Abstract

A global hydrological routing algorithm (HYDRA) that simulates seasonal river discharge and changes in surface water level on a spatial resolution of 5′ long × 5′ lat is presented. The model is based on previous work by M. T. Coe and incorporates major improvements from that work including 1) the ability to simulate monthly and seasonal variations in discharge and lake and wetland level, and 2) direct representation of man-made dams and reservoirs. HYDRA requires as input daily or monthly mean averages of runoff, precipitation, and evaporation either from GCM output or observations.

As an example of the utility of HYDRA in evaluating GCM simulations, the model is forced with monthly mean estimates of runoff from the National Centers for Environmental Prediction (NCEP) reanalysis dataset. The simulated river discharge clearly shows that although the NCEP runoff captures the large-scale features of the observed terrestrial hydrology, there are numerous differences in detail from observations. The simulated mean annual discharge is within ±20% at only 13 of 90 fluvial gauging stations compared. In general, the discharge is overestimated for most of the northern high latitudes, midcontinental North America, eastern Europe, central and eastern Asia, India, and northern Africa. Only in western Europe and eastern North America is the discharge consistently underestimated. Although there appears to be a need for improved simulation of land surface physics in the NCEP product and parameterization of flow velocities within HYDRA, the timing of the monthly mean discharge is in fair agreement with the observations.

Including lakes within HYDRA reduces the amplitude of the seasonal cycle of discharge and the magnitude of the annual mean discharge of the St. Lawrence River system, in qualitative agreement with the observations. In addition, including the wetlands of the Sudd reduces the magnitude of the simulated annual discharge of the Nile River to values in better agreement with observations.

Finally, the impact of man-made dams and their reservoirs on the magnitude of monthly mean discharge can be explicitly included within HYDRA. As an example, including dams and reservoirs on the Parana River improves the agreement of the simulated mean monthly discharge with observations by reducing the amplitude of the seasonal cycle to values in good agreement with the observations.

The results of this study show that, although improvements can be obtained through better representations of flow velocities and more accurate digital elevation models, HYDRA can be a powerful tool for diagnosing simulated terrestrial hydrology and investigations of global climate change.

1. Introduction

The accurate simulation of terrestrial hydrologic systems (rivers, lakes, and wetlands) is an important goal of Earth System Science for at least three reasons. First, the terrestrial hydrologic system provides a strong test of global climate model (GCM) simulations. GCMs have become important tools in determining global policy issues (IPCC 1990). If we intend to use GCMs to simulate the near-surface climate and determine social policy, then we must have a rigorous test of GCM performance. Global datasets of observed river flow (e.g., Vörösmarty et al. 1997) and lake level (e.g., Birkett and Mason 1995; Birkett 1998) are becoming available and, coupled with accurate terrestrial hydrology models, can provide a basis for evaluating GCM performance. Second, modern society is dependent on terrestrial hydrological systems. Rivers, lakes, and wetlands provide water for industry, agriculture, electricity generation, and household use. Future population increases, coupled with changing water availability from climate and land-use changes may greatly stress some water resources. Finally, river discharge is recognized as a potentially important determinant of ocean circulation and therefore global climate (see Broecker et al. 1990; Dümenil and Todeni 1992; Sausen et al. 1994). It is thus important to develop more accurate tools to simulate the global terrestrial hydrologic system.

Recent terrestrial hydrologic modeling efforts have led to improvements in the simulation of catchment-scale runoff through parameterizing the hill-slope hydrology and land surface heterogeneity (Entekhabi and Eagleson 1989; Famiglietti and Wood 1991, 1994a,b; Stieglitz et al. 1997; Nijssen et al. 1997). However, methods for transporting the catchment-scale runoff to the oceans, or inland drainage basins, have received relatively less attention.

Continental-scale river flow and discharge to the ocean have been simulated via relatively coarse resolution (½° to 4°) surface water transport models. These models prescribe river flow directions based on visual inspection of maps and do not include lakes, wetlands, or man-made reservoirs in determining flow velocity (Vörösmarty et al. 1989; Vörösmarty and Moore 1991;Vörösmarty et al. 1991; Liston et al. 1994; Miller et al. 1994; Marengo et al. 1994; Sausen 1994; Russell et al. 1995; Hagemann and Dümenil 1996; Costa and Foley 1997; Nijssen et al. 1997).

River discharge is part of a linked hydrologic network, which includes the transport of water across the land surface and temporary storage of water in lakes, reservoirs, and wetlands. The seasonal cycle of river discharge is a function of the time it takes runoff, generated locally, to traverse the lakes, wetlands, and streams on the way to the oceans or closed interior drainage basins. For example, Hagemann and Dümenil (1996) show that a continental-scale river transport model with no representation of the Great Lakes is inadequate for simulating the monthly mean discharge of the St. Lawrence River in North America. Similarly, reservoirs must be included for accurate simulation of modified river systems (Vörösmarty et al. 1997). Therefore, a global-scale surface hydrology model must integrate river, lake, reservoir, and wetland processes and allow for simultaneous model-generated changes in surface water area, river discharge, and river basin area contributing to discharge. Coe (1997, 1998) presented a terrain-based model (Surface Water Area Model, SWAM) for predicting annual mean river transport and surface water area as a linked hydrologic network at the continental scale.

This study presents a new hydrology routing algorithm (HYDRA) to simulate river discharge, lakes, and wetlands globally on a horizontal resolution of 5′ × 5′ and a 1-h time step. HYDRA differs from SWAM by simulating seasonal variations in river discharge and lake and wetland area, and by allowing for the direct representation of man-made dams and reservoirs. HYDRA is forced with monthly mean or daily runoff, precipitation, and evaporation from either GCM output or observations.

In this study, HYDRA is forced with monthly mean runoff from the National Centers for Environmental Prediction (NCEP) reanalysis dataset (Kalnay et al. 1996). The spatial distribution of lakes and wetlands is prescribed, rather than being predicted, based on the estimates of Cogley (1991). A comparison of the annual and monthly mean river discharge at 90 locations on the world’s major rivers is made to a compilation of river gauge data. This study illustrates that HYDRA can be a useful tool for diagnosing GCM output and assessing possibly important changes in water availability and discharge to the oceans due to anthropogenic, or naturally occurring, climate change.

2. Model description

HYDRA simulates land surface hydrology as a linked dynamical system. Rivers, lakes, and wetlands are defined as a continuous hydrologic network in which locally derived runoff is transported across the land surface in rivers, fills lakes and wetlands, and is eventually transported to the ocean or an inland lake. The model [based on a model described by Coe (1998)] derives the river paths and maximum lake and wetland volumes from digital elevation model (DEM) representations of the land surface topography. This hydrologic network is linked to a linear reservoir model and forced with estimates of runoff over land, and precipitation and surface evaporation over lakes and wetlands.

