Underground water ecological burial depth assessment method fusing CatBoost regression and SHAP attribution

Abstract

The invention belongs to the technical field of ecological and hydrological branches under geophysics, and particularly relates to a groundwater ecological burial depth evaluation method fusing CatBoost regression and SHAP attribution, which comprises the following steps: (1) collecting ecological and hydrological observation data of each sample point in a research area for CatBoost regression modeling, SHAP attribution analysis and groundwater ecological burial depth evaluation; (2) respectively constructing an EVI regression model and an ET regression model by using a CatBoost machine learning algorithm based on the ecological and hydrological basic data in the step (1); (3) respectively carrying out SHAP attribution analysis on the EVI regression model and the ET regression model, and utilizing a Boltzmann function to establish a correlation between the SHAP value of the underground water burial depth in the EVI regression model and the ET regression model and the underground water burial depth; and (4) carrying out inflection point analysis on the SHAP value-underground water burial depth relation curve of the underground water burial depth in the EVI and ET regression models, and carrying out Min analysis on an analysis result to obtain the underground water ecological burial depth. According to the method, the defect of groundwater ecological burial depth analysis in the ecological water demand evaluation research of the arid region is overcome, and an effective tool is provided for groundwater dependent terrestrial ecosystem protection.

Description

Technical Field

[0001] The present invention belongs to the field of geophysical ecology and hydrology branch technology, and specifically relates to a groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution. Background Art

[0002] In the arid inland river basins, water resources are scarce and the ecology is fragile. The ecological depth of groundwater is a key indicator for water resource management, ecological protection and restoration. Due to the special geographical location, climatic conditions and fragile ecological environment in arid areas, groundwater has become a core resource that restricts the social and economic development and ecological and environmental stability of oasis. The groundwater depth affects the ecological characteristics such as vegetation growth and development, coverage, age composition, community structure and biodiversity, and is of great significance to maintaining the structural integrity and functional stability of the oasis ecosystem. The ecological depth of groundwater generally refers to the depth of groundwater that can supply water for the growth and development of different types of terrestrial vegetation such as trees, shrubs and herbs; if the groundwater depth is too large and exceeds this threshold, the capillary rise height of groundwater is difficult to reach the vegetation root layer, and the groundwater is difficult to be absorbed and utilized by vegetation and cannot supply the water required for evapotranspiration. Therefore, it is necessary to propose a groundwater ecological depth assessment method that integrates CatBoost regression and SHAP attribution to provide an effective tool for the protection of groundwater-dependent terrestrial ecosystems and support the construction of ecological civilization and sustainable development in ecologically fragile areas of inland river basins.

[0003] The current methods for assessing the ecological depth of groundwater mainly include three categories: ecological survey statistics and analysis, model calculation analysis, and remote sensing statistical analysis. Specifically, they include a method and device for estimating the ecological groundwater level in arid plains (application / patent number: CN202010331364.3), a method for analyzing the ecological depth of groundwater in arid areas based on joint probability distribution (application / patent number: CN202110590631.3), and a method for determining the optimal groundwater depth of grassland vegetation communities in semi-arid areas (application / patent number: CN202210405935.2). With the rapid development of remote sensing big data, remote sensing statistical analysis methods have gradually been widely used. The remote sensing statistical analysis method uses vegetation indices generated by satellite remote sensing images with different temporal and spatial resolutions, such as the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI), to establish a correlation function between the vegetation index and the groundwater depth. The groundwater depth corresponding to the inflection point detected on the vegetation index-groundwater depth relationship curve is the groundwater ecological depth.

[0004] Existing results still have some shortcomings in the following two aspects: (1) The vegetation index is affected by precipitation, potential evapotranspiration and other hydrological factors. Other hydrological factors may lead to a vague correlation between the vegetation index and groundwater depth, making it difficult to directly establish a significant functional relationship, which affects the assessment of groundwater ecological depth; (2) The correlation between evapotranspiration and groundwater depth is not fully utilized, and the correlation curve between evapotranspiration and groundwater depth is rarely established to detect the inflection point to assess the groundwater ecological depth. Therefore, it is necessary to develop a groundwater ecological depth assessment method that integrates CatBoost regression and SHAP attribution. Summary of the invention

