Validating the runoff from LSP-SVAT models using a global river routing network by one degree mesh

Proc. 13th Conf. on Hydrology, Long Beach, California
Amer. Met. Soc., 319-322, 1997.

Taikan Oki
Laboratory for Atmospheres, NASA/Goddard Space Flight Center, Greenbelt, MD

[!!] This manuscript may be different from the original article.

This document is encoded by Western (Latin-1). Can you display ``±'', ``²'', and ``×'' without any problems?



1. Introduction

The Global Soil Wetness Project (GSWP) has been put into practice under the International Satellite Land Surface Climatology Project (ISLSCP). All the required atmospheric forcing data and physical parameters for Land Surface Parameterizations (LSPs) are given for the period from Jan 1, 1987 to Dec 31, 1988, and the simulated global soil wetness will be examined in the GSWP. River discharges are expected to validate the results from LSPs; however, they are given by point values even though all the data except for river discharge are given as spatial data. This is because there has been no available information about global river basins in 1 degree grids. Thus, a global river channel network (GRCN) of 1 degree grids is necessary for promoting the GSWP.

Consequently, a GRCN in 1ox1o grid boxes was produced in this study. The GRCN was used for the comparison of annual runoff between model simulation and observation, and it was also used for the river routing.


2. Making 1ox1o mesh GRCN

The simple style of channel network was selected for the GRCN, that each grid may have up to 7 inflows from, and only one outflow to its eight neighbor grids.

Since the spatial scale is quite different from previous techniques extracting river channels for relatively small river basins from digital elevation models (DEMs), it may not adequate to apply these techniques for global DEMs. Therefore it is decided that automated river-channel extraction is used only for making base information of the GRCN, and manual corrections will be applied afterword.

The manual correction may be criticized for its subjectivity and a vast consumption of time. However, any result by an automated technique should be validated by some reliable atlases of the world in the case of a GRCN, and if the result has big differences from reality, the result should be corrected manually or the algorithm should be improved. It was judged that it will cost longer time to develop an algorithm which results in a perfect GRCN without manual correction than to use a simple algorithm and spend time for manual correction. In the case of the GRCN, the reality of the result has the higher priority than the subjectivity of the way of making the network.

ETOPO5 (Edwards, 1986) was used in this study as a global DEM. As a simple implementation of an automatic river channel extraction, the outflow direction is determined to the lowest point of its eight neighboring grids, of course, only when the point is lower than the originating point. Details are described in Oki and Sud (1997). The result obtained from the above algorithm showed a fairly good GRCN. Rough stream structure and drainage system to the ocean were well represented by this automatic procedure. The `hollow' grids were found at reasonable areas such as with marshes, lakes or at the confluence of large tributaries.

The GRCN obtained from the automated method was manually corrected referring with published geographical information. The preliminary GRCN obtained by the automatic procedure was illustrated using the General Mapping Tools (GMT, Wessel and Smith, 1995) overlaid with river channels.

Major problems were encountered in the arid area. There are a lot of `wadis' in dry regions, and these streams occasionally contact to the main stream. Sometimes it can be a philosophical issue to decide whether an area should be included in a major river basin or not. Some grids with inland lakes were attributed to river mouths, and river mouths are also arbitrarily put in arid area in order to terminate unnatural river channels.

Figure 1 illustrate a part of the obtained GRCN of major continents. Grids of river mouth are plotted by (star) and other grids are by (dot). The thicknesses of connecting lines correspond to the river order of streams. River mouth, farthest grid from the river mouth (origin of calculating the river length) and gauging stations of identified river basins are marked by (circle), (triangle), and (inverse triangle) respectively. Lake grids are marked by (rectangle), where the sea grids on the land/sea mask were changed into land grids.

Rivers with larger than 100,000 km² area in the GRCN are identified, and river basins over 50,000 km² listed in Korzun (1978) are adopted, too. As a result, 180 rivers are identified. These identified basins cover 83.5 × 106 km² which corresponds to 56% of whole land and 62% of land excluding Antarctica. Most of the uncovered areas are arid regions and these areas should have less influence on the global river runoff. River basin area in the GRCN is compared with published numbers from Matuyama and Oki (1992), Korzun (1978), and Milliman and Meade (1983) (Figure 2). Basically, basin sizes are well reproduced in most rivers, and their differences are within ± 20% compared to previously published numbers.

Currently more than 300 gauging stations are selected and their `own' drainage areas are from less than 10,000 km² to more than 1,000,000 km². The medium size is approximately 110,000 km² while average size is 190,000 km², and standard deviation is 330,000 km². Since the area of 1x1 degree grid box is approximately 10,000 km², the mean size of 190,000 km² is equivalent to that global continents are effectively divided by 4.5 × 4.5 degree boxes. These drainage areas currently cover 60.8 × 106 km² which corresponds to 41% of whole land and 45% of land excluding Antarctica. Further data collection of runoff will increase these numbers.

Figure 3 compares the drainage area of selected gauging stations. The area sizes in the GRCN are generally well corresponding to official numbers associated with the runoff data. The overall accuracy is very good: bias is 9,800 km² or 3.8% and root mean square error is 51,000 km² or 18%.

River length in the GRCN is compared with the numbers from Korzun (1978) and National Astronomical Observatory (1993) (Figure 4). It is reminded again that the GRCN estimated here is just giving information of lateral water movement in global scale, and the stream length in the GRCN should be regarded as imaginary.

However, the imaginary stream length should be related to the real stream length when the GRCN will be used for a `physically based' river routing model. To examine the relation, the longest river length in the GRCN is compared with the representative river length in each river basin. The published river length is generally larger than that in the GRCN, and it should be mainly because that the detailed meandering of river channel is not well represented by 1x1 degree grid boxes. Therefore `meandering ratio' (MR) is defined as a ratio of published river length to that of the GRCN. MR indicates how the real stream length is larger than the length in the GRCN, and most of MRs for large rivers are from 1.0 to 1.6.

