Comparing the Performance of Machine Learning Algorithms for Groundwater Mapping in Delhi

Abstract

The problem of groundwater depletion has arisen as havoc in countries like India due to expanding intensive agriculture, growing population, and burgeoning urban centres. Delhi is one of the greatest urban agglomerations in the country facing severe groundwater depletion, but the robust methods for modelling the groundwater have not yet been adopted for examining the conditions of the groundwater. In such scenarios, accurate modelling of groundwater resources using appropriate techniques and tools is essential. The present study aimed to investigate groundwater level using GIS tools and machine learning algorithms and find the best models for application. The previous studies conducted are purely based on GIS methods without the possibility of accuracy determination of the results. Thus, in this study, boosted regression tree, generalized linear model (GLM), and neural net multi-layer perceptron (NNET-MLP) were applied for modelling the groundwater table in the capital city of India (i.e. Delhi). Anthropogenic, physiographic, meteorological, and hydrological factors like LULC, geology, elevation, slope, aspect, curvature, soil permeability, LST, precipitation, stream power index, and topographic wetness index are supplied as conditioning factors. The performances of the models were compared using area under curve (AUC) plot and correlation (COR). The AUC plot appears well above the diagonal line, showing acceptable results for all the models. The COR is maximum for the NNET-MLP, i.e. 0.93, while minimum value is for GLM, i.e. 0.60. The modelled rasters represented variable groundwater depths, and the mean of each district of Delhi is calculated. This is one of the first studies where GIS and machine learning are integrated to model the groundwater level of Delhi and hence open new prospects for research focussing on the capital of the country.

Full text

Vol.:(0123456789)
1 3
Journal of the Indian Society of Remote Sensing
https://doi.org/10.1007/s12524-023-01789-8
RESEARCH ARTICLE
Comparing thePerformance ofMachine Learning Algorithms
forGroundwater Mapping inDelhi
ZainabKhan1· MohammadMohsin2· SkAjimAli1 · DeepikaVashishtha1· MujahidHusain3· AdeebaParveen1·
SyedKausarShamim1· FarhanaParvin1,4· RukhsarAnjum1· SaniaJawaid1· ZebaKhanam1· AteequeAhmad1
Received: 8 February 2023 / Accepted: 10 November 2023
© Indian Society of Remote Sensing 2023
Abstract
The problem of groundwater depletion has arisen as havoc in countries like India due to expanding intensive agriculture,
growing population, and burgeoning urban centres. Delhi is one of the greatest urban agglomerations in the country facing
severe groundwater depletion, but the robust methods for modelling the groundwater have not yet been adopted for examin-
ing the conditions of the groundwater. In such scenarios, accurate modelling of groundwater resources using appropriate
techniques and tools is essential. The present study aimed to investigate groundwater level using GIS tools and machine
learning algorithms and find the best models for application. The previous studies conducted are purely based on GIS meth-
ods without the possibility of accuracy determination of the results. Thus, in this study, boosted regression tree, generalized
linear model (GLM), and neural net multi-layer perceptron (NNET-MLP) were applied for modelling the groundwater table
in the capital city of India (i.e. Delhi). Anthropogenic, physiographic, meteorological, and hydrological factors like LULC,
geology, elevation, slope, aspect, curvature, soil permeability, LST, precipitation, stream power index, and topographic
wetness index are supplied as conditioning factors. The performances of the models were compared using area under curve
(AUC) plot and correlation (COR). The AUC plot appears well above the diagonal line, showing acceptable results for all
the models. The COR is maximum for the NNET-MLP, i.e. 0.93, while minimum value is for GLM, i.e. 0.60. The modelled
rasters represented variable groundwater depths, and the mean of each district of Delhi is calculated. This is one of the first
studies where GIS and machine learning are integrated to model the groundwater level of Delhi and hence open new prospects
for research focussing on the capital of the country.
Keywords Groundwater· Geographic information system· Machine learning· AHP· Delhi
Introduction
Groundwater is considered as the prime water resource for
humanity in order to meet the needs of various domestic
and commercial purposes (Bidhuri & Khan, 2020). In low-
income areas, groundwater is providing significant supply
of drinking water and it thus plays a crucial role in realizing
the human right to water (Carrard etal., 2019; Grönwall &
Danert, 2020). Even though groundwater forms the largest
freshwater reservoir on the planet, a decline in this precious
resource has been witnessed in the past years (Bhattarai
etal., 2021; Rodell etal., 2009). India is no exception, as
groundwater resources have declined severely both due to
rapidly expanding agriculture and urban centre to support
an ever-growing population (Dangar etal., 2021). However,
the exacerbation of groundwater resource is alarmingly det-
rimental to agriculture and urban centres (Fishman, 2018;
Yar, 2020). Likewise, depleting groundwater is threatening
the availability of drinking water in the megacities and sub-
urbs (Arunprakash etal., 2014; Balan etal., 2012; Sarkar
etal., 2020).
Delhi is one of India's megacities, with a rapid population
increase (Amann etal., 2017). Its groundwater resources
* Sk Ajim Ali
skajimali@myamu.ac.in; skajimali.saa@gmail.com
1 Department ofGeography, Faculty ofScience, Aligarh
Muslim University, Aligarh202002, India
2 James Clark College ofEngineering, University ofMaryland,
CollegePark, MD, USA
3 Department ofGeography, Faculty ofNatural Sciences,
Jamia Millia Islamia, NewDelhi, Delhi110025, India
4 School ofLiberal Arts, Noida International University,
GreaterNoida, UttarPradesh203201, India

Journal of the Indian Society of Remote Sensing
1 3
are also getting depleted for a number of reasons, including
rapid population growth, steadily expanding economic and
industrial activity, intensive agriculture, and various kinds
of human activities (Bierkens & Wada, 2019; Cohen etal.,
2006; Mukherjee etal., 2010). Long-term trend analysis
of Delhi’s groundwater level demonstrated signs of severe
depletion (Roy etal., 2020). Rapid and severe groundwater
depletion has also caused subsidence in the capital (Malik
etal., 2019). Therefore, it is the need of the hour to develop
accurate methods for groundwater modelling and moni-
toring. GIS and remote sensing have been in use for dec-
ades for groundwater prediction (Arefin, 2020; Lee etal.,
2020). Das and Pardeshi (2018) used integration of myriad
parameters influencing the occurrence of groundwater using
GIS environment. Joshi and Gupta (2018) used GIS-based
groundwater modelling to simulate groundwater resource
in Rajasthan. However, machine learning is quite novel
research in the field of groundwater modelling (Cacace etal.,
2013a, 2013b). Trichakis etal., (2011a, 2011b) and Ho etal.,
(2011) used the ANN algorithm to simulate groundwater
level. Rahmati etal., (2019a, 2019b) used algorithms like
PICP, SVM, RF, and kNN in order to predict groundwa-
ter level, while Siade etal., (2020) used Gaussian process
to model groundwater. Mallick etal., (2021a, 2021b) used
coupled machine learning models to predict groundwater
potential. Arabameri etal., 2021 used a novel hybrid model
that combines random subspace (RS), multi-layer perception
(MLP), Naive Bayes tree (NB Tree), and classification and
regression tree (CART) algorithms to map the groundwater
potential (GWP). Some other researchers also employed var-
ious mathematical models like the logistic regression (LR)
(Nguyen etal., 2020), frequency ratio (Guru etal., 2017),
weights of evidence (WoE) Al-Abadi, 2015), analytical hier-
archy process (AHP) Rahmati etal., 2015), certainty factor
(xet al., 2015), and evidential belief function (EBF) Nam-
pak etal., 2014) to assess groundwater potential. However,
Naghibi etal., 2016a, 2016b proved that boosted regression
tree (BRT) is best performing method for groundwater while
RF is worst. Alshehri & Rahman, 2023 utilized gradient
boosting machines (GBM), generalized linear model (GLM)
and convolution neural network (CNN) models for predict-
ing groundwater quality and found GLM to be most accu-
rate. Mohammed adopted multi-layer perceptron artificial
neural network (MLP-ANN) and support vector regression
(SVR) with MLP-ANN and produced most accurate results
for groundwater due to hidden layers.
Many studies have been conducted on groundwater
potential around the world using sophisticated methods of
machine learning (Arulbalaji etal., 2019; Choubin etal.,
2019a, 2019b; Chowdhury etal., 2009; Gnanachandrasamy
etal., 2018). But groundwater studies in Delhi are still based
on the near-obsolete GIS-based techniques and no significant
research work has been done on the predictive modelling
of groundwater as of now. Adhikari and Das (2012) con-
ducted a study on the groundwater quality of Delhi for irri-
gation. Tomer etal. (2019) and Tomer etal. (2021) used
a GIS-based DRASTIC model to assess the groundwater
vulnerability of Delhi. Vishal etal., (2014) used GIS to esti-
mate the groundwater recharge. Roy etal. (2020) used a
statistics-based inverse distance weighting method of inter-
polation. Although the aforesaid studies conducted on the
groundwater of Delhi hardly ever incorporated the impact
of conditioning variables such as LULC, soil permeability,
geology, LST, or precipitation, the application of advanced
machine learning-based predictive modelling in association
with significant anthropogenic, physiographic, meteorologi-
cal, and hydrological conditioning factors has never been
performed before.
Chatterjee etal. 2009 provided an exhaustive study of
groundwater resources of Delhi. Groundwater resources
including availability and extraction are assessed by CGWB
periodically and the latest information available is for the
period 2022. The report of aquifer mapping studies by
CGWB (Kapoor etal., 2016) provides aquifer dispositions
and groundwater management plan based on groundwater
flow modelling. Some multivariate statistical analyses have
been conducted previously for the geochemical assessment
of groundwater of Delhi (Srivastava & Ramanathan, 2008;
Singh etal., 2017), yet their reliability remained question-
able due to lack of standard accuracy assessment.
However, these studies either do not involve prediction
of groundwater scenarios or even if they do, none of such
sophisticated methods were ever applied to analyse and pre-
dict ground water resources for Delhi. Moreover, most of the
previous studies used GIS methods based on multi-criteria
decision-making (MCDM). Variation within Delhi at differ-
ent administrative units too is bound to exist that has been
neglected in previous studies conducted by researchers such
as Pham etal., 2022; Adji & Sejati, 2014; Mukherjee etal.,
2012; Rao and Jugran, 2003 and Reddy etal., 2000. Apart
from that, accuracy assessment is the major limitation of
such studies in the absence of validation of results.
The objective of the present study is to model groundwa-
ter level using inventory groundwater data by the applica-
tion of machine learning algorithms including GLM, BRT,
and neural network multi-layer perceptron NNET (MLP)
in order to predict the groundwater level of Delhi under
spatially variable anthropogenic, physiographic, meteoro-
logical, and hydrological factors along with districts’ mean
groundwater depth so that the most vulnerable districts can
be demarcated. GLM, BRT, and NNET (MLP) are adopted
in the present study over other models due to their abso-
lute variable functionalities, simplicity, and better predict-
ability for the present problem, for instance, GLM is the
most suitable when conditioning variables have chances of
small errors (Armstrong, 1985) such as LST or interpolated

