Hydrological Sciences Journal

ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: http://www.tandfonline.com/loi/thsj20

An integrated, multi-functional approach to water resources management D. G. JAMIESON To cite this article: D. G. JAMIESON (1986) An integrated, multi-functional approach to water resources management, Hydrological Sciences Journal, 31:4, 501-514, DOI: 10.1080/02626668609491070 To link to this article: http://dx.doi.org/10.1080/02626668609491070

Published online: 21 Dec 2009.

Submit your article to this journal

Article views: 275

View related articles

Citing articles: 5 View citing articles

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=thsj20 Download by: [37.44.207.33]

Date: 29 January 2017, At: 16:33

Hydrological Sciences - Journal - des Sciences Hydrologiques, 31, 4, 12/1986

An integrated, multi-functional approach to water resources management* D. G, JAMIESON Thames Water Authority, Road, Reading, Berkshire

Nugent House, RG1 BDB, UK

Vastern

ABSTRACT For the purposes of this paper, water resources management is defined in its broadest terms and deemed to encompass all water-related aspects of river-basin management from long-term planning through to real-time operational control. Bearing in mind the increasing pressure on finite resources, an integrated, multifunctional approach is advocated in which different analytical procedures are used at various stages of decision making. These include mathematical programming, large-scale simulation and optimal control theory. By way of example, the application of these techniques to water resources in the United Kingdom is briefly described.

Une approvhe multifonctionnelle des ressources en eau

intégrée

de la

gestion

RESUME Au sens du présent exposé, la gestion des ressources en eau est définie dans ses termes les plus larges et est censée comprendre tous les aspects se rapportant a. l'eau de la gestion des bassins fluviaux allant de la planification à long terme jusqu'au contrôle opérationnel en temps réel. Tenant compte de la pression toujours plus accrue sur des ressources limitées, une approche multifonctionnelle, intégrée, est préconisée, dans le cadre de laquelle des methodologies analytiques différentes sont employées à divers moments du procédé de prise de décision. Ces méthodologies comprennent les programmes mathématiques, la simulation à grande échelle et la théorie de contrôle optimal. A titre d'exemple, l'application de cette technique aux ressources en eau en Grande-Bretagne est décrite succinctement.

INTRODUCTION Background Traditionally, the water industry in most countries has been fragmented, with responsibility for the various facets of riverbasin management being allocated to different organizations. Although this might suffice whilst demands on the system are small and natural resources are ample to absorb any interaction, there

^Opening lecture. International Symposium on Scientific Basis for Water Resources Management, Jerusalem, Israel, September 1985. 501

502

D.G. Jamieson

comes a time when it is necessary to co-ordinate the activities of these independent organizations by means of, say, a river-basin commission. Thereafter, in some countries, a juncture is reached when pressure on resources becomes so acute and interactions so complex that it becomes necessary to integrate the hitherto separate organizations within a river basin and adopt a multifunctional approach. By and large, the development of analytical procedures for water resources management has been heavily influenced by the institutional arrangements. Initially, techniques were formulated for a single function such as water supply or sewage disposal. Water quantity was usually divorced from water quality and vice versa. Gradually, as these functions began to interfere with each other, static agreements in the form of water abstraction licences, effluent discharge standards etc. came into force which could be included as fixed constraints in the analytical techniques already developed. However, in the subsequent move towards a multi-functional approach where there is no longer a simple objective function, it has become necessary to re-think some of these concepts. The aims of this paper are threefold: firstly, to structure the problem of water resources management and show how the various facets relate to each other; secondly, to indicate the types of analytical techniques available and demonstrate their use; and finally, to compare theory with practice and suggest possible areas for improvement. By way of example, one particular conceptual framework is outlined. Obviously, this is not the only approach but any attempt to summarize the current global state-of-the-art would have been extremely superficial. Even so, it is difficult to do justice to a topic of this magnitude in a single paper: nevertheless, it is hoped there is sufficient detail to appreciate the philosophy advocated. Range of

functions

In the past, the term "water resources" has frequently been interpreted in the restrictive sense to mean the conservation aspects of water supply. For the purposes of this paper, water resources has been taken as the generic name for all water-related functions of which the most common are: (a) water supply, (b) hydropower, (c) navigation, (d) fisheries, (e) recreation and amenity, (f) land drainage, (g) sewage disposal, (h) effluent discharge. Obviously, a river basin managed for one specific function may well provide incidental benefits for some related purposes but conflict with others. However, the various components of water resources can be roughly divided into conservation functions where storage is a benefit and disposal functions where the opposite tends to be true. Conservation functions include water supply, navigation, hydropower, recreation and amenity; disposal functions include

