Comparison of generic simulation models for water resource systems

Environmental Modelling & Software 40 (2013) 214e225

Contents lists available at SciVerse ScienceDirect

Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft

Comparison of generic simulation models for water resource systems Andrea Sulis*, Giovanni M. Sechi Hydraulic Sector, Land Engineering Department, University of Cagliari, 09123 Cagliari, Italy

a r t i c l e i n f o

a b s t r a c t

Article history: Received 12 December 2011 Received in revised form 9 May 2012 Accepted 7 September 2012 Available online 10 November 2012

In water resource systems that frequently experience severe droughts, generic simulation models can provide useful information for developing drought mitigation measures. This paper is about modeling in practice rather than in theory. The emphasis is on the application of generic simulation models to a multi-reservoir and multi-use water system in Southern Italy where frequent droughts over the last two decades have necessitated the use of temporary and unsustainable user-supply restrictions. In particular, AQUATOOL (Valencia Polytechnic University), MODSIM (Colorado State University), RIBASIM (DELTARES), WARGI-SIM (University of Cagliari) and WEAP (Stockholm Environmental Institute) models are considered in a preliminary analysis, which considers series and parallel simple schemes and also evaluates the possibility of alternative plans and operating policies in complex real water system. Each model has its own characteristics and uses different approaches to define resources releases from reservoirs and allocation to demand centers. The proposed model comparison and application does not identify in detail all the features of each model, rather it provides insights as to how these generic simulation models implement and evaluate different operating rules.
2012 Elsevier Ltd. All rights reserved.

Keywords: Decision support systems Water resources management Simulation Optimization

1. Introduction Generic simulation models provide information and insight that can help improve water system management and planning processes. In conditions of drought, simulation models provide an efficient way to reproduce sourceedemand interactions and to predict the impacts of rule modifications, over time and space. This helps set more appropriate drought mitigation measures. Appropriate measures can mitigate the economic, social and environmental impacts of drought. Currently, interventions are largely crisis driven. There is an urgent need (Rossi et al., 2007; Sechi and Sulis, 2009) for more risk-based management approaches to drought planning. Determination of appropriate drought mitigation measures is becoming the primary goal in managing water systems that frequently experience severe droughts. In this context, generic simulation models provide an efficient way to predict the effectiveness and efficiency of alternative mitigation measures. Frequently, generic simulation models are the core of complex decision support systems (DSS). The DSS can assist at different levels of detail, ranging from

* Corresponding author. Tel.: þ39 0706755303; fax: þ39 0706755310. E-mail address: [email protected] (A. Sulis). 1364-8152/$ e see front matter
2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.envsoft.2012.09.012

simple screening models for guiding data collection activities to more complex tools requiring high levels of expertise. These computer-based prediction models can be combined in a mixed optimizationesimulation approach to anticipate the occurrence of drought considering different hydrological scenarios (Pallottino et al., 2005; Sechi and Sulis, 2009). Despite the potential of using scenario optimization in the search for efficient alternatives, full integration between simulation and optimization has not yet been achieved, and real-world applications are frequently applications of generic simulation models. Despite the large number of simulation models available and the perceived value of those models with regard to inform water resource management authorities, there are many improvements that could be made to the work of planners, managers, modelers and analysts in this important area (Assaf et al., 2008). Two decades ago, Loucks (1992) and Simonovic (1992) described the gap between theory and practice in water resources planning and management, and still, models are often not adopted by the intended end users (McIntosh et al., 2011). All models produce simplified representations of real-world systems. What features are incorporated into the model depend in part on what the modelers believe is important. Improving the usefulness, as well as establishing trust and credibility, is important if the models are to be fully understood and accepted by the intended end users.

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

This paper is about modeling in practice rather than in theory. An extended state-of-the-art review on simulation and optimization modeling approaches in reservoir system operation problems is given by Rani and Moreira (2010). The main objective is to illustrate application performances of five generic models for simulating multi-reservoir and multi-use water resource systems: AQUATOOL-SimWin (referred to as AQUATOOL in this paper) (Valencia Polytechnic University) (Andreu et al., 1996), MODSIM (Colorado State University) (Labadie et al., 2000), RIBASIM (DELTARES) (Delft Hydraulics, 2006), WARGI-SIM (University of Cagliari) (Sechi and Sulis, 2009) and WEAP (Stockholm Environmental Institute) (SEI, 2005). Presented models have been applied in the 2009 release version. These models are representative of simulation models used for preliminary analysis of alternative plans and policies on water resources systems. These popular generic simulation models have been implemented world-wide in a large number of water systems. They incorporate most of the desirable attributes of a simulation model. After a short presentation and comparison of the main characteristics and features of each simulation model, we emphasize the application of these simulation models to single-purpose reservoirs in series and in parallel, as well as to a complex multi-reservoir and multi-use real water system in Southern Italy, where frequent droughts over the last two decades have necessitated adoption of temporary and unsustainable user-supply restrictions.

215

operating policies. Operating policies in AQUATOOL, RIBASIM and WARGI-SIM are fixed, whereas operating policies in MODSIM and WEAP are defined as a combination of system states and hydrologic conditions. The most recent version of MODSIM is developed under the MS .NET Framework that allows users to customize MODSIM for specialized operating rules without having to modify the original source code. While these generic simulation models vary with regards to the type and details of the operating policies that they can reproduce, they all include the concepts of priorities and preferences. Each of the five models has a built-in capacity for water quality modeling, but most water quality modeling components and algorithms are relatively simple compared to the state of the art in water quality modeling. In addition to this capacity, MODSIM and WEAP can be linked to a more detailed higher dimensional model (e.g., the US EPA QUAL2E modeling framework, Brown and Barnwell, 1987) to provide highly detailed and comprehensive modeling of water quantity and quality conditions in the system. MODSIM and WEAP can also be linked with the MODFLOW model (Harbaugh et al., 2000), a three dimensional finite difference groundwater model, to study how changes in groundwater levels affect the overall system and vice versa. However, this tight coupling between generic simulation models and MODFLOW is not an easy task because it requires an extensive calibration phase. In AQUATOOL, the user can choose among a spectrum of models to represent groundwater realistically, ranging from a model of reservoir type to a distributed model of a heterogeneous aquifer of irregular shape.

