Professional Documents
Culture Documents
1. Introduction
The water resource issues in the developing countries are usually quite different from
those currently experienced in the industrial world (Brookshire and Whittington, 1993).
The water supply and land management issues in Taiwan are especially acute in the
reservoir area where unexpected drought, soil erosion, nonpoint source pollution,
reservoir eutrophication, and inappropriate development for stock farming or recreation
have continuously lowered the essential level of water quality and quantity. On the
other hand, the rapid increase of population and the driving force of economic growth
further accelerate the need for various land uses within the watershed. To contemplate
the scope of such problems, as experienced in many other developing countries, the
efforts of pursuing integrated optimal planning to achieve the sustainable uses of these
natural resources becomes critical. Many studies have been made of multiobjective
145
0301–4797/95/020145+17 $08.00/0 1995 Academic Press Limited
146 Optimal management by multiobjective programming
2. Model formulation
Since 1973, the U.S. Water Resources Council has recommended the inclusion of
national economic development and environmental quality as two essential non-
commensurable objectives in water and related land resource planning. However, one
N.-B. Chang et al. 147
of the major difficulties in formulating a water resource planning model is the integration
of those non-commensurable objectives or constraints. In this analysis, the basic
difficulties are the attempt to include the environmental factors while lacking a precise
methodology as well as the exact quantitative information on the environmental impacts.
In general, three broad-based objectives related to various land-use programs in a
reservoir watershed can be considered and organized as:
1. Economic development:
National economic development;
Employment level;
Growth rates of national income;
Inflation;
Regional economic development;
Direct income;
Market development.
2. Social welfare:
Social equity;
Economic efficiency;
Income redistribution.
3. Environmental protection:
Water quality impacts control;
Discharge of nitrogen;
Discharge of phosphorus;
Increasing biological oxygen demanding load;
Reservoir safety;
Sediment;
Flood impact on dam.
Thus, the objective function is usually prepared to maximize the economic de-
velopment and social welfare, and minimize the environmental impacts simultaneously
in such a multiobjective analytical framework. The essential elements considering the
institutional framework, social, physical, economic, and environmental limitations in
the constraint set can be briefly described as follows.
Socio-economic factors:
Total land available for development in a reservoir watershed;
Legitimately required area for forest conservation;
Financial viability;
Economic feasibility, such as production cost;
Cultural and aesthetic limitations;
Minimum land requirement for regional agricultural development for self-
sufficiency;
Minimum requirements for recreational areas in regional planning;
Minimum requirements for residential areas in regional planning.
Institutional factors:
Required level of wastewater treatment by law and its related costs;
Other limitations by law or regulation, such as land slope constraint in various
land-use programs.
Natural and environmental factors:
Required area for wildlife conservation and various ecological loading;
148 Optimal management by multiobjective programming
Topographical limitations;
Forest and agricultural uses constrained by soil property in the watershed;
Geological limitations in land development;
Meteorological limitations in land development;
Assimilative capacity for various pollutants;
Limitations of water resource use.
The above framework is simply a definition. In the detailed model formulation, by
considering the disparity and potential in the level of applicability, the structure of this
model should not be unnecessarily complicated, although it is worthwhile to calculate
all the well-known environmental and economic factors accurately. Hence, only major
factors directly related to land-use programs are considered in this analysis. It is also
recognized that optimal use of adequate existing data bases and available methods to
enhance the transparency of environmental impacts and economic costs/benefits is
better than the complete avoidance of considerations, although part of the parameter
values or constraints might not be easily quantified. The selected model formulation is
shown in the following case study.
3. Case study
This case study is prepared for illustration only but may provide some realistic solutions
to aid in the future policy making of the Tweng-Wen reservoir watershed management.
Tweng-Wan Reservoir
Wu-San-Tow Reservoir
Taiwan Strait
Nan-Hwa Reservoir
Jing-Men Reservoir
Taiwan 5 0 5 10 15
km
Figure 1. The geographical location of the Tweng-Wen reservoir and its river basin.
function. The types of environmental impacts selected in this analysis for control as
part of land use planning have to reflect directly the needs in the current watershed,
which consist of the water quality impacts related to the total discharges of nitrogen
(TN), phosphorus (TP), biological oxygen demanding load (BOD), and the sediment
yield from soil erosion.
