Hydrologic Modeling For Flood Warning And Reservoir Management

Juan Bors, Marcelo Uriburu Quirno, Dora Goniadzki, Daniel Vila

Instituto Nacional del Agua (National Water Institute) Ezeiza, ARGENTINA

Abstract An application of hydrologic modeling is implemented for flood warning and reservoir management. The study watershed is the Ibicui River basin (42,900 km2) in Brazilian territory. The Ibicui River is a major tributary of the Uruguay River, which constitutes a bi-national boundary and is crossed by a large hydropower dam downstream from the confluence. A continuous lumped hydrologic model (based on the differential Sacramento) has been calibrated, fed by daily rainfall and potential evapotranspiration (as a function of monthly mean temperature). Rainfall records are not available in real time, which is necessary for operational forecasting. Consequently, it is proposed the use of a satellite rainfall estimation algorithm (the South-American version of the Hydro-Estimator technique). Estimates are readily available in quasi-real time. Monthly mean temperatures are estimated through reanalysis made by NCEP / NCAR. The operational use as a forecasting tool necessarily requires not only observed but predicted inputs, in order to expand the forecast horizon. Rainfall and temperature outlooks for South America, made available by the NCEP, or forecasts with the ETA model are useful for extending the input series. The results are promising in the sense that a reliable forecast may be issued from modeling with input data readily downloadable from the Internet. Sparsity of meteorological networks and unavailability of field records in real time are not a limiting factor for operational hydrology.

Keywords: hydrologic model, continuous modeling, rainfall estimation, flood warning systems, reservoir management

Introduction The National Water Institute (INA) of Argentina is responsible for the operation of the Hydrologic Information and Warning System for the Del Plata Basin in Argentina. The Del Plata Basin is one of the largest watersheds in South America (second after the Amazonia) and in the world. With an area of more than three million square kilometers, it is shared by five countries: Brazil, Paraguay, Bolivia, Uruguay and Argentina. Its main rivers are some of the largest in the world and their floods can have a tremendous socio-economic impact, as observed along the history. The Hydrological Warning System was originated after the devastating floods of 1982-83 and has been providing a permanent service ever since. Through data reception, analysis and transmission, this service is devoted to deal with operational hydrometeorology, basin monitoring, and short-, medium and long-term hydrological forecasts. Through time, consolidation of the system was achieved by improving data collection, diagnosis and forecast and by strengthening the relationship with users. As part of these efforts, this paper presents the application of hydrologic modeling for flood warning and reservoir management. The basin of choice is the Ibicui River basin, a watershed of 42,900 km2 in Brazilian territory. The Ibicui River is the largest tributary of the Uruguay River, a bi-national boundary and is crossed by Salto Grande, a large hydropower dam downstream from the confluence. Future steps are to implement a similar approach to other catchments in the Del Plata Basin, with the purpose of forecasting or for simulation of lateral inflows to larger rivers. A continuous lumped hydrologic model (based on the differential approach of the Sacramento model) has been calibrated, fed by areally-averaged rainfall estimates and temperatures. Although a (relatively reduced) number of rain gauges are installed inside and near the catchment, permanent availability of rainfall records strongly depends on formal agreements with network operators abroad, which conditions the continuous operation. Therefore, these data are not to be supplied in real time, what would be necessary for operational forecasts. Consequently, it is proposed the South-American version of the Hydro-Estimator technique. This is a satellite rainfall estimation algorithm developed by the US National Oceanic and Atmospheric Administration / National

