Available via license: CC BY 3.0
Content may be subject to copyright.
1
Spatial Extent of Future Changes in the Hydrologic Cycle
Components in Ganga Basin using Ranked CORDEX RCMs
Jatin Anand 1, Manjula Devak 1, Ashvani Kumar Gosain 1, Rakesh Khosa 1 and Chandrika
Thulaseedharan Dhanya 1
1 Department of Civil Engineering, Indian Institute of Technology Delhi, New Delhi-110016, India
5
Correspondence to: Dr Chandrika Thulaseedharan Dhanya (dhanya@civil.iitd.ac.in)
Abstract. The negative impacts of climate change are expected to be felt over wide range of spatial scales, ranging
from small basins to large watersheds, which can possibly be detrimental to the services that natural water systems
provide to the society. The impact assessment of future climate change on hydrologic response is essential for the
10
decision makers while carrying out management and various adaptation strategies in a changing climate. While,
the availability of finer scale projections from regional climate models (RCM) has been a boon to study changing
climate conditions, these climate models are subjected to large number of uncertainties, which demands a careful
selection of an appropriate climate model, however. In an effort to account for these uncertainties and select
suitable climate models, a multi-criteria ranking approach is deployed in this study. Ranking of CORDEX RCMs
15
is done based on its ability to generate hydrologic components of the basin, i.e., runoff simulations using Soil
Water Assessment Tool (SWAT) model, by deploying Entropy and PROMETHEE-2 methods. The spatial extent
of changes in the different components of hydrologic cycle is examined over the Ganga river basin, using the top
three ranked RCMs, for a period from January 2021 to December 2100. It is observed that for monsoon months
(June, July, August and September), future annual mean surface runoff will decrease substantially (-50 % to -
20
10%), while the flows for post-monsoon months (October, November and December) are projected to increase
(10- 20 %). While, extremes are seen to be increasing during the non-monsoon months, a substantial decrease in
medium events is also highlighted. The increase in wet extremes is majorly supplemented by the increased
snowmelt runoff during those months. Snowmelt is projected to increase during the months of November to
March, with the month of December witnessing 3-4 times increase in the flow. Base flow and recharge are
25
alarmingly decreasing over the basin. Major loss of recharge is expected to occur in central part of the basin. The
present study offers a more reliable regional hydrologic impact assessment with quantifications of future dramatic
changes in different hydrological sub-system and its mass-transfer, which will help in quantifying the changes in
hydrological components in response to climate change changes in the major basin Ganga, and shall provide the
water managers with substantive information, required to develop ameliorative strategies.
30
Keywords: Climate change, Uncertainty, Soil Water Assessment Tool (SWAT), CORDEX, General Circulation
Model (GCM), Regional Climate Models (RCM), Ganga River.
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
2
1 Introduction
The recent years have seen massive industrialization and the amplified use of fossil fuels, which have prompted
an extraordinary increase in the concentrations of greenhouse gases in the atmosphere. The climate change due to
various anthropogenic activities is considered as “extremely likely”, thereby causing detrimental effects for both
human and natural systems at varied spatial scales (Franczyk and Chang, 2009; IPCC, 2014; Zhang et al., 2012).
5
One of the most important and immediate concerns is the effect of these alterations in water resources systems
both spatially and temporally, especially for the regions where available water reserves are already stressed, due
to population growth, industrial development and agricultural needs. In recent decades, an indubitable relation
between the climate change and water resources together with the potential impacts of climatic change on water
resources and hydrology has gained considerable attention in hydrologic research community (Abbaspour et al.,
10
2009; Akhtar et al., 2008; Amadou et al., 2014; Gardner, 2009). The study of potential impacts of future climate
change on water resources, as a way to recognize suitable mitigation and adaptation strategies holds utmost
importance. There are many observations supporting the evidence of climate change, such as frequent occurrences
of climatic extremes, rising sea level and diminishing snow packs (Amadou et al., 2014; Githui et al., 2009; Shamir
et al., 2015; Yu et al., 2002). Since the climate and hydrologic cycle is inter linked, variability in climate is
15
expected to alter both the timings and magnitudes of surface runoff. Consequently, climate change has important
ramifications on both existing water resources system as well as future water resources planning and management
(Abbaspour et al., 2009; Caballero et al., 2007; Fujihara et al., 2008; Xu et al., 2004). Moreover, the imbalance
between the water demands and water supply has been increasing of late due to change in climate. The above
stated studies confirm that there still exists a huge gap to relate the alterations in climate with water resources to
20
serve the needs of the decision makers.
