**Evaluation of MAR Using Alterative Methodologies**

## **The Economics of Groundwater Replenishment for Reliable Urban Water Supply**

### **Lei Gao, Jeffery D. Connor and Peter Dillon**

**Abstract:** This paper explores the potential economic benefits of water banking in aquifers to meet drought and emergency supplies for cities where the population is growing and changing climate has reduced the availability of water. A simplified case study based on the city of Perth, Australia was used to estimate the savings that could be achieved by water banking. Scenarios for investment in seawater desalination plants and groundwater replenishment were considered over a 20 year period of growing demand, using a Monte Carlo analysis that embedded the Markov model. An optimisation algorithm identified the minimum cost solutions that met specified criteria for supply reliability. The impact of depreciation of recharge credits was explored. The results revealed savings of more than A\$1B (~US\$1B) or 37% to 33% of supply augmentation costs by including water banking in aquifers for 95% and 99.5% reliability of supply respectively. When the hypothetically assumed recharge credit depreciation rate was increased from 1% p.a. to 10% p.a. savings were still 33% to 31% for the same reliabilities. These preliminary results show that water banking in aquifers has potential to offer a highly attractive solution for efficiently increasing the security of urban water supplies where aquifers are suitable.

Reprinted from *Water*. Cite as: Gao, L.; Connor, J.D.; Dillon, P. The Economics of Groundwater Replenishment for Reliable Urban Water Supply. *Water* **2014**, *6*, 1662-1670.

### **1. Introduction**

Growing population is increasing demand for water in many cities. In some arid and semiarid regions, climate change is projected to lead to reduced inflows to surface water reservoirs that have traditionally been the main sources of city water supply [1–4]. Municipal water utilities typically face requirements to ensure that customer water demand is satisfied with a prescribed reliability. For example, Water Corporation, the utility serving Perth, Australia has an objective of ensuring that the annual probability of a complete sprinkler ban is less than 0.5%, or a 1 in 200 year occurrence [5].

Declining, more variable surface water supply and growing demand means that many urban water utilities are contemplating or have already made additional investments in less rain dependent supply sources. For example, Australia's thirty largest utilities invested \$30 billion in new municipal water infrastructure between 2006 and 2012 [6]. Choosing from a range of possible water supply sources, timings and scales to meet supply reliability criteria cost effectively is challenging. Many supply options are long-lived capital assets and they often involve scale economies favouring large increments of investment. However, unknown future inflows and thus unknown supply reliability from existing surface water reservoirs mean that if the future turns out wetter than anticipated, large capital investments can be underutilised and the full capital plus operating cost of small amounts of water supplied to ensure demand is met reliably may be very large [7].

This paper evaluates the economics of aquifer injection and banking of climate independent supplies to enable increased use of groundwater during drought when there is a low surface water supply. Another approach to balancing demand and supply pursued in places such as Arizona involves influencing household conservation ethics for example through landscaping changes that lead to reduced water demand [8]. Conceptually, this water banking strategy would be cost effective because investment can be reduced through a rainfall-independent infrastructure that meets peak demand in drought, but is otherwise left idle. An additional reason to consider storage underground to meet demand during droughts is that evaporative losses from dammed reservoirs can be large in arid and semiarid settings. In contrast in some aquifers, particularly fresh aquifers, there may be potential to store water with little loss [9].

### **2. Case Study Area**

The case study described is based on Perth, Western Australia where demand for water has outstripped conventional supplies [10], and surface water inflows to reservoirs are diminishing due to a changing climate. Perth has a population of 1.8 million with 2008 annual consumption of public supply water of 280 Mm3 /year [11]. Its population is expected to grow to nearly 2.5 million by 2030. The utility providing public water expects demand to grow to between 380 (base case) and 425 Mm3 /year (worst case) by 2030, with actual demand depending on climate driven outdoor consumption growth, success of conservation measures, yard sizes in new housing development and actual population growth [11].

Water is currently provided from three major sources, surface water storages in the hills located to the east of the metropolitan area, regional aquifers located below the metropolitan area, and seawater desalination plants. An important characteristic of the existing surface water supply is that it is highly variable. Perth has experienced a steep change in climate leading to systematically lower inflows in the past 35 years than the mean of the previous 100 years.

While there have recently been new supply investments, additional investments are still required over the coming decades, and an adaptive plan for these investments over the next ten years has been developed by the Water Corporation [12]. Much of the focus for future investment in regional water plans is on two sources of rain independent supplies: seawater desalination and water recycling plants. The Water Corporation has developed an innovative strategy of replenishing confined aquifers with recycled water that has been treated to a very high standard [13] to address an agreed regulatory framework [14]. The utility would then withdraw more groundwater in times of drought. This would increase aquifer net recharge and net extraction in some years but would not increase cumulative net withdrawal of groundwater. Groundwater replenishment has been trialed and proven feasible at small scale (1.5 Mm3 /year) and it is likely it can be upscaled to large facilities with much lower unit costs.

This analysis investigates the cost effectiveness of groundwater replenishment as a potential future supply. The present analysis is built on readily available data and is a generalised approximation of Water Corporation's Perth water supply sources and their potential uses, at annual time scale and aggregated at metropolitan area scale. Still we are able to demonstrate significant potential to meet a supply reliability constraint for Perth with less infrastructure investment and at considerably less cost, when replenishing groundwater and increasing withdrawals in drought is part of the supply solution. The paper concludes with a discussion of the relevance of these results for aquifers with different rates of storage loss.

### **3. Methodology**

Demand for water in the Perth metropolitan area in 2030 is projected to exceed the recent portfolio of supply (a mix of groundwater, surface water, and seawater desalination), as shown in Figure 1a. If managed aquifer recharge is used to create sufficient groundwater credit this can be drawn on in dry years to secure the required supply, as shown in Figure 1a,b as an additional volume to the original groundwater entitlement. Groundwater recharge accumulates through the operation of the installed recharge capacity. The net recharge credit may be discounted annually to allow for losses of recharged water or fresh native groundwater as a consequence of the recharge operation over the losses that would have occurred without it. For example any increase in discharge of groundwater to the ocean due to increased hydraulic gradient attributable to recharge, which would be evaluated for specific recharge proposals, would be included in this depreciation term. In any year, this net recharge credit is diminished by the amount of additional withdrawals to meet water supply shortages over the pre-recharge entitlement.

Modelling was done to simulate two strategies to meet the growth in water demand 2011–2030 and to assess their water supply reliability. Consistent with Water Corporation planning documents we assume that one strategy involves new investments in desalination and in water recycling and water banking. In both treatments groundwater extraction levels in Perth for public water supply are restricted in line with current government regulation. In the without aquifer banking scenario the annual allocated groundwater extraction (120 Mm3 /year) is assumed to be constant across years and supply to meet shortfall is dominated by desalination. In the with aquifer banking scenario any installed recharge capacity is used to replenish groundwater and gain accumulated recharge credits to allow additional extraction, when needed, over and above the pre-existing allocated groundwater extraction.

**Figure 1.** An example of a portfolio of water supply and corresponding recharge credit: (**a**) the varying water supply portfolio to meet demand each year (taken from one Monte Carlo simulation; and (**b**) recharge credit accumulates based on operation of the installed recharge capacity. The net recharge credit available for extraction is the difference between accumulated recharge allowing for depreciation (losses) and accumulated withdrawals to meet water supply shortages.

A Monte Carlo analysis was used to account for variations in the annual amount of surface water available and this depended on inflows in the current and previous years and storage operation rules. In contrast, recycled water and desalinised water can be expected to be available for supply at levels up to plant capacity on a relatively constant basis. Though this is a slight simplification given that plants can experience operational problems or oil spills into ocean water can render a plant unusable for public water supply, these probabilities were considered to be sufficiently small to ignore.

Analysis is a two-step process. The first step is to determine the probability of supply meeting or exceeding demand, for a wide range of possible combinations of levels in investment in desalination and water recycling for groundwater replenishment. The supply, demand comparison algorithm accounted for stochastically varying surface water supply and the dynamics of aquifer

water injection, withdrawal, losses and available recharge credits in the with aquifer banking scenario. For each possible combination of investment in desalination and recycling over a period of 20 years, 10,000 Monte Carlo realisations of surface water availability were run to calculate the percentage of the years that demand exceeds supply.

This process of calculating supply reliability is repeated for the "with-" and the "without aquifer banking" scenario. The results along with estimates of the capital and operating cost associated with a set of possible desalination and water recycling investments produce a set of cost and reliability estimates. These are input into an optimisation that solves for the cost minimising combination of investments that meets specified reliability criterion.

### **4. Case Study Detail**

This analysis is built on readily available data and includes no detail of how Water Corporation and Perth water supply is currently configured and operated. As such, the study should be considered a somewhat stylised demonstration of the significant potential to meet a supply reliability with less infrastructure investment and at less cost, when banking is part of a supply solution.

Estimates of current water supply, and projected 2030 demand were extracted from Water Corporation and State Government reports that are readily available. The Water Corporation [14] estimates 2030 yields from currently existing supply will be 260 Mm3 in its "base case" planning scenario. It estimates 2030 demand for this scenario at 380 Mm3 for the 2030 base case. Thus, there is a "gap" of an average of 120 Mm3 that will have to be filled with new supply investment to meet projected 2030 base case scenario demand.

The stochastic nature of surface water supply was modelled using information from Water Corporation annual reports characterising how much water was actually supplied from surface water storages from 1996 to 2011 [15]. A key feature of stochastic surface water supply that requires consideration in meaningful planning to reliably meet demand is how supply variability can involve multiple year sequences of relatively dry, normal or wet inflow. The length and duration of dry, normal or wet inflow year sequences are key parameters determining the reliability of surface water supply reliability. This is represented by a Markov process [16] and assumes the climate regime of each year switches between three states: high, medium and low supply. We model evolution of these state variables as a discrete Markov chain process where one type of supply year is followed by one of the three possible states based on random probability draws. The probabilities of one state following another are defined with a matrix of transition probabilities for each state variable switching to another state. Note that the Markov chain was used to describe the volume of water supplied by reservoirs in successive years, not the volume of inflow to those reservoirs.

Actual observations of volumes of Perth water supplied from reservoirs from 1996 to 2011 were used to define several levels of supply states and the transition probability matrix between states. Ideally, Markov transition models are based on hydrology and storage operating process models backed by long hydrology time series and future improvement of this study could include such modeling. Still the Markov process approach does provide an opportunity to provide evaluation of reliability and cost effectiveness implications of long dry runs.

Reliability of supply was evaluated with all possible combinations of discrete increments of investment in desalination in 25 Mm3 increments up to 150 Mm3 of capacity above what now exists and discrete increments 10 Mm3 capacity to recycle water up to 80 Mm3 . We assume that the capital cost per Mm3 level of investment in desalination ܿܿ݀ is \$20 million; the capital cost per Mm3 level of investment in recycling capability ܿܿݎ is \$15 million; and the operating costs per Mm3 level of investment in desalination and recycling capability are \$0.8 and \$0.6 million, respectively based on the Science Matters report [17]. While it is true that in some circumstances recharge and withdrawal of water can be much less expensive than desalination. For our case study it is only slightly less expensive because in Perth the recharge water is highly treated prior to aquifer injection. Note that detailed modelling of cost per unit desalination is beyond the scope of this study and only flat estimation of cost is used here. We also model three levels of banked aquifer storage credit loss rate: 1%, 5% and 10% per annum. This represents a range of aquifer loss rates from typical small losses seen in slow moving large regional aquifers to much larger loss rates.

### **5. Results and Discussion**

Results are shown below from simulations with- and without aquifer banking at two reliability criteria levels (95% and 99.5%) and for three annual rates of aquifer recharge credit loss. The minimum costs, optimal choices of Mm3 /yr capacity of desalination (ܦ௧) and water recycling and aquifer replenishment (ܴ௧ ) and estimated reliabilities (ߙ (derived from 10,000 simulations for each scenario are summarised in Table 1.


**Table 1.** Minimum costs, optimal choices, and reliabilities under different model scenarios and reliability requirements.

The results show that highly reliable water supply to meet Perth 2030 urban demand is possible with or without groundwater banking. However, the level of infrastructure investment required and hence cost to achieve a given reliability can be much reduced when aquifer banking is possible. Both 95% and 99.5% supply reliability can be achieved with between 20 and 30 Mm3 less new water supply infrastructure capacity and at all aquifer loss rates considered. Aquifer banking appears to be a particularly attractive strategy especially when losses from banked storage are low. Estimated savings through water banking over strategies without water banking for a 1% aquifer loss rate, over the 20 year horizon exceed A\$1 billion or 37% to 33% of total supply augmentation costs at 95% and 99.5% supply reliability criteria respectively. When the hypothetical recharge credit depreciation rate was increased from 1% p.a. to 10% p.a. savings declined but were still 33% to 31% for the same reliabilities.

Figure 2 presents the trade-offs between cost and reliability (represented by the optimal pareto fronts) for with- and without aquifer banking scenarios and different rates of banked aquifer storage loss. With lower loss rates, the cost effectiveness advantage of ASR is greater, than with higher loss rates. To provide a certain security level of urban water supply, the aquifer banking scenarios outperform the without banking scenario in terms of cost for any given level of reliability.

**Figure 2.** Optimal pareto fronts of different model scenarios for with- and without water banking and showing the effect of rates of storage depreciation between 1% and 10% per annum.

### **6. Conclusions**

A simplified case study based on Perth, Australia shows that an increasing demand for water can be met at the required reliability with less supply infrastructure and at less cost when it is possible to replenish the local aquifer and build a credit that can be drawn on in drought. This is because without such banking, "supply insurance" must be provided for droughts through infrastructure investments that are only infrequently used to achieve the required high reliability of supply at significantly higher average costs of supply. Hence it is demonstrated here that water banking in aquifers in order to provide drought and emergency supplies or "strategic storage" can provide a relatively low-cost insurance for cities with suitable aquifers. The economic analysis shows that aquifer banking provides greatest cost saving where there is little loss of the aquifer banked water. In aquifers with greater loss rates of stored water, the economics are still attractive compared with solutions that exclude water banking. It should also be noted that there are abstraction constraints that can limit the use of banked water in poor years depending on abstraction capacity. Finally, this study can be considered to be a qualitative demonstration of the potential benefit of groundwater banking; additional detailed analyses would be required to estimate benefits for an actual operational model.

