Quantifying the impact of land cover changes on hydrological extremes in India

Several research studies have addressed the effects of future climate changes on the hydrological regime of Mahanadi river basin located in eastern part of India. However, studies investigating the effects of future land cover changes on hydrology are limited owing to the 10 lack of availability of projected land cover scenarios. Our study investigates how the hydrology of Mahanadi river basin would respond to the current and future land cover scenarios under a large-scale hydrological modelling framework. Both historical and future land cover scenarios from the recently released, Land use Harmonisation (LUH2) project for CMIP6, indicates cropland and forest are the major land cover types in the basin with a noticeable 15 increase in the cropland (23.3%) at the expense of forest (22.65%) by the end of year 2100 compared to the baseline year, 2005. A physically semi-distributed model, the Variable Infiltration Capacity has been set up and implemented over the Mahanadi river basin system for the time period 1990-2010. The uncertain model parameters were subjected to Sensitivity Analysis and calibrated within a Monte Carlo framework. The best set of calibrated models 20 obtained is used in conjunction with the harmonized set of present and future land use scenarios from LUH2 at 25 km by 25 km resolution to generate an ensemble of model simulations that captures a range of plausible impacts of land cover changes on discharge and other hydrological components of the basin. Overall, model simulation results indicate an increase in the extreme flows (i.e., 95th percentile or higher) in the range of 0.12 to 21 % at 25 multiple subcatchments within the basin. This increase can be attributed to the direct conversion of forested areas to agriculture (on the order of 30,000 km2) that has reduced the Leaf Area Index and subsequently reduces the Evapotranspiration (ET). These changes ultimately affect other water balance components at the land surface, resulting in an increase in surface runoff and baseflow, respectively. 30 https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c © Author(s) 2020. CC BY 4.0 License.


Context and Background
Land use and land cover change (LULC) induced by the rapid anthropogenic activities, is one 35 of the major causes of change in hydrological and watershed processes (Rogger et al., 2016).
Alterations of existing land cover types and land management practices in a catchment can thereby, significantly modify the rainfall path into runoff by changing the hydrological dynamics such as surface runoff, baseflow, Evapotranspiration (ET), water holding capacity of the soil, interception and groundwater recharge , thus reflecting a change in the water 40 demand (Berihun et al., 2019;Bosch and Hewlett, 1982;Costa et al., 2003;Foley et al., 2005;Garg et al., 2017;Hamman et al., 2018;Mao and Cherkauer, 2009;Zhang et al., 2014). Rapid growth in population in the developing countries has prominent effects on LULC dynamics through deforestation at the expense of increased agricultural production. Deforestation among all other land use changes is the major cause of modifying various hydrological 45 processes such as ET, surface runoff, baseflow and snowmelt processes (Dwarakish and Ganasri, 2015;Gao et al., 2009). The complex relationships between the human induced land cover change and the hydrological processes have gained widespread attention among various scientific communities across the world. In this regard, several studies have been carried out that links the LULC changes and the hydrological dynamics within a river basin 50 (Abe et al., 2018;Behera et al., 2018;Berihun et al., 2019;Chu et al., 2010;Costa et al., 2003;Rogger et al., 2016;Thomson et al., 2018;Wang et al., 2008;Wilk and Hughes, 2002).
However, the exact role of LULC changes in modifying river discharge is still elusive (Rogger et al., 2016) and therefore, remains a challenge to isolate the sole impacts of land use changes on hydrology of a river basin (Tsarouchi and Buytaert, 2018). The challenge also lies in solving 55 these complex processes in a heterogenous catchment coupled with limited hydroclimatological data (Gebremicael et al., 2019;Li and Sivapalan, 2011).