An integrated multi-functional approach to water resources management

503

sewage disposal, land drainage etc. In general, there would seem to be more common interests within each of these groups than between them. However, even within a single grouping, there are likely to be competing demands on scarce resources as in the case of, say, water supply and fisheries or navigation. Therefore water resources management is primarily an exercise in conflict resolution. Scope of

activities

The word "management" has also been given herein a liberal interpretation and taken to mean all administrative aspects associated with the welfare of water resources. More specifically, management is deemed to include: (a) development strategy, (b) system design, (c) operational strategy, (d) real-time control. Each of these aspects operates in a different space/time frame: whereas the development strategy may cover an entire country and have a time horizon measured in decades, real-time control is primarily concerned with one particular subsystem over the next several hours. Nevertheless, it is important to recognize that they are related and form part of the same management continuum and have only been divided in this way for convenience.

DEVELOPMENT STRATEGY Objective The water sector is usually regarded as a service industry which reacts to the requirements of other sectors rather than being the economic driving force. Therefore, in most cases, the objective of water resources planning is to meet the perceived needs in the most cost-effective way, having regard to the environmental implications. Alternatively, if the water resources plan for a country acts as a surrogate for a national development plan, then perhaps benefitmaximization is a more appropriate objective function. Choice

of planning

technique

Whichever objective, the outcome of water resources planning takes the form of a preferred development strategy, covering the next 20 to 30 years. Since the process is subject to considerable uncertainty in terms of future requirements, prevailing economic climate etc., there is a need for intermittent updating and amendment. This implies the use of a computer as the main planning tool since the sheer volume of data, the tedious nature of the analyses and the need to consider alternative scenarios make manual calculations impractical . The planning process on a national scale is characterized by the requirement to screen a large number of options with a view to determining:

504

D.G. Jamieson

(a) which resources should be developed; (b) the areas of need to which they should be assigned; (c) the timing and order of development. To that end, information is required on future demands (water supply, hydropower etc.), the performance of different size sources at various locations, the possible links between sources and demand centres, together with all associated construction and operating costs . Faced with an enormous choice of options for matching resources to needs, a systematic search procedure is necessary if the most cost-effective solution is to be identified. One possibility is the use of mixed integer programming (O'Neill, 1972) in which a section of a cost row might typically comprise: (minimize) ... E.E c s + H p, ,1, . + E, X. q, m l t it it k t *Kt kt k t kt kt + E E v f k t kt kt where Cit = discounted cost of building reservoir i in period t; sit = zero-one variable taking the value of 1 if reservoir i is to be built, in period t, 0 otherwise; p k t = discounted set-up cost of building link k in period t; ljjt = zero-one variable to determine whether a stage of link k is built during period t; Ikt = discounted proportional cost for link k, period t; miç--)- = continuous variable for capacity built, link k, period t; v kt = discounted variable cost of conveying water at unit rate through link k throughout period t; fj^t = continuous variable for flow in link k during period t. Obviously, the impact of water quality can be considered by extending the cost row to include water treatment. However, the main advantage of mixed integer programming rests with the use of the "zeroone" variable for "yes-no" decisions on capital investment, thereby avoiding inadmissible fractional values.

Application

and use

The above technique was used as the basis of the national water resources plan for England and Wales (Water Resources Board, 1973). For that particular study, a 30-year planning horizon divided into 5-year increments was adopted. In this way, it was possible to indicate how the development of resources should evolve during that 30-year period. Figure 1 depicts the outcome in terms of new resources required by the year 2001 for one specific set of planning assumptions. With all demand, resource and costing information held on computer files, it is relatively easy to test the sensitivity of the plan to uncertainty. For example, it may be appropriate to derive alternative plans for different municipal/agricultural/industrial scenarios. Similarly, it is possible to make different assumptions

An integrated multi-functional approach to water resources management

505

on cost information, system performance etc. In this way, the effect of these variables on the preferred development strategy can be assessed, thereby determining the robustness of the solution proposed.