The hydrologic network is derived from two 5′ × 5′ resolution (≈10 km × 10 km at the equator) global DEMs: 1) TerrainBase from the National Oceanic and Atmospheric Administration (NOAA), National Geophysical Data Center [which is based on the original NOAA (1994) product ETOPO5]; and 2) GLOBAL DEM5 (1995) from Geophysical Exploration Technology Ltd. (GETECH 1995) (which is based on ETOPO5 and independent observations). TerrainBase was chosen to define the hydrologic network of North America, Africa, and Australasia. GLOBAL DEM5 was chosen to define the hydrologic network of South America, Europe, Siberia, and Southeast Asia.

The elevations in the DEMs represent an estimate of the average elevation for a given 5′ × 5′ grid cell including lake and ocean bathymetry. Therefore, the DEMs indicate where water may accumulate on the land surface, the path each river will follow to the ocean or inland drainage basin, and the potential land area drained by each river. A complete description of the techniques for defining potential rivers, lakes, and wetlands, including the extensive corrections required are presented by Coe (1998).

The potential surface waters (surface depressions that may fill with water) are defined by iteratively searching the DEM to define all surface depressions (regions where the land surface slopes to a central low elevation rather than draining to the ocean) and the elevation of their outlets. All grid cells sharing a common outlet are summed to create a single potential surface water. Therefore, a potential surface water may consist of many 5′ × 5′ grid cells.

River transport directions are calculated by modifying the DEM land surface so that all grid cells have a unique flow path to an ocean. An initial river transport direction is defined as the direction to the immediately neighboring grid cell (of eight possible neighbors) with the lowest surface elevation. A further step is required to calculate the flow direction of waters through a lake or wetland to the ocean. The transport through the surface waters is determined using an algorithm that fills all depressions in the land surface and slopes the new land surface toward the depression outlet. Then the flow direction is recalculated based on the modified topography. In this way, all grid cells are defined to have a unique flow direction with eventual communication to the ocean.

HYDRA treats the potential surface water volumes as polygons within the hydrologic network. The change in the water volume of each surface water is stored (at each time step) at the cell corresponding to the outlet location. The corresponding surface water area is determined by distributing the change in volume at each time step to the potential surface water area. The change in volume is distributed starting with the nearest low point in the surface depression (if no lake currently exists) or on top of any existing surface water within the depression. In this manner, surface waters grow from an initial low spot in the terrain in agreement with natural processes. Because the total volume is stored and distributed from the outlet location a level water surface is efficiently maintained in large lakes and wetlands despite the different water mass balances of the individual grid cells within the lake or wetland. No stream flow from the outlet of a lake or wetland may occur until the total basin volume has been filled. The predicted surface water area (and surface water volume) will be a unique function of the balance between the streamflow into the surface water and precipitation minus evaporation (PE) over the surface water. In this way, rivers flowing into closed basin lakes and wetlands with no communication to the ocean are explicitly simulated, as are open systems with flow-through to the ocean.

As in numerous other large-scale hydrological studies (e.g., Coe 1998; Vörösmarty et al. 1989) HYDRA uses a linear reservoir model to transport local surface runoff and subsurface drainage through a river network to the oceans or inland basins. The linear reservoir model simulates water transport in terms of river routing directions derived from the local topography, residences times within a grid cell, and effective flow velocities.

The total water entering the hydrologic network at each grid cell is the sum of the land surface runoff (Rs), subsurface drainage (Rd), precipitation (Pw) and evaporation (Ew) over the surface waters, and flux of water from upstream grid cells (m3 s−1). The water transport is represented by the time-dependent change of three water reservoirs. First, the river water reservoir (Wr), which contains the sum of upstream and local water in excess of that required to fill a local surface water depression. Second, the surface runoff pool (Ws), which contains water that has run off the surface locally and is flowing toward a river. Third, the subsurface drainage pool (Wd), which contains water that has drained through the local soil column and is flowing toward a river. All reservoirs are represented in m3 and flow is governed by the following differential equations.

 
formula

Here Aw is an indicator of fractional water area in the grid cell; from 1 (lake, wetland, or reservoir covers entire cell) to 0 (no water present), Ts, Td, and Tr are the residence times (s) of the water in each of the reservoirs. Here Pw and Ew are the precipitation and evaporation rates (m3 s−1) over the surface water, respectively, and Σ Fin is the sum of the fluxes of water (m3 s−1) from the upstream cells.

The local surface and subsurface residence times (Td and Ts) are set to globally constant values for simplicity. Here Td is set to 15 days and Ts to 2 h, similar to the values used by Costa and Foley (1997) to simulate large-scale flow in the Amazon basin.

The stream flow residence time, Tr, is defined as the ratio of the distance between centers of the local and downstream grid cells (D) and the effective velocity of the water (u). To capture the change in velocity as a river flows through a lake, wetland, or reservoir, the effective velocity is calculated differently for those cells with and without surface water. For grid cells without a lake or wetland the effective velocity (u) is proportional to a ratio of the downstream gradient (ic, m m−1) and a reference gradient io (=0.5 × 10−4 m m−1) as by Miller et al. (1994):

 
u = uo1(ic/io)0.5,

where uo1 is the minimum effective velocity of the river (0.8 m s−1).

For grid cells with a lake or a wetland the effective velocity equation was modified to simulate slower transport through the water body. For water cells the effective velocity is proportional to a ratio of a reference volume (1 m of water spread over one grid cell; υ1, m3) and the volume of the entire water body (from one to many grid cells) of which the local cell is part (υt, m3):

 
u = uo2 × (υl/υt)0.5.

Here uo2 is the minimum effective velocity of the river flow through the surface water and was chosen to be 0.1 of uo1 (0.08 m s−1). The factor of 0.1 was derived from scale considerations suggesting that within a grid cell a river is on the order of 0.1 the width of a lake.

These residence time and effective velocity parameters are first approximations of regional water storage and transport terms and result in a reasonable representation of the river flow. However, future studies will need to evaluate more closely the importance of these parameters on the simulation of regional discharge.

Starting with Wr, Wd, and Ws = 0. HYDRA is forced with 0.5° × 0.5° estimates of monthly mean runoff, precipitation, and surface water evaporation converted to daily values and linearly interpolated to the 5′ × 5′ grid of HYDRA. The model solves for dWR/dt, dWd/dt, and dWt/dt for the year with a time step of 1 h until equilibrium is reached (dWR/dt is very small). The predicted river discharge and lake and wetland area and volume represent the surface hydrology in equilibrium with the prescribed mean monthly climate.

3. Application with the NCEP reanalysis dataset

HYDRA can be forced with data from a number of sources. Runoff from land surface or ecosystem models can be used as forcing, in order to investigate the impact of changes, such as land use or vegetation, on river discharge and lake and wetland level. HYDRA can also be forced with GCM output in order to evaluate the model-simulated climate or to investigate the impact of past or future climate changes. In this study we force HYDRA with runoff from the NCEP–NCAR reanalyzed meteorological dataset (Kalnay et al. 1996, to be referred to as NCEP) as an example of the utility of using HYDRA to evaluate GCM performance.

