Integrating numerical simulation and machine learning for identification of groundwater potential zone and its governing factors in the Minqin Basin, Northwest China

doi:10.13745/j.esf.sf.2025.10.6

Abstract

Abstract:

Oases are critical for maintaining ecological security in arid regions, and their stability is heavily dependent on sustainable groundwater resources. The delineation of groundwater potential zones (GPZs) offers decision-making support for optimal water resource allocation in arid areas. This study focuses on the Minqin Basin, a typical inland river oasis in northwest China, and proposes an integrated approach that combines numerical simulation with machine learning to assess groundwater potential in arid regions. Precise spatial distributions of groundwater depth and two key parameters (hydraulic conductivity and specific yield) were derived from groundwater numerical simulation. Considering 18 influencing factors across five categories (meteorology, hydrology, land use, topography, and geology), six machine learning models were employed to evaluate the spatial characteristics of groundwater potential. Results indicate that the LightGBM (Accuracy: 87.87%; F₁ score: 0.716; AUC: 0.943) achieved the best performance, followed by XGBoost and Random Forest, whereas the Support Vector Machine, K-Nearest Neighbors, and BP Neural Network models demonstrated inferior performance. Subsequently, feature importance analysis using tree-based models (Random Forest, XGBoost, and LightGBM) revealed that groundwater depth (weight: 17.1%-18.5%) was the primary controlling factor. Secondary factors include potential evapotranspiration (12.5%-14.2%), precipitation (8.6%-12.5%), NDVI (6.2%-12.8%), and surface elevation (6.7%-11.4%). The proposed methodology establishes a multi-parameter assessment framework for groundwater potential in arid oasis areas, offering a scientific basis for sustainable groundwater management in water-scarce regions.

Key words: groundwater potential, machine learning, numerical simulation, Minqin Basin, arid regions

CLC Number:

P641
P628.3

ZHOU Feiran, YIN Ziyue, SUN Xiaomin, SONG Jian, YANG Yun, WU Jianfeng. Integrating numerical simulation and machine learning for identification of groundwater potential zone and its governing factors in the Minqin Basin, Northwest China[J]. Earth Science Frontiers, 2026, 33(1): 511-522.