Journal of the Indian Society of Remote Sensing
1 3
rainfall rasters. BRT is also very powerful model as it com-
bines the strength of boosting with regression tree giving
it an improved predictive performance (Elith etal., 2008).
Unlike classical statistical ANN, NNET (MLP) is a sto-
chastic approximation yet purely runs on data, proving bet-
ter predictive capacity (Mijwel, 2018; Omrani, 2015). The
adopted models are far more robust than the MCDM, as one
can directly determine the accuracy of the results whereas
the results in MCDM merely rely on the assigned weights
that can be erroneous. It is one of its kind studies on Delhi
to examine the spatial variation in mean groundwater depth
in different districts of Delhi. The study aims to predict the
groundwater table with the best performing machine learn-
ing model using reliable accuracy assessment methods such
as receiver operating characteristic (ROC) and area under
curve (AUC) which is defined as a curve representing the
test of sensitivity or true positive rate versus its 1-specificity
or false positive rate. AUC values along with the generation
of response surfaces are used in the present paper for the
accuracies.
Study Area
Delhi is a small section of the Indo-Gangetic plain (Bray
etal., 2019) which is selected disregarding the basin of
the river Yamuna and its tributaries. Delhi is mostly, if not
completely, under the influence of anthropogenic activities.
The population, land use, permeability of the ground, tem-
perature variation related to urban heat island (Pandey etal.,
2014), and diminished precipitation (Steensen etal., 2022)
are deeply affected by local human interventions. Contigu-
ous countryside has different conditions for these condi-
tioning factors. Therefore, adoption of natural boundary of
catchment of river Yamuna would have averaged out the
pure anthropogenic effects. Hence, it is rational to follow the
administrative boundary of Delhi instead of the catchment
area. Delhi is located at an altitude of 198 to 220m above
mean sea level (MSL). It lies in the centre of the Indian sub-
continent, between the Himalayan and Aravali mountains
(Fig.1). Delhi spans 1485 Km2, comprising both urban and
rural areas (Singh etal., 2010). The urban area is 891.09
Fig. 1 Location of study area

Journal of the Indian Society of Remote Sensing
1 3
Km2 including newer settlements and 593.01 Km2 are occu-
pied by rural dwellings. In addition, Delhi frequently has
millions of migratory populations (Naikoo etal, 2020). The
population of Delhi is estimated to be 15,217,000 persons by
the end of 2023 (census2011.com), out of which 1,700,000
live below the poverty line (Aniruddha Ghosal, 2017). Delhi
only has a small fraction of the water, which is less than
15%. Even these resources are distributed discriminatorily
among the rich and poor. The posh areas are infamous for
using most of its water supply (Babu, 2021), while the mid-
dle class, poor, and homeless have highly dwindling acces-
sibility to water.
Most water sources are obtained from surrounding states.
The Delhi water supply system is under increasing stress
due to the city's tremendous population increase. About 650
million gallons of water are provided to the city each day by
the Delhi Jal Board (DJB), which is in control of the water
supply. However, there is a 250 million gallon/day (MGD)
deficit due to the 900 MGD average daily water demand
(Shekhar & Prasad, 2009). With two neighbouring states,
Haryana to the north, west, and south, and Uttar Pradesh
to the east, it shares boundaries. It includes nine revenue
districts (Bidhuri & Khan, 2020). With typically dry win-
ters lasting from November to January, the city's climate
ranges from humid subtropical to semiarid (DES, 2014).
From January to July, the city's mean temperature ranges
from 14.2 ℃ to 32.2 ℃ (amssdelhi.gov.in). Delhi receives
about 790.8mm of rainfall on an average per year. Novem-
ber and December are the driest months when barely 9mm
of precipitation occurs, while July is the wettest month (cli-
matemps.com).
Database andMethodology
In the present study, total 11 groundwater conditioning fac-
tors were selected which have a direct and indirect effect
on predicting groundwater level. In Fig.2, the methodol-
ogy used in the current research is described in detail. The
groundwater inventory dataset was prepared first using data
derived from India-WRIS. Then, the conditioning factors
include land use land cover (LULC), geology, elevation,
slope, aspect, and soil permeability; unlike groundwater,
subsurface water reservoirs are three-dimensional in nature
with undefined water bodies. The details of these datasets
including variable description and collected sources are
shown in Table1. Some of the significant factors are pur-
posely neglected such as drainage density which was not
adopted as it is useful in areas with greater relief, while
Delhi has minimum elevation of 171m and maximum
elevation of 311m rendering it with a relief being 140m.
The present study prefers geology over geomorphology as
groundwater retention and percolation is more influenced
by internal structure of rocks, voids present in them, and
their mutual connectivity (Wray & Sauro, 2017). Linea-
ment density and drainage density have significant role in
determining the groundwater level. However, these variables
were excluded from the database because lineament density
acts as a conduit for groundwater and have limited role in
groundwater retention, while drainage density has inverse
relation with permeability, but only over pervious surfaces
such as bare soils or sand (Agarwal & Garg, 2016). Delhi
has highly built-up areas and limited significant of drainage
density. Apart from these, previous studies on groundwater
potential of Delhi have demonstrated that lineament density
in Delhi does not express high spatial variability (Singh &
Mukherjee, 2014; Mallick etal., 2021).
Groundwater Inventory
Unlike surface water, subsurface water reservoirs are three-
dimensional in nature with undefined water bodies. The
groundwater inventory data represent the depth of water
table in reference of an aquifer with geo-locations at 88 sites
that are homogeneously distributed over Delhi (Appendix1).
They essentially represent the real-world information of the
concerned event. In the present study, the groundwater table
as the inventory dataset was collected from India-WRIS as
mentioned in Table1. The collected data were processed and
mapped as shown in Fig.3a.
Anthropogenic Variables
There are multiple anthropogenic variables that have poten-
tial to affect the dynamic movement of groundwater. How-
ever, the most significant is LULC that not only controls
potential evapotranspiration (Das etal., 2018), but also
groundwater recharge rate, surface, and subsurface flow
(Owuor etal., 2016). Groundwater quality and harshness
will vary as a result of over usage due to declining agri-
cultural and rising urban land use with population growth.
The present study area offers a variety in LULC classes, i.e.
waterbodies, built-up, vegetation, and cropland (Fig.3b).
Therefore, it is rational to adopt LULC as one of the condi-
tioning variables and examine its role in predictive ground-
water zonation.
Physiographic Variables
In general groundwater takes 20,000years to get recharge
but this duration is highly variable in response to local
geology, elevation, slope, aspect, curvature, and soil per-
meability (Dai etal., 2021; Yifru etal., 2021). Surface
and subsurface geology has a major role in determining
the accessibility of groundwater and its storage capacity
(Maurya etal., 2022; Raad etal., 2022; Rajasekhar etal.,

Journal of the Indian Society of Remote Sensing
1 3
2020). The water yield and recharge rates are governed
by pore size distribution and permeability while eleva-
tion of topography controls the pace of surface runoff
flow at ground level, which affects water permeability in
the earth's strata (Zhang and Li, 2009). The groundwa-
ter table fluctuation has a direct response to the elevation
(Bouwer, 2002). Slope is another key conditioning fac-
tor determining the groundwater resources as slope gov-
erns the surface runoff dynamics and therefore controls
the amount of water percolation (Arya etal., 2020; Jing
etal., 2022). Locations with a high slope have low pros-
pects for recharging since it receives little infiltration or
recharge, all of which impact the volume of water permeat-
ing the earth and thus altering groundwater (Solomon and
Quiel, 2006). The aspect presents slope orientations that
affect the quantity of rainfall, radiation from the sun, wind
velocity, and land use land cover, all of which impact the
volume of water infiltration (Díaz-Alcaide & Martínez-
Santos, 2019). Curvature plays a key role in governing the
surface runoff dynamics and thereby affects the subsurface
inflow of water (Gao etal., 2021; Li etal., 2021). When
it rains, a concave curvature holds onto more water for a
prolonged duration (Lee and Pradhan, 2007; Pothiraj and
Rajagopalan, 2013; Manap etal., 2014). Particularly, in
Fig. 2 Presenting the detail of methodology adopted in the present study

Journal of the Indian Society of Remote Sensing
1 3
Table 1 Domains of groundwater level and variables considered
Selected variables Variable description Source
Inventory dataset
Groundwater Inventory Numerical variable supplied as geospatial points with
normalized values
Point data derived from India-WRIS https:// india wris.
gov. in/ wris/#/ DataD ownlo ad
Anthropogenic factors
Land use/Land cover Categorical variable supplied as AHP values of individ-
ual class of land use/ land cover
Landsat-8 data derived from earth explore of USGS in
the form of multi-layer raster data
https:// scihub. coper nicus. eu/ dhus/#/ home
Physiographic factors
Geology Categorical variable supplied as AHP values of individ-
ual category of Geology
Derived from Bhukosh of Geological Survey of India in
the form of polygons
https:// bhuko sh. gsi. gov. in/ Bhuko sh/ MapVi ewer. aspx
Elevation Continuous variable supplied as normalized between
zero and one
ASTER data collected from earth explore of USGS in
the form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Slope Continuous variable supplied as normalized between
zero and one
ASTER data collected from earth explore of USGS in
the form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Aspect Continuous variable supplied as normalized between
zero and one
ASTER data gathered from earth explore of USGS in the
form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Curvature Continuous variable supplied as normalized between
zero and one
ASTER data derived from earth explore of USGS in the
form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Soil Permeability Categorical variable supplied as AHP values of individ-
ual category of soil permeability
FAO soil polygons with mm/day permeability
Meteorological factors
LST Continuous variable supplied as normalized between
zero and one
Landsat-8 data derived from earth explore of USGS in
the form of multi-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Precipitation Continuous variable supplied as normalized between
zero and one
Cruts 4.05 data derived from National Centre for Atmos-
pheric Science, UK
https:// cruda ta. uea. ac. uk/ cru/ data/ hrg/
Hydrological factors
Stream Power Index Continuous variable supplied as normalized between
zero and one
ASTER data collected from earth explore of USGS in
the form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85
Topographic Wetness Index Continuous variable supplied as normalized between
zero and one
ASTER data collected from earth explore of USGS in
the form of single-layer raster data
https:// cmr. earth data. nasa. gov/ browse- scaler/ browse_
images/ granu les/ G1726 726417- LPCLO UD?h= 85&w=
85

