Spatial prediction of groundwater potential by various novel boosting-based ensemble learning models in mountainous areas

Figures & data

Table 1. BEMs and top performance models that have been used in GWPM.

Studies	Models	Top performance models
Rasool et al. (2022)	ANN, RF, SVM, KNN, NB, XGBoost	XGBoost, RF
Chen et al. (2020)	RF-J48, AB-J48, Bag-J48, RS-J48, Dag-J48	RF-J48, AB-J48
Naghibi et al. (2020)	RF, PRF, XGB-Linear, XGB-DART, XGB-Tree	PRF, XGB-Linear
Naghibi et al. (2016)	BRT, CART, RF	BRT
Chen et al. (2022)	BT, ANN, DLNN, DLT, DB	DB, BT
Sachdeva and Kumar (2021)	GBDT, RF	GBDT
Park and Kim (2021)	RF, GBM, XGBoost	XGBoost
Kamali Maskooni et al. (2020)	LMT, DB, BRT, RF, KNN	LMT, DB
Mousavi et al. (2017)	BRT, FR	BRT
Naghibi and Pourghasemi (2015)	CART, BRT, RF, EBF, GLM	CART, BRT
Prasad et al. (2020)	RF, BRT, RF-SVM	RF
Wei et al. (2022)	BMA, RF, GBDT	GBDT
Al-Fugara et al. (2020)	RF, BRT, SVM, MDA, MARS	MDA, RF
Yen et al. (2021)	AB, Bag, MB, RTF, LWL	ABLWL, MBLWL
Naghibi et al. (2018)	RF, SVM, BRT, FDA, MARS, QDA, LDA, PDA, KNN, ANN	RF, SVM, BRT
Lee et al. (2019)	BT, FR	BT
Yousefi et al. (2020)	RF-BRT, MARS-SVM, FDA-GLM-MDA	RF-BRT
Kim et al. (2019)	BRT, RF, LR	LR
Naghibi et al. (2017b)	FREM, AB, Bag, GAM, NB	FREM, Bag

Note: Bold texts refer to BEMs and abbreviations have been modified from original authors (ANN: Artificial Neural Network; RF: Random Forest; SVM: Support Vector Machine; KNN: K-Nearest Neighbor; NB: Naïve Bayes; XGBoost: Extreme Gradient Boosting; J48: J48 Decision Trees; AB: AdaBoost; Bag: Bagging; Dag: RS: RandomSubSpace; Dag: Dagging; PRF: Parallel Random Forest; XGB-Linear: Extreme Gradient Boosting with linear booster; XGB-DART: Extreme Gradient Boosting with DART booster; XGB-Tree: Extreme Gradient Boosting with tree booster; BRT: Boosted Regression Tree; CART: Classification and Regression Trees; BT: Boosted Tree; DLNN: Deep Learning Neural Network; DLT: Deep Learning Tree; DB: Deep Boosting; GBDT: Gradient Boosting Decision Tree; GBM: Gradient Boosting Machine; LMT: Boosted Regression Tree; DB: Deep Boosting; GLM: General Linear Model; EBF: Evidential Belief Function; BMA: Bayesian Model Averaging; MDA: Mixture Discriminant Analysis; MARS: Multivariate Adaptive Regression Spline; MB: Multi Boost; RTF: Rotation Forest; LWL: Locally Weighed Learning; FDA: Flexible Discriminant Analysis; QDA: Quadratic Discriminant Analysis; PDA: Penalized Discriminant Analysis; FREM: Frequency Ratio Data Mining; GAM: Generalized Additive Model).

Figure 1. Study area.

Figure 2. Methodology framework.

Table 2. Data sources and indicators.