To investigate the impact of climate change on future runoff generation, the reliable simulations from various
General Circulation Models (GCMs), under different scenarios, are used as inputs to hydrological models. These
scenarios are generated by considering the different greenhouse gas emissions and modified initial conditions
(Abbaspour et al., 2009; Serrat-Capdevila et al., 2007). Despite the skill of GCMs at global scales, the spatial
25
resolution of climate variable, such as precipitation, which is an important component for hydrological modeling
of river system, is still too coarse to be used directly as an input to various hydrologic models. (Akhtar et al., 2008;
Fujihara et al., 2008; Orlowsky et al., 2008; Serrat-Capdevila et al., 2007; Steele-Dunne et al., 2008; Troin et al.,
2015). Moreover, GCM outputs have a serious spatio-temporal bias, which needs to be corrected before feeding
to any hydrologic models, as it may lead to unrealistic simulations. In addition, the accuracy of GCMs varies from
30
one climate variable to other. The main sources contributing to the uncertainty of GCM outputs are due to
inadequate knowledge of current system, imperfect representation of various physical processes and incapability
to produce inter-annual and inter-decadal variability (Arora, 2001; Christensen et al., 2008; Troin et al., 2015).
Owing to the coarse resolution of GCMs, downscaling of climate variables becomes necessary. Apart from model
uncertainty, scenario uncertainty and internal variability, the downscaling of these GCM scenarios give birth to
35
one more uncertainty i.e. downscaling uncertainty (Abudu et al., 2012; Liang et al., 2008; Troin et al., 2015).
Hence, the outputs derived from downscaled regional climate models (RCM) at relatively finer scales, may aid to
limit these uncertainties (Shamir et al., 2015; Troin et al., 2015). The finer resolutions coupled with dedicated
physics associated with RCMs can improve the physical processes (Paxian et al., 2016; Vanvyve et al., 2008),
pertaining to the variability in regional climate, especially when applied over large and complex basins like Ganga
40
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
3
river basin. However, the comparison between RCM simulations and observed data demonstrates that some
systematic biases resulting from imperfect adaptation of physical processes; numerical approximations of model’s
equations and other assumptions; still prevails within the climate variables (Eden et al., 2014; Fujihara et al.,
2008). Given the substantial inconsistencies in RCM simulated fields compared to observations, it is often a
prerequisite to process data before it is fed to hydrological models. This prerequisite process referred, as bias
5
correction, is often necessary for a meaningful translation of climate variables, whereby the RCM results featuring
the current prevailing climate are corrected to match with the observations (Johnson and Sharma, 2015; Najafi
and Moradkhani, 2015; Troin et al., 2015). Frequently adopted methods for bias correction are Quantile-Quantile
(Q-Q) mapping for historic scenario and Equidistant Cumulative Distribution Function (EDCDF) matching
method for future scenario (Amadou et al., 2014; Devak et al., 2015; Johnson and Sharma, 2015; Li et al., 2010;
10
Najafi and Moradkhani, 2015; Troin et al., 2015). Since, the simulation of RCMs incorporates various surface
information (soil type, vegetation and hydrology), the selection of RCM is highly region dependent. Hence,
ranking of RCMs is desirable before coupling with any modeling study (Behzadian et al., 2010; Raju and Kumar,
2014; Singh et al., 2016; Srinivasa Raju et al., 2016). The ability of regional climate model to reproduce the
climate conditions, over the current period, forms the basis for the selection of model. Hence, the ranking of
15
RCM’s by comparing observed runoff with the simulated runoff comprises one of the main objectives of this
study.