Journal of the Indian Society of Remote Sensing
1 3
comparison to a convex slope, concave surfaces are more
conducive to the incidence of groundwater (Ao etal.,
2021; Biswas etal., 2020).
Soil is also a major influencing factor affecting the avail-
ability of groundwater (Dar etal., 2021; Golkarian & Rah-
mati, 2018). The soil texture and grain size determine the
fluid movement hence affecting the groundwater recharge
rate as well as the yield of water (Akingboye etal., 2022;
Antia, 2022).
Considering the aforesaid rationales, it is valid to consider
all these six physiographic variables, i.e. geology (Fig.3c),
elevation (Fig.3d), slope (Fig.3e), aspect (Fig.3f), curva-
ture (Fig.4a), and soil permeability (Fig.4b) as key physi-
ographic conditioning variables for groundwater modelling
of Delhi.
Meteorological Variables
Meteorological factors determine the supply and losses of
water in a unit of area (Byers etal., 2020; Su etal., 2019;
Zeng etal., 2019). The precipitation is a direct input of water,
while outflow in the form of runoff and evapotranspiration are
major losses (Aragaw & Mishra, 2022; Kansoh etal., 2020).
While runoff out flow and inflow are depended on aforesaid
anthropogenic and physiographic variables, the evapotranspi-
ration is solely temperature dependent. In the present study,
precipitation and land surface temperature (LST) were selected
as meteorological variables because LST had a significant cor-
relation with both soil moisture and soil temperature (Ali &
Ahmad, 2019a, 2019b, 2020; Patel etal., 2022). From LST, the
approximate groundwater temperature (GWT) can be calcu-
lated, while precipitation directly replenishes the aquifer (Benz
etal., 2015). It is the conditioning variable with utmost signifi-
cance as without precipitation all other conditioning variables
are futile. Therefore, it is logical to consider LST and precipi-
tation as the meteorological conditioning variables (Fig.4c).
LST was calculated using following steps (Ali etal., 2022):
Step 1: Calculation ofTop ofAtmosphere (TOA) Spectral
Radiance
The following equation was used to transform thermal infra-
red digital values into TOA spectral radiance using the radi-
ance rescaling factor (Eq.1).
(1)
TOA (L𝜆)=ML∗Qcal +AL −Qi
Fig. 3 Selected variables for groundwater level mapping (i.e. anthropogenic and physiographic variables)

Journal of the Indian Society of Remote Sensing
1 3
where TOA (Lλ) is the total spectral radiance, ML represents
the band specific multiplicative rescaling factor, Qcal is the
Band 10 of Landsat 8, AL is band specific additive rescaling
factor, and Qi is correction value for Band 10 of Landsat 8.
Step 2: Conversion ofTop ofAtmosphere (TOA)/Spectral
Radiance toBrightness Temperature
Now, the spectral radiance of thermal band was utilized to
convert the radiance into brightness temperature which is
expressed in Eq.2). Most of the studies have found that the
value 0.95 is for vegetated land, while the value 0.92 is for
non-vegetated land (Nichol, 1994).
OR
(2)
BT
=
K2
In (Kl
Lλ
+1
)
−
273.15
BT =K2∕Ln(K1∕TOA +1)) − 273.15.
BT =(1321.0789∕Ln((774.8853∕TOA)+1)) − 273.15.
where T = at-satellite brightness temperature, Lλ = TOA
spectral radiance, K1 = constant band, and K2 = constant
band.
For Landsat 8 OLI, value of K1 for band 10 is 774.8853,
while value of K2 for the same band is 1321.0789.
Step 3: Proportion ofVegetation (Pv)
To estimate the value of Pv, first NDVI was calculated as
expressed in Eq.3.
Then, with the obtained value of NDVI, Pv was esti-
mated as shown in Eq.(4).
where Pv = proportion of vegetation, NDVI = Normalized
Difference Vegetation Index, NDVImin = the NDVI mini-
mum value, and NDVImax = the NDVI maximum value.
(3)
NDVI
=
NIR(Band5)−Red(Band4)
NIR(Band5)+Red(Band4)
(4)
Pv
=
(
(NDVI −NDVImin)
(NDVImax −NDVImin)
)2
Fig. 4 Selected variables for groundwater mapping (i.e. meteorological and hydrological variables)

Journal of the Indian Society of Remote Sensing
1 3
Step 4: Land Surface Emissivity
It is a fundamental characteristic of natural objects and an
important surface parameter obtained from the radiance
of the emitting material as recorded from space. Addition-
ally, it pertains to the average emissivity of a component
of the Earth's surface as determined by NDVI values. It is
shown in Eq.(5).
where E = land surface emissivity, Pv = proportion of veg-
etation, and 0.986 is a constant value.
Step 5: Land Surface Temperature (LST)
This is the final output after following all these steps. It
refers to the average temperature of an object of the exact
surface of the earth calculated from measured radiance,
which is depicted in Eq.6.
where BT = at-satellite brightness temperatures, W = wave-
length of emitted radiance, Ln = the log function, and € =
the land surface emissivity.
Hydrological Variables
The hydrological factors are deeply linked with the
groundwater occurrence (Wang etal., 2022). Stream
Power Index (SPI) is the representation of strength of sur-
face runoff and hence has an inverse relationship with the
groundwater dynamics (Mondal & Mandal, 2020; Wendt
etal., 2021). On the other hand, Topographic Wetness
Index (TWI) as a hydrological factor affecting the distri-
bution of groundwater by representing the moisture con-
tent, saturating areas, and flow accumulation, all of which
determine groundwater (Kalantar-Zadeh etal., 2019). It
is thus, rational to consider the SPI and TWI as the con-
ditioning variables for studying the groundwater of Delhi
(Fig.4d and 4e). The following formula was used to cal-
culate the stream power index (Eq.7):
where
As
is the region of the particular watershed and
𝛽
is
the degree-scaled local slope gradient.
(5)
E=0.004 ∗Pv +0.986
(6)
LST
=
BT
1
+W∗
BT
14380
∗In(E
)
LST =
BT
∕(
1
+ ((
10.895
∗
BT
∕
14, 388
)∗
Ln
(E))).
(7)
SPI =As×tan𝛽
A topographic wetness index calculates how much
water has accumulated at a particular location which can
be defined by the following equation (Eq.8):
where
tan𝛽
is the slope angle at the point and
a
is the total
upslope area draining via a point (per unit contour length).
The
In(
a
tan𝛽
)
index represents both the propensity of
gravity to flow water down slope (represented in terms of
tan𝛽
as an estimated hydraulic gradient) and the propensity
of water to collect at any location in the basin. The perme-
ability, pore water pressure, and impacts on the soil strength
of the material are the main determinants of the infiltration
of water (Poudyal etal. 2010).
Methods
The methodological principles adopted for the prediction
of groundwater are based upon the groundwater inventory
as well as on the conditioning variables. The detail of meth-
odological application and procedure is presented above
(Fig.2).
Collection andPreparation ofConditioning
Variables
Before running the selected machine learning algorithms,
twelve selected conditioning variables were prepared and
processed as input dataset. The inventory data pertaining
the depth of groundwater were extracted from India-WRIS
portal, and the collected data were converted into points.
The geology data were downloaded and converted into ras-
ter. The LULC map was prepared using k-means algorithms
with 0.938% accuracy and 0.9231 Kappa value. The SRTM
DEM was obtained from USGS earth explorer and eleva-
tion map was prepared. Both slope and aspect maps were
prepared from collected DEM using ArcGIS v-10.8, where
z-factor was kept at 0.0001 for slope generation. The soil
permeability data were taken from FAO and converted into
raster. The surface temperature was generated using Landsat
8 OLI and TIRS data. The precipitation data were down-
loaded in the raster format but it was unsuitable to apply
for groundwater prediction. So, the downloaded data was
processed. In this regard, the pixels were converted into
points and then the precipitation values are spatially pre-
dicted using interpolation (i.e. spline method) in the GIS
environment. The curvature, SPI, and TWI were extracted
from SRTM elevation dataset.
(8)
TWI
=In
(
a
tan
𝛽
)

Journal of the Indian Society of Remote Sensing
1 3
Pre‑processing Transformation ofData toSuit
Machine Learning
For machine learning models, it is a prerequisite to normalize
the data and bring it on the same binary scale because machine
learning models identify and function only on the binary data.
Therefore, min–max normalization was applied on the con-
tinuous raster (Table2) and analytic hierarchy process (AHP)
was considered for the categorical raster (Table3).
Normalization
In min–max normalization method, linear transformation of
the original data was performed. Min–max normalization
either stretches or squishes the all the data in a range between 0
and 1. The following equation was used in this regard (Eq.9):
(9)
Xnormalized
=
(X
n
−X
min
)
(X
max
−X
min)
where
Xnormalized
is the normalized Xn,
Xn
is the target
value in the data,
Xmin
is the minimum value in the data, and
Xmax
maximum value in the data
The min–max normalization was applied on the raster rep-
resenting the selected variables which is shown in Table2.
Analytic Hierarchy Process (AHP)
AHP technique is organized and analysed complex deci-
sions using mathematical formulation. It is a multi-criteria
decision-making method that run performance analysis used
in businesses and firms (Görener etal., 2012). Alternatives,
criteria, performance, and weight are the four components of
decision-making that are used to build it. In analytical hierar-
chical process, factors are presented as matrices A1, A2, …An,
while weights are represented as w1, w2, …wn (Eq.10).
where matrix element aij = 1/ aij,; therefore, when i = j, aij
= 1. The values of wi vary from 1 to 9, where 9 represents
absolute importance and 1 represents least importance. The
relative importance of ai and aj is depicted as aij. For cal-
culating the weight, the following matrix was used as shown
in Eq.11.
where wi was calculated using Eq.12:
(10)
A
=
⎡
⎢
⎢
⎢
⎣
1a12 …an
1∕a12 1 …a2n
⋮ ⋮⋱
1∕a1n1∕an21
⎤
⎥
⎥
⎥
⎦
(11)
A
=aij=





w1∕w1w1∕w2…w1∕wn
w2∕w11w2∕w2…w2∕wn
⋮⋮ ⋱
wn∕w1wn∕w2wn∕wn





(12)
w
i=1∕𝜆<
max
∑n
j=
1
aijwj
Table 2 Normalized continuous data
Inventory data/raster
data (numerical)
Actual values Normalized
Values
Min Max Min Max
Groundwater table 0.57 65.5 0 1
Elevation 171 311 0 1
Slope 0 1.12 0 1
Aspect −1 360 0 1
Soil Permeability 3.55 4.43 0 1
LST 30.41 48.7 0 1
Precipitation 2.68 5.26 0 1
SPI −0.87 −14 0 1
TWI 11.39 27.5 0 1
Curvature −0.07 0.06 0 1
Table 3 Normalized categorical
data
P = Producer, T = Total, K = Kappa.
Raster data
(categorical)
Class/rock type AHP weights CI CR Accuracy K
User P T
LULC Crop 0.11 0.19 0.04 0.20 0.18 0.938 0.9231
Built-up 0.33 0.16 0.16
Water 0.35 0.21 0.21
Vegetation 0.16 0.22 0.24
Others 0.05 0.22 0.22
Geology Undivided precambrian rocks 0.41 0.01 0.02 NA NA NA NA
Quaternary sediment 0.48
Quaternary sand dunes 0.11

