Integrated decision support models for sustainable groundwater management in crystalline hard rocks: Implications for sugarcane agriculture

India is second largest sugarcane producing nation with water foot prints ~104 BCM/year of which 80% met from groundwater of deep-wells. The large-scale groundwater management practices were found mostly untenable affecting the sugarcane agrarian and associated sectors. We developed a cell-level (2km x 2km) decision support tool (DST) employing the integrated hydro-geophysical investigations in sugarcane agro-watershed (399 sq.km) beset over granitic aquifer systems of Tattihalla River watershed, Southern India. The refined hydrogeological conceptual model derived from electrical resistivity tomography (ERT) and groundwater level (head) has been accounted into the numerical modelling. The model was run on transient mode for four stress periods during 20152017 and validated with calculated and observed heads. The normalized RMS error 7.57% of the validated model conveys its robustness and estimates groundwater budget at the demarcated 77 cells. The theoretical scenarios for water level projections against the increased groundwater pumping rates of 10%, 25% and 50% were generated. It showed declining trend of water level for projected 10 years period with varied magnitude and vulnerability for drought conditions. The different time periods of water level touching the basement (i.e. dry borewell), infers a distinct hydrogeological property of an individual cell advocating to adapt a cell level management plan. In this article, we explained two cells (Nos. 12 & 60) in detail to show the varying characteristic of aquifer against the different pumping rates.


Introduction
India is agrarian-based nation where water resources are the critical input in nearly all its aspects having a determining effect on the eventual food yield (Sarwar et al. 2010, Dhawan et al. 2017. About 80% of agriculture water requirement is met through groundwater in India (Shrivastava et al. 2011). In the northern India, the potential groundwater resources occur in Indo Gangetic Plain (IGP) with enhanced groundwater recharge (Sonkamble et al. 2020). However, a regional scale studies using GRACE satellite data together with hydrological model have reported a huge loss of groundwater resources in northern India (Tiwari et al. 2009). In peninsular India (Southern 3 is grown over an area of 5.2 Mha requiring about 104 BCM of water/yr (Bhattacharya 2010, Srivastava et al. 2011. Sugarcane producing regions have more than 80% groundwater irrigation through deep-well. As per CGWB (2011) reports, only 162 BCM/yr of groundwater is available for future irrigation and of this about 40 BCM/yr is available in the sugarcane producing states alarming the future water crisis. It necessitates judicious consumption of groundwater resources for sustainable development which is possible through small scale groundwater management plans.
In hard rock terrains, the groundwater management is the key issue as the overexploitation has constrained the groundwater to the subsurface fractures at deeper depths. The large-scale mapping of fractured network is successfully delineated in granitic terrains using airborne electro-magnetic scans (Chandra et al. 2019). However, the complexity of the subsurface fractured network required to be understood at local scale in view of efficient groundwater management. The present situation of groundwater depletion in crystalline hard rock is attributed to inappropriate and largescale management strategies which necessitate a participatory and scenario-based strategy.
To address such challenge, there are few decision support systems (DSS) based model which are computational system that uses data and models interactively to aid in the formulation, analysis, and selection of management strategies (Pierce, 2016). Sophocleous and Ma (1998) provided one of the earliest groundwater DSS that evaluates the impact of salt water intrusion on aquifer yield and Janža MA (2015) suggest DSS for emergency response to groundwater resource pollution in an urban area. The DSS has been applied for optimal estimation of groundwater availability (Uddameri et al. 2013) and groundwater governance for sustainable cities (Howard 2014). Since 1997 interest in DSS applications has been increased (Jamieson 1997) with some lump model without consideration of spatial dimension (Naik and Awahthi, 2003;National Research Council 1997). But these efforts lack the credibility of groundwater model due to heterogenous groundwater system. Recent studies conducted in African, European and Indian basements (e.g, Omorinbola 1982Omorinbola , 1983Wright 1992;Chilton and Foster 1995;Owoade 1995;Wyns et al. 1999;Taylor and Howard, 2000;Maréchal et al. 2004;Dewandel et al. 2006;Krásný and Sharp, 2007;Maréchal et al. 2007, Courtois et al. 2008, Courtois et al. 2008Dewandel et al. 2010Dewandel et al. , 2012Maurya et al. 2021) showed that when hard rocks are exposed to regional and deep weathering processes, the geology can be considered as homogenous, the aquifer is constituted of two main sub-parallel hydrogeological layers, namely the saprolite, a clayey-sandy material, and the fissured layers, generally characterized by a dense horizontal fissuring in the first few meters and a depth-decreasing density of fissures. Based on this conceptual model of hard rock aquifer, a decision support tool (DST) was introduced by Dewandel et al. (2010) for groundwater management at watershed scale using water table fluctuation method and groundwater budget equation which is also a lump model. Their output model produced water level under different pumping and recharge scenarios at the watershed scale, i.e. average value of the watershed. Its field level implications discourage the users to apply the DST on large scale such as watershed level. Mizan et al. 2019b developed a tool to estimate 3-dimensional specific yield and 2-dimensional groundwater recharge in deep weathered crystalline aquifer. It assumes, a net flux of groundwater to the systems is negligible at certain threshold cell size which depends draft and aquifer properties. To take the motivation from here, a groundwater management tool has been developed with the integration of numerical modelling in the present study. The numerical model has been integrated to deal with the hydraulic boundary condition at decisive cell scale which were lacking in the Mizan et al. 2019b tool. The study output of groundwater budget components from the numerical model has been interfaced with Dewandel et al. (2010) model for the generation of the different scenario with changing groundwater abstraction. In the present study, Tattihalla River watershed in Southern India has been selected for the validation of this approach. The area has been discretized into 2 km × 2 km for the decisive groundwater management at cell scale.
The present research work focuses on the development of simple and cost-efficient groundwater management tool for the projection of water level at decisive small scale (2km x 2km) based on the change in groundwater abstraction rate and recharge with the integration of numerical model. In this study we developed an integrated decision support tool (DST) for implementation at farm level groundwater management system. The idea is to make this model more user friendly for the farm groups, practitioners and farm extension service providers who may take their decisions for crop cultivation according to the availability of the groundwater resources.