### **Acknowledgments**

The first author would like to express his gratitude to the support by the Urban Water Theme, CSIRO's Water for a Healthy Country Flagship. Any remaining errors and omissions are the sole responsibility of the authors.

### **Conflicts of Interest**

The authors declare no conflict of interest.

### **References**


## **Economic Assessment of Opportunities for Managed Aquifer Recharge Techniques in Spain Using an Advanced Geographic Information System (GIS)**

### **Enrique Fernández Escalante, Rodrigo Calero Gil, María Á. San Miguel Fraile and Fernando Sánchez Serrano**

**Abstract:** This paper investigates the economic aspects of Managed Aquifer Recharge (MAR) techniques considered in the DINA-MAR (Depth Investigation of New Areas for Managed Aquifer Recharge in Spain) project. This project firstly identified the areas with potential for MAR for the whole of the Iberian Peninsula and Balearic Islands of Spain using characteristics derived from 23 GIS layers of physiographic features, spanning geology, topography, land use, water sources and including existing MAR sites. The work involved evaluations for 24 different types (techniques) of MAR projects, over this whole area accounting for the physiographic features that favor each technique. The scores for each feature for each type of technique were set based on practical considerations and scores were accumulated for each location. A weighting was assigned to each feature by "training" the integrated score for each technique across all the features with the existing MAR sites overlay, so that opportunities for each technique could be more reliably predicted. It was found that there were opportunities for MAR for 16% of the area evaluated and that the additional storage capacity of aquifers in these areas was more than 2.5 times the total storage capacity of all existing surface water dams in Spain. The second part of this work, which is considered internationally unique, was to use this GIS methodology to evaluate the economics of the various MAR techniques across the region. This involved determining an economic index related to key physiographic features and applying this as an additional GIS overlay. Again this was trained by use of economic information for each of the existing MAR sites for which economic data and supply or storage volume were available. Two simpler methods were also used for comparison. Finally, the mean costs of MAR facilities and construction projects were determined based on the origin of the water. Maps of potential sites for Managed Aquifer Recharge (or "MAR zones") in the Iberian Peninsula and Balearic Islands of Spain and the results of the previous economic studies developed at the beginning of the project were used as the foundation for the economic analysis. Based on these data, a new specific mapping of the total expected costs for all "MAR zones" (€/m3 ) was proposed based on the techniques that were considered most appropriate for each Spanish study case. Capital costs ranged from Euro 0.08–0.58 per m3 /year. Overall, this study investigates the opportunity and economic feasibility of implementing new MAR projects and provides support to decision makers in Spain. The novel mapping provides valuable guidance for the future development of Managed Aquifer Recharge projects for water managers and practitioners.

Reprinted from *Water*. Cite as: Escalante, E.F.; Gil, R.C.; Fraile, M.Á.S.M.; Serrano, F.S. Economic Assessment of Opportunities for Managed Aquifer Recharge Techniques in Spain Using an Advanced Geographic Information System (GIS). *Water* **2014**, *6*, 2021-2040.

### **1. Introduction**

This study analyzes the economic aspects in the DINA-MAR project related to the price of Managed Aquifer Recharge (MAR) water. These aspects range from simple ratios to advanced proposals based on GIS. This analysis was conducted to study the feasibility of implementing new building works and to provide support to decision makers in Spain. DINA-MAR (Depth Investigation of New Areas for Managed Aquifer Recharge in Spain) is a project financed by the Tragsa Group with the aim of determining the most suitable areas for MAR and how to implement MAR activities within Spain.

The use of GIS for determining opportunities for MAR is broadly mentioned in hydrogeological literature. Some other approaches have been consulted, especially in papers or reports from Portugal, India, Australia and Italy, which provide a different GIS mapping approach than the one displayed in this article.

A regional scale study was performed by Dudding *et al.*, 2006, [1], for the Melbourne region for ASR potential as well as for depth aquifers.

An explanation of the main features in relation to opportunities for water banking is exposed in Hostetler, 2007 [2], although the aggregated features differs from specific opportunities for MAR.

Some papers from India on GIS approaches have been consulted, as for instance the analysis from Kallalia *et al.*, 2007 [3] (pp. 111–119), for potential wastewater aquifer recharge sites, which assesses mapping MAR opportunities.

A GIS based expert system for selecting recharge methods is reported by Masciopinto *et al.*, 1991 [4] (pp. 331–342). No reference could be found on the previous use of GIS for costing of MAR projects.

The study by Pedrero *et al.*, 2011 [5] (pp. 105–116), describes a GIS-based multi-criteria analysis for site selection of aquifer recharge with reclaimed water. Another regional scale study was performed by Smith & Pollock, 2010 [6], who evaluated the artificial recharge potential for a superficial aquifer by means of GIS in the Perth region.

Three different lines of action have been accomplished and presented in the paper to analyze the economics of MAR.

First, the investment ratios of construction costs to storage volume and the mean life of the existing MAR projects with various techniques were evaluated and compared to dam and irrigation pond costs. Numerous examples were collected for statistical analysis.

Second, an advanced GIS methodology determined the "MAR zones" in Spain. After the identification of these zones, the most ideal devices were identified according to the inventory of 24 categories that were proposed in the project [7] (pp. 303–318).

Third, the origin of the water sources in the above two methods was considered. Water resources originating from either fluvial or sewage waters were then compared. Both of these water sources were budgeted.

The fluvial water is provided by a diversion structure in a river to an adequate aquifer for underground storage. Different premises have been considered according to the available flow, ease of application, suitability studies, feasibility studies and cost including exploitation and maintenance expenses. The sewage water option injects reclaimed water into deep boreholes and wells that are generally located near a sewage treatment plant. Economic studies have considered water flow, tertiary treatment, desalination, method of recharge to aquifers, construction costs, conservation costs, study costs and project costs.

Using the maps of potential sites or "MAR areas" for MAR in the Iberian Peninsula and Balearic Islands of Spain and the results of economic studies as the starting point of this study, we proposed a new specific mapping of the total expected costs for all "MAR zones" (€/m3 ) that depended on the most appropriate device for each case. This novel mapping provides guidelines that are intended to be valuable for water managers and practitioners for future development of Managed Aquifer Recharge projects.

### **2. Materials and Methods**

The methodological approach consisted of a GIS study based on ARC/GIS and DINA-MAP programs. This process determines the most appropriate areas in Spain to apply MAR techniques with potential fluvial or waste waters.

The process is recursive because the method tests different algebraic map options on constructed maps with up to 83 layers and GIS coverage. Various parameters such as permeable outcrop layers, lithology, aquifers, water levels, fluvial riverbeds, water purifying plants, data collection stations with flow-rate measurements, slopes, altitudes, and distance to the coasts have been loaded in the system and taken into consideration (Table 1 and Figure 1).

To identify the MAR zones, 11 chloropeth maps of hydrographic basins were created. An example of the results for one of the most prospective basins is shown in Figure 2. The entire map series is available at DINA-MAR website [8].

This deductive process supported by algebra maps and analysis in GIS has two major drawbacks in information processing: different projection systems and an incorrect boundary overlay of the layers and thematic coverages used. An effort to unify the map was required.

**Table 1.** Relating "Managed Aquifer Recharge (MAR) zones" by hydrographic major basins. Columns: basin name, the MAR zone area contained in the basin, the basin area, the percentage of the basin covered by a MAR zone and the percentage of an individual MAR of the total MAR area.


In total, 23 main layers were employed with the assigned original number as follows:


**Figure 1.** Location map of the operative Managed Aquifer Recharge (MAR) sites in Spain.

**Figure 2.** Example of the distribution of "MAR zones" in the Spanish Jucar basin.

The main objective of this study was to identify a process producing similar results in existing inventories. The "MAR zones" in Spain were defined after several trials. The procedure that best represented these MAR activities in Spain was adopted (detailed explanation of this process in DINA-MAR, 2010 [7]). The pixel size for map overlays was 1 km × 1 km.

To determine the ideal devices for each "MAR zone", an inventory of 24 devices previously proposed (Figure 3) was distributed and classified according to their characteristics and their most suitable environments.

**Figure 3.** Inventory of feasible and applicable MAR devices, modified from Fernández & San Sebastián [9] (pp. 5–6).

Numerous "if-then" conditions were designed into the system for each device or technique to obtain a group of ranked results for each area according to the specific conditions (Table 3).

A system of grades-weights was applied after studying each device individually; these values are presented in the "weight" column in Table 2.

**Table 2.** Initial indicator to determine the suitability of MAR techniques according to costing based on the ratio between the investment costs and the initial storage volume. Mean costs taken from Tragsa Group projects performed for the Spanish Ministry of Agriculture.


After classifying the building projects performed by the Tragsa Group for the Spanish Government according to the origin of the water, a new specific mapping was proposed for total expected costs for all "MAR zones" (€/m3 ). This map depended on the most appropriate device for each case and featured a series of alternatives sorted according to technical suitability and cost.

The final map viewer is called "HydroGeoportal DINA-MAR" and is available at DINA-MAR "Visor cartográfico" website [10].

### **3. Results and Discussion**

### *3.1. Investment Ratios of Building Costs against Storage Volume*

The initial indicator to determine the suitability of MAR techniques according to costs was based on the ratio between the investment costs and the initial storage volume. The mean life of the devices was evaluated and compared to the cost of dams and irrigation ponds that have a 25 year lifespan.

The examples considered in this study were buildings constructed by the Tragsa Group for the Spanish Ministry of Agriculture for 18 irrigation ponds and 16 medium size dams *versus* the ratios for MAR facilities in the Arenales Aquifer (four projects) for surface infiltration facilities and in the Guadiana basin for 25 medium-depth infiltration boreholes.

Data for MAR deep boreholes was collected from Spanish water supply companies.

### Mean Investment Ratios

Data sets were treated by statistical methods (eliminating the maximum and the minimum, *etc.*). The resulting ratios are as described in Table 2.

According to these results, the MAR technique results are rather cheap for basic economic indicators in comparison with other water management techniques.

### *3.2. Advanced GIS Methodology Based on Linear Combination of Map Layer Attributes*

### 3.2.1. Previous Legal Considerations

In Spain, the legal and technical framework is suited to integrate more MAR devices in water management schemes, although several implementation issues remain: Currently, regulations consider MAR as a spill, which is an obstacle to the development and the implementation of this technique. Royal Decree 1620/2007 is too restrictive in terms of water quality whereas the regulations in other countries are more permissive. The laws in these other countries consider the sanitation aspects of MAR and do not regulate several effects such as the changes in sodium concentration during deep injection.

### 3.2.2. Determining "MAR Zones" in Spain

The main aim of this project was to determine the most suitable areas for MAR in Spain (excluding the Canary Islands on which desalination is the typical water management technique). The calculation methodology is summarized in the previous section. A detailed description may be found at DINA-MAR, 2010 [1] (pp. 215–216).

From the results, approximately 16% (67,000 km2 ) of the Spanish peninsular and Balearic Islands territory is suitable for recharge management. The most ideal basins are the Duero and Balearics basins, and the least ideal are the North and Guadalquivir basins.

The determined "MAR zones" or areas notably suitable to apply MAR activities are grouped by hydrographic basins in Table 1.

### 3.2.3. Potential for the MAR Technique in Spain

Based on the premise defined by DINA-MAR that the future of water depends on the storage capacity, the storage potential of currently unsaturated Spanish aquifers was compared to the storage capacity of dams.

Based on the storage in dams in Spain in January 2005, which reached 53,198 hm3 , and the definition of the MAR zones, a GIS was used to compare the capacities based on the water level depth, aquifer permeability and storage coefficients. Spanish subsoil (excluding the Canary Islands) was found to have a storage capacity of, approximately, 2.0 hm3 /km2 in the MAR zones. Therefore, approximately 260% of the stored volume in the dams could be stored in aquifers in safeguarding the quality and utility of the water. Utilizing underground storage would also enable surface occupation of the land.

Despite the uncertainty inherent in the calculations, these figures indicate the high potential for MAR activities in Spain to provide new integrated water management schemes.

3.2.4. Search Criteria Used to Associate Devices with Each "MAR Zone"

With the physical elements well defined and the specifications of the 24 inventoried AR techniques known (Figure 3), determining the most suitable technique was performed by a grades/weights system as the main association criteria. This system was designed and automated in such a way that each device receives a weight according to its suitability. This score is adjusted to the physical characteristics and other indicators with GIS support.

The established grades are the distribution of permeabilities, lithologies, nitrate contaminations, irrigable areas, irrigation origin, proximity to forests, purifying plants (with treatment types), dams (with associated capacities), wetlands, rivers (with average associated flows), distance to the coast, major aqueducts, slope, height, flood risk, water level, water quality, meteorological stations with sufficient rainfall or streamflow and urban areas. The weights range between zero (inadequate) and three (highly favorable).

By establishing a relational structure between physical factors and indicators with GIS support for MAR devices, an association matrix that supplies the HydroGeoportal DINA-MAR (Table 3) was designed and automated.

The weight columns appear to be subjective based on the suitability of each device. Because of the important role that the devices hold in the final ranking, additional criteria are adopted to minimize the subjectivity and are presented as ranges (Table 3, column 3). The ranges have been defined by the breakdown of each "layer" in different classes, generally distinguishing the different major types and establishing relevant groups to work with a reduced number of types. For example, the "water origin" layer distinguishes five types: surface water, groundwater, irrigation returns, water from treatment plants and water from desalination plants.

The weights (Table 3, column 4) appear in hierarchy according to their suitability and fit to the physical characteristics and remaining indicators. The weight assigned to each case and code directly intervenes in the process of SIG calculation because the database is associated with the calculation engine; then, an individual score is assigned to each polygon. For example, the calculation method to score device D1 (infiltration pond) is as follows. First, the fields D1, D2..., D24 are included in the layer in which all layers have been previously crossed to calculate the score for each device in these fields. The crosses table is then connected to the different facilities leader board, starting with the permeability, and D1 is calculated. Successive "joins" must be performed for each of the topics, and the formula of ranges-weights is applied to obtain a final value.