Changes in land use and cropping patterns are modifying the hydrological cycle in many river basins of India. In 1980's, Central, North-Eastern and Peninsular India was bestowed with woody savannas which is mostly forest lands (Paul et al., 2016). Since then, rapid urbanization 60 and agricultural intensification are the constant reasons behind the depletion of natural https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
vegetations and conversion of woody savannas to the cropland. As per the Land Use and Land Cover (LULC) map of 2005, cropland is the dominant land cover type in India. The analysis report of world "greening" from MODIS (2000MODIS ( -2017 showed a significant increase of 82% in the greening over India, which is found to be entirely associated with cropland (Chen et al.,65 2019; IPCC, 2019). Several researchers (Babar and Ramesh, 2015;Bosch and Hewlett, 1982;Gebremicael et al., 2019;Wilk and Hughes, 2002) agree that the expansion in agricultural land at the expense of vegetative cover results in an increase in surface runoff and decreases river discharge in a given watershed. Wilk and Hughes, (2002) found a maximum increase of 19% in the runoff from a tropical catchment in South India due to the expansion in agriculture at 70 the expense of forest and savannas. Babar and Ramesh, (2015) estimated an increase in runoff (0.9%) and decrease in ET (4.5%) due to the conversion of forest to agriculture and built up areas in Ganga basin in India. The population of India is expected to increase by an average of 36% in the 2050's and by 108% at 2090's thereby rendering changes to agriculture and water demand (Jin et al., 2018). Many river basins of India that have undergone drastic 75 land cover transformation over the years have been facing extreme hydrological events like floods and droughts in recent times. Long-term changes in climate and land use are reported as the main reasons causing these hydrological extremes (IPCC, 2019).
Climate and LULC governs the hydrological cycle in a basin through an intricate relationship involving a wide range of interactions among the land surface variables at different spatial 80 and temporal scales. These interactions can be best solved through the implementation of process-based and physically based distributed or semi-distributed hydrological models, representing the land surface characteristics of a heterogeneous catchment, and simulating the multi-layered hydrological processes. Therefore, the selection of an appropriate hydrological model is quite essential. The Variable Infiltration Capacity (VIC) model is a large 85 scale physically semi-distributed land surface model developed by Liang et al., (1994). The ability of the model to simulate the impacts of LULC changes on hydrology are well documented in various research articles (Garg et al., 2017(Garg et al., , 2019Hurkmans et al., 2009;Mao and Cherkauer, 2009;Patidar and Behera, 2019;Zhang et al., 2014).
Eastern part of India is amongst the most rapidly changing landscape over the country, 90 specifically, Mahanadi river basin in Eastern India have undergone drastic land cover changes in the last decades (Behera et al., 2018;Dadhwal et al., 2010). To the best of our knowledge, only one study, Das et al., (2018) predicted the land cover change impacts on the future (year https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. 2025) water balance components of Eastern Indian river basins. Dadhwal et al., (2010) simulated the effects of deforestation and agricultural expansion on the annual streamflow In this study, we address the overall science question: What is the isolated role of LULC change on the water balance of a large river basin? To understand this, two specific research questions arises are: (1) Can we identify changes in surface water balance due to changes in 115 land cover while using a regional hydrological model? and (2) What are the uncertainties associated with model parameters obtained for the regional consequences and how they affect simulated water balance components? This paper specifically focusses on the Mahanadi river basin in India. We identify best ensembles of daily model simulations calibrated within a Monte Carlo Framework, which accounts for the model parameter 120 uncertainties, to evaluate the LULC changes impacts on the hydrology of the basin. The land cover scenarios used in this study represents future changes in the LULC under changing climate (RCP's) and socio-economic conditions (SSP's). This is the first study that uses applications of VIC model in conjunction with gridded land cover forcing's under combined SSP and RCP scenarios in Mahanadi river basin. The outcome of this study is a range of 125 https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
hydrologic predictions associated with the model parameter uncertainties owing to the land cover changes occurring in the Mahanadi river basin, allowing better understanding and implementation of the adaptation and mitigation strategies in the future.

Research Area 130
Geographical Overview The Mahanadi river basin is located in the eastern part of India ( Figure 1) and drains an area of 141,589 km 2 , which nearly accounts for 4.3% of the total geographical area of India. The basin has a varying topography with its lowest elevated area (-17 m) lying in the coastal 135 reaches and the highest elevated area (1323 m) in the northern hills. The basin is characterized by tropical climate zone and receives rainfall from southwest monsoons which commences in June and lasts till September. The average annual rainfall is 1572 mm, with ~ 78% of the total annual rainfall occurring during the monsoon months. The basin is also subjected to spatial variability in terms of receiving rainfall which has resulted in floods in 140 some parts of the basin and drought in others. The mean annual discharge is 1895 m 3 /s. Hirakud dam with a gross storage capacity of 8.136 km 3 is the major hydro project in the river basin constructed in the year 1957 to alleviate the flood problems and to serve multiple other purposes such as irrigation, hydropower generation and supplying drinking water. About 65% of the basin is placed upstream of the dam. Despite its significant storage capacity, the large 145 flows from the catchment upstream as well as from the middle reaches (i.e. between Hirakud dam and Mundali weir) causes devastating floods during the monsoon in the deltaic region of the basin.