Journal of the Indian Society of Remote Sensing
1 3
For consistency measurement, consistency index (CI) was
estimated and consistency ratio was calculated as follows
(Eqs. 13 and 14)
In the present study, online interviews were arranged with
the experts to decide rank scale of AHP and weights were
calculated in order to analyse the significant roles played by
the classes of LULC and categories of rocks in determining
the occurrence of groundwater (Table3).
Rational forSelection ofMachine Learning Models
Generalized Linear Model (GLM)
GLM is founded on regression; thus, it can easily identify
differences between factors (Zhao, 2017). GLM gener-
ates optimal regression model that can predict numerous
events using a variety of linear models. According to several
experts, GLM is most frequently employed for spatial mod-
elling (Keir etal., 2019). Multiple regressions are typically
used by the GLM to improve the accuracy and quality of
the findings since it can clearly show a relationship between
the response and explanatory variables. GLM models allow
us to create a linear relationship between the answer and
predictors irrespective of the fact that their natural associa-
tion is not continuous (Nelder & Wedderburn, 1972). The
response variable is connected to a linear model using a
link function, which enables this. John Nelder and Robert
Wedderburn developed generalized linear models as a means
of combining numerous other predictive methods, such as
linear regression, logistic regression, and Poisson regression.
For the purpose of maximum likelihood estimation (MLE)
of the model parameters, they suggested an iteratively
reweighted least squares method. MLE is still widely used
and is often used as the default method in statistical comput-
ing programmes. Other methods have been developed, such
as least squares fitting to variance stabilized responses and
Bayesian regression.
Boosted Regression Tree (BRT)
BRT is a data mining and machine learning method that
combines decision trees and boosting approaches. It can be
used to solve problems involving regression and classifi-
cation (Youssef etal., 2015). By merging numerous fitted
models, it seeks to improve the effectiveness and predictive
power of one technique (Naghibi etal., 2016a, b). Similar to
(13)
CR
=
CI
RI
(14)
CI
=
(𝜆max −1)
(n−1)
model averaging, boosting is used to integrate the outcomes
of the decision trees. The number of trees, shrinkage (or
learning rate), and interaction depth are some of the model's
characteristics that need to be optimized. The relevance of
trees in the constructed model is defined by shrinkage or
learning rate (Naghibi etal., 2016a, b). The number of nodes
in trees is determined by the depth or intricacy of interac-
tions. Boosted regression tree was chosen as the data mining
method for this task because it can be used to select features
and integrate stochastic gradient boosting to reduce variabil-
ity and prejudices (Abeare, 2009; Naghibi etal., 2016a, b).
The significance of the influencing factors in the modelling
process is also defined by the BRT model.
Neural Network Multi‑Layer Perceptron (NNET‑ MLP)
The most basic form of artificial neural networks (ANNs),
which are models of arithmetic operations and consist of
input, hidden, and output layers, is the multi-layer perception
(Coulibaly etal., 2001). Neurons, the fundamental building
block of ANNs, relate all layers to one another. To forecast
output variables like GWLs, input layers use all the input
variables like temperature, rainfall, etc. Through activation
functions, hidden and output layers manage the weights and
biases derived from input layers. To optimize prediction, the
MLP needs some training data to modify bias and weight
(Elbaz etal., 2019). Modellers have employed a variety of
algorithms, including gradient descent with momentum, the
LM, back propagation, Bayesian regularization, and adap-
tive learning rate back propagation. Krishna etal., 2008
compared a number of training algorithms in the ground-
water modelling of an urban coastal aquifer in the Indian
state of Andhra Pradesh. When contrasted to other learn-
ing algorithms, they discovered that the LM algorithm was
among the best. In groundwater modelling, it is the most
widely used algorithm (Karandish & Šimůnek, 2019). This
algorithm provides greater accuracy in prediction by more
effectively locating the local minima of error functions (Juan
etal., 2005). The LM algorithm was chosen to calculate
the loss function for this study. In this study, different input
variable combinations are used with the MLP to precisely
predict the groundwater level in Delhi.
Accuracy Assessment oftheApplied Models
The measurements of accuracy of the applied models are of
utmost significance. In the present paper, 70–30 split of data
is utilized for training and testing respectively. A graph dis-
playing a categorization model's success across all classifica-
tion thresholds is known as a ROC curve (Receiver Operating
Characteristic). Two factors, True Positive Rate (TPR) and
False Positive Rate, are plotted on this curve (FPR). TPR is

Journal of the Indian Society of Remote Sensing
1 3
described as follows (Eq.15) because recall is a shorthand
for it:
where TP is the true positive and FN is the false negative
FPR is defined as follows (Eq.16):
where FP is the false positive and TN is the true negative
The TPR vs. FPR is plotted on a ROC curve at various clas-
sification levels. More items are classified as positive when the
classification criterion is lowered, which raises the number of
both False Positives and True Positives.
(15)
TPR
=
TP
(TP +FN)
(16)
FPR
=
FP
(FP +TN)
De‑normalization ofModelled Raster
The normalized raster underwent through the machine learn-
ing process must be de-normalized to the range of inventory
database. Therefore, in the present study, the de-normalization
of the output predictive raster was conducted using equation
shown below (Eq.17).
where
DenormRar
is de-normalized raster,
InRarNorm
is the
input of normalized raster,
maxval
is the maximum value of
dataset, and
minval
is the minimum value of dataset.
(17)
DenormRar =InRarNorm ∗(maxval −minval)+minval
Fig. 5 Groundwater occurrence in Delhi prepared using BRT, GLM, and NNET-MLP models

Journal of the Indian Society of Remote Sensing
1 3
Results
Spatial Distribution ofGroundwater Table
The results represent the normalized and de-normalized ras-
ter of all applied machine learning models (Fig.5). Lower
groundwater level possibly but not necessarily represents
disturbed groundwater equilibrium or depleted groundwater
resource, and higher groundwater level can be surmised as
adequate resource of groundwater, though this is purely rela-
tive. When comparing the range of these three models, i.e.
BRT, GLM, and NNET-MLP, it was found out that NNET-
MLP has the widest groundwater depth range with deep-
est point being 63.22m below ground level (mbgl) and the
shallowest point being 0.58 mbgl. The BRT model has the
narrowest range with the deepest groundwater occurrence
being 23.81 mbgl and the shallowest being 8.84 mbgl. The
GLM model depicts a moderate range oscillating between
59.89 to 0.66mbgl. The southern part of the Delhi has the
deepest water table according to all the models and western
part of appears to be better off in terms of depth of water
table. According to the BRT model, all the parts of Delhi
appear to have higher ground water level except the south.
However, GLM model depicts that there is a rough forma-
tion of deep ground water strip from north to south. The
NNET-MLP represents that entire Delhi has a mid-range
depth of groundwater except far west with higher ground-
water and with southern Delhi with deepest occurrence of
ground water. All the models appeared with 100% true posi-
tive value.
The district-wise analysis of the mean groundwater table
also represents the irregular results for all the models. The
BRT represents the shallowest groundwater, while GLM
and NNET-MLP closely follow similar results (Table4).
Table 4 District-wise mean groundwater level of Delhi
Districts BRT GLM NNET-MLP
Central 19.06 49.00 54.67
East 16.82 45.86 42.63
North East 16.65 48.41 44.84
North 15.99 44.40 45.82
North West 16.46 44.35 41.23
New Delhi 18.12 45.43 52.38
South 20.21 52.17 56.30
South West 18.91 45.24 48.08
West 18.25 45.70 43.28
Delhi Mean 17.98 46.14 46.61
Fig. 6 District-wise groundwater depth in Delhi

Journal of the Indian Society of Remote Sensing
1 3
The south Delhi has the deepest groundwater, i.e. 20.21m,
52.17m, and 56.30m as for BRT, GLM, and NNET-MLP
models, respectively. North Delhi has the shallowest ground-
water table as of BRT, i.e. 15.99m, while GLM and NNET-
MLP modelled north-west with the shallowest groundwater
level of 41.23m and 44.35m, respectively.
Unequivocally similar result is presented in Fig.6,
where BRT represents the shallowest groundwater as com-
pared to GLM and NNET-MLP. When compared the mean
groundwater depth of entire Delhi for the three models, it is
found out that according to BRT it is 17.98m, while GLM
and NNET-MLP put forth very similar results with mean
groundwater level of 46.14 and 46.61m, respectively.
Accuracies oftheML Models
The accuracy assessment of the applied models (Table5)
represents that correlation values (represented by COR)
and deviances, while AUC and TSS remain NA using the
DISMO package in RStudio. Based on the obtained COR
values, it can be clearly concluded that the NNET-MLP
is the best performing model for predictive mapping of
groundwater in Delhi. In order to rule out overfitting of data,
the models are separately on training and testing data. The
accuracies of model for training data are not significantly
higher than that of the accuracies of the test data.
ROC Curves, Relative Importance, andResponse Surfaces
Figure7 shows a typical ROC curve of the models naming
BRT, GLM, and NNET-MLP, respectively. Sensitivity of
ROC for GLM (Fig.7a) is near 0.8, while the sensitivity of
ROC curve for BRT is near 1 (Fig.7b). The NNET-MLP
ROC curve depicts sensitivity less than 1 but roughly above
0.9 (Fig.7c).
The relative importance for conditioning variables var-
ies from model to model (Fig.8). LULC appears to be most
importance according to all the models though its value is
highest in GLM, i.e. close to 0.8 (Fig.8a), while smallest as
of NNET-MLP in which it is approximately 0.2 (Fig.8c).
The response curve of the variables (LULC and elevation)
with maximum correlation with groundwater points is
also plotted to for each of the models, i.e. GLM, BRT, and
NNET-MLP, respectively, presented in Fig.9. The response
surfaces appear quite similar to each other except that of
GLM with higher curvature at the z-axis, while the response
surface of NNET-MLP has the minimalistic curvature on
the z-axis.
Discussion
Three models, i.e. BRT, GLM, and NNET-MLP, were taken
into consideration in the present analysis to precisely pre-
dict and model groundwater in Delhi for the first time. The
performance of all these models was compared, and the pre-
diction of mean groundwater level has also been assessed
spatially at every district level of Delhi. Different districts
of Delhi have varying underlying factors and variable depth
of water table at different spatial points. Precise prediction
of groundwater is critical for sustainable development of
groundwater resources. In order to estimate the groundwa-
ter with maximum possible accuracy, application of ML
along with the logical conditioning variables is as essential
aspect. The most widely used modern breakthroughs in the
fourth industrial revolution, machine learning (ML) gives
devices the capability to learn from experience and improve
naturally without being specifically designed (Shorten etal.,
2021 and Sarker etal., 2020). To predict groundwater level,
researchers have employed a number of machine learning
(ML) models, including hybrid ML model (Yang etal.,
2014) methodology to ensemble modelling using spectral
analysis, machine learning and uncertainty analysis (Sahoo
etal., 2017), and random forest (Gaffoor etal., 2022), and
Jyolsna etal. 2021 recently applied a popular machine learn-
ing model, i.e. multi-linear regression (MLR). Furthermore,
groundwater level has also been predicted using various
statistical models (SM) and mathematical models (MM) by
Kenda etal., 2020; Lima etal., 2020; Sierikova etal., 2020;
He etal., 2019; Naji etal., 2016 and Dehn etal., 2005.
However, none of such sophisticated methods were ever
applied to the Delhi where people are battling for water
everyday despite alarmingly exacerbating groundwater
resources (Chatterjee etal., 2009). Some multivariate sta-
tistical analyses have been conducted previously for the
geochemical assessment of groundwater of Delhi (Sriv-
astava & Ramanathan, 2008; Singh etal., 2017) yet their
reliability remained questionable due to lack of standard
accuracy assessment. Most of the significant work related
to groundwater prediction in Delhi was done by CGWB
(Ventral Ground Water Board); however, the work is old
and requires revision (Kapoor etal., 2016). The water table
predicted by CGWB has incorporated many crucial factors
such as LULC, elevation and water bodies but the methods
are not clearly stated and accuracies are uncertain. CGWB
has also prepared the aquifer maps yet the research related
to groundwater should be undertaken more frequently using
advanced methods. Sophisticated studies based on advanced
Table 5 Accuracy assessment of the models
Algorithms AUC COR-Training COR-Testing TSS Deviance
GLM NA 0.610 0.600 NA −0.12
BRT NA 0.680 0.667 NA −0.13
NNET-MLP NA 0.930 0.921 NA −0.04