This process automatically calculates a score for each of the 24 techniques and the highest score determines the most appropriate technique.

The result is a large-scale map ranking the most to the least recommended devices (Figure 4).

The results of these calculations are expressed in the "Favorable Device" map (Figure 4).

This system has enabled several highly ideal MAR zones to be identified. For example, up to 11 MAR devices could be concentrated in the Lower Guadalhorce aquifer (Malaga) when water is withdrawn from the river and a wastewater treatment plant (Figure 5).


**71** 


**Table 3.***Cont.* 


**Table 3.** *Cont.* 


**Table 3.** *Cont.* 


**Table 3.** *Cont.*



**76** 

**Figure 4.** Map of MAR areas and the most appropriate MAR devices. The "HydroGeoportal DINA-MAR" [10] package also provides additional options for each zone.

**Figure 5.** "HydroGeoportal" predicting suitable areas to apply a MAR technique, notably in the Lower Guadalhorce aquifer (Malaga, Spain). The map displays the proposed location of MAR devices obtained through the exposed grades/weights system.

*3.3. Economic Studies for MAR Activities Implementation Based on the Origin of the Water and Its Incorporation into "Hydrogeoportal" Map Viewer* 

An economic study was developed based on the investment ratio or the cost of the device in relation to the recovered water. The ratios for superficial MAR devices are approximately 1/5 of the ratio of the dams, whereas the ratio for ASR is similar to the dams ratio.

The referred study provides two alternatives for decision-making according to the origin of the sources of water, either of fluvial origin or sewage waters.

Table 4 shows the estimation process of the cost intervals. Column 3 differentiates six types according to either the origin of the water or the context in which each device is intended to be implemented. The five distinct classes are as follows: devices in river areas (wells, ponds and canals), dams and dikes in either surface or underground alluvial terrain, urban sustainable drainage systems, drilled wells less than 50 m deep and deep boreholes (deeper than 50 m).

The first alternative diverts running water from a river, channeling the water to an adequate aquifer (underground storage). This technique has several advantages including minimal occupation of the surface, less evaporation, preserved water quality, and the relatively low costs for the storage. For example, from the first row, using a river as a source of intake has a potential cost per action (investment ratio) of close to € 0.20/m3 for an 8 km conduction pipe and the artificial recharge is performed using channels, infiltration ponds and wells. The cost for each activity is estimated to be close to 1.2 M€. Exploitation and maintenance costs have been estimated at € 0.01 m3 /year (real data taken from budgets of building projects performed by the company that the authors work for, in DINA-MAR, 2010 [7]).

The other considered alternative is the direct injection of reclaimed water during managed aquifer recharge (files 5 and 6) using deep injection boreholes and wells. These injection sites are generally located in the vicinity of sewage treatment plants. The water must be tertiary treated, osmotized and inserted into the aquifers. The flow availability is more regular than in the previous alternative. This study considered flows between 50 and 80 l/s to be recharged through 50 m depth wells. Flows exceeding 100 l/s require boreholes approximately 500 m in depth (average values). This technique does not require special water surpluses and can be used for numerous purposes such as irrigation, combating marine intrusion, environmental practices, and industrial supply. The unit cost of investment is € 0.23/m3 (50 m) and € 0.58/m3 (500 m) (tertiary treatment was not considered). An average estimated cost for a 50 m building project is 172,500 €, and 580,000 € is estimated for a borehole 500 m depth plus additional MAR facilities. The estimated costs of conservation per year are € 0.13/m3 (50 m) and € 0.15/m3 (500 m).

The premises considered were the variability of the available flow (100 to 1000 l/s) and the possibility of applying this technique in approximately 16% of the Spanish territory (excluding the Canary Islands). This investigation also considered that the projects must be subject to concessions and require detailed suitability and feasibility studies.

The standards for water quality are ambitious in Spain; therefore, the costs may be lower for countries with less rigorous regulations.


Using the maps of potential "MAR Zones" for Managed Aquifer Recharge in Spain Iberian Peninsula and Balearic Islands (in [8]) as the starting point, a new specific mapping is proposed using the total expected costs for each zone (€/m3 ) that depended on the most appropriate device for each case. The result is a novel map (Figure 6).

**Figure 6.** (**a**) Choroplethic map of "iso-costs" for the best MAR facilities in each "MAR Zone" for Spanish Peninsula and Balearic Islands; (**b**) Detailed view for the East of Madrid province (square in Figure 6a). These results are available at DINA-MAR [8].

(**b**)

### Classes:


This novel mapping provides valuable guidance for future development of MAR projects. Water managers and practitioners are anticipated to be able to utilize these innovative results.

### **4. Conclusions and Comments**

Results show that 16% of the 500,000 km2 area studied using GIS has potential for MAR using a range of techniques adapted to the local situation. In these areas MAR is rather cheap in comparison to surface water storage techniques. The net savings in capital costs if MAR was practiced instead of dams is about 75% for superficial facilities (ponds and channels), about 50% for medium deep wells and 27% for deep boreholes.

Detailed calculations are necessary to support the results and justify future actions. Calculations may be inaccurate, and the resulting figures may cause water managers to consider opportunity costs prior to decision making.

Regarding legality, reviewing current legislation would be desirable (despite the associated difficulty of this goal) because often regulations "fall behind" technological advances. Additionally, the new charges and expenses caused by the economic crisis, some of which may take the form of higher taxes in some communities, have reduced the interest of private investors to undertake MAR projects.

The further understanding of the economics of MAR and an evaluation of the environmental and social effects are necessary. Additionally, the involvement of industry (e.g., agro-industries, desalination agents, waste water treatment agents, and golf courses) in MAR is crucial.

The work presented here could be applied in other countries with appropriate modifications. One aspect to consider in calculations of the "MAR zones" is that the terrain of other countries could vary from the conditions in Spain. The terrain type determines the surface runoff (e.g., plains, plateaus, and moors) and the groundwater flow. Additionally, applying and understanding MAR techniques in heavily deforested areas is desirable according to the results in Figures 2 and 4.

New designs may encompass as many "low cost" devices (example in Figure 7) as possible according to necessities.

**Figure 7.** Example of a "very low cost" domestic MAR device in Madrid.

### **Acknowledgments**

This study was performed and written within the Framework of the R&D DINA-MAR project code 30/13.053 and was financed by the SEPI & Tragsa Group. Special thanks to Peter Dillon for revising the text and to two anonymous reviewers and the journal invited editor for their thoughtful reviews and suggestions.

### **Author Contributions**

The authors participated in different stages during the "Hydro-geoportal" production. Enrique Fernández and María A. San Miguel developed the application and coordinated the different stages. Rodrigo Calero calculated the cost and value of the different options for real-building work budgets. Fernando Sánchez provided the IT input for the GIS to be incorporated into a map viewer and studied its compatibility with the EU INSPIRE Directive.

### **Conflicts of Interest**

The authors declare no conflict of interest.

### **References**


## **A System Dynamics Model to Conserve Arid Region Water Resources through Aquifer Storage and Recovery in Conjunction with a Dam**

### **Amir Niazi, Shiv O. Prasher, Jan Adamowski and Tom Gleeson**

**Abstract:** Groundwater depletion poses a significant threat in arid and semi-arid areas where rivers are usually ephemeral and groundwater is the major source of water. The present study investigated whether an effective water resources management strategy, capable of minimizing evaporative water losses and groundwater depletion while providing water for expanded agricultural activities, can be achieved through aquifer storage and recovery (ASR) implemented in conjunction with water storage in an ephemeral river. A regional development modeling framework, including both ASR and a dam design developed through system dynamics modeling, was validated using a case study for the Sirik region of Iran. The system dynamics model of groundwater flow and the comprehensive system dynamics model developed in this study showed that ASR was a beneficial strategy for the region's farmers and the groundwater system, since the rate of groundwater depletion declined significantly (from 14.5 meters per 40 years to three meters over the same period). Furthermore, evaporation from the reservoir decreased by 50 million cubic meters over the simulation period. It was concluded that the proposed system dynamics model is an effective tool in helping to conserve water resources and reduce depletion in arid regions and semi-arid areas.

Reprinted from *Water*. Cite as: Niazi, A.; Prasher, S.O.; Adamowski, J.; Gleeson, T. A System Dynamics Model to Conserve Arid Region Water Resources through Aquifer Storage and Recovery in Conjunction with a Dam. *Water* **2014**, *6*, 2300-2321.

### **1. Introduction**

Groundwater extraction has enabled significant social development and economic growth, enhanced food security and alleviated drought in many of the world's farming regions [1]. However, if groundwater abstraction exceeds groundwater recharge or decreased baseflow, persistent groundwater depletion or overexploitation problems can occur [2,3]. Groundwater depletion is a significant threat in arid and semi-arid areas, where rivers are usually ephemeral and groundwater is the primary source of water. Consequently, in many arid countries, dams are built on ephemeral rivers to provide farmers with an expanded and reliable source of water. However, the major disadvantages of dams in arid regions are the high evaporation loss from reservoirs and water quality degradation.

An alternative to constructing dams is recharge enhancement [4], a technique used to increase groundwater availability. One well-known recharge enhancement technique is the engineered system of aquifer storage and recovery (ASR), whereby surface water is moved to aquifers via injection wells and serves to bolster groundwater resources. This water can later be recovered for reuse by conventional pumping. The technique was first implemented in 1957 to inject potable water into saline aquifers [5,6].

Given increasing water demand, stresses on supply and wet *versus* dry season water imbalances, managed aquifer recharge (MAR) techniques, including ASR, are likely to become an important component of water projects in arid and semiarid regions [7]. Aquifers offer significant opportunities for underground water storage, reducing the need of high-cost surface reservoirs and storage tanks. Applying MAR techniques can also act to restore a depleted aquifer's functionality [8]. Moreover, MAR can improve agricultural water security, thus improving the livelihood of farmers and providing economic, social and environmental benefits.

In terms of economic benefits, MAR has direct, as well as indirect financial benefits. The costs involved in MAR projects depend on several variables, including location, land prices, method of recharge, geological conditions, design of the entire holistic system, construction costs and initial water quality [9,10]. For two such projects in Australia, the costs of recharge per million liters were 625 USD and 2,000 USD [5,11]. In addition, MAR increases agricultural productivity, which, in turn, improves farmers' livelihood and provides direct benefits, not only at the economic level, but also at the social and cultural levels. A cost benefit analysis developed for a case study in southwest Iran found a 1:1.32 ratio of project investments to agricultural profits, with an estimated payback period of three years [12].

In basins approaching full development of water resources, optimal beneficial use can be achieved by conjunctive use, which involves coordinated and planned operation of both surface water and groundwater development. The concept of conjunctive use of surface water and groundwater is based on surface reservoir impounding stream-flow, which is then transferred at an optimum rate to groundwater storage. Surface storage in reservoirs behind dams supplies most of the annual water requirements, while groundwater storage can be retained primarily for cyclic storage to cover years of subnormal precipitation [13].

There are some successful examples of conjunctive water resources management around the world, such as the elaborate institutional arrangements for conjunctive use and groundwater management in southern California that have been in place since the 1950s [14]. Kern Water Bank (KWB) in California is another successful example. The KWB stores excess water supplies that are available when rainfall or runoff is plentiful by recharging that water through shallow ponds into an aquifer. The stored water is then recovered in times of need by pumping it out with wells [15]. In some cases, treated sewage effluent has been used as the source of water. For example, sewage reclaimed water from an advanced treatment facility is recharged in the wells of the hydraulic barrier constructed to protect the Los Angeles coastal aquifer from seawater intrusion in southern California. Similarly, in the Dan region in Israel, treated sewage effluent from the metropolitan area of Tel Aviv is recharged in sand dunes and then subsequently pumped for various uses [16].

The objective of this study was to determine if ASR, in conjunction with water storage on an ephemeral river, can be an effective water resource management strategy, minimizing evaporative water losses and groundwater depletion rates, while providing water for expanded agricultural activities. The provided framework, based on system dynamics modeling, consists of a dam, recharge wells, extraction wells and water conveyance units, which can be considered as a "Comprehensive Conjunctive Use System" [13]. A modeling framework based on system dynamics modeling was applied to a regional development plan, including both ASR and a dam, and validated through a case study undertaken in the Sirik region of Iran. Given its semi-arid climate and lack of regular surface water, the agricultural production in the Sirik region is heavily dependent on groundwater. Unsustainable groundwater extractions, leading to a declining groundwater table, have threatened both agriculture and local ecosystems. This has led to proposals to build the Merk dam, which would increase the water supply and thereby allow more farms to be irrigated. The effects on groundwater levels of four different ASR schemes were modeled, and in order to assess their respective financial, social and environmental feasibility, each scheme was subjected to a cost/benefit analysis. This analysis considered economic factors, the quantity of water available for environmental flows, the quantity of water to be released from spillways, as well as the social acceptability.

### **2. System Dynamics Modeling in ASR Using a Surface Water Reservoir**

Sustainable water resources management requires a decision-support approach that accounts for dynamic connections between social and ecological systems, integrates stakeholder deliberation with scientific analysis, incorporates diverse stakeholders' knowledge and fosters relationships among stakeholders that can accommodate changing information and changing social and environmental conditions [17]. A system dynamics modeling (SDM) approach has the unique ability to model participatory and stakeholder analysis in water resources and ecological studies [17–21].

Within the few scientific publications that address the application of a system dynamics approach to groundwater issues, groundwater systems are either oversimplified or considered solely as a reservoir. Moreover, in these studies, modeling practices differ substantially from those employed in conventional mathematical groundwater modeling [21–24]. Although such oversimplification (e.g., ignoring the spatial variability of groundwater systems) decreases model runtime, it also decreases model accuracy [24].