Much physical, environmental, and economic data for land resource and reservoir
management has to be compiled to build up the objective function. For instance,
various supporting submodels of point and nonpoint sources of pollution, soil erosion,
and their effect on reservoir water quality have to be investigated in advance. For the
purpose of forecasting soil erosion, several previous studies had to be investigated and
various estimation formula, including the Universal Soil Loss Equation (Wischemeir
and Smith, 1978), are applied in this analysis. Economic database information is mainly
collected from government agencies. Final selections of each parameter value or
submodel need to be reviewed by many disciplines. After such a series of environmental
and economic investigations, the coefficients of the six objective functions are listed in
Table 1 in which the impacts of point and nonpoint sources of pollution reflecting the
current uncontrolled situation are each illustrated as follows.
The above objective functions have to be optimized and coordinated with site
specific information, such as the specified value of land availability, required minimum
forest area, required residential and recreational area, soil property and land slope
limitations, as well as the assimilative capacity of various pollutants loading. In the
case of the determination of assimilative capacity, traditional water quality models and
the standards of lake eutrophication used in Taiwan and Japan are considered. Other
150
Z1 (kg/ha/yr)
Point source — — 66·50 — 38 527·80 18·67
Nonpoint source 1·04 1·73 2·89 0·29 2·89 1·25
Total 1·04 1·73 69·40 0·29 38 531·00 19·92
Z2 (kg/ha/yr)
Point source — — 389·20 — 153 908·30 110·30
Nonpoint source 15·50 25·80 43·00 4·30 43·00 18·60
Total 15·50 25·80 432·20 4·30 153 951·00 128·90
Z3 (kg/ha/yr)
Point source — — 973·00 — 384 771·00 275·83
Nonpoint source 91·4 152·00 253·90 25·40 253·90 109·80
Total 91·4 152·00 1227·00 25·40 385 025·00 385·60
Z4 (ton/ha/yr) 104·20 383·00 263·10 52·80 263·10 96·20
Z5 (capita/ha) 1·60 1·13 102·2 0·00 47·0 2·49
Z6 (104 $/ha) 23·14 16·72 1628·9 0·00 11 998·5 172·08
parameter values are determined according to the government law and regulation, as
well as economic evaluation reports. For instance, 35 778 in the sixth objective function
is the required minimum area for forest conservation in the Tweng-Wen reservoir
watershed. Hence, this preparation for the case without point and non-point sources
of pullution control eventually leads to a model with the first four objective functions
to be minimized and the last two objective functions to be maximized under ten
constraints as listed below:
Subject to:
a. Land availability constraint: The maximum area allowed in developing various land-
use programs is 46 280 ha in this watershed, which is obtained by the subtraction of
the surface area of the reservoir from the entire watershed area.
X1+X2+X3+X4+X5+X6=46 280
b. Forest area constraint: Government regulation requires that the minimum forest area
should be no less than 35 778 ha in this watershed for the purpose of natural resource
conservation.
X1[35 778·00
c. Soil property constraint: The land is categorized into six soil groups in Taiwan and
the first four groups are adequate for the agricultural uses. Thus, 2166·3 ha is the
summation of the area of these first four groups of land in this watershed.
X2Ζ2166·30
d. Minimum agricultural area constraint: According to the current population level in
this region, the minimum agricultural area for self-sufficiency is 164·2 ha.
X2[164·20
e. Minimum residential area constraint: According to the regional development plan,
the minimum required residential area for the estimated population is 42·7 ha.
X3[42·70
f. Land slope constraint: Government regulation requires that the maximum slope
allowed for further development of those uses of agriculture, housing, and stock farming
is 28·7° (i.e. 55%). This generates an area of 4994·4 ha in the Tweng-Wen reservoir
watershed where such a land-use program can proceed.