Indicator types	Indicators	Abbr.	Source	Data type	Scale
Geological indicators	Geology	GEO	RESDC	Shapefile (polygon)	1:250,000
	Lineament density	LD	RESDC	Shapefile (line)	1:250,000
	Distance to faults	DTF	RESDC	Shapefile (line)	1:250,000
	Soil texture	ST	HWSD V1.2	Bil & Mdb	1 km × 1 km
Climate indicators	Ground surface temperature	GST	RESDC	Raster	1 km × 1 km
	Precipitation	PRE	RESDC	Raster	1 km × 1 km
	Evaporation	EVA	RESDC	Raster	1 km × 1 km
Hydrological indicators	Topographic wetness index	TWI	GDEM V2	Raster	30 m ×30 m
	Stream power index	SPI	GDEM V2	Raster	30 m ×30 m
	Drainage density	DD	GDEM V2	Raster	30 m ×30 m
	Distance to rivers	DTRI	NCSGI	Shapefile (line)	1:250,000
	River density	RID	NCSGI	Shapefile (line)	1:250,000
Land use and human activity indicators	Land use/land cover	LULC	FROM-GLC	Raster	30 m ×30 m
	Normalized difference vegetation index	NDVI	RESDC	Raster	30 m ×30 m
	Distance to roads	DTRO	NCSGI	Shapefile (line)	1:250,000
	Road density	ROD	NCSGI	Shapefile (line)	1:250,000
Topographic indicators	Elevation	ELE	GDEM V2	Raster	30 m ×30 m
	Slope gradient	SG	GDEM V2	Raster	30 m ×30 m
	Aspect	ASP	GDEM V2	Raster	30 m ×30 m
	Plan curvature	PLC	GDEM V2	Raster	30 m ×30 m
	Profile curvature	PRC	GDEM V2	Raster	30 m ×30 m
	Terrain ruggedness index	TRI	GDEM V2	Raster	30 m ×30 m
	Topographic position index	TPI	GDEM V2	Raster	30 m ×30 m
	Slope length	SL	GDEM V2	Raster	30 m ×30 m

Note: RESDC: Resource and Environment Science and Data Center; HWSD V1.2: Harmonized world soil database Version 1.2; GDEM V2: Global Digital Elevation Model Version 2; FROM-GLC: Finer Resolution Observation and Monitoring – Global Land Cover; NCSGI: National Catalogue Service for Geographic Information.

Figure 3. Geological parameters. (a) GEO; (b) LD; (c) DTF; (d) ST.

Figure 4. Climate parameters. (a) GST; (b) PRE; (c) EVA.

Figure 5. Hydrological parameters. (a) TWI; (b) SPI; (c) DD; (d) RID; (e) DTRI.

Figure 6. Topographic parameters. (a) ELE; (b) SG; (c) ASP; (d) PLC; (e) PRC; (f) TRI; (g) TPI; (h) SL.

Figure 7. Land use and human activity parameters. (a) LULC; (b) NDVI; (c) DTRO; (d) ROD.

Figure 8. GWP maps. (a) RF; (b) AdaBoost; (c) GBDT; (d) CatBoost; (e) XGBoost; (f) LightGBM.

Table 3. Hyperparameter optimization results of different models.

Model	Hyperparameters	Hyperparameter space	Optimum
RF	n_estimators	hp.choice('n_estimators’,range(1,500))	10
	max_depth	hp.choice('max_depth’,range(1,20))	10
	min_samples_leaf	hp.choice('min_samples_leaf’,range(1,5))	2
AdaBoost	n_estimators	hp.choice('n_estimators’,range(1,500))	56
	learning_rate	hp.uniform('learning_rate’, 0.01, 0.5)	0.0659
	algorithm	hp.choice('algorithm’, ['SAMME', 'SAMME.R'])	SAMME.R
GBDT	num_leaves	hp.choice(‘num_leaves’,range(15,128))	46
	learning_rate	hp.uniform('learning_rate’, 0.01, 0.5)	0.4160
	bagging_freq	hp.choice(‘bagging_freq’,range(4,7))	4
	bagging_fraction	hp.uniform('bagging_fraction’, 0.5, 0.9)	0.6002
	feature_fraction	hp.uniform('feature_fraction’, 0.5, 0.9)	0.5893
CatBoost	learning_rate	hp.uniform('learning_rate’, 0.01, 0.5)	0.2598
	depth	hp.choice('depth’, range(3, 12))	7
	l2_leaf_reg	hp.loguniform('l2_leaf_reg’, −3, 5)	2.7504
	iterations	hp.choice('iterations’, range(100, 1000, 100))	500
	border_count	hp.choice('border_count’, range(32, 256, 32))	224
XGBoost	eta	hp.uniform(‘eta’, 0.001,0.5)	0.3165
	gamma	hp.uniform(‘gamma’, 0.1,0.5)	0.1478
	max_depth	hp.choice(‘max_depth’, [2,3,4,5,6,7,8,9,10])	5
	min_child_weight	hp.choice(‘min_child_weight’, [1,2,3,4,5,6,7,8])	2
	reg_alpha	hp.choice(‘reg_alpha’, [0,0.01,0.05,0.1,0.5])	0.05
	subsample	hp.uniform(‘subsample’, 0,0.9)	0.3809
LightGBM	bagging_fraction	hp.uniform('bagging_fraction’, 0.5, 0.9)	0.6002
	bagging_freq	hp.choice(‘bagging_freq’, range(4,7))	4
	feature_fraction	hp.uniform('feature_fraction’, 0.5, 0.9)	0.5893
	learning_rate	hp.uniform('learning_rate’, 0.01, 0.5)	0.4160
	num_leaves	hp.choice(‘num_leaves’, range(15,128))	46

Table 4. Areas of GWP levels in different models.