Environmental Satellite, Data, and Information Service (US NOAA / NESDIS). These estimates are readily available in quasi-real time. Regarding temperatures as an input for the potential evapotranspiration, only monthly mean values are needed so that they are not a concern in terms of availability. Reanalyses made by NOAA through the National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP / NCAR) provide good estimations and are easily accessible. In order to expand the forecast horizon, forecasted inputs are needed. Different sources for weather forecast are proposed, with different horizons and resolutions. Precipitation forecast, and rainfall and temperature outlooks for South America, developed by NCEP are possible solutions. Study basin The Ibicui River basin, with an area of 42,900 km2, is fully inserted in Brazilian territory (State of Rio Grande do Sul). Its centroid is located at 29 40 S and 55 20 W. The elevation ranges from 100 m to 550 m above mean sea-level. The time of concentration is approximately five days. The mean annual rainfall is about 1,600 mm, smoothly decreasing southward, between 1,800 mm and 1,400 mm (long-term values from the National Water Agency (ANA) of Brazil). The annual rainfall presents a relatively even distribution throughout the year, though autumn and spring are slightly wetter seasons. With a historical mean annual flow of 880 m3/s (650 mm in the year), the specific discharge results in 20.6 l/s/km2 (data source, ANA; period 1955 - 2001). Approximately 40% of the annual rainfall becomes streamflow, the rest being lost as evapotranspiration and groundwater recharge. The mean annual temperature is 18.4C (from reanalysis by NCEP / NCAR, period 1949 - 2005). January is the warmest month (mean monthly temperature, 23C) and July is the coldest (mean monthly temperature, 13C). Mean annual potential evapotranspiration, estimated with the Thornthwaite formula, is approximately 880 mm. The Ibicui River flows westward and it has some major tributaries as the Ibirapuita River and the Santa Maria River. The Ibicui River is the largest tributary of the Uruguay River, which constitutes a bi-national boundary and is crossed by a large hydropower dam 200 km downstream from the confluence, i.e. Salto Grande Dam, whose reservoir has 5500 hm3 of storage. Table 1 summarizes the historical monthly and annual values of mean areal rainfall, temperature and potential evapotranspiration, and discharge at the basin outlet. Table 1 Rainfall, Temperature and Potential Evapotranspiration, and Outflow (long-term statistics) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec AnnualR (mm) 134.9 139.0 149.6 139.2 121.3 120.2 129.6 117.6 148.9 146.0 132.2 121.4 1600.0 T (C) 23.3 22.9 21.5 18.3 15.2 13.2 12.9 14.4 16.3 18.8 21.1 22.8 18.4 PET (mm) 126.5 104.1 94.3 62.6 42.5 30.3 30.3 40.2 52.8 78.1 98.6 121.6 881.9 Q (mm) 23.5 29.9 32.7 59.4 57.9 59.9 75.6 71.9 71.7 73.6 58.9 29.8 646.9 Q (cumecs) 377 531 524 983 927 992 1211 1152 1186 1179 975 478 880 Figure 1 shows the basin divide at the confluence with the Uruguay River, in the context of South America. Figure 2 presents the whole basin with the drainage network and the four rain-gauge stations available. The outlet of the study catchment is in Mariano Pinto, 90 km upstream from the confluence.

Figure 1. Ibicui River Basin in southeastern South America (Google Map, 2005) Figure 2. Ibicui River Basin

Model description The model used for the present study is based on the differential version of the Sacramento Soil Moisture Accounting Model (SAC, Burnash et al., 1973), described in different papers (Georgakakos et al., 1988, Bae and Georgakakos, 1992, 1994 and Sperfslage and Georgakakos, 1996). Still a conceptual spatially-lumped two-layer model of the soil vertical profile, it reduces the SACs complexity, gaining in robustness and ease of use in operational purposes. The fundamental modifications with respect to SAC are: (a) the differential formulation in the state space and (b) the flow routing component. Therefore, the present model proposes two components with a differential formulation for modeling the sequence of processes rainfall-runoff and flow routing. The first component simulates, in a soil column, the hydrological process (soil moisture balance) resulting from two fluxes through the soil-atmosphere interface, i.e. rainfall and evapotranspiration. The soil column is divided into two layers, a fast-responding upper layer and a slow-responding lower one. Precipitation is the primary source of water for both layers whereas water losses are accounted for by evapotranspiration: as evaporation form the upper layer and transpiration from the lower one. Consequently, model inputs for this component are precipitation and potential evapotranspiration. Unlike SAC, no sub-processes are considered within the two layers. Tension and free water contents are not accounted for separately. This formulation of grouping subsurface processes tends to reduce flexibility for applications of fine temporal resolution (hours), such as those required for small and steep catchments, with fast response. However, the time scale of the Ibicui River basin is in the order of days, thus allowing for an adequate use of the proposed model. The second component simulates the routing process along the drainage network by means of a conceptual two-equal-linear-reservoir cascade at the basin outlet. The input to the cascade is given by the first component output, i.e. the runoff volume produced by the rainfall-runoff model. The resulting outflow hydrograph from the cascade is finally the model output. Following the notation by Guetter (2000), the state equations governing the processes read:

INTETPCSRPdt

dX = 11 (1)

GWETPCdt

dX = 22 (2)

( ) 33 XBFSRdtdX += (3)

434 XX

dtdX = (4)

where:

21 / XX : Volume of water in the soil upper / lower layer [L] (state variables),

43 / XX : Water stored in the first / second linear reservoir of the cascade [L] (state variables), P : Rainfall intensity [L / T],