[0005] The purpose of the present invention is to propose a groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution, to make up for the deficiency of precipitation, potential evapotranspiration and other hydrological factors interfering with the analysis of groundwater ecological depth in the existing ecological water demand assessment research in arid areas, to improve the ecological water demand assessment and guarantee method system of groundwater-dependent terrestrial ecosystems in arid areas, to improve the utilization efficiency of ecological and hydrological observation data such as enhanced vegetation index, evapotranspiration, precipitation, groundwater depth, potential evapotranspiration, to provide an effective tool for the protection of groundwater-dependent terrestrial ecosystems, and to provide scientific and technological support for the construction of ecological civilization and sustainable development in ecologically fragile areas of inland river basins.

[0006] The present invention adopts the following technical scheme, integrating CatBoost regression and SHAP attribution groundwater ecological depth assessment method. Collect and organize ecological and hydrological data, use CatBoost machine learning algorithm to respectively construct annual maximum enhanced vegetation index regression model (recorded as EVI regression model) and annual evapotranspiration regression model (recorded as ET regression model); carry out SHAP attribution analysis on EVI regression model and ET regression model respectively, use Boltzmann function to establish the correlation between SHAP value of groundwater depth and groundwater depth in EVI regression model, and the correlation between SHAP value of groundwater depth and groundwater depth in ET regression model; carry out inflection point analysis on the SHAP value of groundwater depth-groundwater depth relationship curve in EVI regression model and the SHAP value of groundwater depth-groundwater depth relationship curve in ET regression model, and carry out Min analysis on the analysis results to obtain groundwater ecological depth.

[0007] The flowchart of the groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution in the present invention is as follows:

[0008] (1) Collect and organize ecological and hydrological observation data such as the annual maximum enhanced vegetation index, annual evapotranspiration, annual precipitation, annual average groundwater depth, and annual potential evapotranspiration at each sample point in the study area for use in CatBoost regression modeling, SHAP attribution analysis, and groundwater ecological depth assessment.

[0009] (2) Based on the ecological and hydrological basic data in step (1), the CatBoost machine learning algorithm is used to construct the EVI regression model and the ET regression model respectively. Specifically:

[0010] ① Taking annual precipitation, annual average groundwater depth, and annual potential evapotranspiration as model inputs, and annual maximum enhanced vegetation index as model output, the CatBoost machine learning algorithm was used to construct the EVI regression model:

[0011] EVI=f CatBoost,EVI (PRE,WTD,PET)

[0012] Where, EVI represents the annual maximum enhanced vegetation index; f CatBoost,EVI () represents the EVI regression model constructed using the CatBoost machine learning algorithm; PRE represents annual precipitation, in mm; WTD represents annual average groundwater depth, in m; PET represents annual potential evapotranspiration, in mm.

[0013] In the process of using the CatBoost machine learning algorithm to build the EVI regression model, 60% of the ecological and hydrological data are used to train the model, recorded as the training set; 20% of the ecological and hydrological data are used to verify the model, recorded as the verification set; and 20% of the ecological and hydrological data are used to test the model, recorded as the test set.

[0014] ② Taking annual precipitation, annual average groundwater depth, and annual potential evapotranspiration as model inputs and annual evapotranspiration as model output, the CatBoost machine learning algorithm was used to construct an ET regression model:

[0015] ET=f CatBoost,ET (PRE,WTD,PET)

[0016] Where, ET represents annual evapotranspiration, unit: mm; f CatBoost,ET () represents the ET regression model constructed using the CatBoost machine learning algorithm; PRE represents annual precipitation, in mm; WTD represents annual average groundwater depth, in m; PET represents annual potential evapotranspiration, in mm.

[0017] In the process of using the CatBoost machine learning algorithm to build the ET regression model, 60% of the ecological and hydrological data are used to train the model, recorded as the training set; 20% of the ecological and hydrological data are used to verify the model, recorded as the validation set; and 20% of the ecological and hydrological data are used to test the model, recorded as the test set.