X2+X3+X5Ζ4994·40
g. Minimum recreational area constraint: According to the regional development plan,
the minimum recreational area required is 137·7 ha.
X6[137·70
152 Optimal management by multiobjective programming
A joint approach in both point and nonpoint source pollution is emphasized given
the fact that the anticipated land-use programs would be mainly agricultural, residential,
recreational, and even stock farming in this watershed. Hence, both the development
and improvement of techniques for controlling point and nonpoint source pollution
are considered in future water pollution control policies, and a different set of objective
functions is prepared in which the removal efficiency and its related engineering costs
for both point and nonpoint sources are assumed. Only the land use activities within
the agricultural, forest, and grass sectors have been considered as nonpoint source
contributors. This estimate is prepared for further analysis, as shown in Table 2. In such
a situation, the fifth and sixth objectives have also been modified by the corresponding
predictions of engineering costs, such as wastewater treatment for point source control
and embankment for nonpoint source control, as well as their direct impacts on
economic income and employment level.
Z∗k (x)−Zk(x)
Z=
Z∗k
N.-B. Chang et al.
Z1 (kg/ha/yr)
Point source — — 53·20 — 30 882·00 18·67
Nonpoint source 0·66 1·10 1·83 0·18 1·83 0·80
Total 0·66 1·10 55·03 0·18 38 883·83 19·47
Z2 (kg/ha/yr)
Point source — — 311·40 — 123 127·00 110·30
Nonpoint source 9·83 16·36 27·26 2·73 27·26 11·79
Total 9·83 16·36 338·66 2·73 123 154·26 122·09
Z3 (kg/ha/yr)
Point source — — 145·95 — 57 716·00 275·83
Nonpoint source 57·95 96·37 160·97 16·10 160·97 69·61
Total 57·95 96·37 306·92 16·10 57 866·97 345·44
Z4 (ton/ha/yr) 66·10 242·80 166·80 33·50 166·80 61·10
Z5 (cap/ha) 1·77 1·3 102·4 0·17 47·2 2·66
Z6 (104 $/ha) 22·07 12·79 1589·3 −0·54 11 794·30 171·09
The six objectives and six decision variables are listed in Table 1.
153
154 Optimal management by multiobjective programming
in which the Z∗k (x) is the maximum value of each individual objective which can be
obtained from the payoff table. Hence, the compromise programming problem is
equivalent to solve the following dimensionless function which is just the relative
measure of the decision maker’s preference:
G] A BH
p a 1/a
Z∗k (x)−Zk(x)
Min da=Min pak
k=1
Z∗k (x)
where
1<a<∞, pak>0
and
p
] p =1
k=1
a
k
The parameter “p” represents the total number of objectives, and pak is the corresponding
weight of each objective. For a=1 (i.e. which is the case of d1), the problem becomes
a linear program and the LINDO software package can be employed as a solver in
this analysis. While a=2 (i.e. which is the case of d2), the solution will be the non-
inferior feasible solution which is closest to the ideal solution Z∗k in terms of a weighted
geometric distance. In this situation, the GINO software package can be employed. If
a=∞ (i.e. which is the case of dx), the model can be transformed into a linear
programming model in which the largest weighted deviation determines the preferred
solution and the situation among trade-off mechanics turns out to be not only competitive
but noncompensatory. The initial effort for solving this model is to apply the solution
technique of compromise programming in the case in which the point and nonpoint
source pollution are uncontrolled in the watershed; and the weight of each objective is
assumed to be equally important in the initial decision making profile. Then, the impacts
by applying pollution control strategies on the reduction of pollutant loading and the
influence of economic development are to be considered and the model will be solved
again by compromise programming.
Besides, for the purpose of sensitivity analysis, the multiobjective simplex method
is also applied to detect the robustness of this model. A FORTRAN program was
adapted from Zeneley (1974) and further modified as a PC version for this analysis.
However, additional solution sets can also be generated based on the multiobjective
simplex method. Final comparison may be placed upon the differences of planning
scenarios between the schemes of applying a single objective and multiple objectives in
dealing with such a complex water resource planning issue.