About 48% of the total area is under agriculture (Figure 2a), out of which 30% is cropped during the kharif season (monsoon season) and 15% is under double or triple irrigation. The with a minimal increase of only 0.3% (NRSC, 2014). In addition, loamy and clayey are the major soil types covering 53.33% and 41.5% respectively of the total basin area (NBBSS-LUP, India).
Approximately 90% of the basin has moderately shallow to deep soil having depth greater 160 than 50 cm.

Model structure and Implementation 165
The VIC-3L model is a semi-distributed macroscale hydrological model which solves either only water balance or full water and energy balance at each grid cell for three soil layers (Cherkauer and Lettenmaier, 1999). The model in water balance mode assumes air temperature to be same as the surface temperature. The most distinguishable features of this 170 model includes, maintaining sub-grid heterogeneity for the vegetation covers and sub-grid variability in soil moisture storage capacity (Liang et al., 1994), causing surface flow considering both infiltration excess and saturation excess (Bao et al., 2011) and occurrence of baseflow from the third soil moisture layer as a non-recession flow (Zhao et al., 1980).
In VIC-3L, direct runoff occurs from the top thin layer. The Middle soil layer allows for diffusion 175 of water to the uppermost soil layer provided the middle soil layer is wetter. Evaporation occurs from all the soil layers and baseflow occurs from the third layer. Sub grid spatial variability in soil moisture storage is represented by a variable infiltration curve where the model assumes that the infiltration capacity is the non-linear function of the soil moisture storage within the grid cell (Liang et al., 1994). More details regarding the structure and 180 formulations of the model can be found in the literature (Gao et al., 2010;Liang et al., 1994).
To obtain the discharge at the outlets of multiple subcatchments, the VIC-3L model is coupled to a stand-alone routing model (Lohmann et al., 1996). Lohmann's routing model follows a simple river routing scheme where runoff and baseflow are first routed to the edge of the grid cells using an instantaneous unit hydrograph and finally transported to the river/channel 185 network using a linearized St. Venant's equation.
The VIC model has been widely applied at multiple scales ranging from global to continental to large scale basins under various land use/climate scenarios (Dadhwal et al., 2010;Hurkmans et al., 2009;Mao and Cherkauer, 2009;Matheussen et al., 2002;Mishra et al., 2010;Yang et al., 2014). Also the model has been used successfully in simulating hydrological 190 https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. processes in many Indian river basins under different environmental conditions (Garg et al., 2017(Garg et al., , 2019Mishra et al., 2008Mishra et al., , 2010Naha et al., 2016;Patidar and Behera, 2019) Table 1) such as initial soil moisture content, Fractional soil moisture content at critical point, Wcr_frac and Fractional soil moisture content at wilting point, Wp_frac are calculated based on average hydraulic properties of USDA soil textural classes (Cosby et al., 215 1984;Rawls et al., 1998;Reynolds et al., 2000). The LULC map which is used in the model runs while performing sensitivity analysis, model calibration and validation is derived from NRSC of year 2005 (scale 1:250000; resolution 56 meters) which was reformatted and reclassified into USGC LULC types as required by the VIC model ( Figure 2a). The root zone depth of each LULC types and the fraction of vegetation roots in each root zone is obtained from the 220 literatures of (Nijssen et al., 1997;Raje et al., 2014;Zeng, 2002). Other vegetation properties in the vegetation library file such as LAI, roughness length, albedo, architectural resistance, https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. stomatal resistance and displacement height are assembled based on Global Land Data Assimilation System (GLDAS). LAI is known to exert strong influence on runoff and ET simulated by the VIC model (Gao et al., 2010;Matheussen et al., 2002). Hence, we derived

Model calibration and validation
VIC-3L model parameters are at first subjected to Sensitivity Analysis (SA) in priory to define 240 the key parameters needed to be calibrated. Table 1 lists all the important VIC-3L model parameters that are either estimated or subjected to SA and model calibration. The uncertain parameters to be included in the SA process and their ranges (Table 1) Table 1 (Demaria et al., 2007;Matheussen et al., 2002;Mishra et al., 2008;Park and Markus, 2014;Shwetha 245 and Varija, 2015; Troy et al., 2008;Xie et al., 2007) and some initial model experiments. The soil parameters Exp and Ksat were assumed to be the same for all three soil layers.