Modeling a reservoir's functions and linking it to an aquifer system while considering various socio-economic factors would constitute a comprehensive and integrated modeling approach. However, at present, there is no comprehensive integrated modeling software that can be used in addressing water resource management problems. On the other hand, system dynamics modeling software packages are flexible and integrated modeling tools, which can be applied to any problem, including participatory modeling and economic analysis. Conventional models, such as MIKE-BASIN (developed by DHI, which is an extension of ArcGIS for integrated water resources management and planning), WEAP (Water Evaluation And Planning system, which is a Windows based decision support system for integrated water resources management and policy analysis) and OASIS (a software program that simulates the routing of water through a water resources system), are all limited to water resources applications [25]. In the proposed groundwater modeling approach described in this paper, a modified spatial system dynamics (MSSD) approach was combined with reservoir function modeling.

Typically, a SDM (system dynamics model) project comprises the following stages: problem definition, system conceptualization, model formulation, model evaluation/testing, policy analysis and implementation [20,26–29]. It is therefore important to determine all system components and their mutual relationships in advance. Table 1 portrays the basic elements that can be found in all system dynamics models and describes each system component.


**Table 1.** Basic components of system dynamics models.

### *System Dynamics Model Conceptualization and Formulation*

The system dynamics model in this study was developed using VENSIM [30] software. The model consists of two key segments, the reservoir model and the groundwater model. The ASR was modeled as a connection between these two segments. By taking into account the relevant components of the surface reservoir, the surface water reservoir segment of the model was the first to be built (Figure 1). This segment included a single level (reservoir), representing the volume of water in the reservoir at each time increment:

```
Reservoir = Inflow + Precipitation í Environmental needs í Outflow í Evaporation í Spillway discharge (1)
```
Precipitation and inflow are the model's two inputs. Precipitation represents the amount of water directly contributed to the reservoir by precipitation and is a function of the monthly precipitation rate and the expanse of water the reservoir represents. This rate was calculated by multiplying monthly precipitation by the reservoir's surface area. Inflow is the river's discharge into the reservoir. The inflow was calculated based on historical hydrological data for the river, imported through the "get Excel" data function in VENSIM.

**Figure 1.** System dynamics model of the surface water reservoir segment.

Evaporation, Environmental needs, outflow and spillway discharge represent the model's outputs. Evaporation, the volume of water evaporated from the reservoir surface at each time step, is a function of monthly evaporation and the reservoir's surface area. This volume was calculated by multiplying monthly evaporation by the reservoir's surface area. Monthly evaporation was derived from historical evaporation data for the study area and was introduced to the model by using the "get Excel" data function. At each time step, the reservoir surface area was taken from a volume-stage-area chart for the reservoir. Environmental needs and outflow were derived based on the allocated environmental needs and the irrigation water demand, respectively. Spillway discharge represents the excess water at each time step that exits the reservoir. Spillway discharge is a function of evaporation, precipitation, inflow, environmental needs, outflow, reservoir and the maximum (Max) capacity of the reservoir:

$$\begin{aligned} \text{If } (\text{Inflow} - \text{Evaporation} - \text{Outflow} + \text{Precipization} + \text{Reservior}) & \times \text{Max capacity, then Spillway} > 0) \\ \text{If } (\text{Inflow} - \text{Evaporation} - \text{Outflow} + \text{Precipization} + \text{Reservori}) & \times \text{Max capacity, then Spillway} = 0) \end{aligned} \tag{2}$$

The groundwater modeling portion of the model was developed according to the spatial system dynamics (SSD) concept of a grid-based interaction of spatially-distributed system dynamics modules [28]. The SSD methodology has been used extensively in ecological modeling [31,32] and combines the powers of temporal and spatial analysis, achieved through systems dynamics and geographic information systems (GIS), respectively. This system was later used to model groundwater flow through compartmental spatial system dynamics (CSSD) [24]. Such a framework was intended to address issues related to groundwater and surface reservoir management.

The "stuck" head of water within the system dynamics model's groundwater modeling segment is presented in Figure 2 and represents the head of water in each cell in the discretized aquifer domain. Each cell has flow toward four adjacent cells, located to its north, south, east or west. The head of water is a function of water flow from or toward the cell, water extraction from the cell, along with direct evapotranspiration and percolation. The water flow is calculated based on Darcy's law. The head of water in the aquifer domain must be calculated based on the "subscript" function in VENSIM. The volume of water having entered or exited from the level is transformed into the head of water by dividing it by the area of the cell and the storage coefficient of the aquifer media. As square-shaped cells are used in this framework to simplify the modeling exercise, the cell area was the square of one side of the cell.

**Figure 2.** System dynamics model of the groundwater-modeling segment. Representing the active or inactive cells in the modeling domain, "active" is an auxiliary variable, which can also serve to define aquifer boundary conditions; S represents aquifer specific yield; Dx represents the length of one side of the cell; and K is the hydraulic conductivity of each cell in the aquifer.

There were seven rates in this segment: water flow to south, water flow to east, water flow from the north, water flow from the west, percolation, evapotranspiration and groundwater extraction. Water flow toward or from adjacent cells were calculated in the first four rates of the last statement, and the three remaining rates account for the boundary flow from the top of the aquifer. Technically, water flow is calculated in two rates: water flow to the south and water flow to the east. Water flow from the west and water flow from the north are water flow to the east and water flow to the south of the previous cell. Based on Darcy's law, water flow is a function of the media's hydraulic conductivity, the head of the water in two adjacent cells and the length of the cell.

The occurrence of direct evapotranspiration from groundwater is a function of ground elevation, head of water and the region's monthly evapotranspiration rate. If the head of water reaches within a certain distance of the ground surface, direct evapotranspiration can occur. This distance varies according to the aquifer media. Groundwater extraction and percolation are introduced to the model according to their monthly rates and pattern in the aquifer domain. In the groundwater modeling approach presented in this paper, the mass conservation concept was applied in each cell of the discretized aquifer (Figure 3).

**Figure 3.** Aquifer discretization and groundwater modeling paradigm in the system dynamics model.

The change in storage in cell *a* is equal to the sum of the flow into *a* minus the sum of flow out of *a* to adjacent cells:

$$\frac{dS\_A}{dt} = Q\_{ab} + Q\_{ac} + Q\_{ad} + Q\_{ae} + Q\_{aB} \tag{3}$$

Where,

ୢୱఽ ୢ୲ is the change in storage through time in cell *a* (L3 ·T<sup>í</sup><sup>1</sup> ); ܳǡ ܳǡ ܳௗǡ ܳ is the flow into *a* from *b*, *c*, *d* and *e*, respectively, (L3 ·T<sup>í</sup><sup>1</sup> ); and ܳ is the sum of boundary flows to cell *a* (L<sup>3</sup> ·T<sup>í</sup><sup>1</sup> ).

All flows are positive for flow into *a* and negative for flow out. Boundary flows are flow terms entering or leaving cell *a*, such as evaporation, evapotranspiration, natural recharge, artificial recharge and groundwater extraction. Ground water flow between two cells, *Qab*, can be described using Darcy's law:

$$Q\_{ab} = \frac{\left(T\_a + T\_b\right)}{2} \cdot \Delta \chi \cdot \frac{\left(h\_a - h\_b\right)}{\Delta \chi} = \frac{\left(T\_a + T\_b\right)}{2} \cdot \left(h\_a - h\_b\right) \tag{4}$$

where,

݄ is the head of water in cell *a* (L);


By substituting Equation (4) and analogous terms for cells *b*, *c* and *d*, Equation (3) can be written as:

$$\frac{ds\_a}{dt} = \frac{(T\_a + T\_b)}{2} \cdot (h\_a - h\_b) + \frac{(T\_a + T\_c)}{2} \cdot (h\_a - h\_c) + \frac{(T\_a + T\_d)}{2} \cdot (h\_a - h\_d) + \frac{(T\_a + T\_e)}{2} \cdot (h\_a - h\_e) + Q\_{aB} \tag{5}$$

Using a finite time step approximation for storage change, adding superscript notation to specify time and converting to matrix form for all possible generic ground water cells, Equation (4) can be rewritten to solve for storage in aquifer cell *i* at time *t* + 1 as a function of storage and head values at time *t*:

$$\mathcal{S}\_{l}^{t+1} = \mathcal{S}\_{l}^{t} + \Delta t \left[ \sum\_{j=1}^{4} (Q\_{lj}^{t}) + Q\_{lB}^{t} \right] \tag{6}$$

where,

ܳ <sup>௧</sup> is the flow in or out of cell *i* from four adjacent cells (L3 ·T<sup>í</sup><sup>1</sup> );

ܳ <sup>௧</sup> is the boundary flow (L3 ·T<sup>í</sup><sup>1</sup> );

ܵ <sup>௧</sup> is the storage of cell *i* at time *t* (L<sup>3</sup> ·T<sup>í</sup><sup>1</sup> );

ܵ ௧ାଵ is the storage of cell *i* at time *t* + 1 (L<sup>3</sup> ·T<sup>í</sup><sup>1</sup> ); and

οݐ is the simulation time step (T).

This is a forward difference explicit solution for calculating groundwater heads in one time step from head values at the previous time step. Aquifer storage (*Si*) is related to aquifer head using the relationship between storage and head in an unconfined aquifer:

$$\mathbf{S}\_{l} = (h\_{l} - Z\_{bed}) \cdot \Delta \mathbf{x}^{2} \cdot \mathbf{S}\_{y} \tag{7}$$

where,

ܵ௬ is the specific yield of the aquifer ;

ܼௗ is the bedrock elevation (L).

Because the forward difference explicit formulation calculates the future state based on the present state, the system of equations can be unstable if the time step is too long relative to the spatial scale and the rate of the movement of water between cells. Therefore, a small time step (such as 0.8 days, as was used in this study) must be used to prevent such a problem.

Having developed a surface water reservoir model and a groundwater model in the system dynamics environment, an ASR segment was added. This can be turned on/off automatically, as needed, in order to quantify the impacts of using ASR for the recharge or extraction of water from the aquifer. In the combined model, as illustrated in Figure 4, the left segment of the system represents the relevant components of the reservoir, while those on the right model groundwater in the aquifer based on the principles explained above.

The connection between the reservoir and the groundwater system is the groundwater recharge rate (GWR), fixed by the rate of water injection determined by the ASR approach. This rate is dependent on the availability of water in the reservoir, the pipeline capacity and the volume of rechargeable water in the groundwater system. As the purpose of the wells is to replenish depleted water, not to raise the water table above its original level, the rechargeable volume is based on the difference between the historical initial level and the actual level of the groundwater table in the cells containing ASR wells.

### **3. Study Area**

### *3.1. Local Setting*

The Sirik region, situated between 26°22' N and 26°43' N lat. and between 57°4' and 57°46' E long., houses an aquifer occupying 65 km2 on the southern edge of Hormuzgan Province, Iran. The Sirik region is semi-arid with mild winters (ഥ ൌ ʹʹǤ͵°) and hot summers (ഥ ൌ ͵ͶǤͳ°). Average humidity ranges from 32.9% in the spring to a maximum of 71.9% in the winter. Mean annual temperature and precipitation are 28.2 °C and 190 mm, respectively, with the most rainfall occurring between October and December. The region has a population of approximately 11,667 people (2010), most of whom are engaged in agricultural production. The total amount of farmed land currently stands at ~1000 ha, with a mixture of vegetables, palm trees and citrus plantations. These crops were used in the modeling of the dam's water resources and were selected based on their acceptance by farmers, as well as their production values. Agriculture is the main source of groundwater extraction, with total pumping amounting to 7.2 × 106 m3 ·y<sup>í</sup><sup>1</sup> . This pumping caused an average decline in groundwater levels of roughly 7 m between 2000 and 2010. It is also important to note that the region's "river" is dry for most part of the year and only experiences flow during flash flood events. Flora and fauna, especially in the southern parts of the region, are more dependent on groundwater discharge than surface water availability. This strong dependency of plants on groundwater is mostly attributable to the fact that in the southern portion of the region, in the absence of surface water, groundwater is near the ground level and thereby available to plants.

The model developed in this study was for the Merk River watershed in the Sirik region; the dam, and the aquifer boundary locations are shown in Figure 5. This watershed drains 745 square kilometers, and the maximum elevation of the watershed is 1950 m above sea level (MASL), while the minimum elevation is 50 MASL. Daily discharge of the Merk River at the Garaik hydrometric station has been measured from 2006 to 2010. The location of the hydrometric station is also shown in Figure 5. Since the measurement's time span was not sufficient for modeling the reservoir, the monthly time series of discharge was constructed for 40 years (1970–2010) by multivariate statistical analysis from nearby hydrometric stations. These analyses and data were derived from the feasibility study of the dam [33]. Subsequently, this monthly time series was used as the input flow to the reservoir model; the time series is shown in Figure 6.

**Figure 5.** Location of the dam's watershed, watershed boundary, aquifer boundary and the Garaik hydrometric station.

**Figure 6.** Time series of discharge at Garaik hydrometric station.

### *3.2. Hydrogeology of the Aquifer*

The primary aquifer in this region is an unconfined and unconsolidated aquifer consisting of quaternary valley terrace deposits and river alluvial deposits (Figure 7a). The piezometric map of the region suggests that there is seepage from the northern sandstone to the aquifer. The bedrock is mostly middle Miocene marl with inter-bedded siltstone and sandstone. In the south, the aquifer is bounded by low permeable mudstone.

**Figure 7.** (**a**) Northeast to southwest cross-section of the Merk aquifer; (**b**) location of wells where pumping tests were conducted and the geological cross-section.

Aquifer hydraulic properties were derived from six pumping tests using the AQTESOLV program with the Neuman method [34]. Figure 7b shows the location of these wells, while Table 2 provides their hydraulic properties and depths. This data was used in the modeling process and adjusted during the calibration process.


**Table 2.** Hydraulic conductivity and specific storage of different regions of the aquifer.

Based on the different soil types and land uses, three recharge zones were assigned in the plain (Figure 8a). Most recharge is due to seepage from sandstone to the aquifer, with some recharge from riverbeds and precipitation. The preliminary estimation of recharge was based on an estimation of water balance components and then adjusted during the model calibration.

Since 2000, 10 observation wells have been installed and water elevation recorded on a monthly basis. This data served in calibrating and validating the groundwater model.