[0018] (3) SHAP attribution analysis was performed on the EVI regression model and the ET regression model, and the Boltzmann function was used to establish the correlation between the SHAP value of groundwater depth in the EVI regression model and the groundwater depth, and the correlation between the SHAP value of groundwater depth in the ET regression model and the groundwater depth. Specifically:

[0019] ① SHAP attribution analysis was conducted on the EVI regression model to calculate the SHAP value of groundwater depth in the EVI regression model. The SHAP value of groundwater depth in the EVI regression model quantifies the impact of groundwater depth on EVI: the larger the SHAP value of groundwater depth in the EVI regression model, the greater the impact of groundwater depth on EVI; the smaller the SHAP value of groundwater depth in the EVI regression model, the smaller the impact of groundwater depth on EVI.

[0020] ② SHAP attribution analysis was conducted on the ET regression model to calculate the SHAP value of groundwater depth in the ET regression model. The SHAP value of groundwater depth in the ET regression model quantifies the impact of groundwater depth on ET: the larger the SHAP value of groundwater depth in the ET regression model, the greater the impact of groundwater depth on ET; the smaller the SHAP value of groundwater depth in the ET regression model, the smaller the impact of groundwater depth on ET.

[0021] ③ Use the Boltzmann function to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the EVI regression model:

[0022] SHAP EVI,WTD =f Boltzmann,EVI (WTD)

[0023] In the formula, SHAP EVI,WTD represents the Boltzmann function fitting value of the SHAP value of groundwater depth in the EVI regression model; f Boltzmann,EVI () represents the Boltzmann function fitted by the SHAP value of groundwater depth and the scatter points of groundwater depth in the EVI regression model, which characterizes the correlation between the SHAP value of groundwater depth in the EVI regression model and the groundwater depth; WTD represents the annual average groundwater depth, in m.

[0024] ④ Use the Boltzmann function to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the ET regression model:

[0025] SHAP ET,WTD =f Boltzmann,ET (WTD)

[0026] In the formula, SHAP ET,WTD represents the Boltzmann function fitting value of the SHAP value of groundwater depth in the ET regression model; f Boltzmann,ET () represents the Boltzmann function fitted by the SHAP value of groundwater depth and the scatter points of groundwater depth in the ET regression model, which characterizes the correlation between the SHAP value of groundwater depth in the ET regression model and the groundwater depth; WTD represents the annual average groundwater depth, in m.

[0027] (4) Carry out inflection point analysis on the SHAP value-groundwater depth relationship curve in the EVI regression model and the SHAP value-groundwater depth relationship curve in the ET regression model, and perform Min analysis on the analysis results to obtain the groundwater ecological depth. Specifically:

[0028] ① The SHAP value-groundwater depth relationship curve in the EVI regression model fitted by the Boltzmann function is inverted S-shaped. The inflection point is determined by the maximum curvature method. Theoretically, there are two inflection points, where the groundwater depth corresponding to the first inflection point is smaller than the groundwater depth corresponding to the second inflection point. On the left side of the first inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the EVI regression model decreases slightly; between the first inflection point and the second inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the EVI regression model decreases sharply, indicating that the impact of the groundwater depth on EVI is sharply weakened; after reaching the second inflection point, the SHAP value of the groundwater depth in the EVI regression model is close to its minimum value, and decreases slightly with the increase of the groundwater depth on the right side of the second inflection point. The groundwater depth corresponding to the second inflection point of the SHAP value-groundwater depth relationship curve in the EVI regression model fitted by the Boltzmann function is taken as the preliminarily determined groundwater ecological depth, denoted as EWTD EVI , unit: m.

[0029] ②The SHAP value-groundwater depth relationship curve of the ET regression model fitted by the Boltzmann function is inverted S-shaped. The inflection point is determined by the maximum curvature method. Theoretically, there are two inflection points, where the groundwater depth corresponding to the first inflection point is smaller than the groundwater depth corresponding to the second inflection point. On the left side of the first inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the ET regression model decreases slightly; between the first inflection point and the second inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the ET regression model decreases sharply, indicating that the impact of groundwater depth on ET is sharply weakened; after reaching the second inflection point, the SHAP value of the groundwater depth in the ET regression model is close to its minimum value, and decreases slightly with the increase of groundwater depth on the right side of the second inflection point. The groundwater depth corresponding to the second inflection point of the SHAP value-groundwater depth relationship curve in the ET regression model fitted by the Boltzmann function is taken as the preliminarily determined groundwater ecological depth, denoted as EWTD. ET , unit: m.