Sensitivity Analysis is performed explicitly for all subcatchments of Mahanadi river basin ( Figure 1) using a Global Sensitivity Analysis (GSA) technique, Elementary Effect Test (EET) (Morris, 1991) and three objective functions: Nash-Sutcliffe Efficiency (NSE), Log 250 transformation of NSE (lnNSE) and Klein-Gupta Efficiency (KGE) were included in the analysis. NSE focuses on high flows (Nash and Sutcliffe, 1970) whereas lnNSE focuses on low flows. KGE is the improved version of NSE, gives equal weight to the high and low flows (Gupta et al., https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.

2009).
Ranking of the uncertain model parameters have been generated according to their relative contribution to the output variability by analyzing the sensitivity indices (Pianosi et 255 al., 2016). Parameters which showed poor performance when tested across all the subcatchments and objective functions were discarded.
The VIC model calibration is performed semi-automatically using a sequence of Monte-Carlo simulation where 10000 near-random parameter sets (influential parameters) were generated from within the specified range using LHSM with uniform distribution. We 260 abstained from calibrating and validating the model for the entire Mahanadi river basin due to the presence of a major water management structure, Hirakud dam at the middle reach of the basin. Instead the model was run, calibrated, and validated daily for each parameter set for the time period (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) at all the subcatchments Basantpur, Kantamal, Kesinga, Salebhata and Sundergarh following 2 years of warm-up period (1988)(1989). Hence, 265 calibrated model parameters vary from one subcatchment to another. Next, a pareto set of solutions (parameters) (Bastidas et al., 1999;Efstratiadis and Koutsoyiannis, 2010) are generated for all the subcatchments according to the various trade-offs among different catchment characteristics through the maximization of NSE. Therefore, to obtain a single parameter set for the entire basin, a pareto rank has been assigned to each of the 10,000 for the time period of  and (2015-2100) respectively (Hurtt et al., 2018) (Table S1 https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author (s) Table   2. The only notable difference is the lack of Barren ground (BG) in LUH2005 compared to about 12% coverage in the NRSC2005 database. However, the areal coverage of the two most important classes in the river basin (DBF and CL) show highly comparable percent values 305 between the two products. Note that we will refer to DBF as Forest (F) Table 3. The percent areas covered by each land use classes in all the subcatchments are shown in Table S3  are mostly within the "good" and "very good" range of NSE in both calibration and validation period according to Moriasi et al., (2007) with the model performing slightly better in the calibration period. The NSE values lying within the "satisfactory" range for Salebhata is the poorest performing station among all subcatchments. The range of NSE'S for the daily 375 calibration and validation at all subcatchments are listed in Table S4 in Supplementary section. The models were able to simulate the daily flows consistently when compared to the observed flows and reproduce the peak flows at different subcatchments in both calibration https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. and validation period (Fig S3 in Supplementary section). The grey lines in the parallel coordinate plot in Figure 6b shows the distribution of input parameters within their variability 380 range and the lines highlighted in black are the parameter sets that have resulted in overall good model performance across the entire Mahanadi river basin obtained through Pareto ranking.

Control case scenario performance
We use the calibrated VIC models with LULC maps from two distinct sources, global LUH2005 385 and regional NRSC2005, which are configured to the model grid resolution. Model performance of LUH2005 in comparison to NRSC2005 helps to evaluate the robustness of the future LUH scenarios prior performing the model simulations using future LUH projections.
The Boxplot in Figure 7  forest. Therefore, the underestimation in the simulated flows using LUH2005 may result from the increasing grasslands which increased LAI, thus resulting in an increase in ET and decrease in surface runoff respectively. The difference in resolution between land cover maps, LUH2005 and NRSC2005 has led to these differences in the aerial coverage of the land cover https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
types. However, in overall, objective function, NSE suggests both distinct land cover maps for 410 the baseline scenario from NRSC and LUH show comparable model performance in the historical period with the model being able to capture the seasonality and Land Use/ Land Cover dynamics while simulating the daily flows.  (Table S5 in Supplementary section). The increase in the annual average of extreme flows in NF and FF scenarios can be attributed to the overall reduction in forest cover by 15.55% and 22.65% and an increment of cropland 13.65% and 23.3% respectively. However, changes in land use area varies from one subcatchment to another (Table S3). Maximum 440 https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License. increment in cropland (37%) at the expense of forest (38%) is observed at Basantpur and the minimum increment in cropland (16%) at the expense of forest (14%) is observed at Salebhata in the FF scenario.