**Figure 8.** (**a**) Different recharge zones in the aquifer; (**b**) elevation-area-volume graph of Merk dam reservoir before and after sedimentation.

### *3.3. Dam/Reservoir Characteristics*

As proposed and if constructed, the Merk dam would be an earth-filled dam with a clay core. The normal elevation would be 91 m above mean sea level (AMSL), and the capacity of the reservoir after maximum sedimentation would be 40 × 106 m3 . The source of water to fill the reservoir would be the Merk River. The river's mean annual stream flow is 25.9 × 106 m3 . An elevation-areavolume chart of this dam (Figure 8b) was used to estimate the rate of evaporation from the reservoir in the system dynamics model developed in our study. This information was derived from a feasibility study report on the Merk dam approved by the Iranian Ministry of Energy [33]. The water supplied from the dam would be conveyed through a pipeline to agricultural areas. According to dam design reports, the mean irrigation demand of the dam's command areas would be 8013 m3 ·ha<sup>í</sup><sup>1</sup> ·y<sup>í</sup><sup>1</sup> , oscillating between 2860 and 12,710 m3 ·ha<sup>í</sup><sup>1</sup> ·y<sup>í</sup><sup>1</sup> . Figure 9a presents a schematic view of the dam, aquifer and agricultural lands [33].

**Figure 9.** (**a**) Schematic view of the proposed system, consisting of the dam, agricultural areas within the aquifer boundaries and a pipeline to convey water from the dam to agricultural areas; (**b**) discretization of the aquifer system and its side views.

Under Iranian governmental regulations [33], a certain percentage of a river's average natural flow must be allowed to remain flowing throughout the river course. This percentage is 10% during the wet seasons and 30% during the dry seasons. Consequently, this amount of water was considered the minimum environmental requirement of the river in the model.

### **4. Methodology**

A conceptual model of the region's groundwater flow was initially developed, then translated to computational form through the use of MODFLOW [35] software. The conceptual model was developed based on information presented in the section "Hydrogeology of the Aquifer". The aquifer was discretized to 45 × 35 cells, with each cell representing an area of 350 m × 350 m (Figure 9b).

The model was calibrated and run using hydrogeological data and aquifer characteristics (Table 2) collected from 2000 to 2005 by regional hydrological experts. The model was then validated using similarly obtained data for the period of 2006 to 2010 using the RMSE performance index. Once the groundwater flow had been modeled using MODFLOW, the information gained was used to build a system dynamics model of the aquifer (Figure 2).

The system dynamics groundwater model was subsequently evaluated against the MODFLOW results. In the next stage, four different ASR implementation scenarios were developed and tested using the comprehensive system dynamics model. The system dynamics model as mentioned formerly has the ability to model concurrently the dam, groundwater system and ASR. Lastly, an economic analysis was undertaken to evaluate each of the different scenarios.

### *4.1. Scenarios*

To assess the best approach to optimize the expanse of land to be converted to new farmland while maintaining appropriate environmental flows from the dam, as well as manageable spillway flows, along with a sustainable groundwater balance, four scenarios were evaluated using the system dynamics model. In all scenarios, the government's goal of adding 1000 ha of new agricultural land was respected. These lands will be referred to as "additional command areas" from now on. In order to gauge its potential economic impact, two different dam heights, resulting in initial reservoir volumes of 20 × 106 m3 or 40 × 106 m3 , were compared in Scenarios 220, 240, 320, 340, 420 and 440, respectively. In the baseline scenario, 1, only the taller dam/larger reservoir option was modeled, and this scenario was represented as 140.

Scenario 1: baseline scenario, in which the dam's effects on the water table are modeled as the dam's implementation is currently proposed (without any inclusion of an ASR approach). Water trapped in the reservoir flows to farmers' fields (old and new) through a constructed irrigation network.

Scenario 2: 40 new injection wells are constructed throughout the region, from which reservoir water is pumped under high pressure into the aquifer. Farmers continue to make use of their existing boreholes for extraction, while also using the injection wells as pumps during recovery periods.

Scenario 3: 40 new high-pressure injection wells are constructed while existing boreholes are shut down, forcing farmers to rely upon the stored water from the new sites. In this scenario, the existing agricultural lands, which were irrigated by farmers' wells, will be rehabilitated. The rehabilitation of the existing lands will add some costs into the project, but on the other hand, will increase the irrigation efficiency and productivity of the farms that will result in more benefit for the project.

Scenario 4: no new high-pressure injection wells are constructed; rather, water from the reservoir flows via gravity into existing borehole wells spread-out across the current 1000 ha of agricultural land. All additional new land is watered directly from the reservoir through a constructed irrigation network.

### *4.2. Economic Analysis*

For economic analysis, a cost/benefit of investment approach was applied, where the net present value of an investment was calculated by using a discount rate and a series of future payments (negative values) and incomes (positive values). Incomes were based on net economic gains of agricultural activities, valued at 3,556 USD ha<sup>í</sup><sup>1</sup> ·y<sup>í</sup><sup>1</sup> , based on average prices of cultivated crops in the region [33]. The cost components of the economic analysis are listed in Table 3.


**Table 3.** Potential costs involved in the aquifer storage and recovery (ASR) project in Sirik, Iran.


**Table 3.** *Cont.* 

### **5. Results**

### *5.1. Results of Aquifer Model Implemented with MODFLOW*

Modelmate software and UCODE were used to calibrate MODFLOW. Recharge, hydraulic conductivity and specific yield were introduced as parameters to Modelmate. The model results were then compared with the head measurement in 10 observation wells across the aquifer. The calibration coefficient was 0.92 in the calibration stage (Figure 10a). For validation, correlation coefficients (R2 ) reached 0.90. The root mean square error (RMSE) at the end of calibration and evaluation of the model was around one meter (Figure 11b). The results show that the conceptual groundwater model is capable of capturing the major processes in the groundwater system in the aquifer.

**Figure 10.** (**a**) Simulated water table *vs.* observed water table for the calibration period.

(**b**) Simulated water table *vs.* observed water table for the validation period.

### *5.2. Comparison of VENSIM/MODFLOW Results*

In this stage, all calibrated data were transferred to the VENSIM software, and this model was run without considering the effect of the dam and ASR system on the aquifer for a period of 10 years to examine whether the groundwater model component of the system dynamics model had the ability to model the groundwater system effectively. Results showed an R2 = 0.95 between the MODFLOW and VENSIM models and an RMSE < 1 m (Figure 11). In Figure 12, the results of the simulation of Scenario 1 and 4 at maximum reservoir capacity are presented.

**Figure 11.** Correlation between MODFLOW results and VENSIM results.

**Figure 12.** Models results for Scenarios 1 and 4: (**a**) accumulative evaporation; (**b**) average water table of the aquifer: and (**c**) water storage in the reservoir.

*5.3. System Dynamics and Economic Analysis Results* 

The results of the modeling with system dynamics and economic analysis are shown in Table 4. All scenarios had the same amount of inflow, as this was generated by the floodwaters captured by the reservoir (Table 4). Water lost to evaporation (106 m3 ) varied greatly amongst the scenarios, with 205.1 lost under the "business as usual" scenario (140), 100.1 and 156.9 under Scenarios 220

and 240, 71.7 and 118.2 under Scenarios 320 and 340 and 98.6 and 153.5 under Scenarios 420 and 440. Environmental flow (106 m3 ) from the dam varied, from 153.4 under Scenario 140, to 130.7 and 144.5 under Scenarios 220 and 240, 117.0 and 130.3 under Scenarios 320 and 340 and 129.9 and 143.6 under Scenarios 420 and 430. Spillway flow from the dam (106 m3 ) also varied, from a high of 423.6 under Scenario 140, to 342.0 and 227.3 under Scenarios 220 and 240, 290.9 and 186.0 under Scenarios 320 and 340 and 340.5 and 222.7 under Scenarios 420 and 440.

The average drawdown of the water table varied from a high of 14.5 m under Scenario 140, to 5.4 m and 3.2 m under Scenarios 220 and 240, 2.4 m and 0.9 m under Scenarios 320 and 340 and 5.3 m and 3.0 m under Scenarios 420 and 440. The total costs of implementation varied, from a low of \$37,296,000 under Scenario 140 (the basic cost of the dam and irrigation network), to \$41,258,000 and \$40,112,000 under Scenarios 220 and 240 (the costs of the dam, irrigation network, 40 new injection wells, as well as the price of pumped water), \$55,983,000 and \$51,082,000 for Scenarios 320 and 340 (the cost of the dam, irrigation network, 40 new injection wells, as well as the price of pumped water) and, finally, \$41,597,000 and \$37,407,000 for Scenarios 420 and 440 (the cost of the dam, irrigation network and modifications to existing boreholes).

**Table 4.** Water supply and economic analysis of scenarios after 40 years of simulation. Note that for Scenarios 2, 3 and 4, each scenario compared two dam heights resulting in initial reservoir volumes of 20 × 106 m3 or 40 × 10<sup>6</sup> m3 .


### **6. Discussion**

### *6.1. Consequences of "Business as Usual"*

From a water management perspective, the proposed standard reservoir and dam system planned for the Sirik region is poorly thought out, given the significant quantity of water lost to evaporation (about 25% more than any other scenario). Furthermore, continued extraction of groundwater with no plan to replenish the aquifer would lead to a water table level drawdown of 14.5 m over the next 40 years, a case that would not only greatly increase the difficulty and cost of pumping water for agriculture and endangering people's livelihoods, but also threaten local wildlife that depend on shallow groundwater levels in the southwest portion of the Sirik region. It is thus suggested that a new paradigm of groundwater management be adopted in the region that makes use of ASR to prevent losses through evaporation and slows the rate of groundwater drawdown.

### *6.2. Scenario Selection Based on Cost/Benefit Analysis*

To decide the most appropriate scenario for the development of Sirik, we rely on a variety of criteria to determine which scenario provides the best return on investment. The first is the cost/benefit analysis, which takes into account the total costs (C) of a scenario, weighed against the expected financial benefits (B) from expanded agricultural production in the region. The two scenarios that provide the greatest return on investment are Scenario 140 and Scenario 440; however, return on investment is not the only criterion for acceptability. Scenario 340 provides the greatest reduction in drawdown over 40 years, at 0.9 m, compared to 3.0 m for Scenario 440 or 3.2 m for Scenario 240. However, scenario 340's slower rate of drawdown comes at an additional cost of \$13,650,000, while only allowing 84.5% of the originally planned environmental flow. Lastly, the need for farmers to shut down their own wells and to rely solely on newly installed high-pressure injection wells poses problems of social acceptability.

Though Scenario 440 has higher rates of evaporation than Scenario 340 and a similar rate of evaporation as Scenario 240, it remains the most cost-effective scenario, providing for a manageable quantity of spillway flow (that when unmanaged can lead to flooding damage), as well as the highest proportion of the original environmental flow (93.57%). As the southern ecosystem that sustains the region's native flora and fauna depends on a shallow groundwater table, it is justifiable to transfer some water from environmental flow into the aquifer in order to maintain upwelling and springs.

Scenario 3 has more benefits and costs than the other scenarios. As it was formerly explained, in this scenario, the irrigation system in existing farmlands should be rehabilitated, so that there would be a cost associated with the rehabilitation that will be added to the base cost of the project. On the other hand, the modified system will elevate crop production, and as a result, benefits would also increase. However, the cost of the project in this scenario outweighs the benefit; thus, the benefit over the cost of this scenario is less than for the other scenarios.

### *6.3. Social Acceptability and Sustainability*

Scenario 440 is the most socially acceptable and sustainable of the solutions, allowing farmers to keep their own wells on their land and for them to be improved at no cost to the farmer. Unlike Scenarios 240 and 340, Scenario 440 does not require the installation of complex high-pressure injection and pumping stations, which require technical upkeep and repairs, but instead makes use of improved boreholes on existing plots. Technical and managerial training programs for farmers would be promoted, in order to provide users with the skills to maintain their own systems and manage water use. Through choosing to work through existing social networks and demonstrating willingness to engage, the project could gain local support from the farmers. This type of public engagement and empowerment is a central tenet of the new paradigm of integrated water resources management and sets the groundwork for farmer-led groundwater management.

### *6.4. Uncertainty Due to Climate Variability and Climate Change*

Although the models benefited from 40 years of historical hydro-climatological time series data, climate variability and climate change results in uncertainties concerning the modeling results of all scenarios. Regarding climate variability, different combinations of wet and dry hydro-climatological input parameters of the model (inflow, recharge, evaporation, *etc.*) will affect the results of each scenario. Nevertheless, since the model input parameters are the same in all scenarios, the variation between scenarios will remain relatively similar to the current study, so the deviation would not be substantial. On the other hand, climate change could have a major impact on the results, since it is believed that the input parameters of the model will no longer remain stationary in the future. It is predicted that climate change will cause more severe extreme events (floods and droughts) in this region [36]. In this situation, conjunctive use should be more beneficial than conventional water management schemes. Conjunctive use (Scenarios 2, 3 and 4) under severe drought conditions is more advantageous than merely relying on surface water.

This study introduced a new modeling tool, which also opens a new avenue to assess uncertainties due to climate variability and climate change in future studies. In order to address uncertainty in future studies, different sets of climate variables (precipitation and temperature) should be derived from downscaled climate change models, and then, this climate data can be used in hydrologic models to estimate discharge in the watershed. The output from the hydrological model can subsequently be used as an input to the SD model to build a set of results. The probability distribution function can then be derived from the results of the SD model to assess the uncertainty associated with climate change.

### **7. Conclusions**

The objective of this study was to examine if ASR, in conjunction with water storage on an ephemeral river, could be an effective water resource management strategy that would minimize both water lost to evaporation and the rate of groundwater depletion, while providing water for expanded agricultural activities. It was determined that this approach can significantly improve the sustainability of groundwater supplies. It must be emphasized that the future development of the Sirik region must include a water management approach of groundwater storage and recovery. In so doing, significant gains can be achieved at a minimal cost. By modeling groundwater flow and whole system dynamics, ASR was shown to be an applicable and beneficial strategy for the well-being of farmers and the region's groundwater system. Without the inclusion of ASR, the region will face grave consequences due to unsustainable exploitation of groundwater. However, through a combination of central technical planning, ASR strategies and farmer engagement and education, the current proposal has the potential to help direct the future development of the region in a sustainable manner.