2. Model characteristics and comparison 3. Model features The five generic simulation models considered in this paper were developed within interactive graphics-based interfaces by public and private organizations. They are all designed to study water related planning and management issues in water systems and to satisfy the needs of those at different levels of the planning and decision-making process (Assaf et al., 2008). Each model has its own special characteristics. However, a feature makes the main difference: AQUATOOL, MODSIM and WEAP apply optimization methods on a single time-period of the simulation, and the results are used as an efficient mechanism for performing the simulation of a single period of water allocation in the system, whereas RIBASIM and WARGI-SIM are simulation-only models based on a more conventional if-then approach. Technically speaking, in MODSIM the flow allocation problem is modeled using a minimum cost flow modeling approach in a simplified way. In WEAP, a standard linear program is used to solve the water allocation problem. This allows the model to consider more complex physical, hydrological, and institutional constraints than the min-cost flow approach. In AQUATOOL, the simulation and management of the surface system are made simultaneously by solving a conservative flow network optimization problem and trying to maximize several objectives. The application of simulation-only models, such as RIBASIM and WARGI-SIM, to complex water systems could enable lower performance system indexes (e.g., vulnerability or reliability at user-defined water supply levels). However, these simulation-only models can better reproduce the operating policies used by water authorities in the resource management of real systems. There is a large variety of operating policies presented in the literature. For single-reservoir systems, operating policies can precisely define how much water to release from the reservoir for all possible combinations of hydrologic and reservoir storage conditions. For multi-reservoir and multi-use systems supply preference and demand priority are frequently included in the

3.1. AQUATOOL 3.1.1. Description AQUATOOL is a generalized DSS developed at the Universidad Politécnica de Valencia (UPV), Valencia, Spain. The model was designed for the operational management and planning stages of decision-making in complex basins comprising multiple reservoirs, aquifers and demand centers. Implemented within the Microsoft Windows Environment, AQUATOOL has been coded in different programming languages, such as Cþþ, Visual Basic and FORTRAN. 3.1.2. Appropriate use The DSS has been upgraded and expanded. It currently consists of several modules including a simulation module (SimWin), a management module for a water resource system that considers the risk of drought (SimRisk) and is based in SimWin, an optimization module with a monthly passage of time (OptiWin) that is more detailed than SimWin, and a simulation module of groundwater that uses the eigenvalues method (AquiVal) to simulate groundwater distribution. The simulation in SimWin is made on a monthly basis, and it allows nonlinear processes, such as evaporation and infiltration, to be adequately shaped. SimWin distinguishes five types of oriented connections that allow the user to reproduce the losses of water, hydraulic connections between nodes, reservoirs and aquifers and flow limitations based on elevation. The optimization of the flow network attempts to minimize several target functions on reservoirs, demands and rivers subjected to the restrictions of mass conservation and to physical capacities. 3.1.3. Training required To effectively use all of the SimWin features, a high skill level and experience in resource modeling is required.

216

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

3.1.4. Documentation Some documentation is available through the UPV website (http://www.upv.es/aquatool/). 3.1.5. Cost The user should contact the UPV Group to inquire about the cost of a license.

3.2. MODSIM 3.2.1. Description MODSIM is a generic system management DSS that was originally conceived in the late 1970s at Colorado State University (CSU), U.S., and has been continuously maintained. MODSIM was developed in the .NET Framework that provides a powerful environment for customization without requiring recoding. MODSIM simulates water allocation in the system at each time step through a sequential solution of a network flow optimization problem where nonlinearities (i.e., evaporation, groundwater return flows, channel losses) are assessed using a successive approximations solution procedure. The problem is solved with the Lagrangian relaxation algorithm, RELAX-IV (Bertsekas and Tseng, 1994). Reservoir balancing routines that allow division of reservoir storage into several operational zones can be used to control the spatial distribution of available reservoir storage. Additionally, operating rules on reservoir regulation and demand allocation can be conditioned based on user defined hydrologic state variables. 3.2.2. Appropriate use MODSIM has been linked with MODFLOW for the analysis of conjunctive use of groundwater and surface resources, as well as to QUAL2E for assessing the effectiveness of pollution control strategies. MODSIM can be applied in an implicit stochastic optimization framework where optimal rules for integrated operation are obtained using the generalized dynamic programming software package, CSUDP. 3.2.3. Training required The use of the main module requires moderate training, whereas external modules are quite hard to use without prior modeling skills. 3.2.4. Documentation Detailed documentation is available through the CSU website (http://modsim.engr.colostate.edu/). 3.2.5. Cost MODSIM can be downloaded free through the CSU website.