The reanalysis procedure, such as that used by NCEP, combines observations with model simulations of the atmosphere and land surface physics to produce internally consistent, spatially and temporally continuous datasets for a suite of atmospheric and surface variables. In the NCEP reanalysis, runoff is entirely model derived as the residual of the precipitation and evapotranspiration. The simulation of evapotranspiration itself is dependent on simplifications of complex land surface physics. Therefore, the runoff may contain significant model-dependant bias (Kalnay et al. 1996). Investigations have confirmed the existence of a bias in the NCEP hydrologic cycle (Higgens et al. 1997; Costa and Foley 1998). HYDRA can provide a simple and robust means of quantifying the bias present in the surface hydrologic balance of the NCEP dataset.

Simulations were performed for the entire globe, with the exceptions of Antarctica, Greenland, the Canadian Arctic Archipelago, and the Middle East because of extreme regional errors in the DEMs. The runoff is the mean monthly NCEP reanalysis product averaged over the period 1979–96. The runoff was converted to the 5′ × 5′ resolution of HYDRA and split into surface and subsurface components based on the calculations of Probst and Sigha (1989). It is assumed that 70% of the NCEP runoff represents subsurface runoff and 30% represents the surface runoff.

To calculate the water budget over the lakes and wetlands, the mean monthly 0.5° × 0.5° precipitation dataset of Legates and Willmott (1990) is used in conjunction with a calculated surface water evaporation term. Estimates of evaporation from surface waters are those used by Coe (1998) and are calculated at 0.5° × 0.5° resolution using the Penman formulation relating equilibrium evaporation from a wet surface to the net surface radiation (Peixóto and Oort 1992). The monthly average estimates of surface temperature from Legates and Willmott (1990) and potential sunshine from Leemans and Cramer (1990) are linearly interpolated to quasi-daily values and used to derive the net surface radiation. The daily equilibrium evaporation values are then summed to provide monthly mean values.

HYDRA may be run either with predicted or prescribed surface waters. Predicted surface waters require a long model run time (on the order of hundreds of years) to allow large inland lakes to come to equilibrium. In this study the surface water areas are prescribed, rather than being predicted, from global estimates of lake area provided by Cogley (1991) on 1° × 1° resolution. The surface water volumes are calculated by distributing the Cogley estimates of lake area into the 5′ × 5′ DEMs and summing the total volume below the outlet elevations of each lake (Coe 1998). A subsequent study will present the results of simultaneous simulations of surface water area and river discharge.

4. Results

Comparisons between simulated mean annual and monthly river discharge are made at 90 locations worldwide to observed values included in the RivDis 1.0 (Vörösmarty et al. 1996) and U.S. Geological Survey discharge datasets. River discharge is generally calculated from a measurement of river water level. The water level is translated into a discharge volume using a rating curve, which is compiled of numerous direct discharge measurements made at as many river levels as possible. The major sources of error in calculating river discharge probably result from direct measurement, which is often made under difficult conditions, and the use of the rating curve, which is dependent upon stream cross-sectional area remaining constant (Cogley 1989). The accuracy of the discharge used in this study is not given in the original data. However, Dickinson (1967) and Cogley (1989) have investigated the potential error in river discharge measurements in detail. Their research suggests that 10%–15% is a reasonable estimate of the error in observed discharge.

The gauging stations in the model were placed at the locations on the simulated rivers that are closest to the latitude and longitude of the actual reported station locations. In general, simulated station locations are within about 10–20 km of the observed (≈1 or 2 grid cells). Exceptions occur where the DEMs do not represent the topography well or within large potential lake basins.

For example, the path of the simulated Danube River across the Hungarian plain does not match the observed path (Fig. 1e). As discussed by Coe (1998), the original DEM simulated a very large lake on the Hungarian plains and the Danube River discharged from this lake into the Baltic Sea. This was corrected to eliminate the erroneous lake and force the Danube to flow into the Black Sea. This correction resulted in a much better representation of the large-scale features of the Danube River system, such as the total area and spatial distribution of land draining to the Black Sea. However, the details of the flow path across the Hungarian plain do not compare well with the observations.

Fig. 1.

(a)–(f) Simulated and observed annual mean discharge. Simulated annual mean discharge is displayed for each grid cell in m3 s−1. Simulated rivers appear as colored lines where the topography used in HYDRA concentrates discharge into the most direct path to the ocean or inland drainage basin. Observational estimates of river discharge (derived from RivDIS 1.0;Vörösmarty et al. 1997) are shown as colored circles centered on the latitude and longitude of each station. The scale for the observed and simulated discharge is identical and presented at the bottom of the figures. The numbers correspond to each observed discharge station and refer to the data presented in Tables 1 and 2. The prescribed lakes and wetlands (Cogley 1991) are shown in dark blue.

Fig. 1.

(a)–(f) Simulated and observed annual mean discharge. Simulated annual mean discharge is displayed for each grid cell in m3 s−1. Simulated rivers appear as colored lines where the topography used in HYDRA concentrates discharge into the most direct path to the ocean or inland drainage basin. Observational estimates of river discharge (derived from RivDIS 1.0;Vörösmarty et al. 1997) are shown as colored circles centered on the latitude and longitude of each station. The scale for the observed and simulated discharge is identical and presented at the bottom of the figures. The numbers correspond to each observed discharge station and refer to the data presented in Tables 1 and 2. The prescribed lakes and wetlands (Cogley 1991) are shown in dark blue.

The Volga and Ural Rivers in eastern Europe are another example of the possible errors in the definition of river flow across the land surface. The observed rivers flow directly into the sea along the northwest and northern shoreline of the sea (respectively). However, the land area to the north of the modern lake shoreline is part of the potentially much larger Caspian Sea. The land surface was reconstructed in the potential lake basin using the technique described in section 2, and the rivers are displaced from their true path across this reconstructed land surface (Fig. 1e). As a result, the location of the simulated Volga Hydro Plant gauging station (number 84) is much farther east than the observed location and the simulated Ural River eventually joins the Volga before emptying into the northeast shore of the Caspian Sea. Other errors in river placement are generally much smaller than the examples given above.

a. Annual mean river discharge

The simulated annual mean discharge is compared globally to observations at 90 selected fluvial gauging stations (Table 1, Figs. 1a–f). In general, the agreement between the observed discharge and discharge simulated with NCEP forcing is poor. The simulated discharge is within ±20% of the observations at only 13 of 90 gauging stations (Figs. 1a–f and 2, Table 1). The simulated annual mean discharge is much greater than observations in many regions of the globe including: 1) northern high-latitude rivers such as the Yukon–Tanana, Mackenzie, Ob, Lena, Kolmya, and Amur; 2) Mid-Continental North American rivers such as the North Saskatchewan, Nelson, and Missouri rivers; 3) northern and southern African rivers such as the Niger, Chari, Senegal, Nile, and Zambezi; 4) most of the eastern European and Asian rivers such as the Don, Neman, Ural, Volga, Ganga, Indus, and Huang He; and 5) South American rivers excluding the Amazon and Orinoco basins such as the Parana, Sao Francisco, and Magdalena. The rivers of western Europe are the only rivers with a consistent discharge underestimate. The Ebro, Loire, Po, Rhine, and Rhone rivers all have simulated discharge 50%–100% less than the observations.