Figures/Tables 10

References 42

[1]	GUO X Y, FENG Q, SI J H, et al. Identifying the origin of groundwater for water resources sustainable management in an Arid Oasis, China[J]. Hydrological Sciences Journal, 2019, 64(10): 1253-1264. DOI URL
[2]	胡汝骥, 樊自立, 王亚俊, 等. 中国西北干旱区的地下水资源及其特征[J]. 自然资源学报, 2002, 15(3): 321-326.
[3]	SONG Y, XUE D, MA B, et al. Farming in arid areas depletes China’s water[J]. Science, 2023, 379(6633): 651.
[4]	陈亚宁, 杨青, 罗毅, 等. 西北干旱区水资源问题研究思考[J]. 干旱区地理, 2012, 35(1): 1-9.
[5]	王玉洁, 秦大河. 气候变化及人类活动对西北干旱区水资源影响研究综述[J]. 气候变化研究进展, 2017, 13(5): 483-493.
[6]	王贵玲, 刘志明, 刘花台, 等. 地下水潜力评价方法[J]. 水文地质工程地质, 2003, 29(1): 63-66.
[7]	SOLOMON S, QUIEL F. Groundwater study using remote sensing and geographic information systems (GIS) in the central highlands of Eritrea[J]. Hydrogeology Journal, 2006, 14(6): 1029-1041. DOI URL
[8]	EL-SOROGY A S, ALHARBI T, AL-KAHTANY K, et al. Identification and validation of groundwater potential zones in Al-Madinah Al-Munawarah, Western Saudi Arabia using remote sensing and GIS techniques[J]. Water, 2024, 16(23): 3421. DOI URL
[9]	SHERBINI R A E, GHAZALA H H, AHMED M A, et al. Mapping groundwater potential zones in the Widyan Basin, Al Qassim, KSA: analytical hierarchy process-based analysis using Sentinel-2, ASTER-DEM, and conventional data[J]. Remote Sensing, 2025, 17(5): 766. DOI URL
[10]	DÍAZ-ALCAIDE S, MARTÍNEZ-SANTOS P. Review: advances in groundwater potential mapping[J]. Hydrogeology Journal, 2019, 27(7): 2307-2324. DOI
[11]	ABRAMS W, GHONEIM E, SHEW R, et al. Delineation of groundwater potential (GWP) in the northern United Arab Emirates and Oman using geospatial technologies in conjunction with Simple Additive Weight (SAW), Analytical Hierarchy Process (AHP), and Probabilistic Frequency Ratio (PFR) techniques[J]. Journal of Arid Environments, 2018, 157: 77-96. DOI URL
[12]	RAZAVI-TERMEH S V, SADEGHI-NIARAKI A, CHOI S. Groundwater potential mapping using an integrated ensemble of three bivariate statistical models with random forest and logistic model tree models[J]. Water, 2019, 11(8): 1596. DOI URL
[13]	MULTANIYA A P, SINHA M K, SAHU K K, et al. Geospatial technique for the delineation of groundwater potential zones using multi-criteria-based AHP and MIF methods[J]. Water Supply, 2024, 4(24): 1024-1047.
[14]	SRINIVASA RAO Y, AND JUGRAN D K. Delineation of groundwater potential zones and zones of groundwater quality suitable for domestic purposes using remote sensing and GIS[J]. Hydrological Sciences Journal, 2003, 48(5): 821-833. DOI URL
[15]	SHINDE S P, BARAI V N, GAVIT B K, et al. Assessment of groundwater potential zone mapping for semi-arid environment areas using AHP and MIF techniques[J]. Environmental Sciences Europe, 2024, 36(1): 87. DOI
[16]	PRASAD P, LOVESON V J, KOTHA M, et al. Application of machine learning techniques in groundwater potential mapping along the west coast of India[J]. GIScience & Remote Sensing, 2020, 57(6): 735-752.
[17]	THANH N N, THUNYAWATCHARAKUL P, NGU N H, et al. Global review of groundwater potential models in the last decade: parameters, model techniques, and validation[J]. Journal of Hydrology, 2022, 614: 128501. DOI URL
[18]	RIAZ M T, RIAZ M T, REHMAN A, et al. An integrated approach of support vector machine (SVM) and weight of evidence (WOE) techniques to map groundwater potential and assess water quality[J]. Scientific Reports, 2024, 14(1): 26186. DOI PMID
[19]	SAMANTARAY S, SAHOO A, GHOSE D K. Assessment of groundwater potential using neural network: a case study[C]. 3rd International Conference on Intelligent Computing and Communication (ICICC). Singapore: Springer Singapore, 2019, 655-664.
[20]	NAGHIBI S A, MORADI DASHTPAGERDI M. Evaluation of four supervised learning methods for groundwater spring potential mapping in Khalkhal region (Iran) using GIS-based features[J]. Hydrogeology Journal, 2017, 25(1): 169-189. DOI URL
[21]	ZABIHI M, POURGHASEMI H R, POURTAGHI Z S, et al. GIS-based multivariate adaptive regression spline and random forest models for groundwater potential mapping in Iran[J]. Environmental Earth Sciences, 2016, 75(8): 665. DOI URL
[22]	SACHDEVA S, KUMAR B. Comparison of gradient boosted decision trees and random forest for groundwater potential mapping in Dholpur (Rajasthan), India[J]. Stochastic Environmental Research and Risk Assessment, 2021, 35(2): 287-306. DOI
[23]	GUO X, GUI X, XIONG H, et al. Critical role of climate factors for groundwater potential mapping in arid regions: insights from random forest, XGBoost, and LightGBM algorithms[J]. Journal of Hydrology, 2023, 621: 129599. DOI URL
[24]	XIONG H, GUO X, WANG Y, et al. Spatial prediction of groundwater potential by various novel boosting-based ensemble learning models in mountainous areas[J]. Geocarto International, 2023, 38(1): 1-32.
[25]	汤奇成, 张捷斌. 西北干旱地区水资源与生态环境保护[J]. 地理科学进展, 2001, 24(3): 226-232.
[26]	刘恒, 钟华平, 顾颖. 西北干旱内陆河区水资源利用与绿洲演变规律研究:以石羊河流域下游民勤盆地为例[J]. 水科学进展, 2001, 12(3): 378-384.
[27]	马兴旺, 李保国, 吴春荣, 等. 绿洲区土地利用对地下水影响的数值模拟分析:以民勤绿洲为例[J]. 资源科学, 2002, 26(2): 49-55.
[28]	马兴旺, 李保国, 吴春荣, 等. 民勤绿洲现状土地利用模式影响下地下水位时空变化的预测[J]. 水科学进展, 2003, 14(1): 85-90.
[29]	贺向丽, 叶懋, 蒋雨彤. 基于潜水-承压水模型的民勤绿洲地下水位预测[J]. 西南石油大学学报(自然科学版), 2015, 37(3): 168-173. DOI
[30]	王晓玮, 邵景力, 甘雨. 基于数值模拟的西北地下水总量控制指标确定研究[J]. 水文地质工程地质, 2017, 44(3): 12-18.
[31]	马金辉, 韩金华, 张艳林. 近10 a来民勤盆地地下水埋深的空间异质性分析[J]. 干旱区地理, 2013, 36(1): 1-7.
[32]	王贵忠, 罗星罡, 徐洲元, 等. 民勤地区水资源特征分析研究[J]. 中国水土保持, 2014(7): 21-22.
[33]	王晓玮, 邵景力, 王卓然, 等. 西北地区地下水水量-水位双控指标确定研究:以民勤盆地为例[J]. 水文地质工程地质, 2020, 47(2): 17-24.
[34]	NIU Z, HE H, ZHU G, et al. An increasing trend in the ratio of transpiration to total terrestrial evapotranspiration in China from 1982 to 2015 caused by greening and warming[J]. Agricultural and Forest Meteorology, 2019, 279: 107701. DOI URL
[35]	彭守璋. 中国1km逐月潜在蒸散发数据集(1901—2023)[Z]. 国家青藏高原科学数据中心, 2024.
[36]	YANG J, DONG J, XIAO X, et al. Divergent shifts in peak photosynthesis timing of temperate and alpine grasslands in China[J]. Remote Sensing of Environment, 2019, 233: 111395. DOI URL
[37]	YANG J, HUANG X. 30 m annual land cover and its dynamics in China from 1990 to 2019[J]. Earth System Science Data, 2021, 13(8): 3907-3925. DOI URL
[38]	宋沛林, 张永强, 姚盼盼, 等. 中国1千米分辨率逐日全天候地表土壤水分数据集(2003—2023)[Z]. 国家青藏高原科学数据中心, 2024.
[39]	李海涛, 许学工, 肖笃宁. 民勤绿洲水资源利用分析[J]. 干旱区研究, 2007, 24(3): 287-295.
[40]	李元红, 王以兵. 石羊河流域高效节水灌溉与农业种植结构调整[J]. 中国水利, 2013, 48(5): 52-54.
[41]	郑永山, 杨正华. 甘肃省石羊河流域地下水管理对策分析[J]. 地下水, 2016, 38(5): 119-121.
[42]	柴会霞, 谭立海, 潘加朋. 气候变化背景下黑河流域水文及水资源特征[J]. 水利水电技术(中英文), 2023, 54(增刊2): 98-104.

