14×10 4 m 3 / d, according to pumping rate Q w Once the source and sink terms of the model are determined, the snowmelt infiltration recharge coefficient of the numerical model will be adjusted based on the measured values ​​from field monitoring. K 1 , K 3 , C d,1 , C d,3 And RCH. Based on Python, modify the corresponding parameters in the MODFLOW input file DRN, and then execute the examples in batches. The parameters in the examples can be freely combined, and 25920 steady-state examples are executed.

[0060] S600: Based on the simulation results of multiple parameter combinations in S500, the SHAP method is used to process the output feature y= of the aforementioned numerical model with n features. f ( p 1, p 2, p 3…, p n An importance and impact analysis was conducted. Based on the analysis, the influence of each parameter in the numerical model on the predicted mine water inflow was quantified. The numerical model was then optimized based on the quantified impact, resulting in a groundwater model that considers the influence of snowmelt recharge. The specific process is as follows: S610: First, calculate the output feature y= f ( p 1, p 2, p 3…, p n Each feature in ) i The SHAP value is: , In the formula: M = {1, 2, 3, ..., n}, representing the set of all features; Not including the first i A subset of features; f S ( p S (Based solely on feature subsets) S The numerical model predicts the output (other features are masked or averaged). f S∪{i} ( pS∪{i} ) to add features i Post-numerical model prediction increment, i.e. i The marginal contribution; f and p These are the importance index function and the corresponding characteristic terms, respectively.

[0061] S620: Then, for each feature i The global importance index is obtained by taking the average absolute value of all samples. I i : .

[0062] S630: Subsequently, the global importance index I i Normalization processing : , In the formula: I i Features i Global importance; Features i The global importance normalized value; N The total number of samples; j For sample index; f i,j For the first i The feature in the first j SHAP values ​​of each sample; n The total number of features.

[0063] S640: Finally, utilize the normalized global importance index. The average marginal contribution of each feature to the prediction results of the groundwater model is measured in different feature subset combinations. The importance of each parameter in the groundwater model is quantitatively ranked, and the groundwater model considering the influence of snowmelt recharge is obtained based on the ranking.

[0064] In this section of 25,920 steady-state simulations, the simulation results for some parameter cases are unreasonable, for example... C d,1 and C d,3 An excessively large value would cause the entire aquifer to dry up, and the water flow in the pumping wells would drop to zero. Therefore, unreasonable parts are filtered out from the simulation results: Let... Q d Between 5×10 3 ~2×10 5 m 3 Between / d. 23706 reasonable cases were selected for parameter sensitivity analysis. Figure 4(This demonstrates the influence of model parameters on the model's inflow rate) K 1 > C d,3 > K 3 > C d,1 >RCH.

[0065] In the above study, Python was used to extract CMA-RA / Land data and calculate snowmelt infiltration recharge (see...). Figure 5 This led to the establishment of an accurate mine water inflow model. Although it was ultimately proven that the permeability coefficient and mine boundary drainage contributed more significantly to the water inflow than snowmelt infiltration recharge, the method proposed in this invention clarified the source and sink terms of the model, which is crucial for constructing an accurate groundwater model and helped determine the range of infiltration recharge intensity values ​​in the parameter sensitivity analysis process.

