Professional Documents
Culture Documents
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
Research papers
A R T I C LE I N FO A B S T R A C T
This manuscript was handled by Emmanouil Freshwater shortages have become much more common globally in recent years. Water resources that are
Anagnostou, Editor-in-Chief naturally available beneath the surface are capable of reversing this condition. Spatial modeling of groundwater
Keywords: distribution is an important undertaking that would aid in subsequent conservation and management of
Groundwater potential mapping groundwater resources. In this study, groundwater potential maps were developed using a machine learning
Support vector regression (SVR) algorithm (MLA) and a deep learning algorithm (DLA), specifically the support vector regression (SVR) and
Convolution neural network (CNN) convolution neural network (CNN) functions, respectively. Initially, 140 groundwater datasets were created
through an extensive survey and then arbitrarily divided into groups of 100 (70%) and 40 (30%) datasets for
model calibration and testing, respectively. Next, 15 groundwater conditioning factors (GCFs), including
catchment area (CA), convergence index (CI), convexity (Co), diurnal anisotropic heating (DH), flow path (FP),
slope angle (SA), slope height (SH), topographic position index (TPI), terrain ruggedness index (TRI), slope
length (LS) factor, mass balance index (MBI), texture (TX), valley depth (VD), land cover (LC), and geology (GG)
were produced and applied for model training. Finally, the calibrated model was validated using both training
and testing data, and the independent measure of the receiver operating characteristic-area under the curve
(ROC-AUC). For validation using training data, the AUC values of CNN and SVR were 0.844 and 0.75, whereas
those of CNN and SVR during validation with the testing data were 0.843 and 0.75. Therefore, CNN has better
predictive ability than SVR. The findings of this study will help policymakers develop better strategies for
conservation and management of groundwater resources.
⁎
Corresponding authors.
E-mail addresses: hr.pourghasemi@shirazu.ac.ir, hamidreza.pourghasemi@yahoo.com (H.R. Pourghasemi), leesaro@kigam.re.kr (S. Lee).
https://doi.org/10.1016/j.jhydrol.2020.125033
Received 22 February 2020; Received in revised form 27 April 2020; Accepted 29 April 2020
Available online 07 May 2020
0022-1694/ © 2020 Elsevier B.V. All rights reserved.
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
recently, integrated RS (remote sensing) and GIS (geographic in- 35°06′52″ N to 35°13′33″N, and the longitude range of 127°01′27″ E to
formation system) approaches have been applied in studies of 127°07′24″ E (WGS_1984_UTM_52). Based on a topographic map of the
groundwater worldwide (Celik and Aslan, 2020; Andualem and study area, the elevation of the Damyang area ranges between 0 m and
Demeke, 2019; Moghaddam et al., 2015; Rahmati et al., 2015; Sener 955 m, with slope angles from 0° to 63.03° (Fig. 4). The Yeongsan River
et al., 2005; Shahid et al., 2000). RS approaches offer inexpensive da- is the main river of the Damyang area. The average annual temperature
tasets with wide coverage of all regions over a short period of time, is 13.52 °C, with a winter average of 2.6 °C and a summer average of
whereas GIS is regarded as the ultimate tool for mapping of CFs, and is 24.8 °C. Mean annual precipitation is 108.58 mm/year, and the rainy
therefore very helpful in assessing groundwater resources (Kharazmi season lasts from June to September (https://web.kma.go.kr/). The
et al., 2018). Several probabilistic and knowledge-based approaches main land cover types in the Damyang area are forest, agricultural land,
have been integrated with GIS techniques to identify possible ground- and inhabited areas, which occupy 59.5, 29.3, and 4.07% of the land
water areas: weight of evidence (WOF: Arabameri et al., 2019), cata- area, respectively. The area includes 61 geologic groups and units, and
strophe theory (Kaur et al., 2020), logistic regression (Lee et al., 2018), the main lithology includes alluvium, granite, and feldspathic gneiss
Shannon’s entropy (Khoshtinat et al., 2019), frequency ratio (FR: (https://www.kigam.re.kr/).
Khoshtinat et al., 2019), index of entropy (Al-Abadi et al., 2016), cer-
tainty factor (CF: Azareh et al., 2019; Hou et al., 2018), evidential belief 3. Methodology
function (EBF: Pourghasemi and Beheshtirad, 2015), and multi-criteria
decision-making (MCDM: Arabameri et al., 2019; Çelik, 2019; Fig. 2 presents the multi-phase methodological framework used in
Chowdhury et al., 2010). this study.