Impact of land use changes
Percent increase of (0-32) % is observed in 'All Cropland' (CL) scenario and percent change in the 'All Forest' (F) scenario ranges from an increase of 0.5% to a decrease of 40%. Percent 445 change in the Grassland (GL) scenario ranges from an increase of 12% to a decrease of 3%.
The median percent change in CL scenario are slightly higher than that of FF scenario whereas the median values suggest negligible change in the GL scenario. The median percent change in F varies from -3% to -12% across the subcatchments. Maximum increase and decrease in percent change are observed in CL and F scenario respectively. In GL scenario, percent 450 decrease in the flows are less than the F scenario whereas percent increase is less than the CL scenario. A maximum increase in the annual extreme flows of 830 cumecs is noticed in CL scenario followed by an increase of 532 cumecs in the FF scenario at Basantpur (Table S5 in Supplementary section).
Because the changes in the extreme and mean flows are of almost similar magnitudes at all 455 the subcatchments, other water balance components are analysed with respect to the entire basin to understand the factors causing changes in the streamflow in overall. Figure 9 shows the percent differences in the water balance components in the NF, FF and the hypothetical scenarios, CL, F and GL. Percent change in the P scenario is negligible, hence not shown in the boxplot. NF scenario depicts an increase in the surface runoff, baseflow and soil moisture 460 content within a range of (1.5 to 9) %, (3 to 26) % and (2 to 7) % respectively followed by decrease in ET within a range of -(1.6 to 3.3) %. FF scenario depicts an increase in the surface runoff, baseflow and soil moisture within a range of (1.5 to 12) %, (4.9 to 32) % and (2.2 to 10) % respectively followed by decrease in ET within a range of -(1.8 to 3.5) %. The median percent change in runoff, ET, baseflow and soil moisture content in the FF scenario are 4%, -465 2.2%, 13% and -4% respectively. Percent change in the CL scenario depicts an increase in the surface runoff, baseflow and soil moisture content within a range of (1 to 20) %, (2 to 50) % and (2.2 to 16) % respectively followed by decrease in ET within a range of (0.5 to7) %.
Reduction in percent change in the F scenario is observed with a decrease in the surface runoff, baseflow and soil moisture content within a range of (1.5 to 12) % , (4.9 to 32) % and 470 (-2.5 to 21) respectively followed by increase in ET within a range of -(1.8 to 3.5) %. The increase in the percent change in the GL scenario for all the components are lesser than the https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.  and 75th percentile to analyse the annual water balance averaged over 20 years. VIC model solves the water balance within the catchment using the Eq. (1) (Gao et al., 2010).
where P, E, R and ∆s/∆t are the precipitation, evapotranspiration, runoff, baseflow and change of water storage respectively. Table 4 indicates that the model is able to estimate all the water 490 budget components and maintain proper closure of the water balance in all the scenarios.
Note that the average annual precipitation is 1318 mm, which is kept constant for all the scenarios.

Model parameter uncertainties
An ensemble of model parameter sets is derived for the entire basin based on pareto ranking 495 to account for the uncertainties in the hydrological components owing to the land use change.
The differences between the minimum and maximum percent changes in annual extreme flows and other water balance components in all the scenarios represents uncertainty in the

Discussion
We carried out sensitivity analysis with 3 objective functions to account for all parameters sensitive to all the flow processes occuring within the basin. However, the objective function used for calibration is based on the application of the model as also followed by Muleta and Nicklow, (2005). Muleta and Nicklow,(2005) also implemented a calibration strategy guided 525 by global sensitivity analysis which reduced the uncertainties in streamflow. In overall, parameter sensitivity results are in accordance with the findings of Demaria et al., (2007) where the parameters Exp and binf were sensitive to the objective function RMSE (focusses on high flows) and parameters, d1 and ksat were slightly sensitive to the streamflow.