Journal of the Indian Society of Remote Sensing
1 3
Fig. 7 Accuracy assessment of all models using ROC plot a GLM, b BRT, and c NNET-MLP

Journal of the Indian Society of Remote Sensing
1 3
Fig. 8 Relative importance of variables a GLM, b BRT, and c NNET-MLP

Journal of the Indian Society of Remote Sensing
1 3
ML methods remain lacking in the modelling of the ground-
water occurrence itself for the national capital. Moreover,
studies are explicitly based on the GIS methods based on
MCDM. These studies provide a hypothetical picture of the
groundwater of despite considering the real-world ground
water depth database. Variation with the Delhi at different
administrative units too are bound to exist that have been
neglected in previous studies conducted by researchers such
as Pham etal., 2022; Adji & Sejati, 2014; Mukherjee etal.,
2012; Rao and Jugran, 2003 and Reddy etal., 2000. Apart
from that, lack of accuracy assessment is a major negative of
such studies in the absence of validation of results.
Hence, considering all these aspects, in this study, we
presented very first predicted groundwater surfaces of Delhi
based on ML algorithms representing the complex inter-
actions of the surface and subsurface variables in order to
evaluate the occurrence of groundwater using the models,
i.e. BRT, GLM, and NNET-MLP. However, it should be
emphasized that the RF model has shown to perform well in
several environmental sectors, including flash flood hazard
assessment, earth fissure hazard prediction, and groundwa-
ter nitrate prediction (Hosseini etal., 2020; Rahmati etal.,
2019a, 2019b and Choubin etal., 2019a, 2019b). Nonethe-
less, compared to other standalone methods, the models
utilized in this study have greater advantages and strengths
which others lack in precisely predicting groundwater of the
study area. The models employed in this study take numer-
ous explanatory variable types (such as continuous and
Fig. 9 Response Surface of all models a GLM, b BRT, and c NNET-MLP

Journal of the Indian Society of Remote Sensing
1 3
classification variables), enhance missing or lost data, and
are not required to transform or remove anomalous and out-
lier data (Knoll etal., 2019; Aertsen etal., 2010; Elith etal.,
2008 and Liaw & Wiener, 2002). They also lack the pre-
analysis necessary to choose variables from a huge range of
predictors, and they expand the variety of classification trees
by randomly choosing predictive factors from the many trees
(Wang etal., 2020; Miraki etal., 2019 and Hepelwa etal.,
2010). The adopted ML methods for the present study also
represented variable relative importance of the conditioning
factors. However, LULC attained maximum relative impor-
tance according to all the ML methods. For instance, LULC
has maximum relative importance as of GLM method, i.e.
close to 0.8 while minimum relative importance values are
modelled according to NNET-MLP method which is roughly
0.2. These models fit and manage the intricate nonlinear
relationship between different variables, and by fitting many
trees, they get beyond the single model's major flaw (poor
prediction performance) (Mosavi etal., 2020; Moghimi
etal., 2017; Naghibi etal., 2017 and Hong etal., 2016).
These are the ML algorithms that are most frequently used
in groundwater modelling (Karandish & Šimůnek, 2019). It
improves prediction accuracy by finding the local minima of
error functions more efficiently (Juan etal., 2005).
In that case, robust methods to model the groundwater in
2-dimensional space is of key significance whose purpose
is met in the present study. The average correlation values
of GLM, BRT, and NNET-MLP are fairly acceptable, i.e.
0.60, 0.68, and 0.93, respectively. The NNET-MLP has the
highest value as shown in Table5. The resulted continuous
layers represent the predicted groundwater table instead of
potential zones of groundwater. The study can be useful for
DJB and other water resource managing government bodies
for sustainable utilization of groundwater resources in Delhi
and prioritizing the water resources in the areas with the
lowest groundwater levels where boring well for domestic
water needs appear non-viable.
Conclusion
In the application of ML algorithms along with correla-
tions, conventional plots, ROC curves, and response curves
for modelling the spatial occurrence of the groundwater
is found to be an effective and reliable method, especially
NNET-MLP that exhibits best COR value. The geospatial
distribution of the groundwater in response to conditioning
variables is found to be fluctuating in space as mean results
of each of the districts of Delhi have been variable meaning
thereby that surface and subsurface conditions deeply affect
the groundwater occurrence in space.
The influence of LULC and elevation was modelled to
be highest in determining the probability of occurrence
of groundwater according to the three utilized algorithms
of ML, i.e. GLM, BRT, and NNET-MLP. The national
capital appears to be affected deeply by the elevation as the
southern Delhi specially the south district has the deep-
est groundwater level according to the all models mean-
ing that the groundwater in south Delhi is either depleted
or is influenced by the elevated local topography. The
next lowest mean groundwater level was found in central
Delhi that can be surmised to either higher population or
impermeable concrete built-up structures. The northern
area such as districts of north and north-west have been
found to have shallower groundwater level in comparison
of other districts surmised as the effected of relative open
permeable surfaces such as agricultural lands and rela-
tively lower altitudes. The study opens up future possibili-
ties of research where impact of aforesaid conditioning
factors along with other possible factors can be modelled
individually and more information can be gathered so
that better groundwater resource management of Delhi is
achieved and more responsible prioritization of resource
can be performed.
Limitation
In this study, 88 points are sufficient and produced reli-
able results, but the predicted groundwater table could be
more accurate if more inventory data points were avail-
able. One of the most crucial variables, i.e. groundwater
withdrawal, is not incorporated in the present study due to
lack of spatial reference points. Only three machine learn-
ing algorithms were used in modelling the groundwater of
Delhi. One can use more machine learning algorithms for
testing their predictive capability in groundwater mapping
as well other relevant studies.
Appendix1
Latitude Longitudes Ground-
water table
(m)
Groundwater
table (m) Nor-
malized
28.85806 77.1963889 8.45 0.121418
28.85139 77.0736111 3.92 0.051618
28.84333 77.1294444 5.31 0.073035
28.83194 77.0083333 25.39 0.382435
28.8225 77.2036111 12.05 0.176888
28.81944 76.9972222 41.99 0.638213
28.81528 77.1516667 8.46 0.121572

Journal of the Indian Society of Remote Sensing
1 3
Latitude Longitudes Ground-
water table
(m)
Groundwater
table (m) Nor-
malized
28.81528 77.1516667 7.63 0.108783
28.81472 77.1975 13.3 0.196148
28.81472 77.1972222 10.45 0.152234
28.78889 77.0291667 10.01 0.145455
28.76889 77.2075 9.04 0.130508
28.75833 77.0625 15.68 0.23282
28.75556 77.0058333 31.91 0.482897
28.75278 77.095 13.81 0.204006
28.75278 76.9666667 8.2 0.117565
28.74 77.2225 44.82 0.681818
28.73639 77.1627778 50.78 0.773652
28.73222 77.1044444 20.97 0.31433
28.72889 77.1469444 8.02 0.114792
28.725 77 63.88 0.975501
28.71944 76.9666667 14.2 0.210015
28.70694 77.025 65.47 1
28.69583 77.2277778 11.52 0.168721
28.69111 77.1238889 31.85 0.481972
28.69028 77.0791667 36.85 0.559014
28.68472 77.2491667 21.83 0.327581
28.68472 77.1994444 13.68 0.202003
28.68222 76.9941667 47.6 0.724653
28.67806 77.0947222 20 0.299384
28.67556 77.0933333 12.78 0.188136
28.67222 77.2305556 23 0.345609
28.66111 77.3030556 6.96 0.098459
28.65556 77.2358333 19.84 0.296918
28.65 77.0166667 28.89 0.436364
28.63917 77.1622222 3.21 0.040678
28.63222 77.0741667 8.56 0.123112
28.63194 77.1986111 8.05 0.115254
28.63194 77.1594444 3.85 0.050539
28.63 77.0913889 8.92 0.128659
28.62806 77.3180556 3.74 0.048844
28.61861 77.1111111 2.18 0.024807
28.61639 77.3044444 9.54 0.138213
28.61528 77.2125 12.94 0.190601
28.615 77.2122222 10.85 0.158398
28.61472 77.0005556 5.69 0.078891
28.6125 77.225 11.91 0.17473
28.60611 77.21 4.21 0.056086
28.60472 77.2661111 25.45 0.383359
28.60417 77.175 3.38 0.043297
28.60389 76.9322222 8.59 0.123575
28.60028 77.2986111 9.01 0.130046
28.60028 77.055 5.51 0.076117
28.59611 77.245 16.19 0.240678
28.595 77.2508333 21.57 0.323575
28.59444 77.2733333 23.54 0.353929
Latitude Longitudes Ground-
water table
(m)
Groundwater
table (m) Nor-
malized
28.59222 77.1275 2.99 0.037288
28.59167 77.2205556 10.56 0.153929
28.59028 77.2125 7.59 0.108166
28.59028 77.1841667 3.39 0.043451
28.59028 77.2163889 17.43 0.259784
28.59028 77.2163889 16.95 0.252388
28.58722 77.3013889 8.97 0.12943
28.58556 77.0261111 5.38 0.074114
28.57861 77.1080556 6.47 0.090909
28.57833 77.1077778 1.76 0.018336
28.57667 76.9141667 6.89 0.097381
28.56667 77.0538889 5.82 0.080894
28.54639 77.0094444 8.4 0.120647
28.54528 77.2022222 3.32 0.042373
28.54306 76.9652778 7.5 0.10678
28.53944 77.1805556 24.51 0.368875
28.53611 76.9047222 53.01 0.808012
28.53583 77.1569444 5.95 0.082897
28.53417 76.9086111 25.86 0.389676
28.52778 77.2266667 1.86 0.019877
28.52472 76.9533333 11.42 0.16718
28.52444 76.9533333 10.87 0.158706
28.51806 76.9041667 54 0.823267
28.50889 77.3405556 11.98 0.175809
28.50611 77.1822222 6.19 0.086595
28.49583 77.2666667 1.53 0.014792
28.48944 77.1458333 1.07 0.007704
28.47694 77.1561111 0.57 0
28.46806 77.15 1.96 0.021418
28.42778 77.2083333 1.66 0.016795
28.42 77.2077778 4.81 0.065331
28.40889 77.1894444 1.37 0.012327
References
Abeare, S. (2009). Comparisons of boosted regression tree, GLM and
GAM performance in the standardization of yellowfin tuna catch-
rate data from the Gulf of Mexico lonline fishery.
Adji, T. N., & Sejati, S. P. (2014). Identification of groundwater
potential zones within an area with various geomorphological
units by using several field parameters and a GIS approach in
Kulon Progo Regency, Java, Indonesia. Arabian Journal of Geo-
sciences, 7(1), 161–172.
Aertsen, W., Kint, V., Van Orshoven, J., Özkan, K., & Muys, B. (2010).
Comparison and ranking of different modelling techniques for
prediction of site index in Mediterranean mountain forests. Eco-
logical Modelling, 221(8), 1119–1130.
Agarwal, R., & Garg, P. K. (2016). Remote sensing and GIS based
groundwater potential & recharge zones mapping using multi-
criteria decision-making technique. Water Resources Manage-
ment, 30, 243–260.