[0030] ③EWTD EVI and EWTD ET Conduct Min analysis, and the result is the groundwater ecological depth:

[0031] EWTD=min(EWTD EVI ,EWTD ET )

[0032] Where, EWTD represents the ecological depth of groundwater, in meters; min() represents the function of taking the minimum value of the set; EWTD EVI It represents the groundwater ecological depth obtained by analyzing the inflection point of the groundwater depth relationship curve between the SHAP value of the groundwater depth in the EVI regression model, in meters; EWTD ET It represents the groundwater ecological depth obtained by analyzing the inflection point of the groundwater depth relationship curve between the SHAP value of the groundwater depth in the ET regression model, in meters.

[0033] Beneficial Effects

[0034] The present invention is a groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution, which makes up for the deficiency of precipitation, potential evapotranspiration and other hydrological factors interfering with the analysis of groundwater ecological depth in the existing ecological water demand assessment research in arid areas, improves the ecological water demand assessment and guarantee method system of groundwater-dependent terrestrial ecosystems in arid areas, improves the utilization efficiency of ecological and hydrological observation data such as enhanced vegetation index, evapotranspiration, precipitation, groundwater depth, potential evapotranspiration, etc., provides an effective tool for the protection of groundwater-dependent terrestrial ecosystems, and provides scientific and technological support for the construction of ecological civilization and sustainable development in ecologically fragile areas in inland river basins. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution of the present invention;

[0036] Figure 2 It is a distribution map of sample points in the Balkhash Lake Basin of the example of the present invention;

[0037] Figure 3 is a correlation diagram between EVI and groundwater depth in the Balkhash Lake Basin of the example of the present invention;

[0038] Figure 4 is a correlation diagram between ET and groundwater depth in the Balkhash Lake Basin of the present invention;

[0039] Figure 5 It is a comparison chart of the simulated value and the measured value of the EVI regression model based on the CatBoost machine learning algorithm in the Balkhash Lake Basin of the example of the present invention;

[0040] Figure 6 It is a comparison chart of the simulated value and the measured value of the ET regression model based on the CatBoost machine learning algorithm in the Balkhash Lake Basin of the example of the present invention;

[0041] Figure 7 It is the SHAP value of groundwater depth and its Boltzmann function fitting diagram in the EVI regression model of the Balkhash Lake Basin of the example of the present invention;

[0042] Figure 8 It is the SHAP value of groundwater depth and its Boltzmann function fitting diagram in the ET regression model of the Balkhash Lake Basin in the example of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below with reference to the accompanying drawings and in combination with examples.

[0044] The application research area of ​​the embodiment of the present invention is the Balkhash Lake Basin in Central Asia; Figure 2 This is a distribution map of sample points in the Balkhash Lake Basin according to an example of the present invention.

[0045] The specific implementation method of the groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution in the Balkhash Lake Basin is as follows:

[0046] Step 1: Collect and organize ecological and hydrological observation data such as the annual maximum enhanced vegetation index, annual evapotranspiration, annual precipitation, annual average groundwater depth, and annual potential evapotranspiration at each sample point in the Balkhash Lake Basin for use in CatBoost regression modeling, SHAP attribution analysis, and groundwater ecological depth assessment.

[0047] Figure 3 This is a correlation diagram between EVI and groundwater depth in the Balkhash Lake Basin of the example of the present invention. It can be seen from the figure that the relationship between EVI and groundwater depth is fuzzy and there is no significant correlation. It is difficult to use traditional methods to construct an EVI-groundwater depth relationship curve to evaluate the groundwater ecological depth.

[0048] Figure 4 This is a correlation diagram of ET and groundwater depth in the Balkhash Lake Basin of the present invention. It can be seen from the figure that the relationship between ET and groundwater depth is fuzzy and there is no significant correlation. It is difficult to directly establish an ET-groundwater depth relationship curve to evaluate the groundwater ecological depth. Therefore, it is necessary to develop a groundwater ecological depth assessment method that integrates CatBoost regression and SHAP attribution to assess the groundwater ecological depth in the Balkhash Lake Basin.