Recently, scholars have recognized the superior efficiency of ma-
chine learning algorithms (MLAs) over statistical approaches (Chen 3.1. Generation of input data
et al., 2019b; Pham et al., 2018). MLAs can manage large datasets that
have diverse formats, are non-separable, and are obtained from dis- 3.1.1. Groundwater inventory map (GIM)
similar sources. Those employed for groundwater potential investiga- A thorough understanding of the associations among groundwater
tion include boosted regression tree (BRT: Naghibi et al., 2018; Naghibi conditioning factors (GCFs) and the potential for groundwater using a
et al., 2016), classification and regression tree (CART: Choubin et al., GIM is extremely important for GMS (Naghibi et al., 2016). Here, a GIM
2019), multivariate adaptive regression spline (MARS: Golkarian et al., is constructed using wide-ranging field measurements. First, it is ne-
2018; Zabihi et al., 2016), support vector machine (SVM: Lee et al., cessary to prepare non-groundwater data that are equivalent to the
2018; Naghibi et al., 2018), random forest (RF: Naghibi et al., 2018), groundwater data used for GMS. Through arbitrary separation, both the
mixture discriminant analysis (Kalantar et al., 2019), and linear dis- groundwater and non-groundwater data are split into calibration and
criminant analysis (Kalantar et al., 2019) and general additive model testing sets in the proportion of 70:30 (Pourghasemi and Beheshtirad,
(GAM: Naghibi et al., 2017). Combinations of two MLAs and fusions of 2015; Termeh et al., 2019). The groundwater and non-groundwater
conservative statistical methods with MLA outperform the precision training data are employed for calibration of the models, while the
attained by any individual MLA (Aghdam et al., 2017; Chen et al., groundwater and non-groundwater testing data are utilized for model
2019b; Dodangeh et al., 2020; Khosravi et al., 2019; Lee et al., 2019). confirmation (Chen et al., 2019c).
However, the predictive outcome of these combined methods relies
solely on the precision of the datasets utilized and the internal structure 3.1.2. Dataset preparation for spatial modeling
of the model (Chen et al., 2019c). Studies using the MLAs listed above It is essential to select appropriate GCFs, because these govern the
have gained prominence, but many similar novel algorithms have also forecasting precision of the models (Garosi et al., 2019). In this study,
been successfully implemented in other fields with high precision, and 15 GCFs were employed for GMS, based on earlier investigations:
therefore should be tested for forecasting groundwater potential. catchment area (CA), convergence index (CI), convexity (Co), diurnal
Among such novel algorithms, support vector regression (SVR) follows anisotropic heating (DH), flow path (FP), slope angle (SA), slope height
the regression strategy of SVM, which has been widely utilized in the (SH), topographic position index (TPI), terrain ruggedness index (TRI),
prediction of flash floods and flood stages (Wu et al., 2019), frequency slope length (LS) factor, mass balance index (MBI), texture (TX), valley
of flooding (Sharifi Garmdareh et al., 2018), landslides (Miao et al., depth (VD), land cover (LC), and geology (GG) (Chen et al., 2019a;
2018), and water levels (Sivapragasam and Liong, 2005), but has not Golkarian et al., 2018; Kalantar et al., 2019; Kim et al., 2018; Lee et al.,
been yet applied for predictive groundwater potential modeling. Ad- 2018; Miraki et al., 2019; Ozdemir, 2011; Rahmati et al., 2018;
ditionally, deep learning algorithms (Hinton and Salakhutdinov, 2006; Rahmati et al., 2016; Zhang et al., 2019).