The future changes in various hydrological components such as runoff, groundwater, and evapotranspiration (ET)
can be projected through hydrological modeling. Efficient and reliable simulation of the different hydro-
meteorological conditions presiding within a watershed and quantification of the interrelationships between
20
topography, soil layer, vegetation and land use, is a vital step. It is also known that a better way of managing and
assessing available water resources is at basin level (Stehr et al., 2008). The belief, that the climate change would
have an evident impact on different hydrologic components, is established by the long-term assessment of
catchment’s water balance (Abbaspour et al., 2009; Serrat-Capdevila et al., 2007). Hence, watershed models are
essential for studying hydrologic processes and their response to different climate changes (Eckhardt and Ulbrich,
25
2003; Li et al., 2009; Zeng and Cai, 2014; Zhang et al., 2012). Several physically based hydrological models have
been recognized and adopted to simulate hydrological processes in a river catchment. Many of the previous studies
were either restricted to the examination of particular components of water balance (For e.g., surface runoff,
groundwater recharge and evapotranspiration) or were focused to a particular event of a year and seasonal
processes (For e.g., high flows, low flows, extreme events or seasonal undulations) (Caballero et al., 2007; Fu et
30
al., 2007; Li et al., 2009; Scibek and Allen, 2006). An integrated hydrological simulation approach could benefit
in understanding the net impact of climate change in a given region (Islam and Gan, 2015; Oguntunde and
Abiodun, 2013).
Following the above discussion, we selected river Ganga as our study basin, to implement the objectives. Being
an agrarian–based country, India’s economy is highly dependent on proper water resources planning and
35
management of various basins. In particular Ganga river basin, covering 33% of the total geographic area of
Indian-sub continent plays a major role in economy. The catchment of Ganga is distributed unequally among
several states and has repetitively facing issues such as floods and drought sequences. Areas, especially, residing
in downstream parts of the basin suffers with reduced surface runoff in the dry season and substantial increased
flows during the wet season (Akhtar et al., 2008; Bharati et al., 2011; Gardner, 2009; Rees et al., 2004; Whitehead
40
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
4
et al., 2015). Furthermore, the Ganga river basin is highly governed by snowfall accumulation and snowmelt
processes within the catchment. Since snowmelt processes provide most of the surface water during non-monsoon
season, its monitoring is important in hydrologic simulations. Any disturbance of the hydrologic characteristics
in snow-dominated areas in this region could have major impacts on the available water resources. Furthermore,
the Ganga river basin, which is already under huge irrigation, industrial and domestic pressure, together with the
5
probable consequences of climate change, demand a proper insight and understanding on future climate and their
effects on available water wealth. Therefore, in Ganga river basin future water resources assessment, minimizing
the uncertainties, is important for decision makers and stakeholders for planning and operation of hydrological
installations under climate change.
Hence, the present study focuses on the assessment of the hydrological impacts of climate change on water
10
resources in Ganga river basin for the near (2021-2060) and far (2061-2100) future using different regional climate
scenarios. The study aims to assess the available water in the basin on account of varying climatic conditions. In
this study, to minimize the climate model uncertainty, firstly the RCMs are ranked based on their ability to mimic
the hydrology of the basin with the help of a multi-criterion decision making method, viz. PROMETHEE.
Secondly, top three ranked RCMs obtained through PROMTHEE-based ranking approach are used to determine
15
the impact of climate change at sub-basin level and at a monthly time step for the Ganga basin. Subsequently,
impact of climate change on various components of water balance of the basin is assessed, which is further used
to quantify the changes in water resources in terms of different hydrologic components.
2 Study area
Ganga river basin, which serves an area of about 1.08 Mkm2 and covers a stretch of 1200km, finally meets the
20
Bay of Bengal in the east. The basin is situated in the northern part of the Indian sub-continent, traces between
the latitudes of 22°30' and 31°30' North and the longitude of 73°30' and 89°00' East (Fig. 1). The catchment area
of 0.86 Mkm2 in India, which is nearly 26 percent of total geographic area of the country, is shared among eleven
states (Himachal Pradesh, Uttarakhand, Uttar Pradesh, Madhya Pradesh, Chhattisgarh, Bihar, Jharkhand, Punjab,
Haryana, Rajasthan, West Bengal, and the Union Territory of Delhi). The primary sources of water in the river
25
are precipitation, base flow, and snowmelt water from the Himalayas. Average annual precipitation varies between
300 mm to 2000 mm, with the western part of the catchment getting less precipitation in comparison to the eastern
part. Most of this rainfall is concentrated in the monsoon months of June to October, creating low surface runoff
conditions in the Ganga river basin and its tributaries, thereby causing dry conditions during non-monsoon periods
(November to May). The temperature during winter season ranges from 2°C to 15°C, while that during summer
30
season varies from 25°C to 45°C (Singh, 1994). The impact of climate change on rainfall and temperature patterns
is expected to be widely varying across the catchment, which in turn will affect the spatial and temporal
distributions of different constituents of water resources.