Table 1.

Annual mean simulated (Sim) and observed (Obs) river discharge (dis; m3 s−1), the % difference, and the peak discharge delay (PDD). The station number identifies each gauging station in Figs. 1a–f.

Annual mean simulated (Sim) and observed (Obs) river discharge (dis; m3 s−1), the % difference, and the peak discharge delay (PDD). The station number identifies each gauging station in Figs. 1a–f.
Annual mean simulated (Sim) and observed (Obs) river discharge (dis; m3 s−1), the % difference, and the peak discharge delay (PDD). The station number identifies each gauging station in Figs. 1a–f.
Fig. 2.

Plot of simulated vs observed annual mean discharge for 90 selected fluvial gauging stations. The dashed lines represent values that differ by ±20% from a 1:1 ratio between simulated and observed.

Fig. 2.

Plot of simulated vs observed annual mean discharge for 90 selected fluvial gauging stations. The dashed lines represent values that differ by ±20% from a 1:1 ratio between simulated and observed.

An examination of the annual mean discharge of the Mississippi River basin illustrates the utility of HYDRA in diagnosing GCM output. Traditionally, comparison of simulated large-scale runoff with observations can be made only for a limited number of sites on a river due to the coarse resolution of the routing models (from ½° to 5°). However, the fine horizontal resolution of HYDRA allows for comparison of a much larger number of sites on a river and therefore a more thorough examination of the simulated water budget can be made. For example, the simulated discharge of the Mississippi River at Vicksburg, Mississippi, is within 1% of the observational estimates (Table 1, Fig. 1a, site 42). However, examination of the discharge of the major tributaries of the Mississippi River shows that the good agreement with observations at Vicksburg is primarily a result of large offsetting errors on the Ohio and Missouri River systems. The discharge of the Ohio River is underestimated compared to observations by about 50% (≈4400 m3 s−1) at Metropolis, Illinois (Table 1, Fig. 1a, site 62), and about 30% at Louisville (site 61). Conversely, the discharge of the Missouri River system is overestimated by about 150% (≈3300 m3 s−1) at Hermann, Missouri (Table 1, Fig. 1a site 43). The discharge overestimate is greater than 3700 m3 s−1 at Nebraska City, Nebraska (site 44), and about 2700 m3 s−1 at Yankton, South Dakota (Table 1, Fig. 1a, site 45). This illustrates that the runoff from the Great Plains of the Dakotas and Montana is greatly overestimated in the NCEP output. Additionally, the discharge of the Mississippi River upstream of the confluence of the Missouri river at Alton, Illinois (Fig. 1a, site 40), is overestimated by about 1200 m3 s−1 (≈50%) despite the simulated Wisconsin River system flowing into the Fox River and Lake Michigan rather than into the Mississippi River at Prairie du Chien, Wisconsin.

There are at least two possible causes of the disagreement between simulated and observed annual mean discharge. First, due to poor definition of the potential river basin area derived from the DEM. If the river basin area does not agree well with observations in humid regions, then the accumulated flux of water within the river basin may not agree with observations. Second, due to poor simulation of runoff by NCEP, any discrepancy between observed runoff and NCEP runoff translates into a difference in the total basin discharge.

The results presented by Coe (1998) show that the DEMs used in HYDRA simulate many of the potential river basin areas in reasonable agreement with the observations. This study, which includes more stations, confirms that there is often good agreement between observed and simulated basin area but that much improvement is possible. Of the 86 stations for which areas are reported, 54 have a simulated potential river basin area within ±20% of the observed. Nine of (Fig. 3, Table 2) the potential basins (Chari, Indus at Kotri; Niger at Milanvilla, Niamey, and Ansongo; North Saskatchewan; Senegal at Bakel and Kayes; and Zambezi) are overestimated because of differences in the definition of river basins. HYDRA defines a potential river basin as the entire land surface that may drain to the ocean regardless of whether runoff is generated. The observed river basins are defined as the land surface producing runoff in the current climate. Therefore, the HYDRA basins include extensive regions that are hydrologically closed in the current climate but may contribute runoff in the NCEP simulation. Fourteen rivers (Columbia at The Dalles, Danube at Bezdan and Bratislava, Garonne, Mekong at Mukdahan and Chiang, Murray, Ob, Ohio at Metropolis, Rhine, Seine, Tanana at Nenana and Tanacross, and the Wisla River at Warsaw) have discharge discrepancies consistent with discrepancies in the simulated basin area.

Fig. 3.

Plot of simulated potential river basin area vs observed river basin area upstream of each of the 90 selected fluvial gauging stations. The simulated potential river basin area represents the entire land surface upstream of a station that would contribute to basin discharge if surface runoff is unlimited. The observed official basin area represents only those regions upstream of a station that contribute some annual or seasonal runoff in the current climate. The dashed lines represent values that differ by ±20% from a 1:1 ratio between simulated and observed.

Fig. 3.

Plot of simulated potential river basin area vs observed river basin area upstream of each of the 90 selected fluvial gauging stations. The simulated potential river basin area represents the entire land surface upstream of a station that would contribute to basin discharge if surface runoff is unlimited. The observed official basin area represents only those regions upstream of a station that contribute some annual or seasonal runoff in the current climate. The dashed lines represent values that differ by ±20% from a 1:1 ratio between simulated and observed.

Table 2.

Simulated (Sim) and observed (Obs) river basin area (km2), and the percent difference. The station number identifies each gauging station in Figs. 1a–f.

Simulated (Sim) and observed (Obs) river basin area (km2), and the percent difference. The station number identifies each gauging station in Figs. 1a–f.
Simulated (Sim) and observed (Obs) river basin area (km2), and the percent difference. The station number identifies each gauging station in Figs. 1a–f.

An independent measure of the agreement of the NCEP runoff to observations can be provided by a comparison of the mean annual runoff of NCEP and UNESCO runoff compiled by Cogley (1989) accumulated over the river basins in HYDRA. Accumulated runoff is calculated by summing the volume of annual mean runoff of each grid cell within a river basin. Therefore, any differences between UNESCO and NCEP runoff is due solely to differences in the data since the areas of summation are identical. The NCEP runoff is within ±20% of the Cogley estimates at only 22 of the 90 basins (Fig. 4, Table 3) and the spatial pattern of the differences between the NCEP and Cogley runoff, with few exceptions, coincide with those noted in the discharge comparisons above. For example, the NCEP runoff in the high northern latitudes, central North America, eastern Europe, India, eastern Asia, and northern Africa is too high, by as much as 100% or greater compared to the Cogley observations (Table 3). While the runoff over western Europe is too low by 20%–50% (see Rhine, Rhone, Po, Loire, and Ebro river basins, Table 3) compared to the observed runoff of Cogley. It should be noted that the consistent underestimation of the runoff of the Danube River is probably a result of the overestimation of the observed runoff by Cogley as noted by Coe (1998).