RESERVOIR STORAGE GROUNDWATER SOURCE RIVER ABSTRACTION POINT f

J .

CENTRE OF DEMAND \ *-

DIRECT SUPPLY LINK

». RIVER LINK V

7

ESTUARY STORAGE

Fig. 1

National planning using mixed-integer programming.

506

D,G. Jamieson

SYSTEM DESIGN

Objective Having used mathematical programming or some other similar technique to screen a large number of options, the next stage is to use what appears to be the preferred least-cost solution as the basis of system design. Here, within this pre-determined configuration, the objective is to identify the size of components such as reservoirs, pumping capacities etc. in order to meet the projected demands with a given reliability. Choice

of design

technique

Whilst the scale is likely to have been reduced from national to that of a region, nevertheless the complexity and interaction between components of a water resources system are such that simulation is probably the only technique suitable for this type of analysis at the present time. Since system reliability is usually a function of extreme events rather than average conditions, there is a need for long-term hydrometric records yet frequently these do not exist. As a result, many systems are designed on the basis of inadequate data. However, over the past two decades or so, considerable effort has been invested in synthetic data generation techniques of the type: Qt

= f{(Tt),(Ht),(St),(At),(et),(j)}

where Q t = flow rate, T^- = trend, H t = persistence, S-t = seasonality, A-t = autoregressive/moving average component, et = random variable, cf> = set of model parameters derived from the historic data. This approach has been extended to take account of water quality variables such as nitrates and chlorides (Page & Warn, 1974). The main advantages of these techniques include utilizing all of the information content in the observed sequences rather than just the extremes and having the ability to test the sensitivity of the preferred solution to the limited historic records. When it comes to assembling the model, there is considerable merit in structuring the simulation in a modular form, thereby enabling different system configurations to be evaluated quickly and effectively. Component models representing the various elements which make up a regional water resources system can be formulated for the most complex arrangement that can be envisaged. These can be used to describe simpler variants by setting appropriate links to zero. Traditionally, system "failure" has been taken to imply complete emptying of storage when in reality, steps would have been taken to offset that possibility some time before it occurred. If "failure" is re-defined to mean the introduction of restrictions on use,

An integrated multi-functional approach to water resources management

507

presumably this would be tolerated more frequently than, say, complete emptying of a reservoir. Therefore, not only would this enable a more confident estimate of system reliability to be made for a given length of historic records, but also different degrees of restriction on use could be considered. For example, in the case of water supply, the design specification might be to meet the demand for 98.0% of the time without any restrictions, 98.5% of the time with only minor restrictions such as a ban on garden-watering, and 99.0% of the time without serious disruption to industrial or domestic use. Simulation is, by itself, an evaluation procedure rather than a design technique and it is usually left to the individual to manipulate the design parameters in determining the component sizes that meet the required performance at minimal cost. Ideally, it should be possible to specify an acceptable reliability and let the model itself derive the appropriate reservoir capacities, pump sizes etc. Attempts to develop a generalized numerical-optimization package for this class of model have met with mixed success (Smith, 1977). Whilst it is possible to formulate an efficient optimization procedure for a specific configuration, there is difficulty in using the same procedure on other systems.

Application

and use

The methodology outlined in the previous sections is currently used for planning the water resources of the Thames basin in the UK. The region supports a population of about 12 million, including that of London, and covers an area of approximately 13 000 km2. Overall, it is estimated that the Thames Water Authority and its agents provide about 3.7 million m3 per day of potable water. In addition, there is a further potential requirement of around 1.4 million m per day for licensed industrial/agricultural abstractions. Approximately 42% of all supply requirements are taken from groundwater whilst the remaining 58% are met by surface water abstraction. The same river network is also used for navigation, land drainage, recreation and amenity, fisheries and the disposal of effluent from some 450 sewage treatment works. Originally, only the water quantity aspects were incorporated in the techniques developed for resource assessment (Sexton et al., 1979). More recently, however, these have been extended to include water quality considerations in general and nitrates in particular (Sinnott & Jamieson, 1982). Figure 2 shows a schematic representation of the component models as used to depict both water quantity and quality. Besides allowing for the interaction of quantity and quality, the model differs from that previously developed in terms of spatial resolution. For example, although many of the reservoirs are interconnected and could be modelled as a lumped subsystem from a water quantity standpoint, water quality considerations dictate a more distributed approach to preserve the option of blending water etc. Moreover, since water quality concentrations change with both space and time, representation of the main river channel is significantly more complex, taking account of the denitrifying effects of river algae.