Journal of the Indian Society of Remote Sensing
1 3
Akingboye, A. S., Bery, A. A., Kayode, J. S., Ogunyele, A. C., Ade-
ola, A. O., Omojola, O. O., & Adesida, A. S. (2022). Ground-
water-yielding capacity, water–rock interaction, and vulner-
ability assessment of typical gneissic hydrogeologic units
using geoelectrohydraulic method. Acta Geophysica, pp. 1–25.
Ali, S. A., & Ahmad, A. (2019a). Mapping of mosquito-borne
diseases in Kolkata Municipal Corporation using GIS and
AHP based decision making approach. Spatial Information
Research, 27(3), 351–372.
Ali, S. A., & Ahmad, A. (2019b). Spatial susceptibility analysis of
vector-borne diseases in KMC using geospatial technique and
MCDM approach. Modeling Earth Systems and Environment,
5(3), 1135–1159.
Ali, S. A., & Ahmad, A. (2020). Analysing water-borne diseases
susceptibility in Kolkata Municipal Corporation using WQI
and GIS based Kriging interpolation. GeoJournal, 85(4),
1151–1174.
Ali, S. A., Parvin, F., & Ahmad, A. (2022). Retrieval of Land Sur-
face Temperature from Landsat 8 OLI and TIRS: A Com-
parative Analysis between Radiative Transfer Equation-Based
Method and Split-Window Algorithm. Remote Sensing
in Earth Systems Sciences, 1–21. https:// doi. org/ 10. 1007/
s41976- 022- 00079-0
Amann, M., Purohit, P., Bhanarkar, A. D., Bertok, I., Borken-Kleefeld,
J., & Cofala, J., etal. (2017). Managing future air quality in
megacities: A case study for Delhi. Atmospheric Environment,
161, 99–111.
Armstrong, B. (1985). Measurement error in the generalised linear
model. Communications in Statistics-Simulation and Computa-
tion, 14(3), 529–544.
Alshehri, F., & Rahman, A. (2023). Coupling machine and deep learn-
ing with explainable artificial intelligence for improving predic-
tion of groundwater quality and decision-making in arid region.
Saudi Arabia. Water, 15(12), 2298.
Antia, D. D. (2022). Provision of desalinated irrigation water by the
desalination of groundwater abstracted from a Saline Aquifer.
Hydrology, 9(7), 128.
Ao, C., Zeng, W., Yang, P., Xing, W., Lei, G., Wu, J., & Huang, J.
(2021). The effects of slope shape and polyacrylamide applica-
tion on runoff, erosion and nutrient loss from hillslopes under
simulated rainfall. Hydrological Processes, 35(4), e14130.
Aragaw, H. M., & Mishra, S. K. (2022). Runoff curve number-potential
evapotranspiration-duration relationship for selected watersheds
in Ethiopia. Modeling Earth Systems and Environment, 8(2),
1899–1910.
Arefin, R. (2020). Groundwater potential zone identification at Plio-
Pleistocene elevated tract, Bangladesh: AHP-GIS and remote
sensing approach. Groundwater for Sustainable Development,
10, 100340.
Arulbalaji, P., Padmalal, D., & Sreelash, K. (2019). GIS and AHP tech-
niques based delineation of groundwater potential zones: A case
study from southern Western Ghats. India. Scientific Reports,
9(1), 1–17.
Arunprakash, M., Giridharan, L., Krishnamurthy, R. R., & Jayaprakash,
M. (2014). Impact of urbanization in groundwater of south Chen-
nai City, Tamil Nadu. India. Environmental Earth Sciences,
71(2), 947–957.
Arya, S., Subramani, T., & Karunanidhi, D. (2020). Delineation of
groundwater potential zones and recommendation of artificial
recharge structures for augmentation of groundwater resources
in Vattamalaikarai Basin. South India. Environmental Earth Sci-
ences, 79(5), 1–13.
Babu, N. M. (2021, September 3). ‘Many in posh areas using 10 times
more water’. The Hindu. Accessed from https:// www. thehi ndu.
com/ news/ cities/ Delhi/ many- in- posh- areas- using- 10- times-
more- water/ artic le362 63518. ece. Accessed on 10.05.2023
Balan, I., Shivakumar, M., & Kumar, P. M. (2012). An assessment of
groundwater quality using water quality index in Chennai, Tamil
Nadu. India. Chronicles of Young Scientists, 3(2), 146–146.
Benz, S. A., Bayer, P., Menberg, K., Jung, S., & Blum, P. (2015). Spa-
tial resolution of anthropogenic heat fluxes into urban aquifers.
Science of the Total Environment, 524, 427–439.
Bhattarai, N., Pollack, A., Lobell, D. B., Fishman, R., Singh, B., Dar,
A., & Jain, M. (2021). The impact of groundwater depletion on
agricultural production in India. Environmental Research Letters,
16(8), 085003.
Bidhuri, S., & Khan, M. M. A. (2020). Assessment of ground water
quality of central and southeast districts of NCT of Delhi. Jour-
nal of the Geological Society of India, 95(1), 95–103.
Bierkens, M. F., & Wada, Y. (2019). Non-renewable groundwater use
and groundwater depletion: A review. Environmental Research
Letters, 14(6), 063002.
Biswas, S., Mukhopadhyay, B. P., & Bera, A. (2020). Delineating
groundwater potential zones of agriculture dominated landscapes
using GIS based AHP techniques: A case study from Uttar Dina-
jpur district. West Bengal. Environmental Earth Sciences, 79(12),
1–25.
Bouwer, H. (2002). Integrated water management for the 21st cen-
tury: problems and solutions. Journal of Irrigation and Drainage
Engineering, 128(4), 193–202.
Bray, C. D., Battye, W. H., & Aneja, V. P. (2019). The role of biomass
burning agricultural emissions in the Indo-Gangetic Plains on
the air quality in New Delhi. India. Atmospheric Environment,
218, 116983.
Byers, E. A., Coxon, G., Freer, J., & Hall, J. W. (2020). Drought and
climate change impacts on cooling water shortages and electric-
ity prices in Great Britain. Nature Communications, 11(1), 1–12.
Cacace, M., Blöcher, G., Watanabe, N., Moeck, I., Börsing, N., Scheck-
Wenderoth, M., etal. (2013a). Modelling of fractured carbonate
reservoirs: Outline of a novel technique via a case study from
the Molasse Basin, southern Bavaria. Germany. Environmental
Earth Sciences, 70(8), 3585–3602.
Cacace, T., Bianco, V., & Ferraro, P. (2013b). Quantitative phase imag-
ing trends in biomedical applications. Optics and Lasers in Engi-
neering, 135, 106188.
Carrard, N., Foster, T., & Willetts, J. (2019). Groundwater as a source
of drinking water in southeast Asia and the Pacific: A multi-
country review of current reliance and resource concerns. Water,
11(8), 1605.
Census of India, 2011.
Chatterjee, R., Gupta, B. K., Mohiddin, S. K., Singh, P. N., Shek-
har, S., & Purohit, R. (2009). Dynamic groundwater resources
of National Capital Territory, Delhi: Assessment, development
and management options. Environmental Earth Sciences, 59,
669–686.
Choubin, B., Mosavi, A., Alamdarloo, E. H., Hosseini, F. S., Sham-
shirband, S., Dashtekian, K., & Ghamisi, P. (2019a). Earth fissure
hazard prediction using machine learning models. Environmental
Research, 179, 108770.
Choubin, B., Rahmati, O., Soleimani, F., Alilou, H., Moradi, E., &
Alamdari, N. (2019). Regional groundwater potential analysis
using classification and regression trees. In Spatial modeling in
GIS and R for earth and environmental sciences (pp. 485–498).
Elsevier.
Chowdhury, A., Jha, M. K., Chowdary, V. M., & Mal, B. C. (2009).
Integrated remote sensing and GIS-based approach for assessing
groundwater potential in West Medinipur district, West Bengal.
India. International Journal of Remote Sensing, 30(1), 231–250.
Cohen, D., Person, M., Daannen, R., Locke, S., Dahlstrom, D., Zabiel-
ski, V., etal. (2006). Groundwater-supported evapotranspiration
within glaciated watersheds under conditions of climate change.
Journal of Hydrology, 320(3–4), 484–500.