Fig. 4.

Plot of the NCEP–NCAR simulated vs Cogley observed cumulative runoff for the 90 selected fluvial gauging stations. Cumulative runoff is calculated as the sum of the runoff at each 5′ × 5′ grid cell upstream of a gauging station. The dashed lines represent ±20% of a 1:1 ratio between simulated and observed.

Fig. 4.

Plot of the NCEP–NCAR simulated vs Cogley observed cumulative runoff for the 90 selected fluvial gauging stations. Cumulative runoff is calculated as the sum of the runoff at each 5′ × 5′ grid cell upstream of a gauging station. The dashed lines represent ±20% of a 1:1 ratio between simulated and observed.

Table 3.

The percent difference between the basin-accumulated NCEP and Cogley annual mean runoff and the NCEP and Leemans and Cramer precipitaiton. NCEP R/P is the ratio of the NCEP simulated runoff to precipitation (runoff ratio), OBS R/P is the ratio of the Cogley runoff to Leemans and Cramer precipitation. The NCEP runoff ratio is greater than one at a few Arctic locations possibly as a result of the snow or permafrost initialization. OBS R/P greater than one is related to the use of two unrelated observationally based datasets. All variables are accumulated over the basins defined in HYDRA.

The percent difference between the basin-accumulated NCEP and Cogley annual mean runoff and the NCEP and Leemans and Cramer precipitaiton. NCEP R/P is the ratio of the NCEP simulated runoff to precipitation (runoff ratio), OBS R/P is the ratio of the Cogley runoff to Leemans and Cramer precipitation. The NCEP runoff ratio is greater than one at a few Arctic locations possibly as a result of the snow or permafrost initialization. OBS R/P greater than one is related to the use of two unrelated observationally based datasets. All variables are accumulated over the basins defined in HYDRA.
The percent difference between the basin-accumulated NCEP and Cogley annual mean runoff and the NCEP and Leemans and Cramer precipitaiton. NCEP R/P is the ratio of the NCEP simulated runoff to precipitation (runoff ratio), OBS R/P is the ratio of the Cogley runoff to Leemans and Cramer precipitation. The NCEP runoff ratio is greater than one at a few Arctic locations possibly as a result of the snow or permafrost initialization. OBS R/P greater than one is related to the use of two unrelated observationally based datasets. All variables are accumulated over the basins defined in HYDRA.

The simulated and observed precipitation and runoff ratio (ratio of basin accumulated runoff to precipitation) for the basins are presented in Table 3. The NCEP runoff ratio is considerably greater than the ratio of the Cogley runoff to Leemans and Cramer precipitation (1990) for many of the rivers in northern high latitudes, midcontinental North America, northern and southern Africa, Eastern Europe, and Central and East Asia consistent with the overestimated discharge in these regions. The precipitation for these same rivers is often greater than the observed, but not always (stations 24, 50–55, 81, and 82). The runoff ratio is exceptionally low for rivers in western Europe and eastern North America (stations 20, 21, 25, 26, 32–34, 61, 62, 68–73, 75, and 80) coincident with the underestimation of discharge compared to the observed. However, the precipitation differences for these regions are not well correlated with the discharge differences. Therefore, although precipitation is often a contributing factor the discrepancies between simulated and observed annual mean discharge appear to be primarily a result of the difficulties involved in accurately capturing the physics of evapotranspiration from vegetated land surfaces.

Within the Mississippi River basin the discrepancies between the simulated and observed runoff ratios agree with the previously presented discharge discrepancies (Tables 1 and 3). For example, the runoff ratio averaged for the entire basin (Vicksburg, station 42) is in good agreement with the observationally based value. However, the simulated runoff ratio is very high for the stations in the Missouri River basin (>0.3 vs about 0.10 observed, stations 43, 44, and 45) and upper Mississippi (>0.35 vs 0.23 observed, stations 40 and 41) consistent with overestimated discharge. While it is relatively low in the Ohio River basin (about 0.3 vs 0.4 observed, stations 61 and 62) in agreement with underestimated discharge at those stations. The results for the Mississippi River suggest that evapotranspiration is underestimated in the continental interior and overestimated in the Ohio Valley.

b. Simulated mean monthly river discharge

Examination of the simulated monthly mean discharge provides a more rigorous test of the accuracy of the NCEP simulated runoff. The simulated peak discharge delay (PDD) of 50 of the 90 stations is within ±1 month of the observed suggesting that the seasonality of the NCEP simulated precipitation and runoff is in fair agreement with the observations. Fifteen of the stations have a PDD within ±2 months, while the remaining 25 stations have a PDD of ±3 months or greater.

An examination of the NCEP-simulated precipitation and runoff suggests that the disagreement of the simulated PDD with observations is primarily a result of three sources of error. First, the NCEP runoff product exhibits extreme seasonality (possibly due to inaccurate snowmelt timing and/or very strong evapotranspiration in all seasons except winter). For example, peak NCEP runoff in mid- to high latitudes and mountainous regions often occurs in winter rather than occurring in the spring snowmelt season or coincident with a simulated summer precipitation maximum (not shown). As a result, many high-latitude rivers and rivers with a large input from mountains have a simulated peak discharge 3 months or more in advance of the observations (see the Amur, Kolyma, Mackenzie, Nelson, North Saskatchewan, Po, Danube, and Columbia rivers, Table 1). Second, the NCEP simulation does not always represent the observed climatology. For example, the PDD of the Murray River, Yukon, Tanana, and Indus rivers disagree with the observations because the simulated precipitation maxima are in a different season than the observations (not shown). For these rivers determining the sign of the disagreement of the PDD is somewhat arbitrary since the difference is due to differences in climatology. Third, the calculation of the river discharge effective velocity based on assumed residence times and the downstream gradient may not adequately represent the stream velocity. Including other factors, such as stream order and sinuosity of the stream may improve the simulation of the timing of peak discharge (Costa and Foley 1997; Vörösmarty et al. 1989).

c. Contribution of lakes and wetlands to river discharge

The observational estimates of river discharge include the effects of lakes, wetlands, and man-made reservoirs. Lakes, wetlands, and reservoirs moderate the seasonal amplitude of the river discharge by storing water during the wet season and releasing it in the dry season. They also affect the annual mean discharge by altering the evaporation and precipitation over the river basin. A feature of HYDRA is that lake and wetland volume and area are explicitly included in the hydrologic network. Therefore, the simulated river discharge includes the attenuation of the flow and changes in annual mean discharge due to water storage, evaporation, and precipitation over the lakes and wetlands. Additionally, reservoir volume and area is easily added to HYDRA.