SR : Surface runoff [L / T], PC : Percolation rate [L / T],

1ET : Evaporation rate from the soil upper layer [L / T], INT : Interflow, flow from the soil upper layer to base flow [L / T],

2ET : Transpiration rate from the soil lower layer [L / T], GW : Groundwater flow [L / T], BF : Base flow [L / T], : Inverse of reservoir recession constant [1 / T]. a) Surface runoff is a direct response to rainfall, produced only by the soil upper layer. It is computed as: ( ) 1011 /. mXXPSR = (5) where 01X , the water-holding capacity of the soil upper layer [L] and 1m , the surface runoff exponent [dimensionless] are two of the model parameters. The other variables are as defined above. b) Surface evaporation rate from the soil upper layer is computed as the product of the potential evapotranspiration rate and the water availability in the layer: ( )0111 /. XXPETET = (6) where PET is the potential evapotranspiration rate [L / T] along the computational time step. c) The process of percolation, which represents the water transfer from the upper to the lower layer of soil, is computed as a non-linear function of their storages (Peck, 1976), such as: ( ) ( )0110222023 /./1.1.. 2 XXXXCXCPC m += (7) where:

02X : The water-holding capacity of the soil in the lower layer [L],

3C : Base flow recession rate [1 / T],

2C : Percolation function coefficient [dimensionless],

2m : Percolation function exponent [dimensionless], These four variables are model parameters while the rest are as previously defined.

Subsurface flow 023.XC corresponds to the lower layer outflow in saturation conditions. The parameters 2C and 2m control the percolation rate when the lower layer is unsaturated. Percolation increases with the ratio 011 / XX , that is, with soil moisture in the upper layer approaching the water-holding capacity of the layer. d) Interflow is taken proportional to the first state variable, such as:

11 XCINT = (8) where 1C , the interflow recession coefficient [1 / T], is a model parameter. The other variables, as previously defined. e) Transpiration from the soil lower layer is computed as:

( )( ) 302212 /. mXXETPETET = (9) where 3m is a model parameter that represents the transpiration function exponent [dimensionless],

and the other variables are as previously defined. f) The groundwater flow is calculated (with predefined variables) as:

23.XCGW = (10) g) The base flow is proposed as a function of both interflow and groundwater flow, such as:

( ) INTGWBF ++= .1 1 (11)where [dimensionless] is a model parameter such that )1/( +GW contributes to the base flow and

)1/(. +GW recharges the aquifer. The other variables are as defined above. h) The flow routing process along the drainage network is modelled by a two equal reservoir cascade, so that:

( ) 33 XBFSRdtdX += (12)

434 XX

dtdX = (13)

3,4=i ,ii XQ = (14)where 3Q and 4Q [L / T] are the outflows from the first and the second conceptual reservoirs, respectively and , the inverse of reservoir recession constant, is a model parameter. The other variables, as defined above. Model implementation The time step adopted for the modeling is one day. It is in accordance with data availability and is sufficiently short compared to the catchment scale (one day is approximately one fifth of the time of concentration) and to the duration of significant storm events (three to four days). Due to the reigning climate characteristics of the region, obviously snowmelt is not a process to be modeled. Similarly, frozen ground effects are not influencing percolation and interflow. Only daily precipitation as rainfall and daily potential evapotranspiration are inputs. Data requirements

Potential Evapotranspiration (PET) The model requires a value of the potential evapotranspiration for each time step (day) along the simulation period. In absence of field measurements of daily PET or even of pan evaporation or other, the PET was estimated dividing an empirical monthly potential evapotranspiration by the number of days in each month. The empirical expression for the potential evapotranspiration was the one by Thornthwaite (1948). Its main independent variable is the monthly mean temperature, which was estimated by means of reanalyses made by NCEP / NCAR. The two-metre temperature is based on a short-term (6 hour) weather forecast. The accuracy of this forecast is related to errors of the initial condition, to accuracy of the forecast model (soil moisture and soil model, surface evaporation and sensible heat flux forecast, forecasted clouds) and to the elevation model employed. There are no observations included in the reanalyses. The only surface land variable used in making the analyses is surface pressure. Above the surface, winds, temperature and humidity are used (Wesley Ebisuzaki, personal communication). The information is readily available on Internet at: http://nomad2.ncep.noaa.gov/ncep_data/index.html. Reanalysis is a cooperation project among the NCEP and the NCAR consisting in the production of an almost-fifty-year record of global analyses of atmospheric fields, as an aid to the research and climate monitoring communities. It involved the recovery of data of different origins as land surface, ship, rawinsonde, pibal, aircraft, satellite, etc. (Kalnay et al., 1996, Kistler et al., 2001). Monthly mean temperatures are downloadable as areal means of a user-defined domain, bounded by parallels and meridians.

