DECISION SUPPORT SYSTEMS FOR INTEGRATED WATER RESOURCES ... · DECISION SUPPORT SYSTEMS FOR INTEGRATED WATER RESOURCES MANAGEMENT WITH AN APPLICATION TO THE NILE ... for water resources

DECISION SUPPORT SYSTEMS FOR INTEGRATED WATER RESOURCES MANAGEMENT WITH AN APPLICATION TO THE NILE BASIN

Aris P. Georgakakos

Professor and Director, Georgia Water Resources Institute,

Georgia Institute of Technology, Atlanta, Georgia 30332-0355, USA

Abstract: This article presents decision support system (DSS) guidelines for water resources planning and management in systems with multiple objectives, multiple disciplines, and multiple decision makers. The guidelines are based on experience with DSS development in Africa, Europe, and the US. A prototype DSS for water resources planning in the Nile Basin is also described. The article advocates that the design, development, and implementation of effective decision support systems bring together disciplines, people, and institutions necessary to address today’s complex water resources challenges. Copyright © 2004 IFAC Keywords: Decision support, system analysis, integrated, forecast, regulation multiobjective, multidimensional, uncertainty.

1. INTEGRATED WATER RESOURCES MANAGEMENT

(IWRM): SCIENCE IN SUPPORT OF PUBLIC

POLICY IWRM is the process of formulating and implementing shared vision planning and management strategies for sustainable water resources utilization with due consideration of all spatial and temporal interdependencies among natural processes and water uses. The IWRM process is conceptualized in Figure 1. The knowledge to support planning and management decisions resides in various disciplines including climatology, meteorology, hydrology, ecology, environmental science, agro-science, water resources engineering, systems analysis, remote sensing, socio-economics, law, and public policy. Public policy actors (such as politicians, judges, government

agencies, financial institutions, Non-Governmental Organizations, citizen groups, industries, and the general public) are often in a position to make critical decisions that reflect society’s shared vision for water resources utilization. Public policy actors develop consensus and decide on shared vision strategies based on information generated and communicated by Decision Support Systems (DSS) and associated processes. Thus, the role of DSS is to leverage current scientific and technological advances in developing and evaluating specific policy options for possible adoption by the IWRM process. DSSs are developed and used by research institutions, government agencies, consultants, and the information technology industry. By its nature, IWRM is a process where information, technology, natural processes, water uses, societal preferences, institutions, and policy actors are subject to gradual or rapid

change. To keep current, IWRM should include a self-assessment and improvement mechanism. This mechanism is indicated by dashed arrows in Figure 1 and starts with monitoring and evaluating the impacts of decisions made. These evaluations identify the need for improvements pertaining to the effectiveness of the institutional set-up, the quality and completeness of the information generated by decision support systems and processes, and the validity and sufficiency of the current scientific knowledge base. IWRM processes can lead to great successes just as they can cause costly failures. In a world where water disputes are on the rise and the delay between science and technology advances and their consideration by management practices widens, IWRM phases important challenges:

• Lack of integrative tools to support planning and management decisions;

• Segmentation of institutions responsible for water resources planning and management;

• Limited participation of stakeholders in decision making processes;

• Lack of disinterested self-assessment and improvement mechanisms;

• Continuing specialization of science and engineering education at the expense of interdisciplinary training.

The article advocates that the process of designing, developing, and implementing effective decision support systems bring together the necessary disciplines, people, and institutions that can address these challenges.

Fig. 1: IWRM: Science in Support of Public Policy

2. DECISION SUPPORT SYSTEMS FOR IWRM

2.1. DSS Elements Decision Support Systems (DSS) are technical tools intended to provide valid and sufficient information to IWRM decision makers. A typical DSS for IWRM includes five main components (Figure 2): data acquisition system, user-data-model interface, database, data analysis tools, and a set of interlinked models. The data acquisition system consists of all means by which generic data are collected and made available to IWRM through the DSS. Data may be collected by conventional sensors (rain-gages, stream-gages, etc.), remote sensors (satellite, radar), as well as by manual compilation efforts (e.g., surveys, interviews,

and literature reviews). The purpose of the user-data-model interface is to (1) transfer the data to the database, and (2) provide easy and meaningful access to data, data analysis tools, and application programs (models). The database is the depository of all data acquired by the data acquisition system and generated by the data analysis tools and application programs. The data analysis tools provide user-friendly means to visualize and analyze various data sets. Geographic Information Systems (GIS) packages are especially important for the visualization and analysis of geo-referenced (spatial) data. Lastly and most importantly, the purpose of the DSS models is to quantify the holistic response of the water resources system to alternative scenarios of basin development, hydrology, water use levels, and management policies.