LeCun et al., 2015) have garnered exceptional interest in recent years CA refers to the region discharging water to the opening of a par-
due to their significant capabilities (Zhang et al., 2019). The convolu- ticular catchment; it is a significant factor in determining groundwater
tional neural network (CNN) is an extensively used DLA that has been potential, because it controls flow accumulation (Freeman, 1991;
employed in various fields including forecasting forest fires (Zhang Gruber and Peckham, 2009). The CA index of the study area ranges
et al., 2019), determining flood conditions (Gebrehiwot et al., 2019; between 30 and 97,590 (Fig. 3a). CI is another important topographic
Sarker et al., 2019), object identification (Radovic et al., 2017), image factor that illustrates the curvature of a slope (San, 2014); it indicates
classification, and segmentation (Längkvist et al., 2016; Maggiori et al., convergence or divergence of a cell, with positive values signifying
2016). CNN employs multiple neuron layers to automatically analyze divergent pixels while negative values denote convergent pixels (Kim
objects (Gebrehiwot et al., 2019). Few previous studies have in- et al., 2018; Rahmati et al., 2018). The minimum convergence indicates
vestigated the efficacy of CNN in groundwater spatial modeling (GMS). a high potential for groundwater (Kim et al., 2018). The CI index varies
Therefore, this study was conducted to compare the predictive abilities between maximum divergence (1 0 0) and maximum convergence
of SVR and CNN in groundwater potential mapping and to assess their (−1 0 0) (Fig. 3b). Co is a vital factor for GMS, as it represents the
performance using ROC-AUCs (receiver operating characteristic – area displacement of slope due to gravity (Pourghasemi et al., 2018). High
under the curve) along with training and testing data. Co leads to high runoff and low infiltration. The Co index in the study
area ranges between 4.2 and 75.1 (Fig. 3c). In this study, DH was used
2. Description of the study area to identify groundwater potential, and was computed as follows
(Böhner and Antonić, 2009):
The Damyang area, which extends over 452 km2, is situated in the
south of South Korea (Fig. 1). It falls within the latitude range of Hγ = cos(γmax − γ ) × arctan(λ ) (1)
2
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
where γmax , γ , and λ denote the aspect with maximum aggregate heat PC ⎞0.4 slope ⎞1,3
LS = ⎛FA × × sin ⎛
excess, slope aspect, and gradient, respectively; the DH index of the ⎝ 22.13 ⎠ ⎝ 0.0896 ⎠ (2)
Damyang area ranges between −0.77 and 0.82 (Fig. 3d). FP was em-
ployed in this investigation to determine the groundwater potential, as where FA and PC indicate flow accumulation and pixel size, respec-
it governs the runoff; the FP index ranges between 0 and 3154 (Fig. 3e) tively (Gayen et al., 2019). The LS index ranges from 0 to 193 (Fig. 3f).
in the study area. The LS factor governs the rapidity of overflow and is MBI is considered a significant factor for predicting the occurrence of
assessed using the following expression (Moore and Burch, 1986): groundwater, and ranges from −0.90 to 1.09 (Fig. 3g). SA is indis-
pensable when forecasting groundwater potential, as it governs the
slope inclination, which influences the rapidity of overflow and
3
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
Fig. 3. GCFs: (a) CA, (b) CI, (c) Co, (d) DH, (e) FP, (f) LS, (g) MBI, (h) SA, (i) SH, (k) TPI, (l) TRI, (n) TX, (m) VD, (n) LC, (o) GG.
infiltration (Adiat et al., 2012; Miraki et al., 2019). In this study, the SA SH value indicates lower recharge, and vice versa (Manap et al., 2013).
index ranges from 0 to 1.17° (Fig. 3h). SH is a significant factor for The SH index in the Damyang area ranges from 0 to 451 (Fig. 3i). TPI is
predicting groundwater potential, as it controls the permeation and an influencing factor of GMS (Arulbalaji et al., 2019); it determines the
runoff processes (Manap et al., 2013; Oh and Pradhan, 2011). A higher variance with elevation of each cell (A0) and the average elevation of
4
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
Fig. 3. (continued)
the surrounding area (A) within a fixed radius (R) (De Reu et al., 2013). to 14.68 (Fig. 3j). TRI is assessed by summing the variance in elevation
among a cell and the surrounding cells. Lower TRI values signify higher
TPI = A0 − A (3)
groundwater potential; here, values range from 0 to 51.49 (Fig. 3l). VD
is computed as the variance between elevation and an interpolated
Positive and negative values of TPI indicate higher and lower ele-
ridge level (Peplau and Conrad, 1989); here, VD values range from 0
vations compared to the mean of their surroundings, respectively
and 353.74 (Fig. 3m).
(Choubin et al., 2019). In this research, TPI values range from −12.02
5
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
6
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
Fig. 5. The CNN method: (a) target and outputs for the training set; (b) targets and outputs for the validation dataset; (c) MSE and RMSE for the calibration dataset;
(d) frequency of errors for the calibration dataset; (e) MSE and RMSE for the validation dataset; (f) frequency of errors for the validation dataset.