Mean areal precipitation (MAP) Inputs of daily mean areal precipitation are required. Rain-gauge data are available from four stations within the catchment divide (see Table 2). The stations belong to ANA in Brazil, who run a telemetric network throughout the country. Mean areal values were computed using Thiessen Polygons. Table 2 Rain gauges in the Ibicui River basin CODE STATION RIVER LAT LON THIESSEN WEIGHT 32757 Rosario do Sul Santa Maria -3015 -5454 41.28% 32756 Manoel Viana Ibicui -2936 -5529 36.83% 32755 Alegrete Ibirapuita -2946 -5547 15.85% 32754 Passo Mariano Pinto Ibicui -2919 -5603 6.04%

Flow rate at the basin outlet The study basin has its outlet at Passo Mariano Pinto, 90 km upstream from the confluence, where a rating curve is available. As explained in the following section, for calibration and validation it is required an observed series of discharge at the watershed outlet. The ANA agency records water stage (h) values twice daily. After transformation to flow rate and averaging, the daily series is obtained. The rating curve is a second order polynomial whose expression for discharge (Q) reads:

( ) 596.51.802.50.597.52 )(2 )(/3 ++= mmsm hhQ for mh 62.9 ( ) 139.260.940.115.690.72 )(2 )(/3 += mmsm hhQ otherwise (15)

Calibration and validation The calibration process is intended to make an accurate estimation of model parameters for a given basin. It consists in successive adjustments of the parameters, starting from an initial set, until the simulated response agrees satisfactorily with observations, and without violating the constraints, assuming they had been previously defined. How good the agreement is is assessed via a model performance criterion. The criterion adopted here was that of minimizing an objective function such as a weighted quadratic error. The calibration was performed with an automatic procedure. The minimization routine employed was the Downhill Simplex Method, while the constraints for the parameters were taken into account via a Penalty Function Method. Table 3 summarizes the ten model parameters, defined previously in the model description section. Table 3 Model parameters PARAMETER 01X

02X 1m 1C 2C 3C 2m 3m

Objective Function The objective function used for the present study was the sum of squares of differences between the observed and the simulated discharges, each term affected by a weighting function that emphasizes the accurate reproduction of observed peak flows. This summation is computed along the entire calibration period, such that:

NWTQQF iN

iisimiobs /.)(

1

2,,

== [ ]21.][ = TLF (16)

where N is the number of time steps of the calibration period, i is the time index such that Ni L1= , iobsQ , and isimQ , are the observed and simulated discharges in step i, and F is the objective function.

NWTi / is a weighting function (Ford et al., 1980) such that:

( )[ ]Q

QQWT iobsi .

.1, += =][ iWT dimensionless (17)

where Q is the mean flow along the calibration period, is a factor that increases the weighting function with decreasing values (not less than unity), thus amplifying the influence of higher discharges. The quantity F reflects the extent to which a model is successful in reproducing the observed flow rates, particularly those above the mean. The value of F is biased by the weighting function since any errors for computed flow rates that exceed the average discharge will be weighted more heavily.

Minimization algorithm: Downhill Simplex Method The minimization of the objective function was performed with the Downhill Simplex Method (Nelder and Mead, 1965). This method is suitable for multi-dimensions, that is, for finding the minimum of a function of more than one independent variable. The method requires only function evaluations, not derivatives. Strictly speaking, in the present case, evaluating the function implies running the hydrological model and computing the objective function (efficiency criterion). Clearly, this method is not very efficient in terms of the number of function evaluations that it requires. However, it works properly for a problem whose computational burden is relatively small, as this is. The method has a geometrical nature. A simplex is the geometrical figure consisting, in N dimensions, of N + 1 vertices and all their interconnecting line segments, polygonal faces, etc. In two dimensions, a simplex is a triangle; in three dimensions, a tetrahedron. For multidimensional minimization, a starting guess is to be provided, that is, an N-vector of independent variables (parameters) as the first point to try. The algorithm is then supposed to make its own way downhill through the N-dimensional topography, until it encounters a (local, at least) minimum. The downhill simplex method must be started not just with a single point, but with N + 1 points, defining an initial simplex. Then the method takes a series of steps, most of them just moving the point of the simplex where the function is largest through the opposite face of the simplex to a lower point. These reflections are constructed to conserve the volume of the simplex. When it can do so, the method expands the simplex in one or another direction to take larger steps. When it reaches a valley floor, the method contracts itself in the transverse direction and tries to seep down the valley. If there is a situation where the simplex is trying to pass through the eye of a needle, it contracts itself in all directions, pulling itself in around its lowest (best) point. Termination criteria are delicate in multidimensional minimization routines. Typically, it can be identified one step of the multidimensional algorithm. It is then possible to terminate when the vector distance moved in that step is fractionally smaller in magnitude than some predefined tolerance. Alternatively, it could be required that the decrease in the function value in the terminating step be fractionally smaller than an also predefined tolerance but for the function (Press et al., 1986).