影响因素	评价指标	指标缩写	原始空间分辨率	数据来源
气象因素	降水	PRE	1 km×1 km	国家地球系统科学数据中心
	实际蒸散发	ET	0.05°×0.05°	Niu等^[34]
	潜在蒸散发	PET	1 km×1 km	彭守璋等^[35]
水文因素	地下水埋深	GWD	800 m×800 m	数值模型
	地形湿度指数	TWI	12.5 m×12.5 m	数字高程模型空间分析
	河流强度指数	SPI	12.5 m×12.5 m	数字高程模型空间分析
	河网密度	DD	1 km×1 km	数字高程模型空间分析
土地利用因素	归一化植被指数	NDVI	30 m×30 m	Parizi等^[36]
土地利用因素	土地利用类型	LUCC	30 m×30 m	Yang等^[37]
地形因素	地表高程	DEM	12.5 m×12.5 m	ALOS PALSAR数据
	坡度	Slope	12.5 m×12.5 m	数字高程模型空间分析
	坡向	Asp	12.5 m×12.5 m	数字高程模型空间分析
	泥沙输送指数	STI	12.5 m×12.5 m	数字高程模型空间分析
	地形位置指数	TPI	12.5 m×12.5 m	数字高程模型空间分析
地质因素	土壤水分	SM	0.05°×0.05°	宋沛林等^[38]
	土壤类型	SC	0.5'×0.5'	世界土壤数据库
	渗透系数	K	—	数值模型
	给水度	SY	—	数值模型