508

D.G. Jamieson a a. < x o DIRECT SUPPLY ARTIFICIAL RECHARGE

g

I!

O til CD I

AQUIFER SUB-SYSTEM

w z

i

i

Z CO I

zI

h\ o a

RESERVOIR SUB-SYSTEM

! !

< tt>_ iI_

o. Fig. 2

RIVER SUB-SYSTEM

System design using simulation.

OPERATIONAL STRATEGY abjective Having defined both the system configuration and the size of components, further detailed consideration of operational management is required. In that respect, it is unfortunate that different aspects of water resources management not only have conflicting interests but also have different criteria for assessing operational performance. Whilst minimizing pumping costs might be the appropriate objective for a pumped-storage reservoir, it may be of little relevance for flood alleviation where the objective is likely to be minimization of flood damage. At least these two objectives have a similar basis, that of cost minimization, whereas others such as effluent disposal might not even be measured in financial terms. Therefore, it is perhaps inappropriate to refer to "optimal" operating rules for water resources management and instead, aim for a "satisficing" approach. In this approach, the state of the system is deemed to be satisfactory for the different facets providing it is within acceptable boundaries for each function. Boundaries are defined in terms of measurable parameters such as depth of water, flow rate, dissolved oxygen, nitrate concentrations etc. If the system state lies within these boundaries, the objective function would be to minimize operating costs. If, however, a particular constraint is, or is about to be, violated, then the specific objective relating to that function would take precedence. For instance, if flooding was imminent (the upper bound constraint on land drainage is about to be broken) then the flood mitigation objective of damage minimization would take precedence over minimizing operating costs. Again, if the depth of water in the river was inadequate (a lower bound constraint on navigation), water supply might have to draw on storage or use an alternative source even

An integrated multi-functional approach to water resources management

though that may be set of constraints insufficient depth alternative source

Choice

of operating

509

more expensive to operate. Only if more than one were likely to be violated simultaneously (e.g. to meet the navigational requirement and no of supply) would priorities have to be assigned.

strategy

At the present time, the size of analytical problem that can be realistically tackled with the necessary level of detail is probably restricted to individual subsystems such as a group of reservoirs used conjunctively or a series of weirs controlling the flow in a river. An entire river basin may contain many such subsystems and it is likely that the sheer scale of a developed river basin would dictate a de-centralized approach to operational control. It will be appreciated, of course, that even if all the individual subsystems were operated in an optimal manner, it does not necessarily follow that the overall system would be optimally controlled. Nevertheless, it is probable that optimal or near-optimal control of subsystems, capable of adapting to meet goals imposed by a more general control strategy, would approximate the optimal control of the entire system to an extent where it made little or no practical difference. To that end, a hierarchical control philosophy has been advocated for managing the resources of an entire basin (Jamieson, 1978). In the light of experience, this concept has been modified to include three tiers of management, namely, strategic, tactical and local control corresponding to Headquarters, Division and Works respectively. Whereas strategic decisions are primarily concerned with defining what is to be achieved, tactical control relates more to how those decisions are to be accomplished. Thereafter, it is left to individual subsystems to implement the instructions received (Fig.3). If, for any reason, a Division or subsystem is unable to comply with the target/instruction, a feedback loop is activated and revised directives issued.

Application and use As part of a continuing research programme, a strategic decision mechanism for operational management of water resources is being developed for the Thames basin as a whole. The model has been formulated as a multi-functional, nonlinear optimization problem, using a monthly timescale. The criteria include both reliability and operational cost objectives for water supply, sewage treatment and water pollution control. In all, the current model comprises some 287 variables and 241 constraints which are solved using a projected Lagrangian algorithm. Tactical decision-making is restricted to the area and functions covered by each of the geographically-based Divisions and operates on a daily timescale, except in an emergency. Normally, the aim would be to meet the targets imposed by the strategic control in the most cost-effective way, whilst keeping the overall system

510

D.G. Jamieson

state within the acceptable range. If, however, one of the boundary constraints defining a maximum or minimum for a particular function has been, or is about to be, breached, provision will be made to switch the objective so as to cater for the short-term emergency. The present prototype model of the lower Thames has been formulated as a multidimensional iterative search procedure which is subject to various legal/physical/operational constraints.