About the study area
Tattihalla River watershed of 399 km 2 located in Haliyal Taluk of Uttar Kannada district, Karnataka State, India has been selected to study the aquifer systems and their dynamic processes with special reference to sugarcane crop ( Fig. 1). The study area falls in the Survey of India (SOI) Toposheet No. 48I/16, 48I/15, 48I11 and lies within geographic coordinates E74°41'39.43" to E74°56'0.00" and N15°11'59.12" to N15°24'58.55". The topographic elevation varies from 484 m to 744 m above MSL with gentle slope towards south (Fig. 2a). The land use and land cover of the study area shows, next to forest land a significant area is belong to agriculture farms (12,209 ha) of which sugarcane cultivation accounts for 1,727 ha (Fig. 2b). Agricultural irrigation in the study area is totally depended on the groundwater resources. A large number of borewells are employed for the irrigation of sugarcane, rice and vegetables. More than 80% groundwater abstraction are used for the sugarcane cultivation by flooding technique.
Geologically the area lies in the Western Dharwar Craton which comes under Chitradurga Group of Archaean to lower Proterozoic age. Major rock types of the area include metamorphosed Quartz-Chlorite Schist, Gabbro/Dolerite dykes and Banded magnetite quartzite intrusions are also exposed in the watershed (Nath et al. 1976;Chadwick et al. 1981Chadwick et al. , 1997Chadwick et al. , 1997. Hydrogeologically, the study area is drained by Tattihalla River drainage network. The aquifers occurs in semi-confined to confined conditions with deep groundwater level. The deep groundwater level below 30-50 m is an indication of over exploitation of groundwater resources and poor recharge process. In the study area the potential groundwater zones are limited to fractures ( Fig. 3e-f). The electrical resistivity tomography (ERT) surveys in the watershed, the soil profile is subdivided into top soil, weathered zone and semi-weathered to fractured zone.