The system dynamics modeling framework developed and implemented in this study was shown to be very effective. Not only groundwater, but a surface water reservoir was modeled in a single program. This modeling approach can be expanded and used in different areas where a combination of groundwater and surface water are considered as sources of a water supply system. Interconnection technologies, such as ASR, can also be addressed in this modeling approach, something not easily accomplished in other modeling frameworks. Although the groundwater modeling portion of the model was developed for an unconfined aquifer, it is relatively simple, using the same mathematical concepts, to develop such a model for a confined aquifer.

Another advantage of such a modeling approach is that groundwater and surface water reservoirs are completely linked to each other and in each time step; each model is updated with the output of the other model. This mutual relationship enables one to solve the problem with greater accuracy and fewer simplifying assumptions.

### **Acknowledgments**

The authors would like to thank David Dean, Nicole Guo and Naeem Ahmed for their assistance in the preliminary stages of this research. The authors would also like to take this opportunity to thank two anonymous editors who improved the quality of this research with their very useful comments.

### **Author Contributions**

Amir Niazi conceived the idea of this research, conducted the SD modeling, economical analysis and writing of the manuscript. Shiv Prasher funded the research and provided recommendations, which improved the system dynamics model. Jan Adamowski provided feedback regarding the SD models and modified the manuscript throughout the project. Tom Gleeson helped conceptualize the groundwater system in the study area, as well as the groundwater modeling in the SD models.

### **Conflicts of Interest**

The authors declare no conflict of interest.

### **References**

1. Giordano, M.; Villholth, K.G. *The Agricultural Groundwater Revolution: Opportunities and Threats to Development*; CABI (Commonwealth Agricultural Bureaux International): London, UK, 2007; p. 419.


## **Assessing the Feasibility of Managed Aquifer Recharge for Irrigation under Uncertainty**

### **Muhammad Arshad, Joseph H.A. Guillaume and Andrew Ross**

**Abstract:** Additional storage of water is a potential option to meet future water supply goals. Financial comparisons are needed to improve decision making about whether to store water in surface reservoirs or below ground, using managed aquifer recharge (MAR). In some places, the results of cost-benefit analysis show that MAR is financially superior to surface storage. However, uncertainty often exists as to whether MAR systems will remain operationally effective and profitable in the future, because the profitability of MAR is dependent on many uncertain technical and financial variables. This paper introduces a method to assess the financial feasibility of MAR under uncertainty. We assess such uncertainties by identification of cross-over points in break-even analysis. Cross-over points are the thresholds where MAR and surface storage have equal financial returns. Such thresholds can be interpreted as a set of minimum requirements beyond which an investment in MAR may no longer be worthwhile. Checking that these thresholds are satisfied can improve confidence in decision making. Our suggested approach can also be used to identify areas that may not be suitable for MAR, thereby avoiding expensive hydrogeological and geophysical investigations.

Reprinted from *Water*. Cite as: Arshad, M.; Guillaume, J.H.A.; Ross, A. Assessing the Feasibility of Managed Aquifer Recharge for Irrigation under Uncertainty. *Water* **2014**, *6*, 2748-2769.

### **1. Introduction**

Water demand continues to grow in order to maintain food security and drinking water supplies, while supplies remain limited from conventional sources. Future water security is threatened in many places, as most suitable locations for large surface storages have already been used [1] and ground water is often being withdrawn at unsustainable rates [2–4]. Among other options of water supply augmentation, such as water recycling, desalination *etc.*, storing more water underground appears to be a potential solution to achieve future water supply goals. For many water stressed areas, water security and reliability do not necessarily depend on the absolute amount of precipitation, but on the fraction of water that is efficiently retained as storage for future use [5].

Water shortages can be eased by storing surplus water underground during wet periods for later use during dry periods. Managed aquifer recharge (MAR) has been used successfully in several countries for the storage and treatment of water [6–9]. Storage of surplus water in aquifers can help minimize evaporative losses and help irrigators to adjust to surface water variability during droughts, provided that MAR is technically feasible and cost effective. The feasibility of MAR and its comparative cost to other alternatives depend on a number of technical and financial factors, such as infiltration, injection and recovery rates, which are dependent on local hydrogeology [10].

A few studies indicate that MAR can achieve more financial value than surface storage and other alternatives [11,12]. However, uncertainty often exists whether it is more cost effective to store water above ground in surface reservoirs or below ground using managed aquifer recharge [13].

Cost-benefit analysis (CBA) provides a comparison of benefits and costs resulting from a proposed policy or investment [14]. Previous studies undertaking CBA of MAR have assumed hydrogeological factors, such as infiltration, injection and recovery rates, to be known [11,12,15]. Overlooking such uncertainties can result in lower than expected operational efficiency and irrigation returns from MAR [16,17]. For example, future returns from MAR may be affected by increases in groundwater pumping cost or reductions in infiltration rates.

An increase in the turbidity of source water due to hydrological variability can significantly increase the cost of infiltration basin maintenance, adding to the cost of water quality treatment for aquifer storage and recovery (ASR) systems. Maliva [16], in this special issue, highlights that assessing such uncertainty is perhaps the most neglected aspect in the economics of MAR.

The primary focus of this paper is to systematically search for conditions under which the requirements for MAR may not be met and failure might occur. Playing such a devil's advocate role has been shown to improve decision making compared to an exclusively expert-driven approach [18]. The approach used identifies thresholds above which MAR is financially better than surface storage and below which it is not. These thresholds (or cross-over points) describe corresponding values of variables at which the net present value (NPV) from MAR and surface storage become equal. All dollar amounts reported in this study are in Australian dollars. An example of a cross-over point for pumping cost is shown in Figure 1, where basin infiltration (red line) and surface storage (green line) options are compared; and where basin infiltration is initially (dashed vertical line) more profitable than surface storage. A cross-over point between the two compared options is possible when the cost of pumping increases from the best guess value of 35 \$/megalitres (ML) to 53.63 \$/ML. This increase in the pumping cost will decrease benefits (NPV) from basin infiltration, such that they become equal to the benefits (NPV) obtained from surface storage. However, aquifer storage and recovery (ASR) always result in an inferior NPV regardless of the pumping cost. There is no cross-over point between ASR and the other alternatives.

**Figure 1.** Illustration of identifying cross-over points for pumping cost when comparing basin infiltration, aquifer storage and recovery (ASR) and surface storage of irrigation water.

At the cross-over point, the decision maker is indifferent to choosing a single option from the two, because their financial returns are equal. In our method, we use computational techniques to identify the cross-over points as values of uncertain variables where the NPV of MAR is exactly equal to the NPV of surface storage of irrigation water. The approach is demonstrated through a case study in a highly developed irrigation region of the lower Namoi catchment in New South Wales, Australia, where irrigation water restrictions motivate the need to consider options to supplement future irrigation supplies, such as MAR.

The suggested approach of identifying cross-over points is beneficial in three ways;


The next section provides an overview of the literature on the feasibility of MAR with a focus on the technical, financial and uncertainty considerations. Section 3 ("Methods") describes the model and tool used to explore cross-over points. In Section 4, an illustrative study in the lower Namoi catchment evaluates the irrigation-related costs and benefits of storing flood water in aquifers compared to surface storages. The analysis of cross-over points in Section 5 provides a discussion of how cross-over points are reached when only a single variable changes, as well as when many variables interact.

### **2. Related Work: Feasibility of Managed Aquifer Recharge**

Assessing the feasibility of MAR requires the integration of many types of data and information from many disciplines (Figure 2). Although carrying out a comprehensive feasibility assessment is essential, the first step in establishing an MAR scheme requires assessing the feasibility of technical and financial factors, to provide a basis for other investigations to proceed.

An overview of the basic requirements and feasibility guidelines for managed aquifer recharge (MAR) is available in [10] and GHD and AGT [19].

### *2.1. Technical Considerations*

Key technical requirements for MAR include hydrogeological assessment of the target aquifer, the availability of surplus surface water and the means to convey it underground. Relevant hydrogeological factors include aquifer storage size, permeability, infiltration, injection and recovery rates and connections with other aquifers [21,22]. High infiltration rates lower the cost of underground storage; for example, a basin infiltration system with high infiltration rates will require a smaller pond area and can be cheaper to construct and maintain than a pond with low infiltration rates.

There are two main types of MAR methods: basin infiltration and aquifer storage and recovery (ASR), each favourable to different hydrogeological conditions. Basin infiltration is suitable to recharge shallow unconfined aquifers with minimal or no treatment of the recharge water. The methods include deep, large diameter isolated wells, infiltration ponds, infiltration galleries, induced bank filtration, leaky and recharge dams and redirecting floodwaters over the wider landscape to supplement areal recharge [7,9]. Some basin infiltration methods require large surface areas and permeable soils to be effective [21,23].

ASR involves the injection and recovery of water using wells; this has the advantage of targeting a desired aquifer for recharge. Thus, zones of saline water or clay layers can be bypassed. However, ASR systems are costly because of the need for bore well construction and water treatment prior to recharging, and if clogging occurs, they are costly to repair. Passive borehole recharge (under gravity) requires limited mechanical assistance, but the infiltration rate is relatively low. Water injection using pumps can greatly improve the rate of aquifer recharge [24,25]; however, the pumps require constant maintenance and are costly to run. The risk of clogging of the surface or well with fine sediments is common to both MAR methods. Solutions to this issue include stabilization of recharge water through settling ponds and treatment of water before recharge.

### *2.2. Financial Considerations*

When the focus is on estimating the total economic benefits of recharge to a region instead of an individual, the benefits of aquifer storage become complex, as this needs to include public good, socio-economic and environmental benefits to a region, which are more difficult to assess and quantify. Maliva [16] in this special issue provides a greater review of the methods and techniques for assessing total benefits from MAR. With a known target volume of storage and recovery, it is easier to quantify the financial benefits, since the goal is the recovery of the stored water, and the volumes recovered accrue to an identifiable person or water utility for a particular use. The financial feasibility of MAR can then be studied in comparison to other water supply and storage alternatives, including surface storage.

The local situation dictates the costs of MAR options, and large variations may occur between localities [10]. For a fair comparison, it is essential to analyse the benefits and costs of MAR and surface storage in the same location, because the comparison of benefits and costs is complicated by the wide range of biophysical, socio-economic and regulatory conditions in which MAR occurs. There is little published analysis of the economic and financial benefits of MAR. From the few published studies, Ross and Arshad [26] compiled and reported the benefits and costs of surface storage and MAR at multiple locations, showing that the costs and returns of MAR options vary substantially.

### *2.3. Uncertainty Considerations*

A number of methods have been used to address uncertainty in cost-benefit analysis. Sensitivity analysis simulates the impact of changes in financial behaviour, such as the change in NPV of an investment due to a change in an input variable, and identifies variables that are of greater concern [27]. Probabilistic analysis provides the combined effect of variables' variability on the financial behaviour [17]. Possibility theory assumes that all values within a certain range are possible, with the exact value being treated as unknown [27].

We focus on cross-over points as one possible means of addressing uncertainty in the cost-benefit analysis of MAR. Identification of cross-over points relies solely on the relationship between variables, such that it requires minimal understanding of the uncertainty of variables. The idea of a cross-over point is sufficiently simple that it has a number of widely used variations; it is also known as a break-even point or switch-over point. However, the term break-even in economics specifically applies to the volume of sales at which profit is zero as revenues cover total cost and is therefore used as a tool to calculate the margin of safety of a single investment [28], rather than comparing alternatives. The concept of a cross-over point is fairly simple with only one or two variables, but the complexity increases in the analysis and interpretation of results as the number of input variables increases [29].

### **3. Methods**

The analyses in this paper are carried out in two steps; in the first step, financial analysis compares the net present value of farm benefits to identify the best among the considered options. In the second step, the break-even analysis of cross-over points is carried out; this involves finding values of variables that will provide exactly the same financial returns from the two compared options. The variables were chosen based on an examination of literature concerning the financial feasibility of MAR. Identifying cross-over points allows the user to understand the minimum conditions required for success and allows measures to be taken to ensure they do not occur.

Financial analysis evaluates whether investment in MAR is worthwhile. Analyses of cross-over points help understand the circumstances when MAR is worthwhile. At the most basic level, MAR is worthwhile when net irrigation returns of MAR exceed those of alternatives. In our example, benefits are determined by the agricultural value of the additional water provided, by saving it from non-productive evaporation. This has been referred to as a "vapour shift" [30] from non-productive evaporation to agriculturally-valuable crop transpiration. Costs are composed of additional pumping to recover recharged water and MAR method-specific capital and ongoing costs of implementation during the life of the project.

To enable the break-even analysis, the financial analysis is programmed as a function in R [31]. As a general purpose statistical programming language, R offers a suite of optimization methods, as well as providing tools for visualization and the means to include a user interface. To identify cross-over points of single variables, other variables are set to fixed values, and the R function *uniroot* [32] is used to identify the value of the variable where the difference in NPV between the two compared options is zero (*i.e.*, ǻNPV(Ĭ) = 0), meaning that the two options have equal NPV. To identify cross-over points involving many variables, we use optimization to identify a crossover point (*i.e.*, a point Ĭ where ǻNPV(Ĭ) = 0) that is closest to the best guess, in the sense of minimizing the maximum of the distances for each variable, expressed in relative terms using userdefined bounds (maxi|Ĭi-Ĭbest,i|/|Ĭbound,i-Ĭbest,i|). This is one possible criterion for selecting cross-over points of concern. Other criteria, including probabilistic ones, would be possible and would usually raise different cross-over points for discussion. The code for the analysis is available online [33]. The cross-over points generated are assessed by comparing them to maximum and minimum values of variables that a decision maker thinks might be possible due to physical, climate or policy change over the analysis period. The resulting judgment of a cross-over point is not perfect and is based on the best available knowledge of the decision maker for each variable.