Fig. 2: Typical DSS Elements In some form, the above-described DSS elements exist in most useful information and modeling tools for water resources planning and management. However, beyond these elements, the effectiveness of a DSS depends on its ability to provide valid, sufficient, and consistent information at all levels of the IWRM process. A generic DSS structure that has the potential to achieve this goal is described next.

2.2. DSS Conceptual Structure for Systems with Multiple Objectives, Temporal and Spatial Scales, and Decision Makers

The key DSS challenge is to support decisions that are made by several decision makers, pertain to various temporal and spatial scales, and concern multiple stakeholder groups and water uses. Experience with the development and application of such decision support systems

in various regions and institutional settings suggests that the multilayer structure illustrated in Figure 3 emerges naturally and can meet this challenge. This DSS structure includes multiple interconnected layers each of which models the system at a particular temporal and spatial scale, addresses a certain subset of objectives, and involves an appropriate subset of decision makers and stakeholder groups. The linkages among and within the layers ensure that (1) system data, models, and outputs provide an integrative understanding of the system response and (2) decision maker choices are prioritized and implemented consistently as the IWRM process evolves. While this structure is generic, the layers appropriate for each application are system specific. In the example of Figure 3, the three modeling layers shown include (1) near real time models (with an hourly time resolution over a horizon of one day), (2) short/mid range models (with a daily, sub-daily, or hourly resolution and a horizon of one month), and (3) long range models (with a 10-day or monthly resolution and a horizon of one to two years). The DSS also includes an assessment model which evaluates the system response under various inflow scenarios, system configurations, demand scenarios, and policy options. This DSS is designed to operate sequentially. In a typical application, the long range models are activated first to consider long range planning issues such as appropriate water conservation strategies for the upcoming one to two years. In carrying out these evaluations, the long range models utilize climate-hydrologic and demand forecasts with a 10 day or monthly resolution. A central part of this analysis is to quantify all tradeoffs in which the planning authorities and system stakeholders may be interested. The tradeoffs quantify the benefit-cost (or impact) relationships among (1) water users and (2) water uses, and delineate the capacity of the system to meet the various demands placed upon it. Interesting tradeoffs pertain to benefits and costs that would accrue to upstream and downstream users or to specific water uses (e.g., power and agriculture, or agriculture and ecosystem health) if relative water shares were to change. The tradeoffs are provided to planning authorities and stakeholders (water, power, environmental protection, etc.) to use in their decision process. After a consensus is reached, key decisions are made on relative water shares, monthly releases, energy generation, lake levels, and reservoir coordination strategies.

The short/mid range models are activated next to consider system operation at finer time scales. The objectives addressed here are more operational rather than planning and may include flood control, power plant scheduling, irrigation management, and environmental flow regulation. This model uses hydrologic and demand (water and power) forecasts with a daily, 6-hour, or hourly resolution and can also quantify the relative importance of upstream versus downstream flooding risks, power generation versus flood control, water supply versus fishery management, and other applicable tradeoffs. Such information is provided to the forum of management authorities and stakeholders to use in their effort to reach consensus on the most preferable operational policy. Such policies are revised as new information on reservoir levels, flow forecasts, and demands becomes available. The model is constrained by the long range planning decisions, unless current conditions indicate that a departure is warranted. The near real time models are activated last to determine the hour to hour operations (e.g., turbine dispatching and flow regulation). All decisions made by the upper levels of the IWRM process are realized at this stage. In developing the above-described DSS, particular attention must be placed on ensuring consistency across modeling layers, both with respect to physical system representations as well as with respect to decisions made. Consistency with respect to decisions is achieved by constraining lower layers to stay within the limits established by the upper layers. Thus, the purpose of the lower layers is to distribute the bulk upper layer decisions (e.g., monthly volumes or energy amounts) at finer temporal scales (e.g., daily, 6-hour, or hourly releases and energy generation) such that system objectives of these finer resolutions are best met. Consistency with respect to system representation is achieved by (1) utilizing models of increasing resolution (temporally and process-wise) and (2) using lower level models to derive (off or on line) aggregate performance functions associated with potential upper layer (bulk) decisions. An example of such an aggregate function is the relationship of power versus plant discharge that provides optimal plant generation as a function of reservoir level and total plant discharge. Such functions can be derived by the real time (turbine dispatching) models by determining the optimal turbine loads corresponding to particular reservoir level and plant discharge combinations. These functions are derived for each system plant and are

provided to the short/mid range models to ensure that they “know” the power that will actually be generated from a particular level of hourly plant release. Similar aggregate performance functions are derived by each modeling layer for each system use and are communicated to the upper DSS layers. In this manner, each layer has an accurate and consistent representation of the benefits and implications of its decisions. The three modeling layers in this particular example address planning and management decisions for a given system configuration. The scenario/policy assessment model addresses

longer term planning issues such as increasing demands, infrastructure change (basin development options), water sharing compacts and policies, potential hydro-climatic changes, and mitigation measures. The approach taken in this DSS layer is to simulate and inter-compare the system response under various inflow, demand, development, and management conditions. Altogether, the DSS provides a comprehensive modeling framework responsive to the information needs of the IWRM process at all relevant time scales.