4.2. Groundwater spatial modeling (MP), low potential (LP), and very low potential (VLP) classes based on
the quantile method (Rahmati et al. 2015; Rahmati et al., 2016). The
In this study, the indices of the generated GMS were categorized GMS indices of CNN ranged from 1.62 to −0.72, with the ranges of
into very high potential (VHP), high potential (HP), moderate potential (1.62)–(0.96), (0.96)–(0.74), (0.74)–(0.49), (0.49)–(0.26), and
7
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
Fig. 6. The SVR method: (a) target and outputs for the training dataset; (b) targets and outputs for the validation dataset; (c) MSE and RMSE for the calibration
dataset; (d) frequency of errors for the validation dataset; (e) MSE and RMSE for the validation dataset; (f) frequency of errors for the validation dataset.
(0.26)–(−0.72) representing the VHP, HP, MP, LP, and VLP classes, Based on the CNN model, the proportions of the total area covered
respectively (Fig. 7). Likewise, the GMS indices of SVR ranged between by the VHP, HP, MP, LP, and VLP categories are 19.96, 20.20, 20.40,
1.45 and −0.25, with ranges for the VHP, HP, MP, LP, and VLP classes 20.12, and 19.32, respectively. Similarly, in the SVR model, 14.84,
of (1.45)–(1.10), (1.10)–(0.83), (0.83)–(0.50), (0.50)–(0.28), and 22.39, 23.09, 20.07, and 19.62 of the area fall into the VHP, HP, MP,
(0.28)–(−0.25), respectively. LP, and VLP classes for groundwater, respectively.
8
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
Model validation using the ROC-AUC curve was carried out with
both training and testing datasets. For AUC using the training and
testing data, the CNN model yielded maximum AUC values of 0.844 and
0.843, while the SVR model yielded AUC values of 0.75 and 0.75, re-
spectively (Figs. 8 and 9). The results of both the training and testing
data AUC values reveal that the CNN has better forecasting accuracy
5. Discussion
9
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
10
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
J. Hydrol.: Regional Studies 24, 100610. Kaur, L., Rishi, M.S., Singh, G., Thakur, S.N., 2020. Groundwater potential assessment of
Ansari, H.R., Gholami, A., 2015. An improved support vector regression model for esti- an alluvial aquifer in Yamuna sub-basin (Panipat region) using remote sensing and
mation of saturation pressure of crude oils. Fluid Phase Equilib. 402, 124–132. GIS techniques in conjunction with analytical hierarchy process (AHP) and cata-
Arabameri, A., Rezaei, K., Cerda, A., Lombardo, L., Rodrigo-Comino, J., 2019. GIS-based strophe theory (CT). Ecol. Ind. 110, 105850.
groundwater potential mapping in Shahroud plain, Iran. A comparison among sta- Kharazmi, R., et al., 2018. Monitoring and assessment of seasonal land cover changes
tistical (bivariate and multivariate), data mining and MCDM approaches. Sci. Total using remote sensing: a 30-year (1987–2016) case study of Hamoun Wetland, Iran.
Environ. 658, 160–177. Environ. Monitor. Assess. 190 (6), 356.
Arulbalaji, P., Padmalal, D., Sreelash, K., 2019. GIS and AHP techniques based delineation Khoshtinat, S., Aminnejad, B., Hassanzadeh, Y., Ahmadi, H., 2019. Groundwater potential
of groundwater potential zones: a case study from southern Western Ghats, India. Sci. assessment of the Sero plain using bivariate models of the frequency ratio, Shannon
Rep. 9 (1), 1–17. entropy and evidential belief function. J. Earth Syst. Sci. 128 (6), 152.
Azareh, A., et al., 2019. Modelling gully-erosion susceptibility in a semi-arid region, Iran: Khosravi, K., et al., 2019. Meteorological data mining and hybrid data-intelligence
investigation of applicability of certainty factor and maximum entropy models. Sci. models for reference evaporation simulation: a case study in Iraq. Comput. Electron.
Total Environ. 655, 684–696. Agric. 167, 105041.
Bayr, U., Puschmann, O., 2019. Automatic detection of woody vegetation in repeat Kim, J.-C., Jung, H.-S., Lee, S., 2018. Groundwater productivity potential mapping using
landscape photographs using a convolutional neural network. Ecol. Inform. 50, frequency ratio and evidential belief function and artificial neural network models:
220–233. focus on topographic factors. J. Hydroinf. 20 (6), 1436–1451.