[0049] Step 2: Based on the ecological and hydrological basic data collected and sorted in the first step, the CatBoost machine learning algorithm was used to construct the EVI regression model and ET regression model of the Balkhash Lake Basin. Specifically:

[0050] ① Taking the annual precipitation, annual average groundwater depth and annual potential evapotranspiration in the Balkhash Lake Basin as the model input, and the annual maximum enhanced vegetation index as the model output, the CatBoost machine learning algorithm was used to construct the EVI regression model. Figure 5 It is a comparison chart between the simulated values ​​and the measured values ​​of the EVI regression model based on the CatBoost machine learning algorithm in the Balkhash Lake Basin of the example of the present invention.

[0051] ② Taking the annual precipitation, annual average groundwater depth and annual potential evapotranspiration in the Balkhash Lake Basin as the model input, and annual evapotranspiration as the model output, the CatBoost machine learning algorithm was used to construct the ET regression model. Figure 6 It is a comparison chart between the simulated values ​​and the measured values ​​of the ET regression model based on the CatBoost machine learning algorithm in the Balkhash Lake Basin of the example of the present invention.

[0052] Step 3: SHAP attribution analysis was conducted on the EVI regression model and ET regression model of the Balkhash Lake Basin, and the Boltzmann function was used to establish the correlation between the SHAP value of groundwater depth in the EVI regression model and the groundwater depth, and the correlation between the SHAP value of groundwater depth in the ET regression model and the groundwater depth. Specifically:

[0053] ① SHAP attribution analysis was carried out on the EVI regression model of the Balkhash Lake Basin, and the SHAP value of groundwater depth in the EVI regression model was calculated.

[0054] ② SHAP attribution analysis was carried out on the ET regression model of the Balkhash Lake Basin, and the SHAP value of groundwater depth in the ET regression model was calculated.

[0055] ③The Boltzmann function was used to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the EVI regression model of the Balkhash Lake Basin. Figure 7 It is the SHAP value of groundwater depth and its Boltzmann function fitting diagram in the EVI regression model of the Balkhash Lake Basin in the example of the present invention.

[0056] ④The Boltzmann function was used to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the ET regression model of the Balkhash Lake Basin. Figure 8 It is the SHAP value of groundwater depth and its Boltzmann function fitting diagram in the ET regression model of the Balkhash Lake Basin in the example of the present invention.

[0057] Step 4: Carry out inflection point analysis on the SHAP value-groundwater depth relationship curve in the EVI regression model and the SHAP value-groundwater depth relationship curve in the ET regression model, and perform Min analysis on the analysis results to obtain the groundwater ecological depth. Specifically:

[0058] ① The relationship curve between the SHAP value of groundwater depth and the groundwater depth in the EVI regression model of the Balkhash Lake Basin fitted by the Boltzmann function is inversely S-shaped. The inflection point is determined by the maximum curvature method. The groundwater depth corresponding to the first inflection point is 5.1m, and the groundwater depth corresponding to the second inflection point is 7.3m. The groundwater depth corresponding to the second inflection point of the relationship curve between the SHAP value of groundwater depth and the groundwater depth in the EVI regression model fitted by the Boltzmann function is used as the preliminary determined groundwater ecological depth, that is, EWTD EVI =7.3m.

[0059] ②The relationship curve between the SHAP value of groundwater depth and groundwater depth in the ET regression model of the Balkhash Lake Basin fitted by the Boltzmann function is inversely S-shaped. The inflection point is determined by the maximum curvature method. The groundwater depth corresponding to the first inflection point is 3.7m, and the groundwater depth corresponding to the second inflection point is 8.6m. The groundwater depth corresponding to the second inflection point of the relationship curve between the SHAP value of groundwater depth and groundwater depth in the ET regression model fitted by the Boltzmann function is used as the preliminary determined groundwater ecological depth, that is, EWTD ET =8.6m.

[0060] ③EWTD EVI and EWTD ETMin analysis was performed, EWTD = min (7.3, 8.6) = 7.3 m, that is, the ecological burial depth of groundwater in the Balkhash Lake basin is 7.3 m.