Fig. 3: A Generic DSS Structure

3. NILE DSS The Nile River Basin (Figure 4) covers about 10% of the African continent and is spread over ten countries (Burundi, Congo, Egypt, Eritrea, Ethiopia, Kenya, Sudan, Tanzania, Uganda, and Rwanda). Almost all Nile water is generated on an area covering 20 percent of the basin, while the remainder is in arid or semi-arid regions. Egypt and Sudan are almost totally dependent on the Nile for their water uses. Most other Nile countries are close to water stress, if not already below the water scarcity threshold of 1000 m3 of water per inhabitant per year. Water stress is compounded by rapid population growth,

occurring at nearly twice the average global rate. Hence severe water scarcity conditions are looming over most Nile countries. Nile Basin economies are heavily dependent on agriculture which accounts for more than half of the gross domestic product and employs more than 80% of the workforce. However, lack of water supply infrastructure, climate variability, and poor cultivation practices have seriously restrained, if not completely halted, economic growth. These complex challenges are at the forefront of an unfolding initiative by the Nile Basin nations to set forth equitable and lasting water development and utilization agreements. The

goal of the Nile Basin Initiative (NBI) is poverty alleviation and sustainable economic growth. Thus, water sharing is intended to facilitate the creation of efficient markets for food and energy and stimulate environmentally-sound industrial and economic growth. However, effective policy dialogue requires that the countries assess and weigh the benefits and impacts of various water development and management strategies accrued to themselves and other Nile partners. Pre-requisite elements in this process are the existence of an institutional cooperative framework, information and modelling systems, and the technical expertise to use them.

The Nile Decision Support System (Nile DSS) is the outgrowth of several projects implemented in the course of the last 10 years. These were collaborative efforts of the Georgia Water Resources Institute at Georgia Tech, the Nile Governments and their agencies, and various international organizations including the Food and Agriculture Organization of the United Nations (FAO) and the World Bank. The Nile DSS includes planning and operational components developed for and used by individual countries as well as basin planners. Operational management systems have been developed and used in Egypt (High Aswan Dam) and Uganda (Lake Victoria), while a planning DSS was recently completed for all Nile countries. In what follows, we describe the elements of the latter. The operational components are being published separately. The following general principles governed the development of the Nile DSS:

• The data of the Nile DSS should be shared and agreed upon by the Nile Basin nations;

• The Nile DSS should be based on

sound and current scientific and engineering approaches able to handle the Nile Basin size, complexity, and range of development and management options; It should also include functionalities useful for users of varying technical backgrounds and experience from novice to advanced;

• The Nile DSS should be a neutral

decision support tool; Thus, its overriding purpose should be to objectively assess the benefits and tradeoffs associated with various water development and sharing strategies that may interest the Nile Basin partners

individually or as an interdependent community of nations;

• The Nile DSS should be sustainable

and adaptable as future needs arise; The implications of this are twofold: First, the Nile DSS should be based on widely supported computational technology and should be expandable to incorporate new data and applications; Second, effective technology and know-how building mechanisms should be implemented during the Nile DSS development as well as for the long term.

The Nile DSS follows the concepts illustrated in Figures 2 and 3 and includes a database, a set of models (for river simulation and management, agricultural planning, hydrologic modeling, and remote sensing), and a user-data-model interface.

Victoria

Lake

Lake Albert

Lake Edward

Lake Kivu

Lake Rudolph

Bor

Sennar

Falls

Gebel El Aulia DamKhartoum

Merowe

AswanToshka Depression

Mediterranean Sea

EGYPT

SUDANERITREA

KENYA

TANZANIA

UGANDA

Lake Kyoga

BURUN DI

Port Said

RUAND A

La ke Ta na

DRC

Sudd

Red Sea

White Nile

Lake Nasser

Blue NileSobat

Equator

Main Nile

Melut

Owen

Pakwatch

ETHIOPIA

Malakal

Khashm El Girba

DiemBorder

Akobo

KaradobiMabil

Mendaia

Pibor

Nimule

Baro

Tonj

Lol

Jur

Atbara

Dongola

Nag Hammadi

Delta Brg.