Materials and methods
The approach of the study involves an integrated hydro-geophysical investigation comprising well inventory, geological and hydrogeological surveys, electrical resistivity tomography and land use land cover. The field acquired hydro-geophysical data has been interpreted and fed to groundwater modelling studies for cell level characterization in term of groundwater budget and simulation of future water level. Further, the aquifer characteristics such as hydraulic conductivity, transmissivity, recharge and pumping data has been fed to decision support tool (DST) for generating the cell level future water level scenarios against varied recharge and pumping patterns. The contour maps prepared using Golden Software Surfer (V. 10.0), groundwater modelling was run on visual MODFLOW and the DST was developed on MS Excel platform. The detailed approach is described in the subsequent sections.

Well inventory
A total of 31 observation wells were established for the periodic monitoring of groundwater levels for four stress periods (pre-and post-monsoon) of two hydrological cycles during April-2015 to May-2017 (Fig. 1). The depth to water level below ground level (bgl) and water level elevation above mean sea level (amsl) map were drawn using the field observed water level data during the study period. The water level was measured using water level 7 sounder (Encardio Rites make, Model: EPP-10/6) (

Groundwater recharge
Groundwater recharge from the rainfall has been estimated on the groundwater resource estimation committee recommendation (GEC 1997). Recharge has been considered 12% of the total rainfall in the hard rock aquifers.
Most of the rainfall (>90%) arrives in the monsoon period, therefore the recharge from rainfall is almost nil in non-monsoon (dry) period. Annual rainfall at Haliyal station varied from 868 to 1377 mm with an average annual rainfall of 1188 mm in the span of last decade . Recharge from the irrigation return flow (IRF) is also computed using a relation shown in Eq. 1 where Q is groundwater abstraction and Cf is return flow coefficient. The return flow coefficient is dependent upon several factors including soil texture & type, types of crop, depth to water table and method of application.
It varies widely (0.15-0.45) for the prevailing three major cropping patterns in the study area, i.e, Rabi, Kharif and Zaid (Umar & Khan 2009). Return flow coefficient value is used 0.15 < Cf < 0.45 based on the groundwater resource estimation committee (GEC 1997) for non-paddy crops.

Aquifer Parameters
The effective groundwater management practice requires a reliable aquifer property such as specific yield and other hydrogeological information (Freeze & Cherry 1979). The hydraulic conductivity (K) has been assigned according to Domenico & Schwartz (1990) for hard rock aquifers as 3x10 -6 to 5x10 -5 for weathered layer & 8x10 -9 to 3E-04 for fractured layer. Another important aquifer property, specific yield (Sy; 2 to 8 %) was assigned into the entire aquifer, using data derived from previous studies (

Well discharge
Well discharge rate changes with time according to the requirement of sugarcane and other crops. April to July is the most consumable period of the groundwater (864 m 3 /day) followed by December to March (576 m 3 /day) and August to November (288 m 3 /day) (Fig. 3c). The pumping scheme information has been collected from the farmers. More than thousands (>1000) pumping borewells are uniformly distributed within the watershed.