[0061] The application of the groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution in the present invention is: it can be applied to assess the groundwater ecological depth in inland river basins in arid areas. The present invention can make up for the deficiency of precipitation, potential evapotranspiration and other hydrological factors interfering with the analysis of groundwater ecological depth in existing arid area ecological water demand assessment research, improve the ecological water demand assessment and guarantee method system of groundwater-dependent terrestrial ecosystems in arid areas, improve the utilization efficiency of ecological and hydrological observation data such as enhanced vegetation index, evapotranspiration, precipitation, groundwater depth, potential evapotranspiration, etc., provide effective tools for the protection of groundwater-dependent terrestrial ecosystems, and provide scientific and technological support for the construction of ecological civilization and sustainable development in ecologically fragile areas of inland river basins.

[0062] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution, characterized in that: The following steps are involved: Step 1: Collect ecological and hydrological observation data of annual maximum enhanced vegetation index, annual evapotranspiration, annual precipitation, annual average groundwater depth, and annual potential evapotranspiration at each sample point in the study area for use in CatBoost regression modeling, SHAP attribution analysis, and groundwater ecological depth assessment; Step 2: Based on the ecological and hydrological observation data in step 1, the CatBoost machine learning algorithm was used to construct the EVI regression model and ET regression model respectively; Step 3: SHAP attribution analysis was performed on the EVI regression model and the ET regression model respectively, and the Boltzmann function was used to establish the correlation between the SHAP value of groundwater depth in the EVI regression model and the groundwater depth, and the correlation between the SHAP value of groundwater depth in the ET regression model and the groundwater depth; Step 4: Carry out inflection point analysis on the SHAP value-groundwater depth relationship curve in the EVI regression model and the SHAP value-groundwater depth relationship curve in the ET regression model, and perform Min analysis on the analysis results to obtain the groundwater ecological depth.

2. The method for assessing groundwater ecological depth by integrating CatBoost regression and SHAP attribution according to claim 1, characterized in that: The step 2 comprises the following steps: ① Taking annual precipitation, annual average groundwater depth, and annual potential evapotranspiration as model inputs, and annual maximum enhanced vegetation index as model output, the CatBoost machine learning algorithm was used to construct the EVI regression model: EVI=f CatBoost,EVI (PRE,WTD,PET) Where, EVI represents the annual maximum enhanced vegetation index; f CatBoost,EVI () represents the EVI regression model constructed using the CatBoost machine learning algorithm; PRE represents annual precipitation, in mm; WTD represents annual average groundwater depth, in m; PET represents annual potential evapotranspiration, in mm; In the process of using the CatBoost machine learning algorithm to build the EVI regression model, 60% of the ecological and hydrological data are used to train the model, recorded as the training set; 20% of the ecological and hydrological data are used to verify the model, recorded as the verification set; 20% of the ecological and hydrological data are used to test the model, recorded as the test set; ② Taking annual precipitation, annual average groundwater depth, and annual potential evapotranspiration as model inputs and annual evapotranspiration as model output, the CatBoost machine learning algorithm was used to construct an ET regression model: ET=f CatBoost,ET (PRE,WTD,PET) Where, ET represents annual evapotranspiration, unit: mm; f CatBoost,ET () represents the ET regression model constructed using the CatBoost machine learning algorithm; PRE represents annual precipitation, in mm; WTD represents annual average groundwater depth, in m; PET represents annual potential evapotranspiration, in mm; In the process of using the CatBoost machine learning algorithm to build the ET regression model, 60% of the ecological and hydrological data are used to train the model, recorded as the training set; 20% of the ecological and hydrological data are used to verify the model, recorded as the validation set; and 20% of the ecological and hydrological data are used to test the model, recorded as the test set.