Bahr El Chazal

Mongala

Roseires

Alexandria

Aqaba

Isna

DamiettaZifta

Rosetta

Cairo

Idfina

Asyut

Fig. 4: The Nile River 3.1. Database and Interface The Nile DST database is an object-oriented, databasing structure developed to (1) house all types of data (existing as well as future) required by a comprehensive water resources decision

support system and (2) to optimize data entry, access, visualization, and analysis. To support the process of water resources planning and management, the data base is designed with the ability to grow, namely, to accept new data, regardless of its type and size. Further, the database tools are capable of visualizing and analyzing the data in efficient and meaningful ways. Database Contents: The Nile DST database is comprised of several national databases and is of considerable size. Each Nile country has painstakingly compiled station data with measurements of more than 30 hydro-climatic parameters. In addition, the project has compiled 10 years of remotely sensed data that covers the entire basin. The temporal resolution of the remotely sensed data is 30 minutes, and the special resolution is approximately 5 km x 5 km. All together, this data represents nearly 37 GB of information. Data Visualization: The data visualization tool in the Nile DSS provides a seamless system to look at all databases. This is a tree-style exploring tool (data tree) that shows the entire contents of the Nile DST database and allows the user to navigate to greater and greater levels of detail. Due to the database size, the data tree is useful in providing a better understanding of the database. Each database has a geo-referenced component and a time series component. The geo-referenced data is viewed in the mapping tool, which holds a GIS. The mapping tool incorporates the ESRI Incorporated Map Objects to ensure compatibility of the system with industry standard GIS files. The time series data is viewed in the charting tool, which features a powerful chart and aggregation and statistics calculators. Together, the charting tool, mapping tool, and the data tree work seamlessly to provide the user with an ability to view any piece of information in the system quickly and meaningfully. Data Analysis: The data analysis tool provides extensive data manipulation capabilities for current and future applications. The user can instruct the Nile DSS to take information from its databases in user specified forms, operate on the data, and construct maps of the output. One example of this is the generation of mean areal precipitation (MAP) and evapotranspiration estimates over user-specified areas. The tool allows the user to save a graphical map of the analysis and revisit/continue the work at a later time.

3.2. River and Reservoir Simulation and Management

The Nile DSS River Simulation and Management system aims at simulating the Nile response under different hydrologic, development, and management scenarios. Thus, its overriding purpose is to objectively assess the benefits and tradeoffs associated with various water development, sharing, and management strategies that may interest the Nile Basin partners individually or as an interdependent community of nations. Tradeoffs exist among water uses in the same country and across the Nile countries. The river basin planning and management Nile DST component has several unique features: This module includes extensive data in five major categories: (a) River network configuration, (b) river hydrology, (c) existing and planned hydro facilities, (d) water use, and (e) reservoir/lake regulation rules. Data can be viewed, added, or modified as necessary through a user-friendly interface. The actual river system is represented by a network of river nodes, reaches, and reservoirs, each with its own attributes. River nodes represent locations of local inflow and/or water withdrawals and returns. River reaches represent physical river segments and their water transport characteristics. Reservoirs represent man-made or natural lakes that may support various water uses including water supply, flood control, drought management, hydropower, and wetland protection, among others. Models: This module integrates streamflow forecasting, river and reservoir simulation, and reservoir management. Ten-day streamflow forecasts are generated at key basin nodes including Lakes Victoria, Kyoga, and Albert, Torrents, Bahr el Ghazal, Sobat, the Blue Nile, Dinder, Rahad, and Atbara. The forecasts have the form of equally likely realizations reflecting historical streamflow characteristics such as seasonal and long-term variability. Streamflow forecasts are generated by the hydrologic watershed models, or via statistical procedures where hydrologic models are unavailable. The river and reservoir routing models simulate the movement of water through the river reaches and quantify transmission losses and time lags. The routing models are based on statistical or physically-based relationships (depending on available information) and incorporate model error characterizations. Reservoir and lake outflow through hydropower facilities and spillways is modeled with

sufficient detail for use in operational applications. The purpose of reservoir management is to determine release sequences from each system reservoir such that sub-basin and basin-wide objectives are met as best as possible. System objectives include meeting water supply targets and avoiding water shortages, minimizing losses, maintaining land use patterns (Sudd), regulating river flows, avoiding spillage, and generating as much firm and average energy as possible. The task of the reservoir control module is complicated by the system size, non-linear response, and intrinsic uncertainties. The optimization process is based on the Extended Linear Quadratic Gaussian (ELQG) control method (developed by Georgakakos and associates, 1987 through present), a trajectory iteration optimization algorithm suitable for multidimensional, dynamic, and uncertain systems. Applications: The Nile-DST river simulation and management model can be used to provide answers to various important questions. Typical applications are listed below: • Value of various regulation, hydro-power,

and irrigation projects along the White, Blue, and Main Nile branches; Such assessments could quantify the incremental benefits from individual development

projects as well as the combined benefits from various project configurations;

• Implications of reservoir regulation policies for local, upstream, and downstream riparians;

• Marginal value (gain or loss) of irrigation with respect to hydropower at various basin locations;

• Irrigation versus hydropower tradeoffs for each nation, region, and the entire basin;