Electrical resistivity tomography (ERT) deriving aquifer geometry
Electrical resistivity tomography (ERT) is an advanced tool which is widely applied to decipher the shallow subsurface (Stummer et al. 2004) on lateral and vertical scale (2D & 3D) (Uhlemann et al. 2017) using multielectrodes. The theory and practical application and interpretation techniques of ERT are well described in the literature (Griffith et al. 1990;Barker 1981;Dahlin 1993;Griffith and Barker 1993;Barker 1996a 1996b;Loke 2001;Dahlin and Zhou 2004). The software program RES2DMOD developed by Loke (1999) was used to calculate the apparent resistivity using the finite difference method. A resistivity meter (ABEM Terrameter LS, Sweden make) was employed in the present case with 41-81 electrodes connected to the meter through a multicore cable having unit electrode spacing of 10 meters. The software known as RES2DINV (Loke 1994(Loke , 2001) is used to prepare the 'Inverse Model Resistivity Sections (IMRS)' and the Iteration RMS (route mean square) error was observed to be below 10%. A total of 13 ERT profiles were carried out at different locations ( Fig. 1 and Table   1) within the watershed covering flood plain, pediplain and pediments zones so that to generate a reliable hydrogeological model. The gradient configuration was used to acquire the resistivity data with maximum AB spacing varies from 400 m to 800 m (depending on the space availability at sites) with maximum depth of investigations range 74-130 m ( Table 1). The IMRS were demarcated with hydro-litho units for delineating the aquifer geometry. The aquifer geometric parameters derived from the ERT were fed to the numerical modelling for revealing the aquifer dynamics. were reference to the topographic elevation generated by DEM for the entire study area (Fig. 2a). These aquifer values have been used to run the model which upon calibration resulted in a model that matches the aquifer systems of Tattihalla River watershed. A calibrated model with 4 stress periods was run for 2 years for estimating the groundwater budget.

Developing the DST
The tool is based on the simple groundwater budget equation and water where R is groundwater recharge; IRF is the irrigation return flow; qin and qout are groundwater flows onto and off the system, E is evaporation, Q is the abstraction of groundwater by pumping, qbf is base flow (groundwater discharge to streams or springs) and ∆S is change in groundwater storage. There is no perennial river present in the watershed, thus base flow (qbf) will be zero. Therefore, Eq. 3 can be rewritten as; The method determines the unknown storage of groundwater is the water table fluctuation method shown in Eq.
4 is, where Sy is called the specific yield (storage) or the fillable porosity of the unconfined aquifer. Several authors (Kayane 1983;Sokolov and Chapman 1974;Sophocleous 1991) distinguish the terms "specific yield" and "fillable porosity". The term specific yield is most often used in connection with unconfined aquifers and is also known as the storage coefficient in case of confined aquifers. ∆h is the change in water level between two seasons.
The two main budget components i.e. groundwater abstraction and recharge were considered as dynamic parameter owing to their high variability on time scale, while all other components were taking as constant throughout the time for the projections of water level. The computation of projected water level ht+1, at time t+1, that of the next season, can be written as shown in Eq. 5 ℎ +1 = ℎ + ∆ℎ Eq. 5 The Eq. 5 can be modified using Eq. 4 as shown below ℎ +1 = ℎ + ∆ / Eq. 6 Where, ht is the water level at time t (last season). Eq. 3 has been used for the computation of ∆S with changing R & Q (considered dynamic components for the generation of scenarios).

Hydrogeological investigations
The periodic monitoring of depth to water level from the 31 observation wells showed the alarming indications with gradual deepening of groundwater head during 2015-2017 ( Fig. 4a-d)

Aquifer geometric parameters
The ERT-11 was carried out at Mugodikoppa village using gradient array with AB spacing 400 m (Table 1 and
The survey area was designated with flat to undulating topography, clayey soil with whitish to grey colour, post harvested sugar cane land, pediplain area situated over Chlorite schist rock. Hydrogeologically, the aquifer systems are under confined conditions of fractured zone. The resistivity image shows the regolith thickness with top soil and weathered zone (20-500 Ωm) up to 13-26 m depth followed by saturated fractured layer (500-2000 Ωm) for varying depth up to 60 m. The shallow hard rock (>2000 Ωm) with varying depth 20-60 m was recorded in the middle to southern part of the profile (Fig. 4b). The area shows potential groundwater sources under confined conditions in fractured layer. These aquifer parameters were accounted as an input to the groundwater numerical modelling.