Böhner, J., Antonić, O., 2009. Land-surface parameters specific to topo-climatology. Dev. Längkvist, M., Kiselev, A., Alirezaie, M., Loutfi, A., 2016. Classification and segmentation
Soil Sci. 33, 195–226. of satellite orthoimagery using convolutional neural networks. Remote Sens. 8 (4),
Brodrick, P.G., Davies, A.B., Asner, G.P., 2019. Uncovering Ecological Patterns with 329.
Convolutional Neural Networks. Trends Ecol. Evol. 34 (8), 734–745. https://doi.org/ LeCun, Y., Bengio, Y., Hinton, G., 2015. Deep learning. Nature 521, 436–444.
10.1016/j.tree.2019.03.006. Lee, S., Hong, S.-M., Jung, H.-S., 2018. GIS-based groundwater potential mapping using
Can, R., Kocaman, S., Gokceoglu, C., 2019. A convolutional neural network architecture artificial neural network and support vector machine models: the case of Boryeong
for auto-detection of landslide photographs to assess citizen science and volunteered city in Korea. Geocarto Int. 33 (8), 847–861.
geographic information data quality. ISPRS Int. J. Geo-Inf. 8 (7). LeCun, Y., Bottou, L., Bengio, Y., Haffner, P., 1998. Gradient-based learning applied to
Çelik, R., 2019. Evaluation of groundwater potential by GIS-based multicriteria decision document recognition. Proceedings of the IEEE 86 (11), 2278–2324.
making as a spatial prediction tool: case study in the Tigris River Batman-Hasankeyf Lee, S., et al., 2019. Sevucas: a novel gis-based machine learning software for seismic
Sub-Basin, Turkey. Water 11 (12), 2630. vulnerability assessment. Appl. Sci. 9 (17), 3495.
Celik, R., Aslan, V., 2020. Evaluation of hydrological and hydrogeological characteristics Lin, L., Weng, F., 2018. Estimation of hurricane maximum wind speed using temperature
affecting the groundwater potential of Harran Basin. Arabian J. Geosci. 13 (4), 1–13. anomaly derived from advanced technology microwave sounder. IEEE Geosci.
Chen, W., et al., 2018. GIS-based groundwater potential analysis using novel ensemble Remote Sens. Lett. 15 (5), 639–643.
weights-of-evidence with logistic regression and functional tree models. Sci. Total Maggiori, E., Tarabalka, Y., Charpiat, G., Alliez, P., 2016. Convolutional neural networks
Environ. 634, 853–867. for large-scale remote-sensing image classification. IEEE Trans. Geosci. Remote Sens.
Chen, W., et al., 2019a. Spatial prediction of groundwater potentiality using ANFIS en- 55 (2), 645–657.
sembled with teaching-learning-based and biogeography-based optimization. J. Manap, M.A., Sulaiman, W.N.A., Ramli, M.F., Pradhan, B., Surip, N., 2013. A knowledge-
Hydrol. 572, 435–448. driven GIS modeling technique for groundwater potential mapping at the Upper
Chen, W., et al., 2019b. Novel hybrid integration approach of bagging-based fisher’s Langat Basin, Malaysia. Arabian J. Geosci. 6 (5), 1621–1637.
linear discriminant function for groundwater potential analysis. Nat. Resour. Res. 28 Miao, F., Wu, Y., Xie, Y., Li, Y., 2018. Prediction of landslide displacement with step-like
(4), 1239–1258. behavior based on multialgorithm optimization and a support vector regression
Chen, W., Tsangaratos, P., Ilia, I., Duan, Z., Chen, X., 2019c. Groundwater spring po- model. Landslides 15 (3), 475–488.
tential mapping using population-based evolutionary algorithms and data mining Miraki, S., et al., 2019. Mapping groundwater potential using a novel hybrid intelligence
methods. Sci. Total Environ. 684, 31–49. approach. Water Resour. Manage. 33 (1), 281–302.