3.3. RIBASIM 3.3.1. Description RIBASIM is a generic model package for simulating the behavior of river basins under various hydrological conditions. It was developed by DELTARES, formerly the DELFT Institute, Delft, The Netherlands. Different scenarios can be easily compared based on user-defined objectives through the powerful graphical interface. The analysis of water demand is extensive (i.e., based on demographic, economic, crop water requirements aspects), and the current and future demands at different horizons can be compared. Crop production and crop damage due to water shortages can be

assessed. RIBASIM provides fixed operating rules based on target storage volumes and multiple zoning. 3.3.2. Appropriate use RIBASIM particularly address the hydrological and hydrographical description of the river-basins and links the hydrological water inputs at various locations with the specific water-users in the supply system. It allows the user to define operating/planning scenarios where each scenario is characterized by a particular operating rule and/or water supply projection. 3.3.3. Training required While RIBASIM is intuitive and easy to use, it requires significant data to perform detailed analysis. 3.3.4. Documentation Documentation can be obtained from DELTARES (http://www. wldelft.nl/soft/ribasim). 3.3.5. Cost Information on the license cost can be obtained from DELTARES (http://www.wldelft.nl/soft/ribasim). 3.4. WARGI-SIM 3.4.1. Description WARGI is a user-friendly tool specifically developed to help users understand the interrelationships between demands and resources for multi-reservoir water systems under drought conditions, such as those that frequently occur in Mediterranean regions. Since the mid-1990s, WARGI has been extended and new modules have been developed by the Water Research Group (WRG) at the Department of Land Engineering, University of Cagliari, Italy. The WARGI modeling capability includes several interrelated macromodules; the main ones are a simulation-only module (WARGISIM), a deterministic optimization module (WARGI-OPT), the reservoir quality optimization module (WARGI-QUAL), and a module of scenario optimization (WARGI-SCEN). The water allocation in WARGI-SIM is simulated using user-defined preferences and priorities. Additionally, the user can define reserved volumes as a fixed function of the period of the year, and the withdrawn water from the reserved zone is decreased to satisfy user-selected high priority demands. 3.4.2. Appropriate use Improvements to the definition of drought mitigation measures and effective linking of these measures with drought indicators are achieved through a full integration of WARGI-SIM and WARGI-OPT. The requirement of massive simulation-optimization runs for the analysis of complex water system under drought conditions are satisfied in a GRID environment. 3.4.3. Training required WARGI-SIM is a relatively simple model that enables nonexperts to understand the main issues and problems of complex water systems. 3.4.4. Documentation Requests for a detailed documentation can be addressed to the authors. 3.4.5. Cost Requests for a non-commercial license can be addressed to the authors.

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

3.5. WEAP

217

4. Comparison of application to single-purpose reservoirs in series and in parallel

3.5.1. Description WEAP is a generic simulation model developed at the Stockholm Environment Institute, Boston, Massachusetts. WEAP model simulations are constructed as a set of scenarios with different simulation time steps. These demand scenarios are applied deterministically to a linear programming allocation algorithm where each demand and source is assigned a user-defined priority. The linear program solves the water allocation problem by trying to maximize satisfaction of demand subject to supply preferences and demand priorities while using reservoir operating policies to minimize the distance to ideal conditions. The water allocation problem is solved at each time step using an iterative, computationally expensive approach. Traditional target storage levels, multiple zones, and reduced releases by a buffer coefficient are implemented in WEAP. Supply balancing within demand centers with the same priority is assured by that approach. A groundwater module in WEAP allows for water transfer between the stream and the aquifer. 3.5.2. Appropriate use The model integrates some physical hydrological processes with the management of demands and infrastructure to allow for multiple scenario analyses, including alternative climate scenarios and changing anthropogenic stressors. The primary aim of the water management analysis in WEAP is the analysis of water demand configuration. 3.5.3. Training required WEAP requires significant data and a moderate amount of experience for a detailed analysis. 3.5.4. Documentation Detailed documentation is available online at the SEI website (http://www.weap21.org). 3.5.5. Cost The cost of WEAP is US $1000 for universities and governments.

In multipurpose multi-reservoir systems there are sometimes conflicting and sometimes complementary multiple purposes served by the water stored in and released from reservoirs. Operating policies define what should be performed for any combination of system states and hydrologic conditions to minimize any necessary deviation from ideal conditions in those systems. Common operating rules for single-purpose reservoirs in series and in parallel can be derived from principles of mathematical optimization and can be supported by system manager experiences. Multipurpose multi-reservoir systems can be seen as an aggregation of reservoirs in series and in parallel (in Fig. 1 indicated with (a) and (b), respectively). Many of the operating rules for multipurpose systems require combinations of operating rules for single-purpose reservoirs in series and in parallel. This is not always the case (Needham, 1998). Nevertheless, the operation of single-purpose reservoirs in series and in parallel is classic in the literature. In addition, it is remarkably easier to analyze generic simulation model results using reservoirs in series and/or parallel schemes. The primary reviews of operating rules are Sheer (1986), Loucks and Sigvaldason (1982), and Lund and Guzman (1999). For reservoirs in series providing water supply, a reasonable objective is to maximize the amount of the water available and the resulting rule is to deplete the downstream reservoir before the upstream reservoir is used to meet downstream demands (Sheer, 1986). A conceptual rule for reservoirs in parallel involves drawing in tandem from each reservoir in a manner that equalizes the probability of reservoir filling for each reservoir (Loucks and Sigvaldason, 1982). This procedure ensures the minimization of expected water wastage. According to Loucks and Sigvaldason (1982), operating rules for reservoirs in series and in parallel may include one or more of the following components: 1. Target storage levels or volumes; 2. Multiple zoning; and 3. Conditional rule curves.

Fig. 1. Configurations of reservoirs in series and in parallel.