Groundwater model
The aquifer system is 2 layered with 3-dimensional discretization of the flow model with varying aquifer depths ( Fig. 6a-b). The thickness of vertical layers is referenced with topographic elevation which controls the hydraulic gradient ( Fig. 6a-b). The boundaries act as no flow zones because of the closed watershed model as a single hydrogeological unit with spatial aquifer dynamics. Contour maps of hydraulic head referenced to mean sea level suggests the groundwater flow direction from north to south direction following the topographic slope (Fig. 2a).
In pumping wells, the casing up to the weathered zone induces the water from the second layer of saturated fractured zone. This acts as screen generating a 3-dimensional flow. Groundwater follows the gradient and the 13 specifics of modelling change in space and time. A transient model with recharge and water bodies as source, and evaporation / evapotranspiration and pumping wells as sinks was conceptualized in this study.

Model calibration and validation
Calibration results of calculated vs. observed groundwater head show the sensitivity to hydraulic conductivity, specific yield and recharge. Parameters were calibrated and a flow model was run to match the water level values collected for four stress period during the field study. Thorough calibration was done for the model, a well estimated value shown in Tables 2 & 3 was considered. With an overall accuracy of 97%, the model was developed using the existing field data which changes over course. The normalized RMS error for the calculated vs. observed groundwater head was recorded 7.57 % indicating the robustness of the model (Fig. 7). This numerical modelling acts as an intermediary step for providing inputs to develop the DST for generation of water level projection scenarios. 14

15
Further, to provide the cell level decisive support tool, the study area has been made into 2 km x 2 km grids (cells) with their intrinsic aquifer properties (Fig. 8). Based on the sugarcane land use, a total of 77 grids were selected for groundwater management at cell scale in Tattihalla watershed (Fig. 8). The cell Nos. 12 and 60 marked on the Figure 8 are presented in this paper for projecting future water level scenarios with various pumping patterns.

Groundwater budget at cell scale
On validation and calibration, these 77 sugarcane specific zones (cells) referred from LULC (Fig. 2b) were selected to estimate the groundwater budget with cell size of 2 km × 2 km. Of the 77, the cell Nos. 12 & 60 marked on Figure 8 have been taken for the detailed explanation in this paper based on the varying hydrogeological behaviour (Fig. 7). In view of the complex aquifer systems with devoid of local or regional hydraulic connections, the small-scale DST would promote the farm level groundwater management. The numerical model assists to select zone budget at cell scale that a user can also choose a cell of individual or village level for the sustainable groundwater management. After creating the cells, the groundwater budget components have been extracted from cell 12 & 60 and imported into developed DST for the generation of possible scenarios. These components are; recharge, groundwater abstraction, lateral in & out fluxes, observed water level, and evaporation. Aquifer geometry and calibrated specific yield are also incorporated in the tool for the generation of possible scenario.
Historical rainfall (10 years) has been considered in a cyclic repetition for future rainfall.