Impacts of flow regulation on wetlands. The Nile-DST utilizes several assessment criteria of interest to the Nile Basin nations. These criteria include: (i) severity and frequency of shortages with

respect to user-specified water supply targets;

(ii) water withdrawals and losses over user-selected regions and times of the year;

(iii) reservoir and lake level drawdown and spillage statistics;

(iv) in-stream flow availability at user-selected river nodes and reaches;

(v) flood and drought severity and frequency; (vi) annual and firm energy generation

statistics; (vii) seasonal and permanent extent of wetlands. •

Fig. 5: The graph on the left depicts the flow frequency curves of the Blue Nile at Khartoum under the

baseline (blue) and the Ethiopian hydropower development scenarios (green). The graph on the right compares annual average energy generation under the same two scenarios. From left to right, the bars correspond to generation in Egypt, Ethiopia, Sudan, and the Lake Victoria Countries.

Annual Energy Comparison

0

5000

10000

15000

20000

25000

30000

35000

Buru

ndi

Con

go

Egyp

t

Eritr

ea

Ethi

opia

Keny

a

Rw

anda

Suda

n

Tanz

ania

Uga

nda

GW

H

BaselineEthiopia Power Development

Flow Frequency Curves at Khartoum

0

100

200

300

400

500

600

700

800

900

0 20 40 60 80 100

Probability of Exceedence (%)

Flow

(mcm

/day

)

BaselineEthiopia Power Development

Figure 5 presents an assessment of the flow and hydropower impacts of two development scenarios in the Blue Nile Basin. The first scenario (blue line) is the baseline of current conditions. The second scenario (green line) assumes that 4 large hydropower projects have been built in Ethiopia and are operated using dynamic inflow forecasts and multi-reservoir control methods. The results show that flow changes can be significant with benefits both at the high and the low ends (reduction of excessive floods as well as low flow augmentation). Furthermore, energy generation in Ethiopia would increase very substantially with the potential to benefit not only Ethiopia but also Sudan and Egypt. Other benefits (not shown) include drastic reductions of the drought risk for Sudan and Egypt. 3.3. Agricultural Planning Agriculture is a major water consumer in the Nile Basin. It is the main source of income for a large part of the population, and many people rely predominantly on crops they grow themselves. Agricultural products are also an important source of foreign currency, as commodities like coffee, cotton, and sugarcane are exported in large quantities and sold on the world market. The Nile Basin contains many regions of high agricultural potential, some of which are already fully exploited, while others await development. Given the importance of agriculture in the basin, decision makers would benefit from reliable assessments regarding potential crop yield in undeveloped lands, irrigation needs, drought vulnerability, and the potential tradeoffs between agriculture and other water uses. The purpose of the agricultural planning model is to address these issues.

Models: The Georgia Water Resources Institute at Georgia Tech has developed a comprehensive Agricultural Planning Model (GT-AgroPlan), which combines state-of-the-science irrigation scheduling and crop yield prediction tools with a user-friendly graphical interface. Physiologically-based crop models form the agronomic simulation core of GT-AgroPlan. These models, adapted from the Decision Support System for Agrotechnology Transfer

(DSSAT; Tsuji et al., 1994), simulate the daily life-processes of crops using input data for soils, meteorology, and genetics. Currently, GT-AgroPlan includes crop models for eleven crops (maize, cassava, groundnuts, wheat, rice, sorghum, millet, barley, potatoes, soybeans, and dry beans), but five additional crops (including sugarcane) have recently been added to the DSSAT and will be incorporated into GT-AgroPlan. The models produce outputs of yield and total irrigation needs as well as daily time-series of soil moisture and biophysical parameters. GT-AgroPlan requires data on soils, terrain, hydro-meteorology, crop characteristics, and agricultural and irrigation practices. All of these data are integrated within a GIS and graphical user interface (GUI). The GUI allows program users to easily select input data scenarios for simulation by the crop models and to view results in onscreen tables and graphs. Applications: GT-AgroPlan has been used for a wide variety of analyses, such as the assessments of gains from irrigation, spatial distribution of irrigation needs, regional drought vulnerability, and regional calorie supply. Other potential GT-AgroPlan applications include investigating the consequences of deficit irrigation relative to water demand and crop production, assessing various irrigation scheduling scenarios, climate variability impacts on agriculture, and determining water-efficient and high-calorie crop combinations. Figure 6 shows a regional assessment carried out in the Lake Victoria Basin (Georgakakos et al., 2000). The purpose was to assess the vulnerability of the agricultural sector to climate variability. The figure includes results for maize assuming only rain-fed conditions. Results are presented for both growing seasons (long rains—March to May; and short rains—October to November) and for typical wet and dry years. The results show that during long rains, Kenya is immune to climate variability while Tanzania, Rwanda, Burundi can experience serious shortages. During the short rains, Kenya becomes most vulnerable with the rest of the basin exhibiting mild potential deficits. Such information is useful in formulating equitable water and benefit sharing strategies.