218

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

When a prescription of the desired storage volumes or levels in each reservoir is introduced in the system operation, reservoir operators are expected to maintain these levels as closely as possible while generally trying to satisfy downstream demands. Multiple zoning defines five storage allocation zones in the total reservoir capacity: the conservation zone (from which requests are normally satisfied), the flood control zone (for storing large unexpected inflows), the spill zone (associated with actual flood damage), the buffer zone (beneath the conservation zone used to satisfy only high priority water demands during dry periods) and the inactive zone (the “dead” zone where withdrawals may not be possible). The conditional rule curve defines reservoir releases as a function of the existing storage volumes and the expected natural inflows for some future months. The well-known SQ type linear decision rule, originally proposed by Loucks (1970), as applied to reservoir j at time step t is as follows:

Rj;t ¼ aj;t Vj;t þ bj;t

tþm X

Ij;i þ cj;t

(1)

i¼t

where Rj,t is the release from the reservoir, Vj,t is the stored volume in the reservoir, Ij,i is the expected natural inflow during the month i within the estimating (m  1) month periods, and aj,t, bj,t and cj,t are coefficient to be assessed. Bhaskar and Whitlatch (1980) tested linear and nonlinear monthly operating rules by regressing the series of releases with reservoir storage at the beginning of each month and inflow at the current month. They found that the linear operating rules are as good as or better than the more complex rules in many cases. All the examined simulation models permit the definition of monthly target storage levels or volumes and multiple zones, eventually further dividing each zone into a user-defined number of subzones (e.g., in MODSIM), through tables and figures. Only three models (MODSIM, WARGI-SIM and WEAP) implement conditional rules, including the SQ type linear decision rule (1). In particular, WEAP uses trial-and-error iterative procedures, which can be time consuming and problematic, or offline multiple regression models to assess the aj,t, bj,t and cj,t coefficients of (1). As an alternative to this simulation-only approach, MODSIM uses the generalized dynamic programming software CSUDP (Labadie, 2003) and WARGI-SIM uses the linear programming module WARGI-OPT (Sechi and Sulis, 2009) that can be applied in a GRID computing environment (Sulis, 2009). Models were applied to single-purpose reservoirs in series and in parallel over a 52-year time horizon with a monthly time step. Fig. 1 shows the two types of system configurations. The main reservoir variables are shown in Table 1. In both configurations, reservoirs 1 and 2 have a total capacity of 260 and 320 106 m3, respectively. Statistical parameters show that the hydrological inputs to the reservoirs were selected to be significantly different to highlight the adequacy of adopted operating rules. The initial reservoir volume was equal to half of the total capacity in both reservoirs. An urban water demand with an annual request of 115.7 106 m3 and a uniform monthly program was modeled. The demand site was connected to the reservoirs with the same supply

preference (MODSIM, RIBASIM and WEAP), transfer cost (AQUATOOL) or weight (WARGI-SIM). Reservoirs in parallel, P1 and P2, had the same priority for filling and no additional components of the operating rule were adopted. Although WARGI-SIM uses a traditional simulation algorithm and WEAP uses a linear optimization module to allocate water in the system, both models correctly reproduced the operating rule by drawing in tandem from each reservoir. As expected, WARGI-SIM and WEAP minimize the spilled water (1.86 106 m3/ month in Table 2), which is the same as maximizing the amount of water available. Fig. 2a and b show that AQUATOOL, MODSIM and RIBASIM discharge water from reservoir P1 first. This reservoir was arbitrarily selected among two reservoirs, having the same priority. In particular, P2 is the reservoir with the larger potential inflow per unit storage volume capacity. Unnecessary spilling occurs in this reservoir (þ17%). In MODSIM, releases from the reservoir in the system follow the order of insertion of arcs outgoing the reservoir nodes. To release in tandem from reservoir P1 and P2, the multiple zoning component was applied to the operating rule. Specifically, in MODSIM, the total capacity in each reservoir must be divided into many subzones (100 in our application) so that water was discharged from reservoir P2 only in the case of having the storage level of reservoir P1 within the same subzone. On the other hand, AQUATOOL and RIBASIM defined three allocation zones (conservation, buffer and inactive zones) within the total capacity of reservoir P1 and P2. As expected, the storage volume time series and total unnecessary spilling simulated by MODSIM are very close to the time series provided by WARGI-SIM and WEAP, whereas the probabilities of reservoir filling for reservoir P1 and P2 remain significantly different in AQUATOOL and RIBASIM. For reservoirs in series, the upstream reservoir S1 has a higher priority of filling than the downstream reservoir S2. No additional components of the operating rule were adopted. As shown in Fig. 3, AQUATOOL, MODSIM, WARGI-SIM and WEAP exactly reproduced the common operating rule for such systems and S2 was depleted before the water in S1 was discharged to meet the downstream urban demand, whereas RIBASIM considers the downstream reservoir S2 as a demand to be supplied by the upstream reservoir S1.

5. Comparison of applications to a multi-reservoir and multipurpose water system To verify the potential for using generic simulation models to define drought mitigation measures, the Agri-Sinni water system (Southern Italy) was considered (Fig. 4). The Agri-Sinni is a multireservoir and multipurpose system located in the Basilicata region. It supplies water to the Puglia and Calabria regions as well. The main reservoirs in the system are Monte Cotugno (capacity of 556 106 m3) and Pertusillo (capacity of 159 106 m3) along the Sinni and the Agri rivers, respectively. Marsico Nuovo and Cogliandrino are single purpose reservoirs (for irrigation and hydroelectric use, respectively) with small total capacities. ENEL, the largest Italian power company, has the authority to manage the Cogliandrino

Table 1 Reservoir data input. Reservoir

Mean inflow (106 m3/year)

Maximum inflow (106 m3/year)

Minimum inflow (106 m3/year)

Initial volume (106 m3)

Maximum capacity (106 m3)

1 2

126.5 15.9

318.6 53.6

24.1 1.7

131 160

260 320

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225 Table 2 Monthly mean volume of spilling for reservoirs in parallel.