Owing to very high computational costs, we abstained from modelling the land cover change 530 impacts at each subcatchments individually. Given that the subcatchments are of conflicting catchment characteristics, there is no feasible point that optimizes all five parameter sets obtained from different subcatchments. Therefore, the idea of pareto ranking which has been https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
applied in many studies (Gupta et al., 2003;Shi et al., 2008) serves the purpose of generating a common set of models by looking for an acceptable trade-off among the calibrated models 535 of all the subcatchments. Our calibration results suggest that the models tend to perform better at the bigger catchments and yielded lower NSE values at the smaller catchments, Salebhata and Sundergarh. These findings are in agreement with some literatures (Kneis et al., 2014;Mishra et al., 2008;Nayak, Venkatesh, Thomas, & Rao, 2010) Table S3 in Supplementary section). Basantpur is the 555 biggest subcatchment which is projected to undergo a maximum expansion in cropland areas (37%) among all the subcatchments. Therefore, future scenarios should reflect more percent change in extreme flows at Basantpur. This is to some extent, supported by our result (Figure   8) as the maximum percent increase in extreme flows from within the range of best ensemble models are observed at Basantpur (21%) followed by Kantamal (13%), Kesinga (12%),  Costa et al., 2003;Dadhwal et al., 2010;Das et al., 2018;Kundu et al., 2017). Kundu et al., (2017) found an increase in runoff and decrease in ET due to the expansion in projected agricultural land in Narmada river basin in India. Das et al., (2018) predicted that 585 deforestation, urbanization and cropland expansion in eastern river basins of India, in the future would increase runoff and baseflow and decrease ET.
The impacts on the annual water balance of the entire basin is, however, small in terms of magnitude. Research elsewhere (Ashagrie et al., 2006;Fohrer et al., 2001;Kumar et al., 2018;Patidar and Behera, 2019;Rogger et al., 2016;Wagner et al., 2013;Wilk and Hughes, 2002) 590 have also reported that the impacts of land cover change on water balance components in a large scale river basin are too small to be detected due to the compensation effects. Patidar and Behera, (2019) in a recent study in a large river basin in India, showed that the impacts of land cover change on ET and runoff cancels out at the basin scale and reported that the conversion of forest to agriculture may not alter the water balance significantly. Our result 595 shows that a major portion of precipitation is contributing to ET in all scenarios (Table 4) which is consistent with Das et al., (2018), Garg et al., (2019) and the impact of land cover change https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.
on ET is more than other water balance components which agrees with the findings of Kundu et al., (2017) in a river basin in India. Moreover, (Garg et al., 2019) found that croplands contributes more to ET than streamflow in a river basin of a similar climate zone in India. 600 The changes in extreme flows and the water balance components are more pronounced in the hypothetical scenarios. All cropland (CL) scenario showed the maximum increase in flows whereas All Forest (F) scenario resulted in the maximum reduction in flows which is in line with previous studies (Ma et al., 2010;Mishra et al., 2010;Wilk and Hughes, 2002). Wilk and Hughes, (2002) reported that the largest increase in runoff resulted from total conversion of 605 the basin to agriculture in in South India. Maximum reduction and increment in baseflow are observed in F and CL scenario respectively which is consistent with the observations in other studies in terms of direction of change , Vano et al., 2006. However, unlike our study, most of these studies reported that an increase in runoff in Grassland (GL) scenario is more than the cropland scenario. These differences can be attributed to the process of 610 generating the hypothetical scenarios. For instance, the forest scenario in Ma et al., (2010) is represented by converting all grassland, barren lands and Croplands only above a certain elevation whereas Mishra et al., (2010) framed the hypothetical scenarios by converting a single grid cell to 100% Cropland, Forest and Grassland. However, more emphasis should be given to the CL scenario as this relates to the major changes occurring in the basin as per the 615 future projected LUH scenarios.
Multiple parameter sets can yield equally good or acceptable model outputs due to the complex interactions among the parameters , known as equifinality, considered as one of the main sources of uncertainty in hydrological modelling (Her et al., 2019). Taking (Garg et al., 2019). The VIC model has been set up and run for the Mahanadi river basin on a regional scale of 5 kms and the modelled discharge are well in agreement with the measured discharge at all the subcatchments considered. The methodological approach used in this study helps to comprehend the possible impacts of changing land cover scenarios within a modelling framework of detailed calibration and sensitivity analysis.   https://doi.org/10.5194/hess-2020-220 Preprint. Discussion started: 6 July 2020 c Author(s) 2020. CC BY 4.0 License.