影响因素	评价指标	指标缩写	原始空间分辨率	数据来源
气象因素	降水	PRE	1 km×1 km	国家地球系统科学数据中心
	实际蒸散发	ET	0.05°×0.05°	Niu等^[34]
	潜在蒸散发	PET	1 km×1 km	彭守璋等^[35]
水文因素	地下水埋深	GWD	800 m×800 m	数值模型
	地形湿度指数	TWI	12.5 m×12.5 m	数字高程模型空间分析
	河流强度指数	SPI	12.5 m×12.5 m	数字高程模型空间分析
	河网密度	DD	1 km×1 km	数字高程模型空间分析
土地利用因素	归一化植被指数	NDVI	30 m×30 m	Parizi等^[36]
土地利用因素	土地利用类型	LUCC	30 m×30 m	Yang等^[37]
地形因素	地表高程	DEM	12.5 m×12.5 m	ALOS PALSAR数据
	坡度	Slope	12.5 m×12.5 m	数字高程模型空间分析
	坡向	Asp	12.5 m×12.5 m	数字高程模型空间分析
	泥沙输送指数	STI	12.5 m×12.5 m	数字高程模型空间分析
	地形位置指数	TPI	12.5 m×12.5 m	数字高程模型空间分析
地质因素	土壤水分	SM	0.05°×0.05°	宋沛林等^[38]
	土壤类型	SC	0.5'×0.5'	世界土壤数据库
	渗透系数	K	—	数值模型
	给水度	SY	—	数值模型

模型	超参数	搜索空间	最优配置
SVC	kernel	[‘rbf’, ‘linear’, ‘poly’]	‘rbf’
	C	loguniform(1e-2, 1e3)	3
	gamma	loguniform(1e-4, 1e1)	0.1
KNN	n_neighbors	[3, 5, 7, 9, 11, 13, 15]	7
	metric	[‘euclidean’, ‘minkowski’, ‘manhattan’]	‘minkowski’
	p	[1, 2, 3]	1
BP	hidden_layer_sizes	[(100,), (50, 50, 50), (200,), (100, 50, 50)]	(100, 50, 50)
	activation	[‘relu’, ‘tanh’, ‘logistic’]	‘relu’
	solver	[‘adam’, ‘sgd’, ‘lbfgs’]	‘adam’
	learning_rate_init	loguniform(1e-4, 1e-1)	0.003
	alpha	loguniform(1e-5, 1e-1)	0.02
RF	n_estimators	randint(100, 1000)	306
	max_depth	range(10, 50)	29
	min_samples_split	randint(2, 10)	4
	min_samples_leaf	randint(1, 10)	2
	min_impurity_decrease	uniform(0, 0.1)	0.0001
XGBoost	n_estimators	randint(100, 1000)	515
	learning_rate	uniform(0.01, 0.3)	0.06
	max_depth	randint(3, 30)	20
	colsample_bytree	uniform(0.5, 1)	0.9
	subsample	uniform(0.6, 1.0)	0.765
	reg_alpha	uniform(0, 1)	0.6
	reg_lambda	uniform(0, 1)	0.5
	early_stopping_rounds	[10]	10
LightGBM	n_estimators	randint(100, 1000)	301
	learning_rate	uniform(0.01, 0.3)	0.204
	max_depth	randint(3, 20)	14
	num_leaves	randint(10, 50)	45
	subsample	uniform(0.7, 1.0)	0.788
	reg_alpha	uniform(0, 1.0)	0.598
	reg_lambda	uniform(0, 1.0)	0.922
	early_stopping_rounds	[10]	10