Overall S y s t e m

Medium-term

Requirements

Medium-termForecasts Physical/Legal Constraints Long-term

Objectives

Strategic

Multi-functional Strategic

Etc.,

Decision-mechanism

( R e g i o n a l HQ)

Model

Monthly

Timescale

Hydrometric Statistics

Operational Feedback

Short-term

Targets

Requirements

Short-term Forecasts

Short-term

Tactical

Multi-functionai

Acceptable Operational Range

Decision-making

( D i v i s i o n a l HQ)

Tactical Model

Daily

Timescale

Objectives

Operational

Feedback

Instructions

C u r r e n t S t a t e Of S u b s y s t e m _ Real-time

Forecasts

On-line Single-function

Subsystem

Contraints

Subsystem

Objective(s)

Control Modules

Actions

Fig. 3

Control

( T r e a t m e n t W o r k s Etc., Hourly

Timescale

Implemented

Hierarchical control as applied to river basin management.

An integrated multi-functional approach to water resources management

511

REAL-TIME OPERATIONAL CONTROL Objective With operational goals being up-dated periodically, it would be left to the small computer controlling each subsystem to decide how to implement the instructions received, having regard to the particular objective of the function in question. For example, in the case of a pumped storage reservoir, the quantity of water to be abstracted from the river would be specified, leaving the control computer to determine which pumps should be used and for how long, so as to minimize energy costs. Similarly, for sewage treatment, effluent discharge standards would be specified, leaving the computer controlling the works to decide the extent of processing necessary to meet the output requirements at minimal operating cost. Choice

of control

procedures

The role of the small computer controlling each subsystem would be to interrogate the telemetry scheme to ascertain the existing state of the system, forecast the future state, decide the optimal or near-optimal course of action, and implement that decision before repeating the whole procedure. In this way, operational decisions are based on the expected future state of the system rather than on the present known state. Since perfect foresight cannot be assumed, self-corrective measures have to be included to compensate for the inevitable errors in the forecast. What few examples there are of feed-forward control systems being used within the water industry tend to base their forecasting procedures solely on simulation. These have usually been adaptations of "explanatory" models in which the algorithms purport to have physical reality in the sense that they attempt to mimic the real system. However, for operational control purposes, the requirement is for a reliable forecast of future events rather than a detailed understanding of the processes involved. For this reason, there has been growing interest in recursive estimation using a Kalman filter approach as a means of improving forecasting procedures (O'Connell,'1980). The merits of the Kalman filter approach include flexible state-space formulation, ability to cater for both system and measurement noise, recursive parameter estimation and optimal forecasts. Operational decisions in water resources management are seldom simple and, even more rarely, optimal. At the present time, dynamic programming in one form or another is perhaps the most commonly proposed decision mechanism for real-time use. However, in common with other optimal control techniques, the scale of problem that can be considered is often curtailed by the limited computing facilities available. Take the case of the Dee Research Programme (Jamieson & Wilkinson, 1972) where in reality the flood mitigation algorithm should have been five state variables and two decision variables. Fortunately, this could be reduced to three state variables and one decision variable without loss of credibility, viz: f (s,q,r) = min {max [R(s),E(q), f (T(s,q,rId))]} m m-i

512

D.G. Jamieson

where fm(.)