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
5
Fig. 1. Location of the Ganga River Basin in India
3 Data description
3.1 Observational datasets
Daily precipitation and temperature data of 0.5° × 0.5° and 1° × 1° spatial resolution, respectively for the period
5
1985-2005, provided by India Meteorological Department (IMD) are used as observed datasets. The gridded daily
rainfall data is developed by incorporating data from more than 3000 rain gauge stations (quality controlled) using
Shepard’s interpolation method (Rajeevan et al., 2006). Gridded daily temperature dataset is developed by
incorporating temperature gauge data from 395 locations using the same interpolation technique (Srivastava et
al., 2009).
10
Observed daily flows at different gauge stations (Fig. 3) for the period 1990 to 2005, acquired from Central Water
Commission (CWC), India are used for the calibration (1990-1999) and validation (2000-2005) of the
hydrological model.
3.2 Other hydrological model inputs
In addition to the meteorological data (mentioned above), model inputs required to run the hydrological model,
15
include the Digital Elevation Model (DEM), land use layer, soil distribution map of the study area (Table 1).
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
6
Table 1. Data description along with its sources used in the SWAT model calibration
Data
Spatial Resolution
Source
Digital Elevation
Model (DEM)
90m × 90m
Shuttle Radar Topography Mission (SRTM)
(http://www2.jpl.nasa.gov/ srtm/)
Soil
1:5,000,000 scale
Food and Agriculture Organization (FAO)
(http://www.fao.org/nr/land/soils/digital-soil-map-of-the-
world/en/)
Landuse
1:250,000 scale
National Remote Sensing Center (NRSC)
3.3 Climate Model data
In this study, simulations of fourteen RCMs from Coordinated Regional Climate Downscaling Experiment
(CORDEX) are extracted (http://cccr.tropmet.res.in/home/ftp_data.jsp). RCMs are chosen based on their data
5
availability over the study region (Table 2). Daily precipitation, maximum and minimum temperature for
historical (1985-2005) and future scenario (RCP 4.5, 2021-2100) are extracted from these RCMs.
Table 2. Details of 14 RCMs considered in this study
S.No.
Institute
RCM
Spatial resolution
1
IAES,GUF
CCLM4
0.45°×0.45°
2
CSIRO
CCAM(ACCESS)
0.5°×0.5°
3
CSIRO
CCAM(CNRM)
0.5°×0.5°
4
CSIRO
CCAM(GFDL)
0.5°×0.5°
5
CSIRO
CCAM(MPI)
0.5°×0.5°
6
CSIRO
CCAM(BCCR)
0.5°×0.5°
7
CCCR, IITM
RegCM4
0.45°×0.45°
8
SMHI
RCA4(CNRM-CERFACS)
0.45°×0.45°
9
SMHI
RCA4(NOAA-GFDL)
0.45°×0.45°
10
SMHI
RCA4(ICHEC)
0.45°×0.45°
11
SMHI
RCA4(IPSL)
0.45°×0.45°
12
SMHI
RCA4(MIROC)
0.45°×0.45°
13
SMHI
RCA4(MPI-M)
0.45°×0.45°
14
GERICS
REMO 2009
0.5°×0.5°
10
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
7
4 Model description
4.1 The hydrologic simulator (SWAT)
One of the promising hydrological models is SWAT, which has been adjudged by researches as computationally
efficient (Abbaspour et al., 2015; Neupane and Kumar, 2015; Shawul et al., 2013). The impacts of climate change
on hydrology of the basin are quantified based on the simulations made by SWAT hydrological model developed
5
by the United States Department of Agriculture (USDA) (Arnold et al., 2013). The SWAT model is a physically
based semi-distributed continuous time model that can operate on a large basin and can simulate several processes
such as flows in rivers, sediment transport and so on, on a daily/sub-daily time step (Arnold et al., 2013; Neitsch
et al., 2002). SWAT simulates various hydrologic processes, including surface runoff generation, using either
SCS curve number method or the Green and Ampt infiltration equation. Model offers vivid methods such as,
10
Hargreaves, Priestley-Taylor or Penman-Monteith methods for the quantification of evapotranspiration.
Groundwater flow, lateral flow and percolation are assessed through mass balance of the underlying system.
SWAT model assess the impact of land use changes on water supplies and erosion in large-scale catchments.