Spilling (106 m3/month)

AQUATOOL, MODSIM and RIBASIM

WARGI-SIM and WEAP

2.17

1.86

reservoir exclusively for hydropower production. The reservoir is operated independently. Four intake structures (Agri, Sarmento, Sauro, and Gannano) were constructed on the main rivers for water diversion. Based on the observed monthly inflows at Monte Cotugno and Pertusillo over the period of 1983e2005, the inflows in other sections of interest in the basin were generated using the distributed runoff model DREAM (Manfreda et al., 2005). The inflow series accurately reproduced the severe water scarcities in the Agri-Sinni system that occurred in the years 1989e1990 and 2001e2002. Table 3 shows the main statistical properties of the inflow series. Urban (Lucano Aqueduct and AQP in Fig. 4), industrial (ILVA), and agricultural (C.B.) demands are 295.8 106 m3/yr, 12.6 106 m3/yr, and 240 106 m3/yr, respectively.

219

Demand nodes in the graph, which were common to all simulation models, represent aggregations of several homogenous urban, industrial or agricultural water requests. The allocation order to follow when supplying those demands was based on the assignment of a decreasing priority Pi (increasing the priority number) to urban (P1 ¼ 1), industrial (P2 ¼ 2) and agricultural (P3 ¼ 3) demands. While in AQUATOOL all demand nodes connected with sources were supplied by those sources, MODSIM, RIBASIM, WARGI-SIM and WEAP required that each demand node had a hierarchical list of sources from which a supply flow could be activated. These lists of sources and priorities for filling reservoirs were established according to the information provided by the system water agency. This system configuration was set up in an Italian National Project (PRIN: “Decision Support Model in complex system management under shortage conditions”). 5.1. Model applications to the basic system configuration In a first application of the simulation models, the conservation and inactive zones were defined in the reservoirs while no further components of reservoir operating rules were introduced. The

Fig. 2. Storage volume time series for reservoirs in parallel.

220

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

Fig. 3. Storage volume time series for reservoirs in series.

results of the first phase simulation help us highlight heavy restrictions operated during shortages and the resulting benefits of management rules that will be introduced in the following paragraph. The release time series simulated by the five models for the Monte Cotugno reservoir (Fig. 5) present similar trends, and they closely reproduce the historical values. In particular, all models simulated the severe restriction of total releases that occurred during the most serious drought in years 2001e2002 when releases were reduced by 55%, with a small exception of RIBASIM that gave a higher reduction of 16% with respect to the historical value. Table 4 shows the mean annual volume of spilling from the system evaluated as the sum of spilling from the Monte Cotugno reservoir and the available resources not diverted at the Gannano intake. As reported in Table 4, the annual mean of spilling is minimized by WARGI-SIM and WEAP, whose values are very close, whereas higher values were obtained by AQUATOOL (þ3.9%), and, more significantly, by MODSIM (þ7.4%) and RIBASIM (þ8.6%). The results previously described for simple systems in series and parallel could partially explain these differences; nevertheless, a more in-depth analysis of diversion utilization and spilling resulting from WARGI-SIM and WEAP was needed.

Fig. 6 shows the simulated time series of monthly spilling at Monte Cotugno and Gannano. WARGI-SIM and WEAP provided similar results with the exception of Monte Cotugno reservoir filling condition (Fig. 7). The occurrence of high flows and flood events highlight the use of different techniques for reproducing operating rules for reservoir filling. Specifically: 1. The simulation-alone algorithm in WARGI-SIM requires a fixed list of reservoirs (i.e., Monte Cotugno) for each source (i.e., Agri, Sauro and Gannano) that can operatively receive the unused resources in all reservoir conditions; 2. The optimization model in WEAP has considerable flexibility for deciding whether incoming withdraw resources at the Agri and Sauro intake structures should be released downstream. Consequently, after having supplied the C.B. Bradano and Metaponto irrigation demand, the WARGI-SIM and WEAP models use different strategies to transfer exceeding resources. The simulation-alone algorithm in WARGI-SIM transfers further available resources from the Gannano intake structure to the Monte Cotugno reservoir even when the Monte Cotugno reservoir is in a filling condition. The optimization model in WEAP releases

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

221

Fig. 4. The Agri-Sinni water system in the WARGI-SIM graphical interface (satellite images courtesy of Google Earth).

downstream the Gannano structure the available resource at Monte Cotugno filling condition. In this study, the water system performance evaluation analysis in the form first suggested by Hashimoto et al. (1982) and recently revised by Sandoval-Solis et al. (2011) was used to compare the supply performances of the models. Selected criteria can capture the main differences of how generic simulation models work more than system-oriented criteria that could describe the behavior of the system in depth but, in some cases, not offering interesting data to compare those models. The performance criteria, CP,l,i were defined for the aggregated water use i (i ¼ 1,2,3 related to urban, industrial and agricultural uses, respectively) when unsatisfactory values were unable to provide

a predefined demand level l (l ¼ 1.0, l ¼ 0.75 and l ¼ 0.5) of the aggregated water demand Dt,i. The time series of monthly simulated values of supplies for each aggregated use, Xt,i, had its own range of satisfactory, Sl,i, and unsatisfactory, Ul,i, values for each threshold demand Dt,l,i ¼ l$Dt,i:

if Xt;i  Dt;l;i

then Xt;i ˛Sl;i else Xt;i ˛Ul;i

PT Table 3 Statistical indexes of inflows in the period 1983e2005. Mean (106 m3/year)

Stand. Dev. (106 m3/year)

Max (106 m3/year)

Min (106 m3/year)

Pertusillo Monte Cotugno Cogliandrino Marsico Nuovo Gannano Agri Sauro Sarmento

212.15 277.60 89.76 7.82 105.54 115.54 50.46 84.10

57.72 106.61 32.12 3.04 88.56 64.43 25.50 38.79

328.54 494.14 147.13 12.91 389.03 241.55 101.31 162.06

118.25 118.45 33.95 2.53 11.72 17.92 11.93 26.42

(2)