Parameter constraints: a Penalty Function Method The model parameters are subject to constraints, that is, physical limitations to the range of feasible values generally imposed by their nature or by the mathematical functions where the parameters are included. Therefore, a certain procedure must be performed in order to consider this restriction. One approach is to incorporate the constraints into the objective function to form a new function which can now be optimized with an unconstrained algorithm, such as the Downhill Simplex used here. Penalty function methods are some of these procedures and they consist in adding a penalty for the violation of constraints. While minimizing the function, the constraint violation is then also minimized. In the limit that the penalty is large compared with the rest of the function, the constraints will eventually be satisfied if possible (Bruen, 1998, Bock, 1998). The parameter constraints are given in Table 4. Different authors give guidance for the definition of feasibility intervals for the parameters (Guetter, 2000, Koren et al., 2000, Bae and Georgakakos, 1994). Table 4 Parameter constraints PARAMETER 01X

02X 1m 1C 2C 3C 2m 3m

UNITS [mm] [mm] [1] [1/day] [1] [1/day] [1] [1/day] [1] [1] 80.0 100.0 1.00 0.00001 100.00 0.0001 0.01 0.010 1.00 1.00 350.0 500.0 3.00 0.01000 400.00 0.2000 6.00 0.400 2.20 5.00

Calibration and validation periods For the calibration and validation processes, data are available along the period 1st September 2004 to 31st January 2006. The period of record was split into three fourths and one fourth, respectively. Although the period is reduced (aspect that will be considered later in the conclusions section), along those 17 months, ten significant flood waves occurred, with peaks well above the normal discharge, and three periods of sustained low waters were observed (two of 3.5 months and one of 1.5 months). Consequently, different hydrologic conditions are found and the parameter estimation is reasonably accurate and reliable.

Model performance in calibration and validation Model performance was assessed through the following statistics: (a) error in mean discharge (relative to mean observed discharges), (b) error in standard deviation of discharges (relative to that of observed discharges), (c) correlation r between modeled and observed discharge series, (d) Root Mean Square Error, RMSE and (e) objective function (as defined early in this section). Additionally, it was computed a measurement of the phase shift between the observed and modeled series: (f) SPEDS (Special Directional Symmetry), whose expression reads:

=

=n

iibn

SPEDS1

.100 [ ] =SPEDS dimensionless (18)with

1=ib for ( )( ) 0. 1,,1,, isimisimiobsiobs QQQQ 0=ib otherwise

where n is the number of time steps of the simulation period, i is the time index such that ni L1= , iobsQ , and isimQ , are the observed and simulated discharges in step i.

The resulting (optimal) parameters are given in Table 5, for a coefficient of the weighting function 5= . A value this high causes the weights to be rather homogeneous along the simulation period,

not assigning particularly heavy weights to peaks. Smaller values should be used when the agreement in peak flows is a priority. Table 5 Values of calibrated parameters

PARAMETER 01X 02X 1m 1C 2C 3C 2m 3m

UNITS [mm] [mm] [1] [1/day] [1] [1/day] [1] [1/day] [1] [1] CALIBRATED VALUE 194.1 397.3 1.55 0.00322 235.67 0.0010 4.87 0.251 1.23 1.35 The model performance for calibration and validation runs with the optimal set of parameters is presented in Table 6. Additionally, the global performance for the entire period was computed with the same set of parameters and is also shown. Graphs of these runs are shown in Figures 3, 4 and 5. Table 6 Model Performance STATISTIC CALIBRATION VALIDATION GLOBAL

(a) obs

obssim

QQQ

-0.27% 10.82% 7.57%

(b) obsQ

obsQsimQ

SSS

,

,, -2.10% -10.05% -0.72% (c) r 96.74% 96.40% 96.42% (d) RMSE 28.66% 45.74% 34.20% (e) OBJECTIVE FUNCTION [ ]])/[( 2daymm 0.0945 0.2307 0.1407 (f) SPEDS 83.50 82.79 83.17