GWP levels	RF	AdaBoost	GBDT	CatBoost	XGBoost	LightGBM
Very low (km^2)	437.47	107.31	88.15	448.28	1.06	538.20
Low (km^2)	539.84	293.43	360.26	544.10	262.63	439.07
Moderate (km^2)	558.74	437.51	711.86	523.82	908.68	425.12
High (km^2)	508.86	438.14	575.72	474.23	267.80	503.27
Very High (km^2)	258.35	1026.88	567.28	312.85	863.11	538.20

Table 5. Field validation results.

Model	Levels	Number of wells/boreholes	Average yields (L/s)
XGBoost	Very low GWP areas	0	/
	Low GWP areas	6	0.9852
	Moderate GWP areas	4	2.3589
	High GWP areas	2	0.4110
	Very high GWP areas	10	13.2727
LGBM	Very low GWP areas	0	/
	Low GWP areas	3	2.7773
	Moderate GWP areas	5	0.7534
	High GWP areas	7	5.7157
	Very high GWP areas	7	13.8267
CatBoost	Very low GWP areas	0	/
	Low GWP areas	2	2.4989
	Moderate GWP areas	8	0.7924
	High GWP areas	2	3.8250
	Very high GWP areas	10	12.9909
GBDT	Very low GWP areas	0	/
	Low GWP areas	0	/
	Moderate GWP areas	6	1.6238
	High GWP areas	10	7.3416
	Very high GWP areas	6	10.9561
AdaBoost	Very low GWP areas	0	/
	Low GWP areas	1	0.0401
	Moderate GWP areas	5	1.8280
	High GWP areas	2	2.0879
	Very high GWP areas	14	9.6814
RF	Very low GWP areas	1	3.3078
	Low GWP areas	4	1.2847
	Moderate GWP areas	7	6.0477
	High GWP	1	15.1999
	Very high GWP	9	9.2127

Figure 9. Model validation results. (a) RF; (b) AdaBoost; (c) GBDT; (d) CatBoost; (e) XGBoost; (f) LightGBM.

Figure 10. Importance analysis of indicators. (a) RF; (b) AdaBoost; (c) GBDT; (d) CatBoost; (e) XGBoost; (f) LightGBM.

Figure 11. Correlation analysis of important indicators. (a) Random points; (b) Pearson correlation coefficient result; (c) Spearman correlation coefficient result; (d) Kendall correlation coefficient result.

Figure 12. The determination of groundwater management priority levels. (a) Population distribution; (b) Areas for high and very high GWP; (c) The matrix for the combination of population and GWP map; (d) Groundwater management priority levels in Luoning County.

Rasool U, Yin XN, Xu ZX, Rasool MA, Senapathi V, Hussain M, Siddique J, Trabucco JC. 2022. Mapping of groundwater productivity potential with machine learning algorithms: a case study in the provincial capital of Baluchistan, Pakistan. Chemosphere. 303(Pt 3):135265. doi: 10.1016/j.chemosphere.2022.135265.

 PubMedGoogle Scholar

Chen W, Zhao X, Tsangaratos P, Shahabi H, Ilia I, Xue W, Wang X, Ahmad BB. 2020. Evaluating the usage of tree-based ensemble methods in groundwater spring potential mapping. J Hydrol. 583:124602. doi: 10.1016/j.jhydrol.2020.124602.

 Web of Science ®Google Scholar

Naghibi SA, Hashemi H, Berndtsson R, Lee S. 2020. Application of extreme gradient boosting and parallel random forest algorithms for assessing groundwater spring potential using DEM-derived factors. J Hydrol. 589:125197. doi: 10.1016/j.jhydrol.2020.125197.

 Web of Science ®Google Scholar

Naghibi SA, Pourghasemi HR, Dixon B. 2016. GIS-based groundwater potential mapping using boosted regression tree, classification and regression tree, and random forest machine learning models in Iran. Environ Monit Assess. 188(1):44. doi: 10.1007/s10661-015-5049-6.

 PubMed Web of Science ®Google Scholar

Chen Y, Chen W, Chandra Pal S, Saha A, Chowdhuri I, Adeli B, Janizadeh S, Dineva AA, Wang X, Mosavi A. 2022. Evaluation efficiency of hybrid deep learning algorithms with neural network decision tree and boosting methods for predicting groundwater potential. Geocarto Int. 37(19):5564–5584. doi: 10.1080/10106049.2021.1920635.