T 3. The optimal solution in the case of the inclusion of each individual objective independently
T 4. The payoff table for the case of uncontrolled pollution in compromise programming
Z1 Z2 Z3 Z4 Z5 Z6
I −46 135 −638 677 −3 658 556 −4 351 748 62 137 923 901
II −46 349 −640 000 −3 662 281 −4 352 400 62 453 928 938
Z∗k −46 135 −638 677 −3 658 556 −4 351 748 62 453 928 938
Alternative land management and control policies are therefore generated by the
allowed land-use activities and watershed characteristics. Table 5 lists the optimal
solutions using a compromise programming scheme. It shows that different values of
the exponent adopted in the compromise programming model do not obviously influence
the final optimal solution. Additional solution sets, generated by using the multiobjective
simplex method, also show similar results. By considering the fact that the objectives
in this analysis do not compensate each other, the solution set of dx is chosen as the
basis for further analysis. Table 6 thus demonstrates that the solution sets would not
differ much if only one objective in each broad-based group of objectives was selected
to form a simplified model. Once artificial control of either point or nonpoint sources
are included, the solutions would suggest quite a different profile, as shown in Table
N.-B. Chang et al. 157
T 7. The dx solution based on two different combinations of point or nonpoint source
control
7. Larger residential areas and forest are then allowed by dramatically reducing the
area of grass land. Therefore, corresponding trade-offs for each alternative plan and
the selection of final land management and pollution control strategy with respect to
water quality desired, engineering or production cost, economic productivity, and other
technical limitations can be achieved by the decision maker.
T 9. Sensitivity analysis of resource availability based on the dx solution in the multiobjective
optimization model
T 10. Sensitivity analysis of resource availability based on the dx solution in the single
objective optimization model (objectives 1–4)
T 11. Sensitivity analysis of resource availability based on the dx solution in the single
objective optimization model (objectives 5–6)
system can be shown, as in Table 9. It is observed that the allowable changes of the
initial available resources do present big differences between single and multiple
objective programming models, as is evident from Tables 10 and 11. This implies that the
N.-B. Chang et al. 159
5
4·8
4·6 Multiple objectives
4·2
4
3·8
3·6
3·4
3·2
3
–16 –14 –12 –10 –8 –6 –4 –2 0
Coefficient (kg/ha)
Figure 2. The variations of parameter values versus the relative change of decision variables in a single (the
second objective) and multiobjective programming model.
3·593
3·591
Single objective
3·587
3·585
3·583
3·581
3·579
3·577
3·575
20 25 30 35 40 45 55 55
Coefficient (104 $/ha)
Figure 3. The variations of parameter values versus the relative change of decision variables in a single (the
sixth objective) and multiobjective programming model.
4. Discussion
Several alternative planning scenarios in the above analysis have provided necessary
information for the decision maker in selecting the most desirable management plan
160 Optimal management by multiobjective programming
for the Tweng-Wen reservoir watershed. Overall, final compromised solutions suggest
that increasing the residential area is feasible if pollution can be controlled properly in
these new communities, but livestock husbandry can not be allowed under any cir-
cumstance within the Tweng-Wen reservoir watershed. A more practical approach can
be gained through an interactive procedure in this analysis, in which the decision maker
may proceed from one non-inferior solution to another by comparing each land
management alternative and evaluating trade-offs between individual goals. But, the
inherent complexity in comparing the environmental impacts of very different natures
to the technico-economic values in this analysis may decrease the confidence in applying
its optimal solutions. However, such an analysis is significant specifically for those
developing countries with a highly populated community and limited land space,
although the decision makers usually tend to ignore such a dilemma.
A more complex analytical system is needed in the future because Taiwan is also
evolving from a development of land-use in watershed and conservation of reservoir
water quality to a water economy (water right trade-off) and river basin system planning
by a conjunctive surface-ground water reallocation. Further extension work can be
focused on the combination with fuzzy systems theory and welfare economics in such
an application. Therefore, a fuzzy multiobjective linear or mixed integer programming
model with adequate value functions related to various water consumptions could be
an advanced tool in solving future water resource problems in developing countries
(Wright et al., 1983; Xiang et al., 1992). Further, the modeling efforts can be combined
with a geographical information system (GIS) to help decision makers evaluate the
socio-economic and ecological impacts of various planning scenarios in the framework
of a decision support system (DSS) (Stansbury et al., 1991).