Journal of the Indian Society of Remote Sensing
1 3
Coulibaly, P., Anctil, F., Aravena, R., & Bobée, B. (2001). Artificial
neural network modeling of water table depth fluctuations. Water
Resources Research, 37(4), 885–896.
Dai, X., Xie, Y., Simmons, C. T., Berg, S., Dong, Y., Yang, J., etal.
(2021). Understanding topography-driven groundwater flow
using fully-coupled surface-water and groundwater modeling.
Journal of Hydrology, 594, 125950.
Dangar, S., Asoka, A., & Mishra, V. (2021). Causes and implications
of groundwater depletion in India: A review. Journal of Hydrol-
ogy, 596, 126103.
Dar, T., Rai, N., & Bhat, A. (2021). Delineation of potential ground-
water recharge zones using analytical hierarchy process (AHP).
Geology, Ecology, and Landscapes, 5(4), 292–307.
Das, P., Behera, M. D., Patidar, N., Sahoo, B., Tripathi, P., Behera, P.
R., etal. (2018). Impact of LULC change on the runoff, base flow
and evapotranspiration dynamics in eastern Indian river basins
during 1985–2005 using variable infiltration capacity approach.
Journal of Earth System Science, 127(2), 1–19.
DES (2014). Department of Economics and Statistics.
Díaz-Alcaide, S., & Martínez-Santos, P. (2019). Advances in ground-
water potential mapping. Hydrogeology Journal, 27(7),
2307–2324.
Elbaz, K., Shen, S. L., Zhou, A., Yuan, D. J., & Xu, Y. S. (2019).
Optimization of EPB shield performance with adaptive neuro-
fuzzy inference system and genetic algorithm. Applied Sciences,
9(4), 780.
Elith, J., Leathwick, J. R., & Hastie, T. (2008). A working guide to
boosted regression trees. Journal of Animal Ecology, 77(4),
802–813.
Fishman, R. (2018). Groundwater depletion limits the scope for adapta-
tion to increased rainfall variability in India. Climatic Change,
147(1), 195–209.
Gaffoor, Z., Pietersen, K., Jovanovic, N., Bagula, A., Kanyerere, T.,
Ajayi, O., & Wanangwa, G. (2022). A comparison of ensemble
and deep learning algorithms to model groundwater levels in a
data-scarce aquifer of Southern Africa. Hydrology, 9(7), 125.
Gao, Z., Niu, F., Wang, Y., Lin, Z., & Wang, W. (2021). Supraper-
mafrost groundwater flow and exchange around a thermokarst
lake on the Qinghai-Tibet Plateau. China. Journal of Hydrology,
593, 125882.
Ghosal, A. (2017, May 16). Delhi per capita income 3-times more than
rest of india: Even as city retains top spot, over 17 lakh are BPL.
Indian Express. New Delhi. Accessed from https:// india nexpr ess.
com/ artic le/ cities/ delhi/ delhi- per- capita- income- 3- times- more-
than- rest- of- india- even- as- city- retai ns- top- spot- over- 17- lakh-
are- bpl- 46576 33/. Accessed on 09.05.2023
Gnanachandrasamy, G., Zhou, Y., Bagyaraj, M., Venkatramanan, S.,
Ramkumar, T., & Wang, S. (2018). Remote sensing and GIS
based groundwater potential zone mapping in Ariyalur District,
Tamil Nadu. Journal of the Geological Society of India, 92(4),
484–490.
Golkarian, A., & Rahmati, O. (2018). Use of a maximum entropy
model to identify the key factors that influence groundwater
availability on the Gonabad Plain, Iran. Environmental Earth
Sciences, 77(10), 1–20.
Görener, A. (2012). Comparing AHP and ANP: an application of stra-
tegic decisions making in a manufacturing company. Interna-
tional Journal of Business and Social Science, 3(11), 194–208.
Grönwall, J., & Danert, K. (2020). Regarding groundwater and drink-
ing water access through a human rights lens: Self-supply as a
norm. Water, 12(2), 419.
Guru, B., Seshan, K., & Bera, S. (2017). Frequency ratio model for
groundwater potential mapping and its sustainable management
in cold desert, India. Journal of King Saud University-Science,
29(3), 333–347.
He, X., Wu, J., & He, S. (2019). Hydrochemical characteristics and
quality evaluation of groundwater in terms of health risks in
Luohe aquifer in Wuqi County of the Chinese Loess Plateau,
northwest China. Human and Ecological Risk Assessment: An
International Journal, 25(1–2), 32–51.
Hong, H., Pourghasemi, H. R., & Pourtaghi, Z. S. (2016). Landslide
susceptibility assessment in
Hosseini, F. S., Choubin, B., Mosavi, A., Nabipour, N., Shamshirband,
S., Darabi, H., & Haghighi, A. T. (2020). Flash-flood hazard
assessment using ensembles and Bayesian-based machine learn-
ing models: Application of the simulated annealing feature selec-
tion method. Science of the Total Environment, 711, 135161.
Jing, X., Li, L., Chen, S., Shi, Y., Xu, M., & Zhang, Q. (2022). Straw
returning on sloping farmland reduces the soil and water loss
via surface flow but increases the nitrogen loss via interflow.
Agriculture, Ecosystems and Environment, 339, 108154.
Juan, Y., Bo-Ming, Y., Bin, Z., & Ming-Tao, H. (2005). A geometry
model for tortuosity of streamtubes in porous media with spheri-
cal particles. Chinese Physics Letters, 22(6), 1464.
Jyolsna, P. J., Kambhammettu, B. V. N. P., & Gorugantula, S. (2021).
Application of random forest and multi-linear regression methods
in downscaling GRACE derived groundwater storage changes.
Hydrological Sciences Journal, 66(5), 874–887.
Kalantar-Zadeh, K., Tang, J., Daeneke, T., O’Mullane, A. P., Stewart,
L. A., & Liu, J., etal. (2019). Emergence of liquid metals in
nanotechnology.ACS Nano, 13(7), 7388–7395.
Kansoh, R., Abd-El-Mooty, M., & Abd-El-Baky, R. (2020). Computing
the water budget components for lakes by using meteorological
data. Civil Engineering Journal, 6(7), 1255–1265.
Kapoor, U., Chakraborty, D., Kumar, J., Chandra, R., Nayak, S., &
Kapoor, S. (2016). (rep.). Aquifer Mapping and Ground Water
Management Plan of Nct Delhi. Delhi, Delhi: Central Ground
Water Board.
Karandish, F., & Šimůnek, J. (2019). A comparison of the HYDRUS
(2D/3D) and SALTMED models to investigate the influence of
various water-saving irrigation strategies on the maize water
footprint. Agricultural Water Management, 213, 809–820.
Kenda, K., Peternelj, J., Mellios, N., Kofinas, D., Čerin, M., &
Rožanec, J. (2020). Usage of statistical modeling techniques in
surface and groundwater level prediction. Journal of Water Sup-
ply: Research and Technology-AQUA, 69(3), 248–265.
Keir, G., Bulovic, N., & McIntyre, N. (2019). Stochastic modeling of
groundwater extractions over a data-sparse region of Australia.
Groundwater, 57(1), 97–109.
Knoll, L., Breuer, L., & Bach, M. (2019). Large scale prediction of
groundwater nitrate concentrations from spatial data using
machine learning. Science of the Total Environment, 668,
1317–1327.
Krishna, B., Satyaji Rao, Y. R., & Vijaya, T. (2008). Modelling ground-
water levels in an urban coastal aquifer using artificial neural
networks. Hydrological Processes: An International Journal,
22(8), 1180–1188.
Lee, S., Hyun, Y., Lee, S., & Lee, M. J. (2020). Groundwater potential
mapping using remote sensing and GIS-based machine learning
techniques. Remote Sensing, 12(7), 1200.
Li, H., Wang, W., Fu, J., Chen, Z., Ning, Z., & Liu, Y. (2021). Quanti-
fying the relative contribution of climate variability and human
activities impacts on baseflow dynamics in the Tarim River
Basin, Northwest China. Journal of Hydrology: Regional Stud-
ies, 36, 100853.
Lianhua County (China): a comparison between a random forest data
mining technique and bivariate and multivariate statistical mod-
els.Geomorphology,259, 105–118.
Liaw, A., & Wiener, M. (2002). Classification and regression by ran-
domForest. R News, 2(3), 18–22.

Journal of the Indian Society of Remote Sensing
1 3
Lima, A. B. S., Batista, A. S., de Jesus, J. C., de Jesus Silva, J., de
Araújo, A. C. M., & Santos, L. S. (2020). Fast quantitative detec-
tion of black pepper and cumin adulterations by near-infrared
spectroscopy and multivariate modeling. Food Control, 107,
106802.
Malik, K., Kumar, D., & Perissin, D. (2019). Assessment of subsidence
in Delhi NCR due to groundwater depletion using TerraSAR-X
and persistent scatterers interferometry. The Imaging Science
Journal, 67(1), 1–7.
Mallick, J., Naikoo, M. W., Talukdar, S., Ahmed, I. A., Rahman, A., &
Islam, A. R. M. T., etal. (2021a). Developing groundwater poten-
tiality models by coupling ensemble machine learning algorithms
and statistical techniques for sustainable groundwater manage-
ment. Geocarto International, 1–27.
Mallick, J., Talukdar, S., Alsubih, M., Almesfer, M. K., Shahfahad,
Hang, H. T., & Rahman, A. (2021b). Integration of statistical
models and ensemble machine learning algorithms (MLAs) for
developing the novel hybrid groundwater potentiality models:
a case study of semi-arid watershed in Saudi Arabia.Geocarto
International, 1–32.
Manap, M. A., Nampak, H., Pradhan, B., Lee, S., Sulaiman, W. N.
A., & Ramli, M. F. (2014). Application of probabilistic-based
frequency ratio model in groundwater potential mapping using
remote sensing data and GIS. Arabian Journal of Geosciences,
7, 711–724
Maurya, P. K., Ali, S. A., Zaidi, S. K., Wasi, S., Tabrez, S., Malav, L.
C., etal. (2022). Assessment of groundwater geochemistry for
drinking and irrigation suitability in Jaunpur district of Uttar
Pradesh using GIS-based statistical inference. Environmental
Science and Pollution Research, 1–25.
Mijwel, M. M. (2018). Artificial neural networks advantages and dis-
advantages. Retrieved from LinkedIn https// www. linke din. com/
pulse/ artifi cial- neura lnet. Retrieved on 12.05.2023
Miraki, S., Zanganeh, S. H., Chapi, K., Singh, V. P., Shirzadi, A., Sha-
habi, H., & Pham, B. T. (2019). Mapping groundwater potential
using a novel hybrid intelligence approach. Water Resources
Management, 33(1), 281–302.
Moghimi, A., Pourreza, A., Zuniga-Ramirez, G., Williams, L. E., &
Fidelibus, M. W. (2017). A novel machine learning approach
to estimate grapevine leaf nitrogen concentration using aerial
multispectral imagery. Remote Sensing, 12(21), 3515.
Mondal, S., & Mandal, S. (2020). Data-driven evidential belief func-
tion (EBF) model in exploring landslide susceptibility zones for
the Darjeeling Himalaya. India. Geocarto International, 35(8),
818–856.
Mosavi, A., Ardabili, S., & Varkonyi-Koczy, A. R. (2020). List of deep
learning models. InInternational conference on global research
and education(pp. 202–214). Springer, Cham.
Mukherjee, P., Singh, C. K., & Mukherjee, S. (2012). Delineation of
groundwater potential zones in arid region of India—a remote
sensing and GIS approach. Water Resources Management, 26(9),
2643–2672.
Mukherjee, S., Shah, Z., & Kumar, M. D. (2010). Sustaining urban
water supplies in India: Increasing role of large reservoirs. Water
Resources Management, 24(10), 2035–2055.
Naghibi, S. A., Ahmadi, K., & Daneshi, A. (2017). Application of
support vector machine, random forest, and genetic algorithm
optimized random forest models in groundwater potential map-
ping. Water Resources Management, 31(9), 2761–2775.
Naghibi, S. A., Pourghasemi, H. R., & Dixon, B. (2016a). GIS-based
groundwater potential mapping using boosted regression tree,
classification and regression tree, and random forest machine
learning models in Iran. Environmental Monitoring and Assess-
ment, 188(1), 1–27.
Naghibi, S. A., Pourghasemi, H. R., & Dixon, B. (2016b). GIS-based
groundwater potential mapping using boosted regression tree,
classification and regression tree, and random forest machine
learning models in Iran. Environmental Monitoring and Assess-
ment, 188, 1–27.
Naikoo, M. W., Rihan, M., & Ishtiaque, M. (2020). Analyses of land
use land cover (LULC) change and built-up expansion in the
suburb of a metropolitan city: Spatio-temporal analysis of Delhi
NCR using landsat datasets. Journal of Urban Management,
9(3), 347–359.
Naji, L., Tawfiq, M., & Jabber, A. K. (2016). Mathematical Modeling
of Groundwater Flow. C Glob. J. Eng. Sci. Res., 3, 2348–8034.
Nampak, H., Pradhan, B., & Abd Manap, M. (2014). Application of
GIS based data driven evidential belief function model to pre-
dict groundwater potential zonation. Journal of Hydrology, 513,
283–300.
Nelder, J. A., & Wedderburn, R. W. (1972). Generalized linear mod-
els. Journal of the Royal Statistical Society: Series A (general),
135(3), 370–384.
Nguyen, P. T., Ha, D. H., Avand, M., Jaafari, A., Nguyen, H. D., &
Al-Ansari, N., etal. (2020). Soft computing ensemble models
based on logistic regression for groundwater potential mapping.
Applied Sciences, 10(7), 2469.
Omrani, H. (2015). Predicting travel mode of individuals by machine
learning. Transportation Research Procedia, 10, 840–849.
Owuor, S. O., Butterbach-Bahl, K., Guzha, A. C., Rufino, M. C., Pel-
ster, D. E., Díaz-Pinés, E., & Breuer, L. (2016). Groundwater
recharge rates and surface runoff response to land use and land
cover changes in semi-arid environments. Ecological Processes,
5(1), 1–21.
Pandey, A. K., Singh, S., Berwal, S., Kumar, D., Pandey, P., Prakash,
A., ... & Kumar, K. (2014). Spatio–temporal variations of urban
heat island over Delhi. Urban Climate, 10, 119–133.
Patel, N. R., Mukund, A., & Parida, B. R. (2022). Satellite-derived
vegetation temperature condition index to infer root zone soil
moisture in semi-arid province of Rajasthan. India. Geocarto
International, 37(1), 179–195.
Pham, Q. B., Kumar, M., Di Nunno, F., Elbeltagi, A., Granata, F.,
& Islam, A. R. M., etal. (2022). Groundwater level prediction
using machine learning algorithms in a drought-prone area.Neu-
ral Computing and Applications, 1–23.
Poudyal, C. P., Chang, C., Oh, H. J.,& Lee, S. (2010). Landslide sus-
ceptibility maps comparing frequency ratio and artificial neural
networks: a case study from the Nepal Himalaya. Environmental
Earth Sciences, 61, 1049–1064.
Raad, S. M. J., Leonenko, Y., & Hassanzadeh, H. (2022). Hydrogen
storage in saline aquifers: Opportunities and challenges. Renew-
able and Sustainable Energy Reviews, 168, 112846.
Rahmati, O., Choubin, B., Fathabadi, A., Coulon, F., Soltani, E., Sha-
habi, H., etal. (2019a). Predicting uncertainty of machine learn-
ing models for modelling nitrate pollution of groundwater using
quantile regression and UNEEC methods. Science of the Total
Environment, 688, 855–866.
Rahmati, O., Golkarian, A., Biggs, T., Keesstra, S., Mohammadi, F., &
Daliakopoulos, I. N. (2019b). Land subsidence hazard modeling:
Machine learning to identify predictors and the role of human
activities. Journal of Environmental Management, 236, 466–480.
Rajasekhar, M., Gadhiraju, S. R., Kadam, A., & Bhagat, V. (2020).
Identification of groundwater recharge-based potential rainwater
harvesting sites for sustainable development of a semiarid region
of southern India using geospatial, AHP, and SCS-CN approach.
Arabian Journal of Geosciences, 13(1), 1–19.
Razandi, Y., Pourghasemi, H. R., Neisani, N. S., & Rahmati, O. (2015).
Application of analytical hierarchy process, frequency ratio, and
certainty factor models for groundwater potential mapping using
GIS. Earth Science Informatics, 8, 867–883.
Reddy, G. P., Mouli, K. C., Srivastav, S. K., Srinivas, C. V., & Maji,
A. K. (2000). Evaluation of ground water potential zones using