 Web of Science ®Google Scholar

Sachdeva S, Kumar B. 2021. Comparison of gradient boosted decision trees and random forest for groundwater potential mapping in Dholpur (Rajasthan), India. Stoch Environ Res Risk Assess. 35(2):287–306. doi: 10.1007/s00477-020-01891-0.

 Web of Science ®Google Scholar

Park S, Kim J. 2021. The predictive capability of a novel ensemble tree-based algorithm for assessing groundwater potential. Sustainability. 13(5):2459. doi: 10.3390/su13052459.

 Web of Science ®Google Scholar

Kamali Maskooni E, Naghibi SA, Hashemi H, Berndtsson R. 2020. Application of advanced machine learning algorithms to assess groundwater potential using remote sensing-derived data. Remote Sens. 12(17):2742. doi: 10.3390/rs12172742.

 Google Scholar

Mousavi SM, Golkarian A, Naghibi SA, Kalantar B, Pradhan B. 2017. GIS-based groundwater spring potential mapping using data mining boosted regression tree and probabilistic frequency ratio models in Iran. AIMS Geosci. 3(1):91–115. doi: 10.3934/geosci.2017.1.91.

 Web of Science ®Google Scholar

Naghibi SA, Pourghasemi HR. 2015. A comparative assessment between three machine learning models and their performance comparison by bivariate and multivariate statistical methods in groundwater potential mapping. Water Resour Manage. 29(14):5217–5236. doi: 10.1007/s11269-015-1114-8.

 Web of Science ®Google Scholar

Prasad P, Loveson VJ, Kotha M, Yadav R. 2020. Application of machine learning techniques in groundwater potential mapping along the west coast of India. GISci. Remote Sens. 57(6):735–752. doi: 10.1080/15481603.2020.1794104.

 Web of Science ®Google Scholar

Wei A, Li D, Bai X, Wang R, Fu X, Yu J. 2022. Application of machine learning to groundwater spring potential mapping using averaging, bagging, and boosting techniques. Water Supply. 22(8):6882–6894. doi: 10.2166/ws.2022.283.

 Google Scholar

Al-Fugara A, Pourghasemi HR, Al-Shabeeb AR, Habib M, Al-Adamat R, Ai-Amoush H, Collins AL. 2020. A comparison of machine learning models for the mapping of groundwater spring potential. Environ Earth Sci. 79(10):19.

 Web of Science ®Google Scholar

Yen HPH, Pham BT, Van Phong T, Ha DH, Costache R, Van Le H, Nguyen HD, Amiri M, Van Tao N, Prakash I. 2021. Locally weighted learning based hybrid intelligence models for groundwater potential mapping and modeling: a case study at Gia Lai province, Vietnam. Geosci Front. 12(5):101154. doi: 10.1016/j.gsf.2021.101154.

 Web of Science ®Google Scholar

Naghibi SA, Pourghasemi HR, Abbaspour K. 2018. A comparison between ten advanced and soft computing models for groundwater qanat potential assessment in Iran using R and GIS. Theor Appl Climatol. 131(3–4):967–984. doi: 10.1007/s00704-016-2022-4.

 Web of Science ®Google Scholar

Lee S, Hyun Y, Lee M-J. 2019. Groundwater potential mapping using data mining models of big data analysis in Goyang-si, South Korea. Sustainability. 11(6):1678. doi: 10.3390/su11061678.

 Web of Science ®Google Scholar

Yousefi S, Sadhasivam N, Pourghasemi HR, Nazarlou HG, Golkar F, Tavangar S, Santosh M. 2020. Groundwater spring potential assessment using new ensemble data mining techniques. Measurement. 157:107652. doi: 10.1016/j.measurement.2020.107652.

 Web of Science ®Google Scholar

Kim J-C, Jung H-S, Lee S. 2019. Spatial mapping of the groundwater potential of the geum river basin using ensemble models based on remote sensing images. Remote Sens. 11(19):2285. doi: 10.3390/rs11192285.

 Google Scholar

Naghibi SA, Moghaddam DD, Kalantar B, Pradhan B, Kisi O. 2017b. A comparative assessment of GIS-based data mining models and a novel ensemble model in groundwater well potential mapping. J Hydrol. 548:471–483. doi: 10.1016/j.jhydrol.2017.03.020.

 Web of Science ®Google Scholar

Data availability

The authors do not have permission to share data.