The St. Lawrence River system in North America is a good example of the impact of large lakes on the discharge of a river system. The observed annual mean discharge of the St. Lawrence River at Cornwall, Ontario (downstream of the Great Lakes, site 79, Fig. 1a), is about 7500 m3 s−1. The monthly discharge varies from the annual mean by only about 20% throughout the year (Fig. 5), despite strong monthly differences in land surface runoff within the basin. The small variation from the annual mean is a result of the large volume of the Great Lakes upstream of Cornwall. The Great Lakes store the runoff from the basin as small variations in lake level and release it throughout the year. A previous attempt to simulate the monthly discharge of the St. Lawrence River using linear reservoir models (Hagemann and Dümenil 1996) failed to capture the characteristic pattern of discharge because lakes were not included.

Fig. 5.

Mean monthly and annual discharge hydrograph (m3 s−1) of the St. Lawrence River basin at Cornwall, Ontario (site 79, Fig. 1a), for observed (diamonds), simulated without the Great Lakes (triangles), and simulated with the Great Lakes (squares).

Fig. 5.

Mean monthly and annual discharge hydrograph (m3 s−1) of the St. Lawrence River basin at Cornwall, Ontario (site 79, Fig. 1a), for observed (diamonds), simulated without the Great Lakes (triangles), and simulated with the Great Lakes (squares).

In addition to the simulation with the prescribed modern mean annual Great Lakes, a second simulation was performed with HYDRA in which there were no Great Lakes. The climatological forcing is identical in both experiments. A comparison of these simulations illustrates the importance of including surface waters in river simulations. Without the Great Lakes in the model, the prescribed surface and subsurface runoff are transported rapidly across the land surface. As a result, the simulated hydrograph of the St. Lawrence river at Cornwall, Ontario (Fig. 5), has a strong seasonal signal in qualitative agreement with the simulation by Hagemann and Dümenil (1996). The maximum simulated monthly discharge occurs in April and is about 27 000 m3 s−1, while the minimum occurs in October and is almost 0. With the Great Lakes included in HYDRA, the seasonality of the discharge is dramatically reduced and the agreement with observations greatly improved. The difference between the maximum and minimum discharge with the lakes included is about 4000 m3 s−1, which is in much better agreement with the observed seasonal discharge difference of about 1500 m3 s−1 (Fig. 5). The seasonality of the discharge is still out of phase with the observations suggesting that either the residence time of the water within the Great Lakes is not yet well simulated or that the extreme seasonality of the NCEP runoff results in an early spring discharge maximum.

Including the Great Lakes also improves agreement of the simulated annual mean discharge with observations. The Great Lakes have a large impact on the annual mean water budget of the St. Lawrence river basin through evaporation from the about 240 000 km2 of surface water area. Without the Great Lakes the annual mean discharge at Cornwall, Ontario, is about 55% greater than the observed. When lakes are included the annual mean discharge is only about 15% greater than the observed discharge (Fig. 5).

Similarly, extensive wetlands in Sudan have a strong impact on the annual mean discharge of the Nile River. In addition to the simulation with the prescribed modern mean annual lakes and wetlands, a second simulation was performed with HYDRA in which there were no wetlands in the Sudd. Including the observed wetlands of the Sudd reduces the annual mean discharge of the Nile River at Dongola, Sudan (site 57, Fig. 1d), by almost 1250 m3 s−1 compared to the simulation without wetlands. This discharge reduction represents almost 50% of the annual mean observed discharge of the Nile River and illustrates the importance of including wetlands in simulations of river discharge.

d. Contribution of dams and reservoirs to river discharge

Although few river systems have lakes and wetlands of the scale of the Great Lakes or the Sudd, many large river systems of the world have been significantly modified with dams and reservoirs. As a result, the seasonal amplitude and the annual mean discharge of most large rivers systems is reduced from the natural discharge. In addition to lakes and wetlands, dams and reservoirs may be explicitly included in HYDRA in a very simple manner.

The Parana River in southern South America is one of the most strongly controlled large river systems on earth. The observed annual mean discharge of the Parana river at Guaira, Brazil (24.13°S–54.21°W; site 66, Fig. 1c), is about 8593 m3 s−1 with a maximum discharge of about 13 300 m3 s−1 in March and a minimum of 5200 m3 s−1 in August (Fig. 6). The relatively high observed discharge during the May–October dry season is a result of the discharge of water, which has been stored in the reservoirs during the wet season.

Fig. 6.

Mean monthly and annual discharge hydrograph (m3 s−1) of the Parana River basin at Guaira, Brazil (site 66, Fig. 1c), for observed (diamonds), simulated without reservoirs (squares), and simulated with large reservoirs included (triangles).

Fig. 6.

Mean monthly and annual discharge hydrograph (m3 s−1) of the Parana River basin at Guaira, Brazil (site 66, Fig. 1c), for observed (diamonds), simulated without reservoirs (squares), and simulated with large reservoirs included (triangles).

Fig. 1.

(Continued)

Fig. 1.

(Continued)

A second simulation, with identical forcing as the simulation without reservoirs, was performed in which a number of the largest reservoirs on the Parana and its tributaries were represented (see appendix). Without the reservoirs the timing of the mean monthly discharge is in good agreement with the observations (Fig. 6). However, the amplitude of the discharge is in poor agreement with the observations because the surface and subsurface runoff is not stored in reservoirs. The simulated hydrograph at Guaira, Brazil, has a very strong seasonal signal with a maximum monthly discharge of about 35 000 m3 s−1 (greater than three times the observed maximum) occurring in February. The minimum discharge is almost 0 occurring during the dry austral winter months (June–October).

Including reservoirs dramatically reduces the seasonality of the discharge on the Parana and improves the agreement with the observations (Fig. 6). In the simulation with reservoirs, the maximum simulated discharge is reduced by about 40% (to 21 100 m3 s−1) compared to the simulation without reservoirs. The reduced maximum discharge is in much better agreement with the observed discharge maximum of about 13 300 m3 s−1. The winter discharge minimum is increased from near 0, without the reservoirs, to about 5500 m3 s−1, with the reservoirs, in good agreement with the observations (Fig. 6). The Parana River illustrates the importance of including dams and their reservoirs in future global discharge simulations.

5. Discussion and conclusions

In this study a hydrologic routing algorithm (HYDRA), which integrates river, lake, and wetland processes in a single hydrologic network, has been presented. HYDRA can be forced with monthly mean or daily runoff, precipitation, and evaporation from either observations or model output to simulate river discharge, and lake and wetland levels. This study represents an initial global application of HYDRA to diagnose the accuracy of the NCEP reanalysis runoff and illustrates the utility of HYDRA for evaluating GCM simulations of land surface hydrology.