The Nt periods of successive unsatisfactory Xt,i for the criterion (2) were then evaluated in the total time length, T, and reliability and vulnerability indices were defined as:

Reliability : CR;l;i ¼

Stations

and Zt;l;i ¼ 1 and Zt;l;i ¼ 0

Vulnerability : CV;l;i

t¼1

Zt;l;i

T 9 8 Dt;l;i  Xt;i > > > > > > = > Nt > > > > ; : t˛T

(3)

(4)

Reliability and vulnerability were calculated in the 1983e2005 simulation period (T ¼ 264 months). As expected, the RIBASIM model, using allocation rules based only on a hierarchical order of demand priorities, obtained higher vulnerabilities to those

222

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

WARGI-SIM RIBASIM

Historical AQUATOOL

400

MODSIM WEAP

6

3

Supply [10 m ]

300

200

100

2003-2004

2002-2003

2001-2002

2000-2001

1999-2000

1998-1999

1997-1998

1996-1997

1995-1996

1994-1995

1993-1994

0

[Year] Fig. 5. Simulated and historical releases from Monte Cotugno reservoir in the period 1993e2004.

demands with lower priorities (Table 5). The absence of any reservoir release rule determined CV,1.0,3 (irrigation vulnerability) higher than other uses in all model simulations. The urban vulnerability values CV,1.0,1 obtained by models having a network flow optimization algorithm engine (MODSIM and AQUATOOL) were the same. These values were very close to the results of WARGI-SIM, which uses a simulation-alone algorithm for reproducing allocation rules based on priorities and preferences, ranging from 0.42 to 0.46. WEAP using a linear programming model gave a higher vulnerability value (CV,1.0,1 ¼ 0.54), which was significantly lower than RIBASIM (CV,1.0,1 ¼ 0.78). Despite the similarities in the vulnerability values, CV,1.0,i of AQUATOOL, MODSIM and WARGI-SIM, those models used different techniques for reproducing allocation rules in the Agri-Sinni system. Specifically: 1. The management optimization of surface water in the system was made once every month in MODSIM and AQUATOOL. Deficit costs in the objective function were calculated from demand priorities (P1 ¼ 1, P2 ¼ 2 and P3 ¼ 3); 2. In WARGI-SIM, the simulation process allocates the available resources by first considering only urban demands (P1 ¼ 1), then the ILVA industrial demand (P2 ¼ 2) and finally irrigation demands (P3 ¼ 3), following the hierarchical order of demand priorities. MODSIM supplied the different demands having the same priority from each reservoir zone simply by following the order of node insertion in the system graph, and an equitably share of water shortage among those demands during drought events was then

Table 4 Annual mean volume of spilling.

assured only by introducing a really high number (i.e.: 100) of multiple subzones in the conservation storage zones of Monte Cotugno and Pertusillo reservoirs. Reliability was then briefly assessed from WARGI-SIM and WEAP results. Table 6 shows that urban reliability values CR,1.0,1 and CR,0.75,1 were lower than industrial values CR,1.0,2 and CR,0.75,2 for WARGI-SIM and WEAP results. This was unexpected given the priorities attached to the demand sites. This is because the ILVA industrial demand could be supplied by several sources of the AgriSinni, whereas one of the urban demands (AQP) has only the Pertusillo reservoir as source. As expected, reliability values for urban CR,1.0,1 and CR,0.75,1 and industrial CR,1.0,2 and CR,0.75,2 in WARGI-SIM were lower than in WEAP and equal or higher for CR,0.5,i. This comparison shows that WARGI-SIM reduced the risk and cost of large shortages, at a cost of more frequent small shortages compared to WEAP that uses a linear objective function weighted on deficit costs. Finally, reliability values for agricultural use CR,l,3 were equal in WARGI-SIM and WEAP for all l levels, the agricultural demands being supplied after demands with higher priorities (Pi < 3).

5.2. Second model application No operating rules were implemented in the first generic application of the simulation models; therefore all models reproduced the unsustainable condition of urban use during the two severe drought events in the Agri-Sinni system (1988e1990 and 2001e2002). The results highlight that these two 2-year droughts accounted for more than 80% of the total water shortage over the total simulation period. A second application of the models introduced operating rules to reduce the scarcity impacts. According to the system water agency, the objective was to minimize the urban and industrial vulnerabilities with respect to a single month (T ¼ 1 in (4)) and the entire demand level (l ¼ 1):

Water loss (106 m3/year) AQUATOOL MODSIM RIBASIM WARGI-SIM WEAP

93.99 97.22 98.26 90.49 90.69

CVðT ¼ 1Þ;1;i

9 8 Dt;l;i  Xt;i > > > > > > = < Dt;i ¼ max > > Nt > > > > ; :

(5)

Fig. 6. Simulated spilling volumes at the Monte Cotugno reservoir (a) and Gannano intake structure (b).

RIBASIM

WARGI-SIM

MODSIM

WEAP

AQUATOOL

500

300

6

3

[10 m ]

400

200

100

0 Oct-83

Oct-86

Oct-89

Oct-92

Oct-95

Oct-98

[time] Fig. 7. Simulated storage volumes at Monte Cotugno reservoir.

Oct-01

Oct-04

224

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

Table 5 Vulnerability values for different uses.