[0066] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, characterized by: The process includes data acquisition, creation of data extraction points, data processing, determination of source and sink terms, dynamic simulation, and model optimization. The specific details of each step are as follows: A. Data Acquisition: Obtain borehole data and field monitoring data on snowmelt infiltration and groundwater level for the study area, and obtain the corresponding CMA-RA / Land dataset for the study area from the National Meteorological Science Data Center; B. Creating data extraction points: Using geographic information system software, data extraction points are created for the study area; C. Data Processing: Extract the time series of meteorological variables corresponding to each data extraction point created in step B from the CMA-RA / Land dataset in batches, and perform spatial interpolation and unit conversion. D. Determine source and sink terms: Based on the meteorological variable time series extracted in step C, and combined with the groundwater depth conditions in the study area, determine the relationship between groundwater recharge, runoff and discharge, and determine the source and sink terms including rainfall, evaporation and snowmelt equivalent water volume. E. Dynamic Simulation: A three-dimensional geological structure model is constructed based on borehole data of the study area, and a groundwater model is established based on the three-dimensional geological structure model. The source and sink terms determined in step D are used as input parameters of the groundwater model, and the initial conditions and boundary conditions of the groundwater model are set. Multiple sets of parameter combinations are obtained by freely combining the aforementioned input parameters and set conditions. The groundwater model executes batch calculations based on multiple sets of parameter combinations to obtain dynamic groundwater data. Then, the snowmelt infiltration recharge coefficient of the groundwater model is adjusted according to the field monitoring measured values ​​obtained in step A. F. Model Optimization: Based on the groundwater dynamic data obtained in step E and the adjusted groundwater model, the SHAP method is used to optimize the groundwater model. n The output feature y= of each feature f ( p 1, p 2, p 3…, p n An importance and impact analysis was conducted. Based on the analysis, the influence of each parameter in the groundwater model on the prediction results of mine water inflow was quantified. The groundwater model was then optimized based on the quantified impact, resulting in a groundwater model influenced by snowmelt recharge, which was used to predict mine water inflow.

2. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 1, is characterized in that: In step A, the CMA-RA / Land dataset corresponding to the study area is retrieved and downloaded from the National Meteorological Science Data Center website, and then Earth science data visualization software is used to identify the variable names and units contained in the dataset.

3. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 1, is characterized in that: In step B, geographic information system software is used to deploy spatially distributed scattered points within the study area, and the distance between any two scattered points is not less than the spatial resolution of the CMA-RA / Land dataset. Then, a shapefile-formatted extraction point file extrac_point.shp is generated.

4. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 1, is characterized in that: The specific process of step C is as follows: C10. Read the extrac_point.shp file and assign a unique identifier to each extraction point; then read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset. C20. Calculate the extraction points using the nearest neighbor interpolation method. P s ( λ s , φ s ) and each grid point in the CMA-RA / Land dataset ( λ i , φ j The Euclidean distance is used to select the grid point index with the smallest distance. i \* ,j \* ), index the aforementioned nearest grid point ( i \* , j \* The physical quantity values ​​of ) V i\*,j\* ( t k Directly assign to the extraction point V s ( t k ); in,( λ s , φ s ) is the extraction point P s latitude and longitude, ( λ i , φ j ) is the grid number in the CMA-RA / Land dataset. i Line number j The latitude and longitude of the grid points V s ( t k The physical quantity corresponding to the nearest grid point at that grid point and time is... t k The value, subscript " i \* , j \* "" indicates the nearest grid point in the CMA-RA / Land dataset to that extraction point; C30. The units for both the instantaneous value of snowmelt equivalent water volume SWE_inst and the average value SWE_tavg are changed from kg / m³. 2 Convert to mm, and change the units of rainfall (TotalPrecip_tavg) and evaporation (Evap_tavg) from kg / (m²) 2 ·s) is converted to mm / d.

5. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 4, is characterized in that: In step C10, Python code is used to read the extrac_point.shp file containing geographic coordinates using the GeoPandas package, and a unique identifier is assigned to each extracted point. Subsequently, the netCDF4 package is used to read the time, SWE_inst, SWE_tavg, Evap_tavg, SnowCover_tavg, SnowDepth_tavg, and TotalPrecip_tavg variables from the CMA-RA / Land dataset. In step C20, the extraction point is first calculated. P s ( λ s , φ s ) and each grid point in the CMA-RA / Land dataset ( λ i , φ j Euclidean distance: Then find and extract points. P s ( λ s , φ s Recent grid point index: Then index the nearest grid point ( i \* ,j \* The corresponding physical quantity value V i\*,j\* ( t k Assign it directly to the extraction point: V s ( t k )= V i\*,j\* ( t k ); where the superscript (s) represents the time series at the extraction point Ps.

6. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 1, is characterized in that: In step D, the equivalent water volume of rainfall, evaporation and snowmelt in the extracted meteorological variable time series is compared, and the relationship between groundwater recharge, runoff and discharge is determined in combination with the groundwater depth conditions of the study area to determine the composition of source and sink terms. If the study area is an arid to semi-arid region and the groundwater depth is greater than 5 m, the composition of source and sink terms may not include rainfall.

7. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 1, is characterized in that: In step E, the groundwater flow in the groundwater model is represented by the three-dimensional transient Darcy flow equation: ， In the formula, ▽ represents the Nabra operator; S s The water storage rate is expressed in units of 1 / m. H Water head, in meters (m); K This is the permeability tensor, in m / d. W The unit time source and sink terms include infiltration recharge, river recharge, mine boundary drainage, and well pumping, in units of 1 / d; t is time, in units of days. Among them, the source and sink terms W per unit time are represented by the grid cells (i,j, in the groundwater model) k The part marked ) is represented as: ， In the formula, i , j , k Number the grid points; Q p ( i , j , k , t This represents the amount of rainfall / snowmelt infiltration replenishment. Q h ( i , j , k () represents the boundary recharge flow rate for a given head. Q w ( i , j , k () represents the flow rate of the pumping well; Q d ( i , j , k ) represents the drainage volume at the mine boundary; Δ V i,j,k This represents the volume of the grid cell.

8. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 7, is characterized in that: In step E, the initial and boundary conditions of the groundwater model are set as follows: Initial conditions: , Given head boundary: , Waterproof boundary: , In the formula: H 0 represents the initial water level of the model, in meters. For a steady-state model, the initial water level is set to the ground elevation, while for a non-steady-state model that is simulated step by step, the initial water level is the result of the previous step. H river Γ1 represents the river level in meters (m); Γ2 represents the Dirichlet and Newman boundaries, respectively; and n represents the normal unit vector of the boundary.

9. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in any one of claims 1 to 8, is characterized in that: In step E, the source and sink terms determined in step D are first standardized by Z-score to eliminate the influence of dimensions, and then used as input parameters for the groundwater model. Subsequently, based on the groundwater dynamic simulation results of multiple parameter combinations, a numerical model to replace the groundwater model is constructed using the random forest machine learning method.

10. The method for constructing a groundwater model based on CMA-RA / Land data to assess the impact of snowmelt recharge, as described in claim 9, is characterized in that: In step F, the SHAP method is used to process the output feature y= of the aforementioned numerical model, which has n features. f ( p 1, p 2, p 3…, p n The importance and impact of each parameter in the numerical model are analyzed to determine its influence on the predicted mine water inflow. Based on this analysis, the numerical model is optimized to obtain a groundwater model that considers the impact of snowmelt recharge. The specific process is as follows: F10. First, calculate the output feature y= f ( p 1, p 2, p 3…, p n Each feature in ) i The SHAP value is: ， In the formula: M = {1, 2, 3, ..., n}, representing the set of all features; Not including the first i A subset of features; f S ( p S (Based solely on feature subsets) S The numerical model prediction output; f S∪{i} ( p S∪{i} ) to add features i Post-numerical model prediction increment, i.e. i The marginal contribution; f and p These are the importance index function and its corresponding characteristic terms; F20, then, for each feature i The global importance index is obtained by taking the average absolute value of all samples. I i : ， F30, subsequently, the global importance indicators I i Normalization processing : ， In the formula: I i Features i Global importance; Features i The global importance normalized value; N The total number of samples; j For sample index; φ i,j For the first i The feature in the first j SHAP values ​​of each sample; n The total number of features; F40. Finally, utilize the normalized global importance index. The average marginal contribution of each feature to the prediction results of the groundwater model is measured in different feature subset combinations. The importance of each parameter in the groundwater model is quantitatively ranked, and the groundwater model considering the influence of snowmelt recharge is obtained based on the ranking.

Images