模型	超参数	搜索空间	最优配置
SVC	kernel	[‘rbf’, ‘linear’, ‘poly’]	‘rbf’
	C	loguniform(1e-2, 1e3)	3
	gamma	loguniform(1e-4, 1e1)	0.1
KNN	n_neighbors	[3, 5, 7, 9, 11, 13, 15]	7
	metric	[‘euclidean’, ‘minkowski’, ‘manhattan’]	‘minkowski’
	p	[1, 2, 3]	1
BP	hidden_layer_sizes	[(100,), (50, 50, 50), (200,), (100, 50, 50)]	(100, 50, 50)
	activation	[‘relu’, ‘tanh’, ‘logistic’]	‘relu’
	solver	[‘adam’, ‘sgd’, ‘lbfgs’]	‘adam’
	learning_rate_init	loguniform(1e-4, 1e-1)	0.003
	alpha	loguniform(1e-5, 1e-1)	0.02
RF	n_estimators	randint(100, 1000)	306
	max_depth	range(10, 50)	29
	min_samples_split	randint(2, 10)	4
	min_samples_leaf	randint(1, 10)	2
	min_impurity_decrease	uniform(0, 0.1)	0.0001
XGBoost	n_estimators	randint(100, 1000)	515
	learning_rate	uniform(0.01, 0.3)	0.06
	max_depth	randint(3, 30)	20
	colsample_bytree	uniform(0.5, 1)	0.9
	subsample	uniform(0.6, 1.0)	0.765
	reg_alpha	uniform(0, 1)	0.6
	reg_lambda	uniform(0, 1)	0.5
	early_stopping_rounds	[10]	10
LightGBM	n_estimators	randint(100, 1000)	301
	learning_rate	uniform(0.01, 0.3)	0.204
	max_depth	randint(3, 20)	14
	num_leaves	randint(10, 50)	45
	subsample	uniform(0.7, 1.0)	0.788
	reg_alpha	uniform(0, 1.0)	0.598
	reg_lambda	uniform(0, 1.0)	0.922
	early_stopping_rounds	[10]	10

模型方案	评价指标	传统机器学习模型			集成学习模型
模型方案	评价指标	SVC	KNN	BP	RF	XGBoost	LightGBM
方案A	OA/%	78.55	80.19	80.08	85.66	86.88	87.87
	F₁	0.481	0.588	0.582	0.692	0.725	0.716
	AUC	0.772	0.816	0.832	0.914	0.951	0.943
方案B	OA/%	73.40	79.89	78.45	82.30	85.23	85.57
	F₁	0.131	0.558	0.533	0.595	0.686	0.705
	AUC	0.626	0.807	0.800	0.875	0.920	0.917
方案C	OA/%	74.12	77.07	75.18	83.46	77.85	79.56
	F₁	0.008	0.509	0.326	0.632	0.448	0.523
	AUC	0.686	0.722	0.732	0.874	0.811	0.820