AQUATOOL MODSIM RIBASIM WARGI-SIM WEAP

CV,1.0,1

CV,1.0,2

CV,1.0,3

0.42 0.42 0.78 0.46 0.54

0.67 0.67 1.00 0.70 0.80

0.98 0.98 0.81 1.00 1.00

Table 6 Monthly reliability for different uses in WARGI-SIM and WEAP. Urban

CR,1.0,i CR,0.75,i CR,0.5,i

Industrial

Agricultural

WARGI-SIM

WEAP

WARGI-SIM

WEAP

WARGI-SIM

WEAP

91.67 94.32 97.73

92.05 95.08 97.73

94.70 95.08 95.45

96.21 96.21 96.21

90.53 91.67 93.56

90.53 91.67 93.56

A hedging rule was introduced in all software to reduce agricultural releases to save water for higher priority uses in the following periods. Each simulation model has its own way to reproduce this reservoir operating rule at Pertusillo and Monte Cotugno. Briefly: 1. In AQUATOOL and RIBASIM at each arc entering in a demand node we associated a monthly alarm indicator containing

Table 7 Vulnerabilities values at single month for different uses, considering the reservoir hedging rules. CV(T

CV(T

¼ 1),1.0,1

0.16 0.12 0.57 0.28 0.39

¼ 1),1.0,2

1.00 1.00 1.00 0.00 0.85

RIBASIM

CV(T

¼ 1),1.0,3

1.00 0.98 0.79 1.00 0.91

WARGI-SIM

In some cases, procedures were extremely sensitive and computationally expensive. Table 7 summarizes results for CV(T ¼ 1),1.0,i values. In these results, only WARGI-SIM minimized both urban and industrial demands, whereas the hedging rule in RIBASIM and WEAP did not reduce significantly the urban and industrial vulnerabilities. AQUATOOL and MODSIM could only save water for urban demands. In particular, AQUATOOL, MODSIM, RIBASIM and WEAP showed CV(T ¼ 1),1.0,2 higher than 0.85. It should be noted that the total releases in AQUATOOL, MODSIM and WEAP were reduced and the reduced supplies were allocated to the demands according to their priorities; in other words, the procedure did not allow for water saving to decrease CV(T ¼ 1),1.0,2. Finally, the application of the proposed hedging rule in RIBASIM did not efficiently restrict supply to agricultural uses and CV(T ¼ 1),1.0,3 was significantly lower than that obtained using other simulation models (CV(T ¼ 1),1.0,3 ¼ 0.79). Fig. 8 shows the simulated time series of monthly storage volumes at Monte Cotugno when releases were temporarily decreased by the hedging rule during the severe drought events. As expected, the time series of all models are close with significant exceptions during drought events in 1988e1990, 1992e1993 and

MODSIM

WEAP

AQUATOOL

500

400

300

6

3

[10 m ]

AQUATOOL MODSIM RIBASIM WARGI-SIM WEAP

a restriction coefficient for the agricultural use and a target volume of half of the conservation volume in both reservoirs as a trigger for the restriction rule; 2. In WARGI-SIM when the storage volume in a reservoir (Pertusillo or Monte Cotugno) was within a reserved volume that was equal to half of the conservation volume, releases were decreased to supply only the urban demands and the ILVA industrial demand; 3. In MODSIM and WEAP, a conditional rule curve was introduced that defined reservoir releases from Monte Cotugno and Pertusillo as a function of existing storage volume (WEAP), and a function of existing storage volume and inflow into the reservoir (MODSIM), when the storage volume was within the buffer volume, which was equal to half of the conservation zone. The coefficients of these linear functions for each reservoir were obtained using a trial-and-error procedure.

200

100

0 Oct-83

Oct-86

Oct-89

Oct-92

Oct-95

Oct-98

Oct-01

[Time] Fig. 8. Simulated storage volumes at Monte Cotugno reservoir operated through a hedging rule.

Oct-04

A. Sulis, G.M. Sechi / Environmental Modelling & Software 40 (2013) 214e225

2001e2002. In those cases, the storage volume trend in Monte Cotugno emphasized the use of different techniques for reproducing the hedging rule that reduces reservoir releases when storage volume is within the buffer zone. In particular, the most conservative rules were those implemented in AQUATOOL and RIBASIM, which use static hedging coefficients, whereas the release reduction in MODSIM and WEAP were adaptive. This could justify the fact, noticeable in Fig. 8, that reservoir storage remains considerably high even during a drought when considering AQUATOOL values. 6. Conclusions Five popular generic simulation models were synthetically illustrated in this paper and their features were compared via application to a complex water system in Southern Italy. While RIBASIM and WARGI-SIM use simulation-only algorithms in a traditional if-then approach, AQUATOOL, MODSIM and WEAP additionally employ optimization methods for the single period as an efficient tool for performing simulations. Demand priority is a common concept in all simulation models. AQUATOOL, MODSIM and WEAP extend this concept to reservoir filling. Each model has its own way to reproduce operating rules in the reservoir. A traditional hedging rule was implemented to reduce releases and save water for high priority demands during the drought periods. The results demonstrated that simulation-only models, such as WARGISIM, can provide system performance criteria values close to or higher than simulation models aided by optimization models. They also can facilitate greater adherence to water allocation in real world management of water systems. Further research is needed to estimate and evaluate water management of the system based on indexes that represent aggregate measures of performance criteria, such as environmental sustainability or drought risk indexes. Nevertheless, the second model application, which introduced operating rules, highlights the potential use of a simulation approach to reduce the environmental impacts of drought. This approach could also be used to estimate drought mitigation measures under different hydrologic conditions, considering the alternation of the hydrological scenarios due to climate change. Acknowledgments This research was developed with the financial support of the Italian Ministry of Education, University and Research, under the PRIN 2005 Project, “Decision Support Model in complex system management under shortage conditions” and the Regione Autonoma della Sardegna, under the PO Sardegna FSE 2007e2013 (L.R.7/ 2007 e Promozione della Ricerca Scientifica e dell’Innovazione Tecnologica in Sardegna). References Andreu, J., Capilla, J., Sanchìs, E., 1996. AQUATOOL, a generalised decision-support system for water-resources planning and operational management. Journal of Hydrology 177 (3e4), 269e291. Assaf, H., van Beek, E., Borden, C., Gijsbers, P., Jolma, A., Kaden, S., Kaltofen, M., Labadie, J.W., Loucks, D.P., Quinn, N.W., Sieber, J., Sulis, A., Werick, W.J., Wood, D.M., 2008. Generic simulation models for facilitating stakeholder involvement in water resources planning and management: a comparison, evaluation, and