Fig. 6: Agricultural Assessments in the Lake Victoria Basin 3.4. Hydrologic Modeling Hydrologic models are necessary to translate climatic forcing (rainfall and temperature) to lake rainfall, evaporation, and watershed inflow. The temporal scales of interest are ten days for water resources planning and one day for operational applications; the applicable spatial scale varies from a few hundred to several thousand square kilometers.

Models and Applications: The Nile DSS and its derivatives include Sacramento type watershed models (Peck, 1976) that simulate the hydrologic processes of surface and subsurface runoff through physically-based conceptual elements. Such models are very useful for applications that require flow predictions over daily to sub-daily time steps. Another hydrologic model type that is also part of the Nile DSS is especially suited for medium size (1,000 to 10,000 km2) to large watersheds (larger than 10,000 km2), as well as for weekly, ten-day, and monthly time steps. The model has been developed by Georgakakos and Yao, 2000, and has successfully been tested in the Southeastern US and elsewhere. Figure 7 shows a comparison of observed and simulated flows for the Nzoia basin in Kenya and indicates that predicted and observed streamflows exhibit good agreement. Some hydrologic model features are summarized below:

• Input data include watershed rainfall

(weekly, ten-day, or monthly) and temperature or evapotranspiration. Model calibration additionally requires observed streamflows at the watershed outlet. A record of at least 10 years is recommended. Shorter records can still be used but may not capture decadal climate variability and basin response;

• Model output includes streamflow at the watershed outlet and spatially-averaged soil moisture. The latter is also important for agricultural planning applications;

• The model (a) simulates the watershed water balance processes (rainfall, evapotranspiration, streamflow, and soil moisture storage) on weekly, ten-day, or monthly time steps, and (b) uses a historical analog approach to determine streamflow from current conditions of rainfall, evapotranspiration, and soil moisture;

• A particular basin can be sub-divided into smaller sub-basins leading to a semi-distributed model implementation. However, this would be appropriate for sub-basins with streamflow measurements;

• This model is also used to assess the sufficiency of hydrologic data and develop data monitoring plans.

Fig. 7: Observed vs. Simulated streamflows for Nzoia Basin, Kenya An additional hydrologic module has recently been added to simulate the seasonal and permanent extent of the Sudd wetlands (Sutcliffe and Parks, 1999) in southern Sudan. The wetlands are critical resources for the local population, fauna, and flora, and can adversely be impacted by water conservation projects and reservoir regulation policies. The wetland module enables the DSS river and reservoir simulation and management models to identify policies that minimize wetland impacts . 3.5. Remote Sensing Rainfall drives the response of the Nile Basin at all spatial and temporal scales, and its reliable estimation is a requisite DSS component. Traditional rainfall estimation techniques rely on conventional raingages. However, in many areas of the world, including large portions of the Nile Basin, the raingage network is sparse, and satellite data provide an attractive alternative. The purpose of this Nile-DSS model is to combine existing raingage data with data from operational satellites (currently received by the Nile Basin hydro-meteorological agencies) to provide reliable rainfall estimates over the Nile sub-basins.

Models: At present, no universal method exists for satellite-based rainfall estimation, but several techniques have been developed and are currently used. Georgia Tech’s (GT’s) experience in the Lake Victoria Basin (De Marchi and Georgakakos, 2000) has shown that

rain from convective storms (most common over the Nile Basin) can successfully be determined for daily time intervals using a combination of Meteosat visible and infrared (VIS/IR) images. This method also captures rain produced by non-convective cells, though with somewhat less accuracy. For a 10-day time scale, this approach provides reliable total rainfall estimates with correlation to ground stations as high as 0.8 to 0.9. The principle idea of the GT rainfall estimation approach is to identify characteristic “signatures” of rain-producing events in the VIS/IR data signals. More specifically, a typical tropical convective storm exhibits a certain visible and infrared signal pattern. Initially, the cloud is low and relatively thin, resulting in a high infrared temperature and low visible signal. As the convective storm matures and its top approaches the tropopause, the cloud gets thicker and colder. This leads to a rising visible count and falling infrared temperature. Then, as the convective cell dissipates, the visible count decreases while the infrared temperature increases. This pattern represents a distinct signature for all rain-producing cells. The Georgia Tech rainfall estimation method exploits this characteristic relationship between rain-producing storms and Meteosat infrared (IR) and visible (VIS) images in a two-step procedure. First, the existence of convective cells (which produce the highest rain rates) is

determined for each satellite pixel (5x5 km2) by screening the IR/VIS signals exhibiting these rain-indicating signatures. Two specifically trained neural networks (one for daytime and one for nighttime) have been developed for this task. Then, based on the temporal and spatial distribution of convective cells and other characteristics of the cloud system, the daily mean areal precipitation (MAP) is estimated over the areas of interest. The method has been calibrated and validated for three years in several locations around the Nile Basin. These tests show that the approach can successfully identify strong convective cells and provide reliable MAP estimates for daily time steps (correlation with ground measurements of 0.7), ten-day time steps (correlation of 0.80 – 0.9), and monthly time steps (correlation of 0.87 – 0.98). The good performance of this approach is attributed to the use of the full VIS/IR signal information rather than simply considering specific threshold levels.