Choubin, B., et al., 2019. Regional groundwater potential analysis using classification and Moghaddam, D.D., Rezaei, M., Pourghasemi, H., Pourtaghie, Z., Pradhan, B., 2015.
regression trees. In: Spatial Modeling in GIS and R for Earth and Environmental Groundwater spring potential mapping using bivariate statistical model and GIS in
Sciences. Elsevier, pp. 485–498. the Taleghan watershed, Iran. Arabian J. Geosci. 8 (2), 913–929.
Chowdhury, A., Jha, M.K., Chowdary, V., 2010. Delineation of groundwater recharge Moore, I.D., Burch, G.J., 1986. Physical basis of the length-slope factor in the universal
zones and identification of artificial recharge sites in West Medinipur district, West soil loss equation 1. Soil Sci. Soc. Am. J. 50 (5), 1294–1298.
Bengal, using RS, GIS and MCDM techniques. Environ. Earth Sci. 59 (6), 1209. Mueller, J., Park, J., Sahu, R., Varadharajan, C., Arora, B., Faybishenko, B., Agarwal, D.,
De Reu, J., et al., 2013. Application of the topographic position index to heterogeneous 2019. Surrogate Optimization of Deep Neural Networks for Groundwater Predictions.
landscapes. Geomorphology 186, 39–49. arXiv preprint arXiv:1908.10947.
Ding, A., Zhang, Q., Zhou, X., Dai, B., 2016. Automatic recognition of landslide based on Naghibi, S.A., Moghaddam, D.D., Kalantar, B., Pradhan, B., Kisi, O., 2017. A comparative
CNN and texture change detection. In: 2016 31st Youth Academic Annual Conference assessment of GIS-based data mining models and a novel ensemble model in
of Chinese Association of Automation (YAC). IEEE, pp. 444–448. groundwater well potential mapping. J. Hydrol. 548, 471–483.
Dodangeh, E., et al., 2020. Integrated machine learning methods with resampling algo- Naghibi, S.A., Pourghasemi, H.R., Abbaspour, K., 2018. A comparison between ten ad-
rithms for flood susceptibility prediction. Sci. Total Environ. 705, 135983. vanced and soft computing models for groundwater qanat potential assessment in
Freeman, T.G., 1991. Calculating catchment area with divergent flow based on a regular Iran using R and GIS. Theor. Appl. Climatol. 131 (3–4), 967–984.
grid. Comput. Geosci. 17 (3), 413–422. Naghibi, S.A., Pourghasemi, H.R., Dixon, B., 2016. GIS-based groundwater potential
Garosi, Y., Sheklabadi, M., Conoscenti, C., Pourghasemi, H.R., Van Oost, K., 2019. mapping using boosted regression tree, classification and regression tree, and random
Assessing the performance of GIS-based machine learning models with different ac- forest machine learning models in Iran. Environ. Monit. Assess. 188 (1), 44.
curacy measures for determining susceptibility to gully erosion. Sci. Total Environ. Nair, V., Hinton, G.E., 2010. Rectified linear units improve restricted boltzmann ma-
664, 1117–1132. chines. In: Proceedings of the 27th international conference on machine learning
Gayen, A., Pourghasemi, H.R., Saha, S., Keesstra, S., Bai, S., 2019. Gully erosion sus- (ICML-10), pp. 807–814.
ceptibility assessment and management of hazard-prone areas in India using different Neshat, A., Pradhan, B., Pirasteh, S., Shafri, H.Z.M., 2014. Estimating groundwater vul-
machine learning algorithms. Sci. Total Environ. 668, 124–138. nerability to pollution using a modified DRASTIC model in the Kerman agricultural
Gebrehiwot, A., Hashemi-Beni, L., Thompson, G., Kordjamshidi, P., Langan, T.E., 2019. area, Iran. Environ. Earth Sci. 71 (7), 3119–3131.
Deep convolutional neural network for flood extent mapping using unmanned aerial Nosrati, K., Van Den Eeckhaut, M., 2012. Assessment of groundwater quality using
vehicles data. Sensors 19 (7), 1486. multivariate statistical techniques in Hashtgerd Plain, Iran. Environ. Earth Sci. 65 (1),
Ghorbanzadeh, O., Blaschke, T., Gholamnia, Kh., Meena, S.R., Tiede, D., Aryal, J., 2019. 331–344.
Evaluation of Different Machine Learning Methods and Deep-Learning Convolutional Oh, H.-J., Pradhan, B., 2011. Application of a neuro-fuzzy model to landslide-suscept-
Neural Networks for Landslide Detection. Remote Sens. 11 (2), 196. https://doi.org/ ibility mapping for shallow landslides in a tropical hilly area. Comput. Geosci. 37 (9),
10.3390/rs11020196. 1264–1276.