225

identification of future needs in environmental modelling, software and decision support (3): the state of the art and new perspective. In: Jakeman, A., Voinov, A., Rizzoli, A.E., Chen, S. (Eds.), IDEA Book Series. Elsevier, pp. 229e246. Bertsekas, D., Tseng, P., 1994. RELAX-IV: a Faster Version of the RELAX Code for Solving Minimum Cost Flow Problems. Completion Report Under NSF Grant CCR-9103804. Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA. Bhaskar, N., Whitlatch, E., 1980. Derivation of monthly reservoir release policies. Water Resources Research 16 (6), 987e993. Brown, C.L., Barnwell Jr., T.O., 1987. The Enhanced Stream Water Quality Models QUAL2E and QUAL2E-UNCAS Documentation and User Manual. U.S. Environmental Protection Agency, Environmental Research Laboratory, Athens, Georgia, EPA/600/3e85/040, 455 pp. Delft Hydraulics, River Basin Planning and Management Simulation Program, 2006. In: Voinov, A., Jakeman, A.J., Rizzoli, A.E. (Eds.), Proceedings of the IEMSs Third Biennial Meeting: “Summit on Environmental Modelling and Software”. International Environmental Modelling and Software Society, Burlington, Vermont, USA. Harbaugh, A.W., Banta, E.R., Hill, M.C., McDonald, M.G., 2000. MODFLOW-2000, the U.S. Geological Survey Modular Ground-water Model e User Guide to Modularization Concepts and the Ground-water Flow Process. U.S. Geological Survey Open-file Report 00-92, 121 pp. Hashimoto, T., Stedinger, J.R., Loucks, D.P., 1982. Reliability, resiliency, and vulnerability criteria for water resource system performance evaluation. Water Resources Research 18 (1), 14e20. Labadie, J.W., Baldo, M.L., Larson, R., 2000. MODSIM: Decision Support System for River Basin Management: Documentation and User Manual. Colorado State University and U.S. Bureau of Reclamation, Ft Collins, CO. Labadie, J., 2003. Generalized Dynamic Programming Package: CSUDP, Documentation and User Guide. Department of Civil Engineering, Colorado State University, Ft. Collins, CO. Loucks, D.P., 1970. Some comments on linear decision rules and chance constraints. Water Resources Research 6 (2), 668e671. Loucks, D.P., 1992. Water resource systems models: their role in planning. Journal of Water Resources Planning and Management 118 (3), 214e223. Loucks, D.P., Sigvaldason, O.T., 1982. The operation of multiple reservoir systems. In: Kaczmarek, Z., Kindler, J. (Eds.), Multiple-reservoir Operation in North America. International Institute for Applied Systems Analysis. Laxemburg, Austria. Lund, J., Guzman, J., 1999. Derived operating rules for reservoir in series or in parallel. Journal of Water Resources Planning and Management 125 (3), 143e153. Manfreda, S., Fiorentino, M., Iacobellis, V., 2005. DREAM: a distributed model for runoff, evapotranspiration, and antecedent soil moisture simulation. Advances in Geosciences Vol. 2, 31e39. McIntosh, B.S., Ascough, J.C., Twery, M., Chew, J., Elmahdi, A., Haase, D., Harou, J.J., Hepting, D., Cuddy, S., Jakeman, A.J., Chen, S., Kassahun, A., Lautenbach, S., Matthews, K., Merritt, W., Quinn, N.W.T., Rodriguez-Roda, I., Sieber, S., Stavenga, M., Sulis, A., Ticehurst, J., Volk, M., Wrobel, M., van Delden, H., ElSawah, S., Rizzoli, A., Voinov, A., 2011. Environmental decision support systems (EDSS) development e challenges and best practices. Environmental Modelling & Software 26 (12), 1389e1402. Needham, J.T., 1998. Application of Linear Programming for Flood Control Operations on the Iowa and des Moines Rivers, MS thesis, University of California, Davis. Pallottino, S., Sechi, G.M., Zuddas, P., 2005. A DSS for water resources management under uncertainty by scenario analysis. Environmental Modelling & Software 20, 1031e1042. Rani, D., Moreira, M., 2010. A survey and potential application in reservoir systems operation. Water Resources Management 24 (6), 1107e1138. Rossi, G., Vega, T., Bonaccorso, B., 2007. Methods and Tools for Drought Analysis and Management. Springer, 418 pp. Sandoval-Solis, S., McKinney, D.C., Loucks, D.P., 2011. Sustainability index for water resources planning and management. Journal of Water Resources Planning and Management 137 (5), 381e390. Sechi, G.M., Sulis, A., 2009. Water system management through a mixed optimization-simulation approach. Journal of Water Resources Planning and Management 135 (3), 160e170. SEI Stockholm Environment Institute, 2005. WEAP: Water Evaluation and Planning System, User Guide. Somerville, MA. Sheer, D., 1986. Dan Sheer’s Reservoir Operating Guidelines. School of Civil and Environmental Engeneering, Cornell University, Ithaca, NY. Simonovic, S.P., 1992. Reservoir systems analysis: closing gap between theory and practice. Journal of Water Resources Planning and Management 118 (3), 262e280. Sulis, A., 2009. GRID computing approach for multireservoir operating rules with uncertainty. Environmental Modelling & Software 24 (7), 859e864.