Figure 3. Ibicui River. Hydrographs of calibration run

Figure 4. Ibicui River. Hydrographs of validation run

Figure 5. Ibicui River. Hydrographs of global run Operational use for flood warning and reservoir management

General aspects The Ibicui River is the largest tributary of the Uruguay River and although it usually conveys a relatively reduced percentage of the Uruguay discharge, the supply of water volumes may be substantive during floods (which not necessarily concur with floods of the Upper Uruguay River). The drainage basin basically presents a response by events, which means that soil moisture condition depends largely on recent rainfall (the previous few days) and the response to storm events is rather fast. Along prolonged low-water periods in the Uruguay River basin, the Ibicui River supplies volumes that can attenuate the deficiency of the Upper Uruguay River, and sustain the Salto Grande reservoir inflow. During floods, an accurate estimation of volumes and inflows to the Uruguay River is required for planning the reservoir operation and for water stage forecast downstream. Required lead-times are about two to three days. A good approximation to the flood wave shape is necessary and implies the need of a hydrological model adequate for fast changing hydrographs. In a typical forecast session, the forecaster may combine meteorological data from several sources. The simulation begins at some point in the past for which watershed conditions are known, that is, the state variables have a known (or assumed) value, the initial condition. Starting at that date, the run uses as input a combination of available meteorological data and a weather forecast, for the early and the future part of the simulation period, respectively. Basically, the weather forecast is one of rainfall and temperature. The operator can use precipitation and temperature forecasts from different sources, weighting them adequately. In general, rainfall forecasts have a large uncertainty and, therefore, it is convenient to set different scenarios with their own weight. An evaluation of the ensemble of results will help in the forecast definition. Model runs necessarily should be carried out on a regular basis, with daily updating and possible forecast re-definitions. Regarding the simulation period, the early part of it has to be at least as long as the memory length of the system. In practice, the initial condition will be either assumed or the result of a previous run and, therefore, its accuracy may be rather low, thus requiring a warm-up period at least that long. The length of the future period has to be set with caution. It should not be too long since the reliability of the weather forecast diminishes fast after a few days (even the forecast of daily totals of rainfall). Ideally, a forecasted storm event should be fully covered so that the resulting streamflow hydrograph

reaches the peak. This precaution is particularly important in the case of two successive events, when the resulting flood waves tend to merge into a large one. Large watersheds usually contain reservoirs regulating the flow of rivers for meeting different objectives as flood control, hydropower production, irrigation, etc. A successful operation usually counts on reliable flow forecasts. Some studies (Georgakakos et al., 1995) showed that conceptual hydrologic models (possibly coupled with routing models) are more successful in assisting reservoir management decisions than simple statistical forecast models using flow rate as the predictor (Georgakakos, 1995). Reservoir operations normally require forecasting streamflows. The procedure is basically the same as for flood warning, except for the fact that now the runs are not focused on storms but rather are made on a permanent basis. The operator must review the operational criteria to be used at reservoirs and power plants during the forecast period and enter generation schedules and water demands. The user can try several operational schemes, review the results, and possibly redefine reservoir operations. In the case of Salto Grande, the operators necessarily have to include the evaluation and forecasting of the Ibicui streamflow for a proper reservoir operation, particularly during periods of either water abundance or scarcity.

Input of temperature As stated above, the daily potential evapotranspiration is computed by dividing the monthly estimate into the number of days of the month. Therefore, for the future part of the input, mean temperature for the months involved has to be forecasted. The NCEP daily update a temperature outlook for the subsequent 14 days (starting from the current date), divided into two periods of 7 days each, for which mean temperature is predicted. An outlook of this type is useful for monthly mean temperature prediction. Although the sensitivity of the model to the daily PET is not too high, this approach is finer than simply inputting the climatic PET for the month. The grid size is 0.5 by 0.5 degrees latitude / longitude. This spatial resolution is sufficiently high for this purpose. The maps present a color scale, set every 5 degrees Celsius, which is a bit too coarse though still adequate.