Evaluation of the annual mean discharge simulated with the NCEP runoff at 90 fluvial gauging stations indicates that, in general, the annual mean discharge is poorly related to the observations. The discharge is generally greater than observations (by 100% or more) due to underestimated evapotranspiration and overestimated precipitation for most regions of the globe (northern high latitudes, midcontinental North America, eastern Europe, India, western Asia, and northern Africa). The discharge is consistently less than the observations (by about 50%) only in western Europe and eastern North America. The simulated peak discharge delay is within 1 month of the observed at 50 of the 90 stations, indicating that despite differences on the annual mean, the runoff climatology was well simulated in many regions. However, The tendency to overestimate the seasonality of the runoff resulted in many basins having a peak discharge delay greater than ±3 months. The annual and monthly mean discharge results confirm the results of Kalnay et al. (1996); that entirely model-derived variables, such as runoff, differ significantly from the observations.

This study also indicates that HYDRA represents a logical next step in representations of large-scale terrestrial hydrology. By including lakes, wetlands, and reservoirs in the simulated hydrologic network of HYDRA, previously ignored aspects of terrestrial hydrology are accounted for. For example, large lakes and wetlands, such as the Great Lakes of North America and the Sudd on the Nile River, have a strong impact on the magnitude of the monthly and annual mean discharge to the ocean. Including lakes and wetlands within HYDRA reduces the amplitude of the seasonal cycle of discharge and the magnitude of the annual mean discharge in qualitative agreement with the observations.

The impact of man-made dams and their reservoirs on the magnitude of monthly mean discharge can be explicitly included within HYDRA. Including dams and reservoirs on the Parana River improves the agreement of the simulated mean monthly discharge with observations by reducing the amplitude of the seasonal cycle.

HYDRA can provide an important tool to the earth system science community. For example, HYDRA can be used to diagnose the accuracy of paleo-climate simulations on GCMs by comparing observed paleo-lakes with those simulated by HYDRA when forced with hydrologic output from a GCM (e.g., Coe 1997; Broström et al. 1998). The model can also be coupled to land surface and ecosystem models at various horizontal resolutions or forced with GCM output to investigate the potential impact of future climate and land use changes on river discharge, lake and wetland level and area, and reservoir storage and power generation capabilities (e.g., Costa and Foley 1997). Finally, river discharge and lake and wetland areas from HYDRA can be used as boundary conditions for more realistic GCM simulations (e.g., Hostetler and Giorgi 1992; Russell et al. 1995; Coe and Bonan 1997).

Improvements in HYDRA can be obtained through more accurate digital elevation model representations of the land surface and automated techniques for correction of river basin definition. The DEMs used to define river basins and potential lakes and wetlands are still somewhat too coarse to resolve many important topographic features, such as extensive wetland complexes in Siberia. They also have a number of errors in the very high northern latitudes and mountainous regions, which result in inaccurate representation of some river basins (Ob and Mekong). The techniques described by Graham et al. (1999) to correct errors in the definition of river basin area and flow path may significantly reduce problems associated with DEM errors. In addition, there are a number of new high-resolution (1 km or finer) and corrected 5′ resolution DEMs being created, which may lead to improved representation of the land surface within HYDRA. Additional improvements to the simulated monthly discharge can also be made through a more thorough consideration of discharge residence time and flow velocity (e.g., Hagemann and Dümenil 1996).

Note: HYDRA and the derived land surface input parameters are freely available to the scientific community. The model and input parameters can be adapted easily to coarser resolutions for use offline or to be incorporated in GCM or land surface models. Contact the author at the listed address for the web site address and information.

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Fig. 1.

(Continued)

Table 1.

(Continued)

(Continued)
(Continued)
Table 2.

(Continued)

(Continued)
(Continued)
Table 3.

(Continued)

(Continued)
(Continued)

Acknowledgments

I would like to thank Jon Foley and John Lenters for providing useful comments on this manuscript. I would also like to thank Veronica Fisher for support in developing the figures. This research was supported through the National Aeronautics and Space Administration Grant NAG5-3513 and the Climate People and Environment Program of the Institute for Environmental Studies, University of Wisconsin—Madison.