Journal of the Indian Society of Remote Sensing
1 3
remote sensing data-A case study of Gaimukh watershed,
Bhandara District, Maharashtra. Journal of the Indian Society
of Remote Sensing, 28(1), 19–32.
Rodell, M., Velicogna, I., & Famiglietti, J. S. (2009). Satellite-based
estimates of groundwater depletion in India. Nature, 460(7258),
999–1002.
Roy, S. S., Rahman, A., Ahmed, S., & Ahmad, I. A. (2020). Alarm-
ing groundwater depletion in the Delhi Metropolitan Region: A
long-term assessment. Environmental Monitoring and Assess-
ment, 192, 1–14.
Sahoo, D., Pham, Q., Lu, J., & Hoi, S. C. (2017). Online deep learning:
Learning deep neural networks on the fly.arXiv preprint arXiv:
1711. 03705.
Sarkar, T., Kannaujiya, S., Taloor, A. K., Ray, P. K. C., & Chauhan,
P. (2020). Integrated study of GRACE data derived interan-
nual groundwater storage variability over water stressed Indian
regions. Groundwater for Sustainable Development, 10, 100376.
Sarker, I. H., Kayes, A. S. M., Badsha, S., Alqahtani, H., Watters, P.,
& Ng, A. (2020). Cybersecurity data science: An overview from
machine learning perspective. Journal of Big Data, 7(1), 1–29.
Shekhar, S., & Prasad, R. K. (2009). The groundwater in the Yamuna
flood plain of Delhi (India) and the management options. Hydro-
geology Journal, 17(7), 1557–1560.
Shorten, C., Khoshgoftaar, T. M., & Furht, B. (2021). Deep Learning
applications for COVID-19. Journal of Big Data, 8(1), 1–54.
Siade, A. J., Cui, T., Karelse, R. N., & Hampton, C. (2020). Reduced‐
dimensional Gaussian process machine learning for groundwa-
ter allocation planning using swarm theory.Water Resources
Research,56(3), e2019WR026061.
Sierikova, E., Strelnikova, E., Pisnia, L., & Pozdnyakova, E. (2020).
Flood risk management of Urban Territories. Ecology Environ-
ment and Conservation, 26(3), 1068–1077.
Singh, A., & Mukherjee, S. (2014). Groundwater Exploration: Geo-
physical, Remote Sensing, and GIS Techniques. Handbook of
Engineering Hydrology: Fundamentals and Applications, 207.
Singh, C. K., Kumar, A., Shashtri, S., Kumar, A., Kumar, P., & Mal-
lick, J. (2017). Multivariate statistical analysis and geochemical
modeling for geochemical assessment of groundwater of Delhi,
India. Journal of Geochemical Exploration, 175, 59–71.
Singh, Y. K., De Waele, B., Karmakar, S., Sarkar, S., & Biswal, T. K.
(2010). Tectonic setting of the Balaram-Kui-Surpagla-Kengora
granulites of the South Delhi Terrane of the Aravalli Mobile Belt,
NW India and its implication on correlation with the East Afri-
can Orogen in the Gondwana assembly. Precambrian Research,
183(4), 669–688.
Srivastava, S. K., & Ramanathan, A. L. (2008). Geochemical assess-
ment of groundwater quality in vicinity of Bhalswa landfill,
Delhi, India, using graphical and multivariate statistical methods.
Environmental Geology, 53, 1509–1528.
Steensen, B. M., Marelle, L., Hodnebrog, Ø., & Myhre, G. (2022).
Future urban heat island influence on precipitation. Climate
Dynamics, 58(11–12), 3393–3403.
Su, L., Miao, C., Duan, Q., Lei, X., & Li, H. (2019). Multiple-wavelet
coherence of world’s large rivers with meteorological factors and
ocean signals. Journal of Geophysical Research: Atmospheres,
124(9), 4932–4954.
Tomer, T., & Katyal, D. (2021). Assessment of Groundwater Vulnera-
bility to Pollution by using GIS based DRASTIC Model in Delhi
Region.IWRA (India) Journal,10(1), 8–11.
Tomer, T., Katyal, D., & Joshi, V. (2019). Sensitivity analysis of
groundwater vulnerability using DRASTIC method: A case
study of National Capital Territory, Delhi, India. Groundwater
for Sustainable Development, 9, 100271.
Trichakis, I. C., Nikolos, I. K., & Karatzas, G. P. (2011a). Artificial
neural network (ANN) based modeling for karstic groundwa-
ter level simulation. Water Resources Management, 25(4),
1143–1152.
Trichakis, I., Nikolos, I., & Karatzas, G. P. (2011b). Comparison of
bootstrap confidence intervals for an ANN model of a karstic
aquifer response. Hydrological Processes, 25(18), 2827–2836.
Vishal, V., Kumar, S., & Singhal, D. C. (2014). Estimation of ground-
water recharge in national capital teriitory, Delhi using ground-
water modeling.
Wang, L., Li, P., Duan, R., & He, X. (2022). Occurrence, va factors and
health risks of Cr6+ in groundwater in the Guanzhong Basin of
China. Exposure and Health, 14(2), 239–251.
Wang, W., Kiik, M., Peek, N., Curcin, V., Marshall, I. J., Rudd, A. G.,
... & Bray, B. (2020). A systematic review of machine learn-
ing models for predicting outcomes of stroke with structured
data.PloS one,15(6), e0234722.
Wendt, D. E., Van Loon, A. F., Scanlon, B. R., & Hannah, D. M.
(2021). Managed aquifer recharge as a drought mitigation strat-
egy in heavily-stressed aquifers. Environmental Research Letters,
16(1), 014046.
Wray, R. A., & Sauro, F. (2017). An updated global review of solu-
tional weathering processes and forms in quartz sandstones and
quartzites. Earth-Science Reviews, 171, 520–557.
Yang, X., Liu, D., & Wang, D. (2014). Reinforcement learning for
adaptive optimal control of unknown continuous-time nonlinear
systems with input constraints. International Journal of Control,
87(3), 553–566.
Yar, P. (2020). Urban development and its impact on the depletion of
groundwater aquifers in Mardan City. Pakistan. Groundwater for
Sustainable Development, 11, 100426.
Yifru, B. A., Chung, I. M., Kim, M. G., & Chang, S. W. (2021). Assess-
ing the effect of land/use land cover and climate change on water
yield and groundwater recharge in East African Rift Valley using
integrated model. Journal of Hydrology: Regional Studies, 37,
100926.
Youssef, A. M., Pourghasemi, H. R., Pourtaghi, Z. S., & Al-Katheeri,
M. M. (2015). Landslide susceptibility mapping using random
forest, boosted regression tree, classification and regression tree,
and general linear models and comparison of their performance
at Wadi Tayyah Basin, Asir Region. Saudi Arabia. Landslides,
13(5), 839–856.
Zeng, X., Zhang, J., Yu, L., Zhu, J. X., Li, Z., & Tang, L. (2019).
A sustainable water-food-energy plan to confront climatic and
socioeconomic changes using simulation-optimization approach.
Applied Energy, 236, 743–759.
Zhao, J. (2017). Reducing bias for maximum approximate conditional
likelihood estimator with general missing data mechanism. Jour-
nal of Nonparametric Statistics, 29(3), 577–593
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds
exclusive rights to this article under a publishing agreement with the
author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of
such publishing agreement and applicable law.