Input of precipitation The hydrologic model has been calibrated using rainfall data from rain-gauges. However, in the operational phase, a limitation arises since field measurement availability is not guaranteed in real time and, therefore, the early part of the run may not be fed with rain records. Precipitation estimates from satellite data are an option for flood forecasting. The South American version of the NOAA / NESDIS Hydro-Estimator (HE) satellite rainfall estimation technique is available for use and readily downloadable from the Internet. It was developed as an automated technique to assist forecasters in monitoring precipitation at different scales, particularly for regions where rain-gauge networks are sparse or data are not available in real time and where no weather radars are present, as is the study basin. This estimation technique uses an empirical relationship between rain rate and cloud-top temperature (GOES 8 - 10.7 m channel) as the basis for the initial rain rate estimates. It then adjusts the rain rate assigned to each pixel according to the temperatures of surrounding pixels. This allows discerning raining and non-raining pixels. It also helps focus rainfall estimate totals into more clearly defined maximums. It has a major overall applicability, particularly considering its high frequency (a minimum of one estimation every three hours), high spatial resolution (pixel size is 4 km) and broad coverage. When available, precipitation data from rain-gauges may be used for correcting the HE estimation (Vila et al., 2003, Vila and Lima, 2004). Regarding the future part of the run, unlike what was stated for temperature, the quality of the precipitation forecast should be carefully considered. The NCEP issue maps of medium-range forecast of precipitation (through the Global Forecast System, GFS http://wxmaps.org/pix/sa.vv.html) for the subsequent six days (current date plus five days). The map color scale is set at variable intervals of rainfall, usually between 10mm and 25 mm of accumulated precipitation for the upper end of the scale (days of heavy rains). NCEP also issue a precipitation outlook for South America for the subsequent 14 days (starting from the current date), divided into two periods of 7 days each, for which accumulated rainfall is predicted (http://wxmaps.org/pix/prec8.html). A certain criterion is required for distributing the variable along the 7-day period in the model time intervals. The grid size is 0.5 by 0.5 degrees latitude / longitude. This spatial resolution is sufficiently high for the size of the study basin. Color scale in the map is set at non-equal intervals, from 15mm to 50mm, for increasing rainfalls.

A combined use of the two forecasts can extend the future input to two weeks, gradually loosing reliability, particularly after the sixth day. One option is the use of the ETA-SMN model, which is an adjustment of the ETA model (Mesinger et al., 1988) for Southern South America made by the National Weather Service of Argentina (SMN). The horizontal resolution is given by a grid size of 25 km (roughly 0.25 by 0.25 geographical degrees). The model domain is bounded by the coordinates 14 - 65 S, 30 - 91 W. The boundary conditions are taken from the GFS-NCEP model, mentioned above. The model forecasts the subsequent four days of rainfall and other variables (starting from current days). It can be visited at http://www.meteonet.com.ar/?mod=dpd&id=1. Conclusions and remarks The results are promising in the sense that a reliable operational forecast may be issued from modeling with input data readily downloadable from the Internet. Sparseness of meteorological networks and unavailability of field records in real time are not a limiting factor for operational hydrology. It has been observed that the model represents faithfully the physical processes and the daily water balance. It is capable of simulating adequately the streamflows, both during floods and low waters, with daily resolution. The model shows a fast reaction to heavy-rain events, especially when the precipitation pattern affects the whole catchment. The peak error was in almost all cases reasonably reduced and the timing has always been very well resolved, important features when the focus is set on operational forecasting of flood waves. Additionally, the agreement was also very high during low flow regimes and the total volume was well quantified. Therefore, it is concluded that the model is suitable for operational use. Although the time series is relatively short (only 17 months), which would invite to future recalibrations, it is worthwhile noting that receding limbs are almost all very well simulated and this is an indication of a close representation of the soil layer characteristics. Rather, where future efforts should be devoted, according to the authors belief, is in investigating possible improvements in the representation of the spatial distribution of the daily precipitation field (which would ensure a better estimation of the mean areal precipitation). Further research will be directed to implement an operation of the model in updating mode, which is known to outperform the regular operation in simulating both peak magnitude and timing, leading to improved forecasts. In this paper, it has been presented the implementation of a forecasting tool for a particular study basin. However, the methodology is expected to be implemented in different catchments in the Del Plata Basin in Argentina; in some of them, to develop a forecast tool and in some others, to improve the estimation of lateral inflows to the huge Parana River, where a hydraulic model is currently in use. References Bae, D.H., and Georgakakos, K.P., 1992: "Hydrologic modeling for flow forecasting and climate studies in large drainage basins," IIHR Rep. 360, Iowa Institute of Hydraulic Research and Department of Civil and Environmental Engineering, University of Iowa, Iowa City, 252 pp.

Bae, D.H., and Georgakakos, K.P., 1994: "Climatic Variability of Soil Water in the American Midwest: Part 1. Hydrologic Modeling," Journal of Hydrology, 162, 355-377.