5. Conclusions
This study has led to an intellectually appealing approach for integrated watershed
resource management. Social, physical, economic, environmental, and institutional
considerations are tied together to find the comprehensive solutions and support
guidelines in the water resources and water quality regional planning in a watershed
area. Based on the methodology used in this analysis, in order to maintain the essential
quality of water supply and achieve regional development targets, the decision maker
would thereby optimally manage watersheds so that soil erosion, point and nonpoint
source pollution can be reduced to a required level while regional income and em-
ployment levels can be maximized simultaneously.
References
Brookshire, D. S. and Whittington, D. (1993). Water resources issues in the developing countries. Water
Resource Research 29(7), 1883–1888.
Cohon, J. L. (1978). Multiobjective Programming and Planning. San Diego: Academic Press.
Das, P. and Haimes, Y. Y. (1979). Multiobjective optimization in water quality and land management. Water
Resource Research 15(6), 1313–1322.
Glover, F. and Martinson, ?. (1987). Multiple-use land planning and conflict resolution by multiple objective
linear programming. European Journal of Operational Research 28, 343–350.
Goicoechea, A. and Duckstein, L. (1976). Multiobjective programming in watershed management: a study
of the Charleston watershed. Water Resource Research 12(6), 1085–1092.
Goodman, A. S. (1984). Principles of water resources planning. New Jersey: Prentice Hall.
Leone, A. and Marini, R. (1993). Assessment and mitigation of the effects of land use in a lake basin (Lake
Vicoin in central Italy). Journal of Environmental Management 39, 39–50.
N.-B. Chang et al. 161
Loucks, D. P., Stedinger, J. R. and Haith, D. A. (1981). Water Resource Systems Planning and Analysis. New
York: Prentice-Hall.
Ridgley, M. A. and Giambelluca, T. W. (1992). Linking water-balance simulation and multiobjective
programming: land-use plan design in Hawaii. Environment and Planning B: Planning and Design 19,
317–336.
Stansbury, J., Woldt, W. and Bogardi, I. (1991). Decision support system for water transfer evaluation. Water
Resource Research 27(4), 443–451.
Tecle, A. (1992). Selecting a multicriterion decision making technique for watershed resource management.
Water Resource Bulletin 28(1), 129–140.
Van, D. A. and Nijkamp, P. (1976). A Multiobjective Decision Model for Regional Development, Environmental
Quality Control and Industrial Land Use. Papers of the Regional Science Association, Vol. 36, pp. 35–57.
Water Resources Council (1973). Principles and standards for planning water and related land resources,
Federal Registration, 38 (174), Part III, 24,77,24,869, 10 September.
Wischmeier, W. H. and Smith, D. D. (1978). Predicting rainfall erosion losses—a guide to conservation planning.
Agricultural Handbook 537, U.S. Department of Agriculture, Washington, D.C., USA.
Wright, J., ReVell, C. and Cohon J. (1983). A multiobjective integer programming model for the land
acquisition problem. Regional Science and Urban Economics 13, 31–53.
Wu, S. L. (1993). Evaluation of land-use programs in a reservoir watershed by multiobjective programming.
Masters thesis, Graduate Institute of Environmental Engineering, National Cheng-Kung University,
Tainan, Taiwan.
Xiang, W. N., Gross, M., Fabos, J. G. and MacDougall, E. B. (1992). A fuzzy-group multicriteria decision
making model and its application to land-use planning. Environment and Planning B: Planning and Design
19, 61–84.
Zeleny, M. (1974). Linear Multiobjective Programming, pp. 197–220. New York: Springer-Verlag.
Zeleny, M. (1976). Multiple Criteria Decision Making. New York: McGraw-Hill.