REFERENCES

REFERENCES
Birkett, C. M., 1998: Contribution of the TOPEX NASA radar altimeter to the global monitoring of large rivers and wetlands. Water Resour. Res., 34, 1223–1239.
——, and I. M. Mason, 1995: A new global lakes database for a remote sensing program studying climatically sensitive large lakes. J. Great Lakes Res., 21, 307–318.
Broeker, W. S., T.-H. Peng, J. Jouzel, and G. L. Russell, 1990: The magnitude of global fresh-water transports of importance to ocean circulation. Climate Dyn., 4, 73–79.
Broström, A., M. T. Coe, S. P. Harrison, R. Gallimore, J. E. Kutzbach, J. Foley, I. C. Prentice, and P. Behling, 1998: Land surface feedbacks and palaeomonsoons in northern Africa. Geophys. Res. Lett., 25, 3615–3618.
Coe, M. T., 1997: Simulating continental surface waters: An application to Holocene northern Africa. J. Climate, 10, 1680–1689.
——, 1998: A linked global model of terrestrial hydrologic processes:Simulation of modern rivers, lakes, and wetlands. J. Geophys. Res., 103, 8885–8899.
——, and G. B. Bonan, 1997: Feedbacks between climate and surface water in northern Africa during the middle Holocene. J. Geophys. Res., 102 (D10), 11 087–11 101.
Cogley, J. G., 1989: Runoff from the world’s landmasses: Amounts and uncertainties at 2-degree resolution. Trent Climate Note 89-3, 25 pp. [Available from the Department of Geography, Trent University, Peterborough, ON K9J 7B8, Canada.]
.
——, 1991: GGHydro—Global hydrological data release 2.0. Trent Climate Note 91-1, 13 pp. [Available from the Department of Geography, Trent University, Peterborough, ON K9J 7B8, Canada.]
.
Costa, M. H., and J. A. Foley, 1997: The water balance of the Amazon basin: Dependence on vegetation cover and canopy conductance. J. Geophys. Res., 102, 23 973–23 989.
——, and ——, 1998: A comparison of precipitation datasets for the Amazon basin. Geophys. Res. Lett., 25, 155–158.
Dickinson, W. T., 1967: Accuracy of discharge determinations. Hydrology Paper 20, 54 pp. [Available from Colorado State University, Fort Collins, CO 80523.]
.
Dümenil, L., and E. Todeni, 1992: A rainfall-runoff scheme for use in the Hamburg climate model. Advances in Theoretical Hydrology: A Tribute to James Dooge, European Geophysical Society Series on Hydrological Sciences, J. P. O’Kane, Ed., Vol. 1, Elsevier, 129–157.
Entekhabi, D., and P. S. Eagleson, 1989: Land surface hydrology parameterization for atmospheric general circulation models including subgrid scale spatial variability. J. Climate, 2, 816–831.
Famiglietti, J. S., and E. F. Wood, 1991: Evapotranspiration and runoff from large areas: Land surface hydrology for atmospheric general circulation models. Surv. Geophys., 12, 179–204.
——, and ——, 1994a: Multiscale modeling of spatially variable water and energy balance processes. Water Resour. Res., 30, 3061–3078.
——, and ——, 1994b: Application of multiscale water and energy balance models on a tallgrass prairie. Water Resour. Res., 30, 3079–3093.
Geophysical Exploration Technology (GETECH), 1995: Global Digital Terrain Model: Global DTM 5. Tech. Rep., 18 pp. [Available from the Department of Earth Sciences, University of Leeds, Leeds LS2 9JT, United Kingdom.]
.
Graham, S. T., J. S. Famiglietti, and D. R. Maidment, 1999: 5-minute, ½-degree, and 1-degree data sets of continental watersheds and river networks for use in regional and global hydrologic and climate system modeling studies. Water Resour. Res., 35, 583–587.
Hagemann, S., and L. Dümenil, 1996: Development of a parameterization of lateral discharge for the global scale. MPI Report No. 219, 32 pp. [Available from the Max Planck Institute for Meteorology, Bundestrasse 55, 20147 Hamburg, Germany.]
.
Higgens, R. W., Y. Yao, and X. L. Wang, 1997: Influence of the North American monsoon system on the U.S. summer precipitation regime. J. Climate, 10, 2600–2622.
Hostetler, S. W., and F. Giorgi, 1992: Use of a regional atmospheric model to simulate lake–atmosphere feedbacks associated with Pleistocene lakes Lahontan and Bonneville. Climate Dyn., 7, 39–44.
IPCC, 1990: The IPCC Scientific Assessment. Cambridge University Press, 365 pp.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Leemans, R., and W. P. Cramer, 1990: The IIASA database for mean monthly values of temperature, precipitation, and cloudiness on a global terrestrial grid. IIASA Working Paper WP-90-41, 62 pp. [Available from the International Institute for Applied Systems Analysis, A-2361 Laxenburg, Austria.]
.
Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climate, 10, 111–127.
Liston, G. E., Y. C. Sud, and L. Dümenil, 1994: Evaluating GCM land surface hydrology parameterizations by computing river discharges using a runoff routing model: Application to the Mississippi basin. J. Appl. Meteor., 33, 394–405.
Marengo, J. A., J. R. Miller, G. L. Russell, C. E. Rosenzweig, and F. Abramopoulos, 1994: Calculations of river runoff in the GISS GCM: Impact of a new land-surface parameterization and river routing model on the hydrology of the Amazon River. Climate Dyn., 10, 349–361.
Miller, J., G. Russell, and G. Caliri, 1994: Continental scale river flow in climate models. J. Climate, 7, 914–928.
Nijssen, B., D. P. Lettenmaier, X. Liang, S. W. Wetzel, and E. F. Wood, 1997: Streamflow for continental-scale river basins. Water Resour. Res., 33, 711–724.
Peixoto, J. P., and A. H. Oort, 1992: Physics of Climate. American Institute of Physics, 520 pp.
Probst, J.-L., and N. Sigha, 1989: Estimation de l’ecoulement superficiel et de sa charge en suspension sur quelques grands bassins fluviaux du monde. C. R. Acad. Sci. Paris, Ser. II, 309, 357–363.
Row, L. W., 1994: Terrain Base. NOAA/NGDC, CD-ROM Release 1.0.
Russell, G. L., J. R. Miller, and D. Rind, 1995: A coupled atmosphere–ocean model of transient climate change. Atmos.–Ocean, 33, 683–750.
Sausen, R., S. Schubert, and L. Dümenil, 1994: A model of the river runoff for use in coupled atmosphere–ocean models. J. Hydrol., 155, 337–352.
Stieglitz, M., D. Rind, J. Famiglietti, and C. Rosenzweig, 1997: An efficient approach to modeling the topographic control of surface hydrology for regional and global climate modeling. J. Climate, 10, 118–173.
Vörösmarty, C. J., and B. Moore III, 1991: Modeling basin-scale hydrology in support of physical climate and global biogeochemical studies: An example using the Zambezi River. Surv. Geophys., 12, 271–311.
——, ——, A. L. Grace, P. Gildea, J. M. Melillo, B. J. Peterson, E. B. Rastetter, and P. A. Steudler, 1989: Continental scale models of water balance and fluvial transport: An application to South America. Global Biogeochem. Cycles, 3, 241–265.
——, ——, ——, B. J. Peterson, E. B. Rastetter, and J. M. Melillo, 1991: Distributed parameter models to analyze the impact of human disturbance of the surface hydrology of a large tropical drainage basin in southern Africa. Proc. Symp. on Hydrology for Water Management of Large River Basins, Vienna, Austria, Int. Assoc. Hydrol. Sci., 233–244.
——, B. M. Fekete, and B. A. Tucker, 1996: Global river discharge database (RivDIS Version 1.0), Vol. 0: Introduction, overview, and technical notes. International Hydrological Program Tech. Document in Hydrology SC. 96/WS/26. [Available from UNESCO, Division of Water Sciences, 1 rue Miollis, 75732 Paris, France.]
.
——, K. Sharma, B. M. Fekete, A. H. Copeland, J. Holden, J. Marble, and J. A. Lough, 1997: The storage and aging of continental runoff in large reservoir systems of the world. Ambio, 26, 210–219.

APPENDIX

Including Dams on the Parana

The structure of HYDRA allows for easy addition of reservoirs or other surface waters not represented in the digital elevation models. The new reservoir volume and corresponding area are specified at the outlet of the reservoir to the river system. Once specified, the new surface waters are a dynamic part of the hydrology model, impacting the water budget of the river system and fluctuating in area and volume with seasonal and interannual changes in the basin water budget.

The reservoirs on the Parana were represented in HYDRA by summing the known potential volumes and areas of six large man-made reservoirs upstream of Guaira, Brazil (24.13°S, 54.21°W), to obtain a volume of about 5 × 1010 m3 of water and a surface area of about 3 × 109 m2. Because there are many more reservoirs on the Parana for which volume and area information were not available, the volume and area were arbitrarily doubled to a maximum volume of 1 × 1011 m3 and area of about 6 × 109 m2. These final values were used as a rough estimate of the maximum amount of water stored on the Parana river behind dams and were entered into the boundary conditions of HYDRA at a location on the Parana River near Guaira, Brazil. In addition, the seasonal fluctuations of the reservoirs during wet and dry seasons (respectively) and, therefore, their impact on the discharge from the reservoir was parameterized by prescribing the percent of the total potential reservoir that may be full for each month. The monthly percentages are derived by first prescribing the reservoir to be at maximum capacity when the mean monthly flux into the basin is greatest and at a minimum (10% of the total) when the flux in is at a minimum. Then the reservoir capacity is linearly interpolated between the minimum and maximum months.

Footnotes

Corresponding author address: Dr. M. Coe, Institute for Environmental Studies, University of Wisconsin—Madison, 1225 Dayton St., Madison, WI 53706.