Bock, R.K., 1998: Penalty Function, in http://rkb.home.cern.ch/rkb/AN16pp/node205.html

Bruen, M., 1998: Lecture Notes for the M.Sc. Course in Hydrologic Computation, for the International Postgraduate Course in Hydrology, National University of Ireland, Galway.

Burnash, R.J.C., Ferral, R.L., and McGuire, R.A., 1973: "A Generalized Streamflow Simulation System - Conceptual Modeling for Digital Computers," U.S. Department of Commerce, National Weather Service and State of California, Department of Water Resources.

Ford, D.T., Morris, E.C. and Feldman, A.D., 1980: Corp of Engineers experience with automatic calibration of a precipitation-runoff model in Water and Related Land Resource Systems (Y. Haimes and J. Kindler, eds.) Pergamon Press, New York, p. 467-476, 1980.

Georgakakos, K.P., 1995: Real-Time Prediction For Flood Warning And Management. U.S.- Italy Research Workshop on the Hydrometeorology, Impacts, and Management of Extreme Floods. Perugia (Italy), November 1995.

Georgakakos, K.P., Bae, D.H., Mullusky, M.G., and Georgakakos, A.P., 1995: "Hydrologic

Variability in Midwestern Drainage Basins: Diagnosis, Prediction and Control," Chapter II.2 in Global Change II, A Midwest Perspective, E. Folk (ed.), SPB Academic Publishing, Amsterdam, The Netherlands, 61-90.

Georgakakos, K.P., Rajaram, H. and Li, S.G., 1988: On Improved Operational Hydrologic Forecasting of Streamflows. IIHR Technical Report No. 325. Iowa Institute of Hydraulic Research, The University of Iowa, Iowa City, Iowa, 52242, 162 pp.

Guetter, A. K., 2000: Hydrological Modeling and Hydrological Warning Systems. Lecture Notes. Advanced Course in Numerical Modeling. Hydrological Modeling, Lima, Peru, 03-14/July/2000

Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y., Leetmaa, A., Reynolds, B., Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K.C., Ropelewski, C., Wang, J., Jenne, Roy, Joseph, Dennis, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bulletin of the American Meteorological Society: Vol. 77, No. 3, pp. 437472.

Kistler, R., Kalnay, E., Collins, W., Saha, S., White, G., Woollen, J., Chelliah, M., Ebisuzaki, W., Kanamitsu, M., Kousky, V., van den Dool, H., Jenne, R., Fiorino, M., 2001 The NCEPNCAR 50Year Reanalysis: Monthly Means CDROM and Documentation. Bulletin of the American Meteorological Society: Vol. 82, No. 2, pp. 247268.

Koren, V.I., Smith, M., Wang, D., Zhang, Z., 2000: Use of Soil Property Data in the Derivation of Conceptual Rainfall-Runoff Model Parameters, 15th Conference on Hydrology, AMS, January 9-14,2000, Long Beach, CA, USA.

Mesinger, F., Janji, Z.I., Nikovi, S., Gavrilov, D. and Deaven, D.G., 1988: The step-mountain coordinate: Model description and performance for cases of alpine lee cyclogenesis and for a case of an Appalachian redevelopment. Mon. Wea. Rev., 116, 1493-1518.

Nelder, J.A. and Mead, R., 1965: A simplex method for function minimization. Computer Journal, 7:308--313, 1965.

Peck, E.L., 1976: Catchment Modeling and Initial Parameter Estimation for the National Weather Service River Forecast System, NOAA Technical Memorandum NWS HYDRO-31, NWS, NOAA, DOC, Washington D.C., June 1976.

Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T., 1986: Numerical Recipes. The Art of Scientific Computing (FORTRAN version), ISBN 0-521-38330-7, Copyright (C) Cambridge University Press, 1986-1989.

Sperfslage, J.A., and Georgakakos, K.P., 1996: Implementation and Testing of the HFS Operation as Part of the National Weather Service River Forecast System (NWSRFS). HRC Technical Report No. 1. Hydrologic Research Center, San Diego, California, 92130, 213 pp.

Thornthwaite, C.W., 1948: An approach toward a rational classification of climate. Geographical Review 38:55-94.

Vila, D.A., Scofield, R., Kuligowski, R. and Davenport C., 2003: Satellite Rainfall Estimation Over South America: Evaluation Of Two Major Events. NOAA Technical Report NESDIS N 114, U.S. Department of Commerce, pp 1-17

Vila D.A. and Lima A., 2004: Satellite Rainfall Estimation Over South America: The Hydroestimator Technique. Proceedings of the 14th International Conference on Clouds and Precipitation, Bologna, Italy, July 2004.