Goldman, M., Neubauer, F., 1994. Groundwater exploration using integrated geophysical Owen, R., Gwavava, O., Gwaze, P., 2006. Multi-electrode resistivity survey for ground-
techniques. Surv. Geophys. 15 (3), 331–361. water exploration in the Harare greenstone belt, Zimbabwe. Hydrogeol. J. 14 (1–2),
Golkarian, A., Naghibi, S.A., Kalantar, B., Pradhan, B., 2018. Groundwater potential 244–252.
mapping using C5. 0, random forest, and multivariate adaptive regression spline Ozdemir, A., 2011. GIS-based groundwater spring potential mapping in the Sultan
models in GIS. Environ. Monit. Assess. 190 (3), 149. Mountains (Konya, Turkey) using frequency ratio, weights of evidence and logistic
Gruber, S., Peckham, S., 2009. Land-surface parameters and objects in hydrology. Dev. regression methods and their comparison. J. Hydrol. 411 (3–4), 290–308.
Soil Sci. 33, 171–194. Panahi, M., et al., 2019. GIS-Based SWARA and Its Ensemble by RBF and ICA Data-Mining
Gu, J., et al., 2018. Recent advances in convolutional neural networks. Pattern Recogn. Techniques for Determining Suitability of Existing Schools and Site Selection of New
77, 354–377. School Buildings, Spatial Modeling in GIS and R for Earth and Environmental
Hinton, G.E., Salakhutdinov, R.R., 2006. Reducing the dimensionality of data with neural Sciences. Elsevier, pp. 161–188.
networks. Science 313 (5786), 504–507. Peplau, L.A., Conrad, E., 1989. Beyond nonsexist research: the perils of feminist methods
Hou, E., Wang, J., Chen, W., 2018. A comparative study on groundwater spring potential in psychology. Psychol. Women Quarterly 13 (4), 379–400.
analysis based on statistical index, index of entropy and certainty factors models. Pham, B.T., Hoang, T.-A., Nguyen, D.-M., Bui, D.T., 2018. Prediction of shear strength of
Geocarto Int. 33 (7), 754–769. soft soil using machine learning methods. Catena 166, 181–191.
Kalantar, B., et al., 2019. Optimized conditioning factors using machine learning tech- Pourghasemi, H., Gayen, A., Park, S., Lee, C.-W., Lee, S., 2018. Assessment of landslide-
niques for groundwater potential mapping. Water 11 (9), 1909. prone areas and their zonation using logistic regression, logitboost, and naïvebayes
11
M. Panahi, et al. Journal of Hydrology 588 (2020) 125033
machine-learning algorithms. Sustainability 10 (10), 3697. image recognition. arXiv preprint arXiv:1409.1556.
Pourghasemi, H.R., Beheshtirad, M., 2015. Assessment of a data-driven evidential belief Sivapragasam, C., Liong, S.-Y., 2005. Flow categorization model for improving fore-
function model and GIS for groundwater potential mapping in the Koohrang casting. Hydrol. Res. 36 (1), 37–48.
Watershed, Iran. Geocarto Int. 30 (6), 662–685. Smola, A., Schölkopf, B., 2004. A tutorial on support vector regression, statist. Comput
Pourghasemi, H.R., Mohammady, M., Pradhan, B., 2012. Landslide susceptibility map- 14, 199–222.
ping using index of entropy and conditional probability models in GIS: Safarood Su, H., Li, X., Yang, B., Wen, Z., 2018. Wavelet support vector machine-based prediction
Basin. Iran. Catena 97, 71–84. model of dam deformation. Mech. Syst. Signal Process. 110, 412–427.
Radovic, M., Adarkwa, O., Wang, Q., 2017. Object recognition in aerial images using Szegedy, C., et al., 2015. Going deeper with convolutions. In: Proceedings of the IEEE
convolutional neural networks. J. Imag. 3 (2), 21. Conference on Computer Vision and Pattern Recognition, pp. 1–9.
Rahmati, O., et al., 2018. Groundwater spring potential modelling: Comprising the cap- Termeh, S.V.R., et al., 2019. Optimization of an adaptive neuro-fuzzy inference system for
ability and robustness of three different modeling approaches. J. Hydrol. 565, groundwater potential mapping. Hydrogeol. J. 27 (7), 2511–2534.