3. The groundwater ecological depth assessment method integrating CatBoost regression and SHAP attribution according to claim 1 is characterized in that: The step three comprises the following steps: ① SHAP attribution analysis was conducted on the EVI regression model to calculate the SHAP value of groundwater depth in the EVI regression model; the SHAP value of groundwater depth in the EVI regression model quantified the impact of groundwater depth on EVI: the larger the SHAP value of groundwater depth in the EVI regression model, the greater the impact of groundwater depth on EVI; the smaller the SHAP value of groundwater depth in the EVI regression model, the smaller the impact of groundwater depth on EVI; ② SHAP attribution analysis was conducted on the ET regression model to calculate the SHAP value of groundwater depth in the ET regression model; the SHAP value of groundwater depth in the ET regression model quantifies the impact of groundwater depth on ET: the larger the SHAP value of groundwater depth in the ET regression model, the greater the impact of groundwater depth on ET; the smaller the SHAP value of groundwater depth in the ET regression model, the smaller the impact of groundwater depth on ET; ③ Use the Boltzmann function to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the EVI regression model: SHAP EVI,WTD =f Boltzmann,EVI (WTD) In the formula, SHAP EVI,WTD represents the Boltzmann function fitting value of the SHAP value of groundwater depth in the EVI regression model; f Boltzmann,EVI () represents the Boltzmann function fitted by the SHAP value of groundwater depth in the EVI regression model and the scatter points of groundwater depth, which characterizes the correlation between the SHAP value of groundwater depth in the EVI regression model and groundwater depth; WTD represents the annual average groundwater depth, in m; ④ Use the Boltzmann function to fit the correlation function between the SHAP value of groundwater depth and groundwater depth in the ET regression model: SHAP ET,WTD =f Boltzmann,ET (WTD) In the formula, SHAP ET,WTD represents the Boltzmann function fitting value of the SHAP value of groundwater depth in the ET regression model; f Boltzmann,ET () represents the Boltzmann function fitted by the SHAP value of groundwater depth and the scatter points of groundwater depth in the ET regression model, which characterizes the correlation between the SHAP value of groundwater depth in the ET regression model and the groundwater depth; WTD represents the annual average groundwater depth, in m.

4. The method for assessing groundwater ecological depth by integrating CatBoost regression and SHAP attribution according to claim 1, characterized in that: The step 4 comprises the following steps: ①The SHAP value-groundwater depth relationship curve in the EVI regression model fitted by the Boltzmann function is inverted S-shaped. The inflection point is determined by the maximum curvature method. Theoretically, there are two inflection points, where the groundwater depth corresponding to the first inflection point is smaller than the groundwater depth corresponding to the second inflection point. On the left side of the first inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the EVI regression model decreases slightly. Between the first inflection point and the second inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the EVI regression model decreases sharply, indicating that the influence of the groundwater depth on EVI is sharply weakened. After reaching the second inflection point, the SHAP value of the groundwater depth in the EVI regression model is close to its minimum value, and decreases slightly with the increase of the groundwater depth on the right side of the second inflection point. The groundwater depth corresponding to the second inflection point of the SHAP value-groundwater depth relationship curve in the EVI regression model fitted by the Boltzmann function is taken as the preliminarily determined groundwater ecological depth, denoted as EWTD EVI , unit: m; ②The SHAP value-groundwater depth relationship curve in the ET regression model fitted by the Boltzmann function is inverted S-shaped. The inflection point is determined by the maximum curvature method. Theoretically, there are two inflection points, where the groundwater depth corresponding to the first inflection point is smaller than the groundwater depth corresponding to the second inflection point. On the left side of the first inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the ET regression model decreases slightly. Between the first inflection point and the second inflection point, as the groundwater depth increases, the SHAP value of the groundwater depth in the ET regression model decreases sharply, indicating that the influence of the groundwater depth on ET is sharply weakened. After reaching the second inflection point, the SHAP value of the groundwater depth in the ET regression model is close to its minimum value, and decreases slightly with the increase of the groundwater depth on the right side of the second inflection point. The groundwater depth corresponding to the second inflection point of the SHAP value-groundwater depth relationship curve in the ET regression model fitted by the Boltzmann function is taken as the preliminarily determined groundwater ecological depth, denoted as EWTD. ET , unit: m; ③EWTD EVI and EWTD ET Min analysis is performed, and the result is the groundwater ecological depth: EWTD=min(EWTD EVI ,EWTD ET ) Where, EWTD represents the ecological depth of groundwater, in meters; min() represents the function of taking the minimum value of the set; EWTD EVI It represents the groundwater ecological depth obtained by analyzing the inflection point of the groundwater depth relationship curve between the SHAP value of the groundwater depth in the EVI regression model, in meters; EWTD ET It represents the groundwater ecological depth obtained by analyzing the inflection point of the groundwater depth relationship curve between the SHAP value of the groundwater depth in the ET regression model, in meters.

Images