### **4. The Study Area: Lower Namoi**

In many parts of Australia, overdraft of aquifers is resulting in falling groundwater levels in the shallow, unconfined systems and decreasing groundwater pressures in the deep confined and semi-confined systems [34]. In response to the groundwater overdraft, the New South Wales (NSW)

government has reduced current groundwater entitlements in its stressed aquifer systems [35]. For the lower Namoi catchment, a highly developed cotton irrigation district in NSW, this cutback translates to a reduction of 21 gigalitres (GL)/year in groundwater entitlements for irrigation by 2015 and beyond. Groundwater in the Namoi River catchment supports an irrigation industry worth in excess of \$380 million per annum [36]. All irrigation water is stored and routed from surface storages before application to the field. On-farm water storages within the lower Namoi range from conventional single-cell to advanced multi-cell farm dams. The typical Namoi valley farm holds enough water in storage to complete one full year of irrigation. Conservative estimates suggest that the total capacity of on-farm storages in the cotton industry could be on the order of 3150 gigalitres (GL). Evaporative losses from these surface storages are significant. On average, from surface water storages, evaporative losses range from 1200 to 1800 mm/year [37], which constitute 35% to 50% losses from surface water storage volumes.

To tackle the problem of reduced allocation and evaporative losses, improving water use efficiency at the farm level is an obvious option. This will include installing drip irrigation systems, lining water courses and further improving the design of surface storages to minimize evaporative and seepage losses. Improving water use efficiency needs to be a stepwise approach. Another potential option to reduce evaporative losses is to store water underground in aquifers using managed aquifer recharge. Recently, several studies have highlighted the potential of a regional-scale MAR project in the lower Namoi. Arshad *et al.* [38] indicated that a significant volume of water could be available from large floods for MAR while still satisfying environmental flow and ecological requirements. Similarly, Rawluk *et al.* [20] showed a high level of social acceptability for an MAR project in the study area.

### *4.1. The Analytical Framework for Financial Analysis*

The study undertakes an analysis to estimate irrigation-related costs and benefits for a typical irrigation farm in the lower Namoi. The analysis considers a cotton irrigation farm, which has three different scenarios for the storage of flood water: surface storage in farm dams, aquifer storage using basin infiltration and ASR using existing wells. All of the surface water allocations, including flood water, is stored in farm dams before application to the fields. Owing to limited water availability, less than 20% of the available land is irrigated, and irrigated land in each year is variable. Irrigated cotton Bt (*Bacillus thuringiensis*) and faba bean *(Vicia faba L)* are the sustainable summer and winter rotations that provide the highest net income per megalitre (ML) of irrigation water applied [39]. It is assumed in the analysis that all required irrigation infrastructure, such as surface storage and the irrigation water delivery network, are already built for the entire irrigation land, as this is a common practice in the study area. The annual irrigation water allocation from all sources for an average cotton farm in the lower Namoi is approximately 1350 ML. However, in this analysis, we only consider and report irrigation costs and returns of 200 ML of flood water, which is only 25% of recent statutory flood water allocations in the study area. The analysis assumes 40% evaporative losses, taking into account current estimates in the study area [37].

Storage and recovery of water underground will require new infrastructure and additional costs, as reported in Section 4.3. Farm economic data, such as the variable cost of farm inputs, cotton prices and gross margins from irrigated and dryland, are adopted from Powell and Scott [39]. The analysis only considers farm-related costs and revenue and does not monetize any socio-economic or environmental cost or revenue that may occur as a result of a change in the water storage option.

### *4.2. Infiltration and Injection Rates That Can be Possible in Lower Namoi*

Infiltration and injection rates can highly affect the usefulness of any aquifer recharge and storage facility. Bouwer [40] provides typical infiltration rates for surface infiltration systems in the range from 0.3 to 3 m per day (m/day) with relatively clean and low turbidity river water. For systems that are operated year-round, long-term infiltration rates vary from 30 m/year to 500 m/year, depending on soil type, water quality and climate. In the lower Namoi, the infiltration rate of 0.2 m/day is considered to be likely achieved in many locations.

ASR can achieve injection rates from 0.5 to 8 megalitres per day (ML/day) per borehole (1 megalitre = 1000 cubic meter = 0.8107 acre foot). In the absence of accurate well injection rates based on field monitoring, Pyne [41] observed that injection rates increased with increasing aquifer transmissivities. For the lower Namoi, Williams *et al.* [42] reported that the alluvial aquifers that are primarily tapped for irrigation extraction are associated with the semi-confined Gunnedah and Cubbaroo formations and have transmissivities in the range of 1000–2000 square meters per day (m2 day<sup>í</sup><sup>1</sup> ). The yields from bores tapping these aquifers vary up to 250 litres per second in the Gunnedah Formation at depths of 60–90 m and in the deep Cubbaroo Formation at depths of 80–120 m. The shallow Narrabri Formation has transmissivities less than 250 m2 day<sup>í</sup><sup>1</sup> . For this study, an assumed injection rate of 25 L per second (2.2 ML/day) is considered likely for an ASR well.

### *4.3. Estimation of Costs and Benefits*

Cost estimates of aquifer recharge are scarce and can vary considerably with location. Itemized costs for this study were estimated by combining current market rates of earthworks, services and materials for water infrastructure projects in Australia and were adjusted to the local situation in the lower Namoi. Cost estimates were also compared with published data and technical reports of Khan *et al.* [12], Dillon *et al.* [10] and Pyne [13].

Capital costs of basin infiltration were estimated by assuming an infiltration rate of 0.2 m/day and calculating the required land area to achieve 2 ML of recharge per day. The target volume of harvested flood water of 200 ML would, on average, appear in four or more events in a flood year. An infiltration pond with a surface area of 1 ha and an infiltration rate of 0.2 m/day would recharge 50 ML of floodwater in a single cycle of 25 days. The size of the basin here has therefore been designed to operate only for 100 days, in 4 cycles of 25 days each, allowing rest and maintenance. The analysis assumed 40% evaporative losses from surface storage and a 5% MAR loss rate. The MAR loss rate is the percent of water lost during aquifer recharge and recovery from basin infiltration and ASR and can be expressed as:

$$\text{MAR loss rate} = \left(1 - \frac{\text{groundwater volume recovered}}{\text{Initial water volume used for storage}}\right) \text{\textsuperscript{6}}$$

In the base case, surface storage of flood water, the costs considered are the cost of harvesting 200 ML of flood water and the cost of farm dam annual maintenance. The capital cost of basin infiltration includes the cost of earth works and pipes. Ongoing costs include operation and maintenance of water harvesting and recovery and the cost of basin annual maintenance. An existing bore is assumed to be available for recovery after basin infiltration or for injection and recovery in ASR. The capital cost of an ASR facility on existing farms with a bore primarily includes installing a coagulation and filtration pre-treatment facility. Ongoing operation and maintenance costs for ASR include well maintenance, flood water harvesting, water treatment and water recovery. The analysis assumed a 30-year lifespan for surface storage and basin infiltration and 20 years for ASR, with a 7% uniform discount rate for all options. All capital cost estimates are exclusive of land value. Table 1 summarizes the levelised costs of 200 ML of flood water with each water storage option. Levelised costs are annual unit costs obtained by amortising capital cost components over their expected working life, adding the annual operation, maintenance and management cost and dividing by the annual volume of supply, as defined in Dillon *et al.* [10].


**Table 1.** Levelised costs (\$/ML) of surface storage and MAR methods in lower Namoi. Adapted from Dillon and Arshad [43]. ASR, aquifer storage and recovery; ML, megalitre.

Note: Totals may not match due to rounding.

With the additional water saved through MAR, farmers in our example have the choice to irrigate additional land with cotton, faba bean or some combination of the two crops that yields the highest returns. Value brought by the MAR water under each option is estimated from the useable volume of flood water, after evaporative and recovery losses, times the gross margin per ML of mixed cropping of cotton and faba bean on equal land areas. On average, for a typical lower Namoi irrigation farm, the average gross margins for cotton and faba bean are estimated as 310 \$/ML and 435 \$/ML, respectively. It is assumed that cotton and faba bean are planted on the same land area, as they are summer and winter crops, respectively. Allocating the water accordingly yields an average gross margin of 342.3 \$/ML and a net margin of 230 \$/ML after subtracting overhead costs. In the analysis, we assume that additional irrigation with the saved water is not going to increase the overhead cost, as the farm size is large enough (1200 ha) and irrigated land cropped each year is variable depending on water availability. In this analysis, we use gross margins as the irrigation returns, which is the total revenue minus the variable cost of production. Table 2 presents the value of crop that can be grown with the useable volume in each case.


**Table 2.** Irrigation benefits: value of the crop under each water storage option. Adapted from Powell and Scott [39] and Arshad *et al.* [44].

Note: Totals may not match due to rounding.

### *4.4. Results of Financial Cost-Benefit Analysis*

A long-term trajectory of the difference of the discounted benefits and discounted costs of the three water storage options is expressed in Figure 3 as net present value using the fixed data in Table 2.

**Figure 3.** Net present value (NPV) of surface storage, basin infiltration and aquifer storage and recovery options.

The results show that MAR using the basin infiltration method will yield 11% more value than surface storage of irrigation water. ASR using existing wells appears to be uneconomical, with 64% less value than surface storage, mainly due to the high capital and water treatment costs required for an ASR system.

The cost and additional value of basin infiltration is highly dependent on the infiltration rates; as infiltration rates increase, the capital costs decrease, and the value of saved water increases. Conversely, as infiltration rates decrease, the capital cost increases, and the additional value of basin infiltration decreases. With a reduction in the infiltration rates, a cross-over point is reached, where the additional value brought by basin infiltration becomes zero and its NPV is exactly equal to that of surface storage. The following section expands the analysis to explore cross-over points of infiltration rates and other variables.

### **5. Identification of Cross-Over Points in a Single Variable**

In single variable analysis, the aim is to identify how far a single variable needs to change to reach a cross-over point for the two compared options. A cross-over point may not always exist; there might be situations where the cross-over point falls outside the minimum or maximum limits considered for the analysis or when the change in the cost or benefit is in the same direction. Such a situation is noted with the use of the acronym, NA, for not applicable, in the tables and following text. A cross-over point for basin infiltration and surface storage occurs when their NPVs are equal; and similarly, for ASR and surface storage, as well as basin infiltration and ASR. Figure 1 showed the cross-over point for pumping cost. Figure 4 illustrates cross-over points for basin capital cost.

**Figure 4.** NPV for varying basin capital cost in three water storage scenarios, showing cross-over points at intersections between lines.

A cross-over point between basin infiltration and surface storage is possible when the basin capital cost increases from 363 \$/ML to 466.69 \$/ML. That increase in the capital cost will equate to the NPV of the two compared options. Similarly, a cross-over point between basin infiltration and ASR is possible when the basin capital cost increases from 363 \$/ML to 1085.55 \$/ML. No cross-over point is identified between surface storage and ASR (it is NA). The increase in the basin capital cost may result from increases in the price of services and materials or the need to construct a larger pond due to a reduction in infiltration rates. Rather than drawing these curves for

**116**

every variable, the values of the cross-over points are reported in Table 3 and discussed in the following text.

Table 3 lists cross-over points for 14 variables when each is varied separately. These cross-over points represent the minimum requirements for MAR to be preferred to surface water storage, assuming that the values of other variables listed in the table remain fixed. For example basin infiltration is financially better than surface storage when pumping cost does not exceed 53.63 \$/ML or the surface evaporation rate does not fall below 34%, and so on. The variables selected are the most important when undertaking a financial comparison of surface storage with the two MAR options. In the following section, we discuss the basis of how these cross-over points may be reached in reality for each single variable.


**Table 3.** Single variable cross-over points in three scenarios.

### *5.1. Discussion of Single Variables*

### 5.1.1. Pumping Costs and Surface Evaporation Rates

A cross-over point between surface storage and basin infiltration is possible when pumping costs increase by 53% to become 53.63 \$/ML; an increase in the cost of pumping will cause an increase in the cost of agricultural production and a decrease in farm benefits (NPV) from basin infiltration. A cross-over point between basin infiltration and ASR is NA, because the rate of increase in pumping cost applies to both aquifer storage options. Similarly, there is no cross-over point between surface storage and ASR, as the lowest possible pumping cost considered in the analysis (6.25 \$/ML) will not make ASR financially superior or equal to surface storage.

Low surface evaporation rates will make surface storage financially superior to MAR, as less water will be lost from surface storage, making more water available and resulting in larger benefits. A cross-over point between surface storage and basin infiltration is possible when evaporation rates decrease by 15%, from 40%, to become 34%. For evaporation rates, a cross-over point between surface storage and ASR is possible when evaporation rates increase to 74%, whereas the cross-over point between basin infiltration and ASR is NA.

### 5.1.2. Basin Capital Cost, Basin Infiltration Rate and Basin Maintenance Rate

An increase in basin capital cost will increase the overall cost and lower the benefits with a concomitant decrease in NPV. For the basin capital cost, a cross-over point between surface storage and basin infiltration is possible when the capital cost of basin infiltration increases from 363 \$/ML to 466.69 \$/ML.

A decrease in the infiltration rates will recharge less water per unit area of infiltration basin, requiring a large infiltrating pond area with larger capital cost, or with decreased infiltration rates, less water will infiltrate and be stored underground. A cross-over point between surface storage and basin infiltration is possible when infiltration rates drop from 0.2 m/day to 0.16 m/day. Similarly, a cross-over point between basin infiltration and ASR is achieved when infiltration rates drop from 0.2 m/day to 0.07 m/day. An increase in the basin maintenance rates will increase the overall cost of basin infiltration, reducing NPV in comparison to the compared options. A cross-over point between surface storage and basin infiltration is possible when basin maintenance rates increase from 10% to become 15%. The three considered variables do not apply when comparing surface storage and ASR, such that the corresponding cross-over points are NA.

### 5.1.3. MAR Loss Rate

Increasing the MAR loss rate makes MAR financially less attractive, because it reduces the volume of water recovered and the resulting benefits, though some pumping cost is saved, as less water is recovered with an increase in the MAR loss rate. In other words, a higher MAR loss rate represents a lower recoverability and, therefore, lower useful storage [22,45]. For benefits to be realized, the volume of water that is not recovered from storage must be less than evaporation losses. This applies to both MAR methods when compared to surface storage. A cross-over point between basin infiltration and surface storage is possible when the MAR loss rate reaches 11%. A cross-over point between basin infiltration and ASR is NA.