Initially, SWAT simulates each hydrological response unit’s (HRU) water balance to estimate the amount of water
available for each sub-basin’s main channel at a given step, which is then routed to determine the movement of
15
water through the river system towards the basin outlet.
5 Methodology
For the present climate simulation, bias corrected forcings are used as an input for calibrated SWAT model, which
is used further for the ranking of the RCMs based on runoff. In the future climate simulations, the top three ranked
RCMs are used to simulate the impact of climate change. Figure 2 illustrates the schematic diagram of the
20
methodology adopted in this study, to achieve the objectives.
Fig. 2. Schematic diagram of the methodology adopted in the study
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
8
5.1 Bias correction method
Biases are responsible for poor characterizations of inter annual variability, which is crucial while assessing the
impacts of climate change on water resources availability and infrastructure. The quantile-quantile (Q‐Q)
transformation, also referred to as quantile-quantile mapping or quantile matching, is a widely adopted technique
to bias correct the different climatic variables. Quantile-quantile mapping scales the climate variables according
5
to the observed daily series. This method aims to make the statistical distribution of a climate variable as close as
possible to the statistical distribution of the observed variable. Either an empirical distribution or fitted probability
distribution is adopted in Q-Q mapping to match the observed and modeled rainfall quantiles (Amadou et al.,
2014; Johnson and Sharma, 2015; Li et al., 2010, Troin et al., 2015). In addition to the simplistic application,
quantile-quantile mapping corrects the full distribution of precipitation amounts including the extreme wet and
10
dry events, while maintaining the rank correlation between modeled outputs and observations. This method also
takes care of the biases, if any skewness is present in the rainfall distribution. However, an underlying assumption
of this method is that the climate distribution does not change much over time i.e., stationarity exists in terms of
the variance and skewness of the distribution, and only the mean changes. Hence, Equidistant Cumulative
Distributive Function matching (EDCDFm) method proposed by Li et al. (2010) is deployed in this study, which
15
considers the change of distribution in future. Thus, limitation of Q-Q mapping is dealt by EDCDFm, which adjust
the CDF of the model future projections based on the difference between CDFs of model and observed for the
considered baseline period. EDCDF method has been widely used in several studies related with climate change
and bias correction (e.g., Devak et al., 2015; Wang et al., 2014).
5.2 Ranking of RCMs
20
5.2.1 Performance indicators
Different performance indicators are widely used to analyze the performance of the modeled data with respect to
the observed time series. Five statistical indicators, namely Nash-Sutcliffe efficiency (NSE), coefficient of
determination (R2), normalized root mean square error (NRMSE), absolute normalized mean bias error (ANMBE)
and average absolute relative error (AARE) are considered among the numerous performance indicators available
25
(Moriasi et al., 2007). Nash-Sutcliffe efficiency (NSE) is a normalized statistic that determines the degree of
residual variance compared to the measured data variance. NSE indicates how well the computed output matches
the observed data along the 45o line (Moriasi et al., 2007). Coefficient of determination (R2) determines the degree
of collinearity between measured and computed data. R2 can range from 0 to 1 with higher values indicating good
correlation. Usually higher values of R2 ( >0.5) are considered satisfactory (Moriasi et al., 2007). NRMSE is a
30
measure of the difference between the measured data and the model simulated data. Typically, smaller values of
NRMSE suggest better model simulation (Raju and Kumar, 2014). AARE is the average of the absolute values of
relative errors. Smaller values indicates better model simulation (Raju and Kumar, 2014). The details of the
aforementioned statistical performance measures are tabulated in Table 3.
35
Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-189, 2017
Manuscript under review for journal Hydrol. Earth Syst. Sci.
Discussion started: 29 May 2017
c
Author(s) 2017. CC-BY 3.0 License.
9
Table 3. Details of performance measures used in the study
S.No.
Performance Measures
Formula
Ideal
value
1
Nash-Sutcliffe
efficiency (NSE)
n
ii
n
iii
XX
YX
NSE
1
2
1
2
1
1
2
Coefficient of
Determination (R2)
2
11
2
2
11
2
111
2
n
ii
n
ii
n
ii
n
ii
n
ii
n
ii
n
iii
YYnXXn
YXYXn
R
1
3
Normalized Root Mean
Square Error (NRMSE)
X
YX
n
NRMSE
n
iii
5.0
1
2
1
0
4
Absolute Normalized
Mean Bias Error
(ANMBE)
X
YX
n
ANMBE
n
iii