Projection of water level
Upon incorporation of all the groundwater budget components into the tool, the projection of water level was performed using different groundwater abstraction scenarios. The hydraulic model was linked with the historical rainfall values of the area for future projection of natural recharge. Natural recharge for future projection has been assigned 12% of historical rainfall according to groundwater estimation committee (GEC 1997). The scenario modules allow creating different theoretical scenarios to test the impact of different managements strategies on groundwater resources. Figure 9a-b shows the variations in projected water level under different groundwater abstraction rate for cells 12 & 60. In cell 12, the projected water level against the increased groundwater abstract of 10%, 25%, and 50% indicate the declining trend (Fig. 9a). In Scenario 1, groundwater abstraction has been increased 10% in cumulative form for 10 years. The responded groundwater depletion was observed starting with time but yet to reach the basement of aquifer (Fig. 9a). In scenario 2, the 25% increased groundwater abstraction observes the intense depletion of water level tending toward the basement. In scenario 3, the 50% increase in groundwater abstraction rate resulted rapid declining of water level touching the basement (well goes dry) in 2025 ( Fig. 9a).
In cell 60, the projected water level against the increased groundwater abstraction of 10%, 25%, and 50 % showed similar fashion like cell 12 with varying magnitude of time. In scenario 1 for 10% increase in groundwater abstraction rate, the projected water level trend showed poor declining trend. In scenario 2, the depletion of water level has started but it was not much significant in comparison with cell 12 against 20% increase in groundwater pumping. After that, scenario 3 for 50% increase in groundwater pumping, records some abrupt depletion trend of water level but it was not drying the aquifer in projected 10 years of period (Fig. 9b). Overall, in comparison with cell 12, the cell 60 shows the declining trend of water level for scenario 2 & 3 but with drastic variation in magnitude of depletion.
Similarly, water level of all the 77 cells have been projected against 10%, 25% and 50% increased pumping rates to evaluate the futuristic groundwater condition within the watershed (Fig. 10). Since each cell has different topographic elevation and aquifer basement, the aquifer is not discretized into the layers. With the increase of 10% groundwater abstraction rate, all the cells show almost similar trend of water level (Fig. 10a). While the trend of projected water level show abrupt changes in some cells with 25% and 50% abstraction rate (Fig. 10b-c). These abrupt water level behavioural changes are observed with accelerated declining trend against 50% increased abstraction rate (Fig. 10c). It suggests the insignificant effect of 10% increased pumping rate on the groundwater level as compared to 25% & 50% increased pumping rates. Over all, it implies that the variation in aquifer behaviour on spatial scale suggest to opt cell level decision instead of watershed level or large scale. Further, it was observed that each cell has individual consumption of groundwater for irrigation purpose where the higher groundwater abstracted cells showed more depletion of groundwater with 25% and 50% pumping scenarios ( Fig.   10b-c). Projection of water level for all the cells (77) provides the complete picture of groundwater resource condition in the next decade which would be very helpful for policymakers to prepare the plans to tackle the worst scenarios. DST has been developed specially to test the impact of changing groundwater abstraction rate on groundwater levels at cell scale in the sugarcane agriculture dominant areas. The DST employed water level projections brings new insight to improve the groundwater management strategies. The information generated from DST can guide farmers or crop planner to plan for efficient and sustainable ground water management for sugarcane crops and also spot the water scarcity due to high groundwater consumption.
The water level scenarios have been created according to the groundwater abstraction rate. The result shows severe depletion of water level in some area of the watershed in the upcoming decade. For the detailed explanation, the projection of water level particularly for two cells (12 & 60) (Fig. 9a-b) are presented to demonstrate the importance groundwater management at cell scale. The basement (hard rock) depth in cell 12 is at ~50 m depth, whereas in cell 60 it is encountered at 46 m depth indicating more scope for the aquifer system in cell 12 ( Fig. 9ab). But in contrary, the cell 12 has been found more vulnerable to drought conditions due to the dominancy of sugarcane farms as recorded form LULC (Fig. 2b). It advocates, each cell requires to manage distinctively based on its availability of groundwater resources, and the same prescription of groundwater management strategies may

20
The specific cases at cell 12 and 60 shows similar pattern of water level against the increased groundwater abstraction rates of 10%, 25%, and 50%. However, the temporal magnitude of water level suggests drastic variations among these two cells which decide their ability to cope up with drought conditions. The cell 12, despite having greater aquifer scope, is observed to be more vulnerable to drought due to dominancy of sugarcane farms and sensitive aquifer dynamics. With 50% increased groundwater pumping rates the cell 12 touches basement resulting dry bore well in September 2025, whereas cell 60 continuous to yield groundwater beyond the year 2025.
The scenarios generated results across the cells (77 cells) decipher the isolated aquifer systems with varying dynamics behaviour. It advocates to adapt small scale decision support tools at farm level or village level for achieving the concept of sustainable groundwater management.

Declarations
Funding: The study was funded by World Wide Fund for Nature-India (WWF-India)

Conflicts of interest/Competing interests:
We have no conflict of interest associated with this paper Availability of data and material: Can be shared upon request to corresponding author