### 5.1.4. ASR Water Treatment Cost and ASR Maintenance Rates

A cross-over point for ASR maintenance rate is not possible when ASR is compared with surface storage and basin infiltration. Even its cheapest possible value, when considered alone, does not achieve an NPV equal or superior to basin infiltration and surface storage. The ASR water treatment cost only has a cross-over point if the treatment cost decreases by 91% to 13.25 \$/ML. Increases in both variables increase the cost of ASR and, hence, (further) diminish its advantage over the other options.

### 5.1.5. Price of Cotton and Faba Bean

A decrease in the price of cotton and faba bean will influence the benefits of all three water storage options and lower NPVs for each case. A cross-over point for the price of cotton and the price of faba bean between surface storage and basin infiltration is possible when the price of cotton drops from \$538 per bale to \$475.64 per bale, and the price of faba bean drops from \$348 per tonne to \$229.52 per tonne, which are 11% and 34% drops from the best guess values, respectively. A cross-over point for the cotton and faba bean price is possible between surface storage and ASR when the price of cotton rises to \$1,155.22 per bale, an increase of 114%. No cross-over point between surface storage and ASR is possible with the highest price considered possible for faba beans.

### 5.1.6. Discount Rate and Project Lifespan

An increase in the discount rate tends to increase the levelised cost of the two MAR options, in particular through the basin capital cost and the capital cost of establishing an ASR treatment facility. This will result in lower NPVs from the two MAR options. A cross-over point between surface storage and basin infiltration is possible at a discount rate of 13%, while there is no cross-over point between surface storage and ASR. Because ASR is already more expensive than surface storage, a higher discount rate will make ASR even more expensive, while the lowest considered discount rate of 1% will not be able to raise the NPV of ASR to be equal or superior to surface storage. Similarly, a lower discount rate will make basin infiltration more favourable than ASR, so no cross-over point is possible.

Lowering the lifespan of an option increases its levelised cost, such that the NPV of that particular option is lowered. A cross-over point between surface storage and basin infiltration is possible when the lifespan of surface storage increase from 30 years to 48.16 years or the lifespan of the basin infiltration drops from 30 years to become 23.51 years. Similarly, a cross-over point between surface storage and ASR exists when the lifespan of surface storage drops to 5.57 years. A cross-over point between basin infiltration and ASR is possible when the lifespan of the basin infiltration drops to 6.69 years. No cross-over point for the lifespan of ASR is possible when compared with basin infiltration and with surface storage options.

### *5.2. Changes in Cross-Over Points Due to Interactions between Variables*

The values at which cross-over points occur are affected by the values of other variables, so it is important to consider interactions between variables. Every variable that either increases or decreases changes the financial advantage of MAR in comparison to surface storage. We describe the advantage of MAR in terms of change in the position (value) of cross-over points with respect to the best guess. The interaction of two variables can bring a cross-over point closer or further to the best guess. Two variables can interact in a way that they can increase, decrease or balance the effect of each other on the resulting advantage of MAR, depending on whether changes in the variable increase or decrease the financial advantage of MAR.

A cross-over point that moves away from the best guess value indicates increasing financial advantage for MAR. Conversely, when it moves closer to the best guess, the financial advantage decreases. The movement of a cross-over point closer to the best guess reveals situations where the benefits of MAR are reduced and could ultimately have equal benefits to surface storage when the cross-over point coincides with the best guess value.

Figures 5 and 6 illustrate with examples where the advantage of MAR over surface storage changes due to the interaction of variables. This is expressed through changes in the cross-over point of the MAR loss rate.

Given that increased costs reduce the relative benefit of MAR, when costs increase, the cross-over point for the MAR loss rate moves closer to the best guess value (Figure 5). Similarly, lower prices of crops decrease the benefit of MAR, and the cross-over point moves closer to the best guess (middle bar in Figure 6). When costs and prices both increase, the cross-over point can move closer or further from the best guess, depending on the level of change in costs and prices (bottom bar in Figure 6).

**Figure 5.** Plot of the MAR loss rate when costs increase. An example of a cross-over point moving toward the best guess.

**Figure 6.** Plot of the MAR lose rate when costs and prices change. An example of the cross-over point changing position when costs and prices both increase.

### *5.3. Assessing the Risk of Attaining Cross-Over Points*

Uncertainty in the financial assessment of MAR can be assessed by evaluating whether the scenarios described by the cross-over points identified are likely to be experienced in reality. If this occurs, then MAR may not be financially attractive. Alternatively, other measures may need to be taken to avoid situations leading to the cross-over point. Note that initial financial analysis suggests that basin infiltration is a favourable investment. As mentioned in the Introduction, the aim of this analysis is therefore to play the devil's advocate, that is to systematically search for reasons that requirements may not be met and that failure might occur.

While cross-over points could be assessed probabilistically, a simple approach is to say that a cross-over point is of greater concern if it is closer to the best guess value. This implies that investment in the MAR infrastructure is at greater risk of not making additional profits than surface storage because the return from MAR becomes closer to that of surface storage. On the other hand, the value of a cross-over point may fall outside the bounds (minimum and maximum limits) that are considered to be of concern, in which case, the analysis suggests that the minimum requirements will be met.

Following this approach, Table 4 shows the cross-over point of greatest concern when surface storage and basin infiltration are compared. The point was identified by simultaneously varying all of the variables and searching for a combination where each variable is closest to the best guess, relative to bounds. The bounds were defined by the authors based on an understanding of the factors influencing the variables, taking into account the expected variability, considering the lack of complete the knowledge of hydrogeological variables and the actions that can be taken to manage these concerns. In interpreting the results, the combination of values is assessed, not just each variable separately, and the reasons for the bounds selected are explained.


**Table 4.** Cross-over point of greatest concern with basin infiltration *vs.* surface storage, using a subset of variables.

Table 4 shows that the values of cross-over points are very close to the best guess and, hence, are of concern. The point of greatest concern describes a scenario of particularly unfavourable conditions, namely when all of the variables interact and change simultaneously. The scenario of greatest concern describes a situation where pumping costs have increased and the prices of cotton and faba bean have decreased. Basin capital cost turns out to be higher than expected, as well as the MAR loss rate. The lifespan of the basin infiltration project is marginally shorter than that of a surface storage project. Other variables remain at their best guess.

Individually, all variables of the scenario appear to be of great concern. However, in reality it is unlikely that all variables change at once and result in the situation described in Table 4. We analyse groups of variables to assess whether or not the generated scenario is possible, what mitigation options might prevent this cross-over point from occurring and what adaptation actions might be taken if it the scenario described by the cross-over point does occur.

### 5.3.1. Pumping Costs and Surface Evaporative Rates

The cross-over point of this variable is very close to the best guess value and, hence, may be reached. Based on historical trends, energy costs are expected to increase in the future, despite efficiency improvements in pumping technologies. However, the effect of higher pumping costs may be balanced or outweighed if there is an increase in the price of cotton and faba bean in the future. In addition, if farming becomes uneconomical at some stage, it is possible that government might provide subsidies for pumping to maintain agricultural production and preserve the livelihoods of farmers. Using alternate sources of energy, such as wind and solar, can be cheaper mitigation options in the future. High head gravity feed systems can be designed in certain cases to avoid pumping costs [9].

Surface evaporative rates are expected to increase under climate variability and change [46]. Evaporative rates may also be higher for farms where surface storage is shallow and depending on the water colour and turbidity. Higher evaporative rates will favour MAR, so this is unlikely to be a reason not to proceed with MAR. Reducing evaporative losses from surface storage at costs cheaper than those of setting up a basin infiltration system could have been a reason not to proceed with basin infiltration.

### 5.3.2. Basin Capital Cost, Infiltration Rate and Basin Maintenance Rate

The increase in the basin capital cost seems likely to occur if the investment is delayed, as the cost of labour, construction materials and energy prices for earth moving machinery are expected to rise due to inflation and other economic factors. Similarly, the values of basin infiltration rates and basin maintenance rates exactly coincide with the current best guess estimates, and hence, the two variables are of great concern. The infiltration rate is a function of a number of variables, with water quality a major factor.

A few mitigation options exist to avoid increases in basin capital cost. Field trials and geophysical investigations can help find suitable sites where high infiltration rates can be achieved. Basin maintenance is related to the amount of silt and other suspended and organic matter contained in the floodwater. Basins can be sized to allow rest and maintenance. In the lower Namoi catchment, floodwater already passes through a *de facto* two-stage sediment and silt removal process. Firstly, it is retained in large public dams before release, thereby reducing heavy silt loads; secondly, at the farm level, floodwater is kept in farm dams as temporary storage before recharging begins. The two-step sediment removal process can be advantageous in lowering the cost of basin maintenance.

### 5.3.3. MAR Loss Rate

In the lower Namoi, more than four decades of groundwater pumping have dropped the water levels, and in many places, rivers and streams (naturally) recharge groundwater [47], such that useful storage exists at a large scale. At the farm scale, while water may not physically stay within a farmer's land and, as such, is not physically stored, the system of surface and groundwater water rights means that injected or infiltrated water could, in principle, be allocated to the farm anyway, in a form of "regulatory storage" [22]. This results in potentially extremely high recovery rates (95%) and low loss rates, as a farmer benefits from contributing water to a common pool rather than being restricted to physically retrieving the water that they recharged. The loss rate determined by regulation could however be affected by a number of broad-scale issues. For example, the MAR loss rate can be of concern for locations where surface water and groundwater connectivity exists and where streams and rivers gain groundwater from aquifers, which is rare in lower Namoi. Low recovery is possible only in aquifers that contain brackish or high salinity water, due to the mixing of fresh recharge water with the native high salinity groundwater. This may occur in some parts of the lower Namoi, particularly areas where drops in groundwater hydraulic heads have resulted in the mixing of saline and freshwater within different layers of aquifers. In areas of excessive groundwater extraction, groundwater hydraulic heads can drop and allow saline water to

enter into pumping wells [42], thereby increasing the salinity levels of the recovered water and resulting in less recovery of the volume of freshwater recharged initially.

### 5.3.4. Price of Cotton and Faba Bean

Cross-over points for the price of cotton and faba bean are not likely to occur, and they are not of greater concern. The future price of cotton is expected to remain stable or increase because of ongoing demand and an established linkage of the Australian cotton industry to overseas markets, where demand exists and can be expected to grow. In the future, with limited irrigation water availability at the global scale, international prices of cotton are expected to rise, rather than decrease. Other cotton producing and competing countries, such as China, Pakistan and Egypt, are likely to become more water stressed in future. Additionally, with world population growth continuing unabated, a higher demand for cotton is expected. The price of faba bean is also expected to increase in the future; however, a drop in the price of faba bean is also possible whenever supply exceeds the local demand. A change in the price of faba bean is not a major concern, because it is a local crop mainly used for cattle and human consumption and has limited potential for export in national and international markets. Faba bean is not a major source of farm revenue, and if at some point, there is an oversupply and a drop in price occurs, faba bean can be replaced with some other high value crop. Any rise in the sale price of both cotton and faba bean would also compensate for increases in pumping costs and other MAR infrastructure costs.

### 5.3.5. Discount Rate and Lifespan of Projects

A 7% discount rate coincides with the current best guess and is highly likely to occur and is therefore of great concern. Discount rates of more than 7% will make MAR financially unattractive. As this may occur if the cost of borrowing capital is high, farmers may search for financing at lower rates or governments may assist farmers to set up special MAR grants or loans involving the least possible interest rates. Cross-over points for the lifespan of surface storage and basin infiltration almost coincide with the best guess (30 years) and are of great concern. The lifespan of basin infiltration can be enhanced by drying of basins, frequent scarping of accumulated silt layers and controlling weed growth.

### **6. Conclusions**

Break-even analysis of cross-over points is one way of assessing the financial performance of MAR under uncertainty. Cost-benefit analysis of surface storage and MAR helps to compare options in financial terms, but results cannot be relied upon completely without due consideration of uncertainty. Our approach to addressing uncertainty is to undertake a financial cost-benefit analysis by analysing a range of values for influencing variables and to establish thresholds (cross-over points) where financial returns from surface storage and MAR are equal. Once the thresholds are established, mitigation options can be identified and put in place to avoid variables reaching identified thresholds.

The analysis of cross-over points can be undertaken to identify minimum requirements under which MAR can be more advantageous than surface storage, and this was illustrated for the lower Namoi. For this catchment, MAR using basin infiltration can be financially superior to surface storage, but this depends on the selection of a suitable site where a high infiltration rate, low loss rates and other minimum requirements can be achieved. Further exploration of MAR through field trials and geo-physical investigation is suggested in areas of lower Namoi. MAR can be a potential option to achieve future water supply goals, provided that it is technically feasible and more financially viable than surface storage.

### **Acknowledgments**

This research was partly funded by the Australian Government's Endeavour Scholarships and Fellowships programme, the National Centre for Groundwater Research and Training (NCGRT), Australia. The authors thank Tony Jakeman and Michael Asher of Australian National University and two reviewers for their comments on this paper. We also thank Ejaz Qureshi of the Australian Centre of International Agriculture Research (ACIAR), Peter Dillon from the *Commonwealth Scientific and Industrial Research Organisation (*CSIRO) and David Pyne, an independent consultant and pioneer of ASR technology, for discussions of the topic in personal communications.

### **Author Contributions**

Muhammad Arshad is the lead author and contributed in developing the introduction, feasibility of MAR and the text and financial analysis in the case study section. Joseph Guillaume contributed to the text on methods, development of the code for consideration of uncertainty in the cost-benefit analysis of cross-over points. Joseph also produced major graphics for the paper. Andrew Ross contributed to the text on abstract, introduction, conclusion and editing the paper. The analysis of cross-over points was carried out jointly by Muhammad Arshad and Joseph Guillaume.

### **Conflicts of Interest**

The authors declare no conflict of interest.

### **References**