Figure 8 illustrates the value of satellite-based rainfall estimation over Lake Victoria. The top graphs present rainfall estimates based on raingages (located at the lake shores), while the lower graphs make use of the GT satellite rainfall estimation scheme. The resulting rainfall distributions are rather different with the satellite estimates being considerably wetter. The lake level changes experienced during the same period provide strong evidence of the accuracy of the satellite estimation procedure. The above-described Nile DSS modules can express the response of the Nile system in terms of river flow, water supply, food production, and energy generation. Building on these developments, the next modeling phase will introduce models that can translate physical outputs into economic and social benefits and impacts. Furthermore, a water quality component is planned to enable fully integrated assessments.

Fig. 8: Rainfall Estimation over Lake Victoria with and without Satellite Information

4. CONCLUSION Decision support systems are integral parts of IWRM processes facilitating the use of science and technology advances in public policy. Although generic DSS development principles exist, much is system specific and a thorough understanding of the interdependence of natural processes, water uses, institutional setting, and decision maker objectives is necessary for a successful DSS design. Participatory DSS Development: As indicated earlier, DSS generate a wide array of quantitative system response measures including water use tradeoffs, risks, and benefits. A key effort during the DSS development phase is to determine the necessary and sufficient information that decision makers need to make good decisions. This information set is expected to vary by decision type (planning, management, or near real time), management agency, and stakeholder group, and significant interaction should take place with the decision makers defining the most suitable informational content and form. In view of the inter-disciplinary DSS application scope, the formation of an inter-disciplinary DSS expert group is recommended to participate in all DSS development phases, to ensure the effective transfer of the new technology to user groups, and to contribute to DSS sustainability beyond the development phase. Institutional Implementation Aspects: Decisions in the Nile Basin involve a multitude of agencies and involve various stakeholders. The DSS provides some useful information to all parties. However, some modeling components are more relevant to particular decision maker groups, and there is a need to consider how to best implement this system in view of the current institutional framework. The DSS structure of Figure 3 suggests that higher DSS modeling layers are more suitable for agencies with planning functions (e.g., environmental, water, and energy planning departments), while lower DSS layers are more suitable for agencies with operational mandates (e.g., flood control, power scheduling, and water distribution). This issue should be addressed in close collaboration with all relevant agencies and stakeholders by reviewing past practices and experience, and by developing a shared vision institutional decision framework. In the assessment phase of the project, the operational experience with the DSS and the agreed-upon inter-agency decision process should be documented and possible enhancements should be made.

Long Term Sustainability: The effective use of any DSS requires that a sufficient number of skilled "in-house" staff be trained. Those that will be using the DSS should be involved in its development to ensure the adequacy of data, models, assessment capabilities, and user interfaces, and to understand all underlying assumptions and limitations. “Know-how” transfer should be facilitated through various means, including technical reports, instructional manuals, video tapes describing model theory and use, training missions to centers where the DSS technology is developed or implemented, attendance of graduate level courses at academic institutions, and workshops for high level government officials as well as agency engineers. On a longer time scale, the key to sustainable water resources management lies in the existence of well educated, trained, and motivated people. Thus, partnerships between applied research centers and professional organizations that create competent human resources should be an integral part of IWRM processes.

ACKNOWLEDGEMENTS The work referred to herein has been sponsored by various organizations. Specifically, I am grateful to FAO, World Bank, and US Agency for International Development (AID) for funding and continuing to support my involvement in various DSS development efforts. Several of my graduate students and associates contributed to the development of the Nile DSS including Huaming Yao, Kelly Brumbelow, Carlo De Marchi, Stephen Bourne, Lori Visone, and Amy Tidwell. I am very fortunate to have the opportunity to teach and learn from such gifted individuals. Many people from the Nile Basin Countries have contributed to the Nile DSS. I specifically wish to thank the team of National Modelers from all ten Nile countries that have worked closely with us in developing and testing the various Nile DSS modules. Their continued interest, collaboration, and friendship are highly rewarding. Lastly, I wish to thank Professor Rodolfo Soncini-Sessa for inviting me to participate at the IFAC workshop on Modeling and Control for Participatory Planning and Managing Water Systems that was held in Venice, Italy.