248–261. Vapnik, V.N., 1998. Statistical learning theory. Adaptive and learning systems for signal
Rahmati, O., Pourghasemi, H.R., Melesse, A.M., 2016. Application of GIS-based data processing. Commun. Control 2, 1–740.
driven random forest and maximum entropy models for groundwater potential Wang, J., Li, L., Niu, D., Tan, Z., 2012. An annual load forecasting model based on support
mapping: a case study at Mehran Region, Iran. Catena 137, 360–372. vector regression with differential evolution algorithm. Appl. Energy 94, 65–70.
Rahmati, O., Samani, A.N., Mahdavi, M., Pourghasemi, H.R., Zeinivand, H., 2015. Wang, Y., Fang, Z., Hong, H., 2019. Comparison of convolutional neural networks for
Groundwater potential mapping at Kurdistan region of Iran using analytic hierarchy landslide susceptibility mapping in Yanshan County, China. Sci. Total Environ. 666,
process and GIS. Arabian J. Geosci. 8 (9), 7059–7071. 975–993.
Sajedi-Hosseini, F., et al., 2018. A novel machine learning-based approach for the risk Wu, J., et al., 2019. Flash flood forecasting using support vector regression model in a
assessment of nitrate groundwater contamination. Sci. Total Environ. 644, 954–962. small mountainous catchment. Water 11 (7), 1327.
Sameen, M.I., Pradhan, B., Lee, S., 2020. Application of convolutional neural networks Yu, H., Ma, Y., Wang, L., Zhai, Y., Wang, X., 2017. A landslide intelligent detection
featuring Bayesian optimization for landslide susceptibility assessment. Catena 186, method based on CNN and RSG_R. In: IEEE International Conference on Mechatronics
104249. and Automation (ICMA). IEEE, pp. 40–44.
San, B.T., 2014. An evaluation of SVM using polygon-based random sampling in landslide Zabihi, M., Pourghasemi, H.R., Pourtaghi, Z.S., Behzadfar, M., 2016. GIS-based multi-
susceptibility mapping: the Candir catchment area (western Antalya, Turkey). Int. J. variate adaptive regression spline and random forest models for groundwater po-
Appl. Earth Obs. Geoinf. 26, 399–412. tential mapping in Iran. Environ. Earth Sci. 75 (8), 665.
Sarker, C., Mejias, L., Maire, F., Woodley, A., 2019. Flood mapping with convolutional Zahedi, S., Azarnivand, A., Chitsaz, N., 2017. Groundwater quality classification deriva-
neural networks using spatio-contextual pixel information. Remote Sens. 11 (19), tion using multi-criteria-decision-making techniques. Ecol. Ind. 78, 243–252.
2331. Zeiler, M.D., Fergus, R., 2014. Computer Vision–ECCV 2014: 13th European Conference,
Sener, E., Davraz, A., Ozcelik, M., 2005. An integration of GIS and remote sensing in Zurich, Switzerland, September 6-12, 2014. Proceedings, Part I, 818: 833.
groundwater investigations: a case study in Burdur, Turkey. Hydrogeol. J. 13 (5–6), Zhang, G., Wang, M., Liu, K., 2019. Forest fire susceptibility modeling using a convolu-
826–834. tional neural network for Yunnan Province of China. Int. J. Disaster Risk Sci. 10 (3),
Shahid, S., Nath, S., Roy, J., 2000. Groundwater potential modelling in a soft rock area 386–403.
using a GIS. Int. J. Remote Sens. 21 (9), 1919–1924. Zhang, X., Wang, J., Zhang, K., 2017. Short-term electric load forecasting based on sin-
Sharifi Garmdareh, E., Vafakhah, M., Eslamian, S.S., 2018. Regional flood frequency gular spectrum analysis and support vector machine optimized by Cuckoo search
analysis using support vector regression in arid and semi-arid regions of Iran. Hydrol. algorithm. Electr. Power Syst. Res. 146, 270–285.
Sci. J. 63 (3), 426–440. Zuo, R., Xiong, Y., Wang, J., Carranza, E.J.M., 2019. Deep learning and its application in
Simonyan, K., Zisserman, A., 2014. Very deep convolutional networks for large-scale geochemical mapping. Earth-Sci. Rev.
12