= minimum cost of damage attainable between now and the time horizon; s,q and r = reservoir water level, downstream flow and rate of release from storage, respectively; R(s) and E(q) = damage cost functions at vulnerable locations; T(s,q,r|d) = values of s,q,r adjusted according to the decision d. Moreover, it was subsequently established that physical constraints on opening and closing the sluice gates were such that the decision mechanism could be further simplified by actually avoiding having to solve the dynamic programming algorithm, thereby lending support to the concept of near-optimal control. Application

and use

As part of the same research programme mentioned previously, a series of real-time operational control procedures are being developed for different aspects of water resources management. Rather than repeatedly developing similar procedures for the same type of subsystem at different locations, the intention is to formulate generalized computer packages which can be used throughout the region with minimal adaptation. In the first instance, three such modules are being developed: (a) river management (optimal or near-optimal control of releases, impoundments etc.); (b) water supply management (optimal or near-optimal control of water treatment and distribution); (c) sewage disposal management (optimal or near-optimal control of sewage treatment and effluent discharges). Other modules such as aquifer management are scheduled for later development. Initially, these procedures could be used for operating individual subsystems. If, however, they were to be developed within a common framework, as is the case, this preserves the option of subsequently linking appropriate modules to the strategic and tactical decision-mechanisms, thereby introducing integrated operational control on a region-wide basis (Fig.4).

CONCLUSIONS It is hoped that this brief paper has demonstrated that there are considerable advantages in adopting an integrated approach to water resources management. On the planning side, these include the prospect of reduced construction costs when compared with an unstructured, piecemeal approach, particularly if the same facilities can serve more than one purpose. Similarly, on the operational side, there are advantages arising from the rational allocation of scarce resources between competing interests and minimizing the degree of conflict between irreconcilable functions. These advantages manifest themselves in terms of improved performance and reduced operating costs. However, such benefits will only be achieved through the systematic application and use of improved analytical techniques.

An integrated multi-functional approach to water resources management

513

TO REGIONAL CONTROL C E N T R E -

METEOR -

-yy

OLOGICAL OFFICE DIVISIONAL CONTROL

tr

CENTRE

' MULTI-PURPOSE RESERVOIR

PUMPEDSTORAGE

MODULE

RESERVOIR MODULE

±

RIVER MANAGEMENT MODULE

>t'

±

\

WATER SUPPLY

\

t

K UNCQNFlNED

WATER SUPPLY

tr

SEWAGE

SEWAGE

TREATMENT

AQUIFER MODULE

MODULE

MODULE

URBAN STORMWATER MODULE

Fig, 4

TREATMENT MODULE

^

t

URBAN STORM-

t

WATER MODULE

(COMBINED

(SEPARATE

SYSTEM)

SYSTEM)

Real-time operational control of water resources systems.

There is, of course, already a substantial body of technical knowledge available to those with the responsibility for managing water resources. However, the actual take-up rate of such knowledge is comparatively small, with many practitioners preferring to rely on outdated techniques or previous experience. The situation is not helped by the fact that some of the theoretical techniques proposed are inflexible, unrealistic or wholly dependent on data which do not exist. As a consequence, theory and practice are in danger of going their separate ways. Clearly, that trend should be reversed with increased emphasis being placed on the application side. To that end, more effort should be spent on adapting existing methodologies to real situations, particularly for use in a multifunctional environment.

REFERENCES Jamieson, D.G. (1978) Formulating operational policy for a riverbasin management authority. NATO Advanced Institute on Systems Analysis and Reservoir Management, Coimbra, Portugal. Jamieson, D.G. & Wilkinson, J.C. (1972) River Dee Research Program 3. A short term control strategy for multi-purpose reservoir systems. Wat. Resour. Res. 8(4), 911-919.

514

D.G. Jamieson

O'Connell, P.E. (1980) Real-time Hydrological Forecasting and Control (Proc. 1st International Workshop, July 1977). Institute of Hydrology, Wallingford, Oxfordshire, UK. O'Neill, P.G. (1972) A mathematical programming model for planning a regional water resource system. J. Instn Wat. Engrs 26, 47-61. Page, C. & Warn, A.E. (1974) Water quality considerations in the design of water resource systems. Wat. Res. 8, 969-975. Sexton, J.R., Cook, D.J. & Jones, A.E. (1979) Water Resources Model: An Introduction. Thames Water Publication, Reading, UK. Sinnott, C.S. & Jamieson, D.G. (1982) River basin planning for control of nitrate pollution. Wat. Sci . Tech. 14, 245-252. Smith, D.K. (1977) An approach to the optimisation of simulation models of water resource systems. PhD thesis, Univ. of Lancaster, UK. Water Resources Board (1973) Water Resources in England and Wales. HMSO, London. Received Open for

17 October discussion

1985; accepted 15 April until 1 June 1987.

1986.