REFERENCES Brumbelow, K., and A. Georgakakos (2000). An Assessment of Irrigation Needs and Crop

Yield for the United States under Potential Climate Changes. Journal of Geophysical Research – Atmospheres 106(D21), 27383-27406, 2001.

Brumbelow, K. and A. Georgakakos, "Agricultural

Planning and Irrigation Management: The Need for Decision Support,” The Climate Report, 1(4), 2-6, 2001.

De Marchi, C., and A. Georgakakos, “A satellite-

based rainfall estimation method for the Lake Victoria basin,” Remote Sensing and Hydrology 2000, Proceedings of a symposium held at Santa Fe, New Mexico, USA, April 2000, (IAHS Publ. no. 267, 2001.

Georgakakos, A., and D. Marks D (1987). A New

Method for the Real-Time Operation of Reservoir System, Water Resources Research, 23 (7), 1376-1390

Georgakakos, A., (1989a). Extended Linear Quadratic Gaussian Control for the Real-Time

Operation of Reservoir Systems. Esogbue A. In the book entitled DYNAMIC PROGRAMMING FOR OPTIMALWATER RESOURCES SYSTEMS ANALYSIS. 329-360. Prentice Hall Publishing Company, New

Jersey. Georgakakos, A., (1989b). Extended Linear

Quadratic Gaussian (ELQG) control: Further extensions, Water Resources Research, 25 (2), 191-201.

Georgakakos, A. (1991). Computer-Aided Management of the Southeastern U.S.

Reservoir System. Decision Support Systems, D.P. Loucks, ed., NATO ASI Series, Vol. G 26, 407-428

Georgakakos, A., (1993). Operational tradeoffs in\

reservoir control. Water Resources Research, 29 (11), 3801-3819

Georgakakos, A., and H. Yao (1995). A Decision

Support System for the Equatorial Lakes. Technical Report No. GIT/CEE-HYDRO-95-7, School of Civil and Environmental Engineering, Georgia Tech, Atlanta.

Georgakakos, A., H. Yao, and Y. Yu (1995) Control Models for Hydropower System Analysis and Operation, Technical Report, School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, 158.

Georgakakos, A., H. Yao, and Y. Yu (1997a). A Control Model for Dependable Hydropower Capacity Optimization. Water Resources Research, 33 (10), 2349-2365.

Georgakakos, A., H. Yao, and Y. Yu (1997b). Control Models for Hydroelectric Energy Optimization. Water Resources Research, 33 (10), 2367-2379.

Georgakakos, A., H. Yao, and Y. Yu (1997c). A Control Model for Hydroelectric Energy Value Optimization. ASCE Journal for Water Resources Planning and Management, 123 (1), 30-38.

Georgakakos, A., H. Yao, M. Mullusky, and K. Georgakakos (1998). Impacts of Climate Variability on the Operational Forecast and Management of Midwestern Water Resources Systems. Water Resources Research, 34(4), 799-821, 1998.

Georgakakos, K., N. Graham, and A. Georgakakos, “Can forecasts accrue benefits for reservoir management? The Folsom Lake Case Study,” The Climate Report, 1(4), 7-10, 2000.

Georgakakos, A., H. Yao, K. Brumbelow, C. DeMarchi, S. Bourne, and M. Mullusky (2000). The Lake Victoria Decision Support System, prepared for the Food and Agriculture Organization of the United Nations, Technical Report No. GWRI-2000-1, Georgia Water Resources Institute and Georgia Tech.

Georgakakos, A., H. Yao, K. Brumbelow, C. DeMarchi, S. Bourne, L Visone, and A. Tidwell (2003). The Nile Decision Support Tool, prepared for the Food and Agriculture Organization of the United Nations, 6 Technical Reports, Georgia Water Resources Institute and Georgia Tech.

Georgakakos, A., and H. Yao (2000). Climate Change Impacts on Southeastern U.S. Basins, 1, Hydrology. USGS Open File Report #00-334, July 2000.

Peck, E.L., (1976). Catchment Modeling and Initial Parameter Estimation for the National Weather Service River Forecast System, NWS HYDRO-31, Office of Hydrology, NWS, Silver Spring, Maryland.

Sutcliffe, J.V., and Y.P. Parks (1999), The Hydrology of the Nile, IAHS Press, Institute of Hydrology, Wallingford, IAHS Special Publication no. 5.

Tsuji, G.Y., G. Uehara, and S. Balas (eds.) (1994).

Decision Support System for Agrotechnology Transfer, Version 3. University of Hawaii, Honolulu.

Yao, H, and A. Georgakakos, "Assessment of Folsom Lake Response to Historical and Potential Future Climate Scenarios," Journal of Hydrology, 249, 176-196, 2001.