MRs of river basins smaller than 500,000 km² scatter more. This may not be only due to the meandering of stream. For some of these small rivers, even the rough structure of the main stream is not well reproduced by 1 × 1 degree grid boxes, and it makes the accuracy worse. One can apply MRs depending on its area size, or use different values in each river basin. Overall mean MR of 1.40, or 1.28, which is averaged only for rivers larger than 500,000 km², may be used universally as a first order approximation.


3. Comparing annual runoff

Annual runoff simulated by SSiB (Mocko et al., 1997) was compared with observations (Figure 5). A template of one degree mesh drainage areas was constructed using the GRCN for the stations where discharge data are available in 1987 and 1988. Weighted mean of the total runoff simulated by SSiB was calculated according to the template. Monthly mean discharges (m3/s) from 52 stations are converted to runoff (mm/year) by their official drainage areas.

The correspondences between simulated and observed are fairly well, but the scatters are not small. There may be some reasons why they do not match exactly.

One reason should be the data quality. Runoff may be influenced by human activities, such as reservoirs and regulations. River water may moisten the surface soil at downstream either by irrigation or naturally, and it will change the water balance by the evapotranspiration of runoff water generated at upstream. Precipitation is prescribed in the simulation, and its accuracy is critical for the output, as well.

The location of drainage area may be shifted in the GRCN. If the spatial gradient of water balance is large, the comparison becomes poor for that station. Actually, precipitation is approximately 800 mm/y but runoff is approximately 50 mm/y for the Chari river basin. On the contrary, annual runoff is almost the same as annual precipitation or more than that for the Irrawaddy, the Brahmaputra and the Columbia river basins. Further investigations are required to identify the cause of these unrealities.

Of course, the land surface parameterization models may have problems. The SSiB tends to produce higher annual runoff compared to the observation in dryer areas. Yet it is still unclear which is right, because irrigation may be active in these regions.


4. Linear river routing model

A linear river routing model was applied for the total runoff from the SSiB simulation (Mocko et al., 1997) in order to compare the monthly runoff with observation.

Conservation of river water storage can be written as

[equation 1]

where Src is the river water storage, DIN is the sum of inflow from neighboring grids and generated runoff. The flow direction from the GRCN was used to calculate the DIN. DOUT=c Src is the outflow through river, and c = 1/k by Liston et al. (1994) or c = u/d by Miller et al. (1994), where k is the time constant, u is the effective velocity and d is the distance between grids. Let Ct= exp(-c dt) and,

[equation 2]

is derived with constant during t = t0 to t0+ dt. As a preliminary test, c = u/d was used and a universal meandering ratio of 1.4 was applied for the calculation of d. Effective velocity u is the only tuning parameter.

As for the time step, u dt < d should be satisfied. In polar regions, d is less than 20,000 m for 1o distance in east-west direction, then dt should be smaller than 3 hour for u= 2.0 m/s.

Spin up was done for the year of 1987 until the Src of the first day becomes within ± 5% of the first day of the previous year for more than 95% grid boxes. Residential time of river water is very short, and two years of spin up was sufficient for most cases.

Figure 6 is an example of the results for Vicksburg in the Mississippi river basin. Introducing the routing model improves the shape of hydrograph and the peak value corresponds best by the effective velocity of u= 2.0 m/s. Actually, u= 2.0 m/s is faster than the optimal value of 0.3 m/s by Kanae et al. (1995), but their GRCN is by 5o× 5o degree grid boxes and the river channel length may not be adequately represented. In the case of low flow, it is not nicely reproduced even for u= 2.0 m/s. This will be improved by introducing a ground water reserver as Liston et al. (1994) did. The comparisons of other gauging stations are basically similar to the result of Vicksburg.


5. Summary

A global river channel network (GRCN) in 1o× 1o was produced. Areas of whole river basin and drainage area for selected gauging stations were within 20% differences compared to published or official numbers. Meandering ratio (MR), the ratio of actual river length to river length on the GRCN was defined and examined. As a result, MR averaged for all available data is 1.40 and it is 1.28 for rivers larger than 500,000 km².

The template obtained from the GRCN was used to take mean values of simulated runoff by SSiB (Mocko et al., 1997) over drainage areas, and annual mean was compared with observed discharge. A linear river routing model was introduced in order to compare the model simulation in monthly basis with observation. A preliminary result showed that the model simulation is not bad, and effective flow velocity of m/s gives good correspondences. Further detailed examination will be carried out for the data quality control and routing model development, and different runoff data produced by other land surface parameterizations will be examined.



Edwards, M. H. (1986)
Digital image processing of local and global bathymetric data, Master's thesis, Washington University, Department of Earth and Planetary Sciences, St. Louis, Missouri, pp. 106.
Kanae, S., T. Oki, and K. Musiake (1995)
Hydrograph estimations by flow routing modelling from atmospheric general circulation model output in major basins of the world, In 2nd International Study Conference on GEWEX in Asia and GAME, 154-157.
Korzun, V. I. (1978).
World Water Balance and Water Resources of the Earth, Volume 25 of Studies and Reports in Hydrology, UNESCO, pp. 587.
Liston, G. E., Y. C. Sud, and E. F. Wood (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(3), 394-405.
Matsuyama, H. and T. Oki (1992)
GRDC and the runoff data, J. Japan Soc. Hydrol. & Water Res., 5(2), 65-70.
Miller, J. R., G. L. Russell, and G. Caliri (1994)
Continental-scale river flow in climate models, J. Climate, 7, 914-928.
Milliman, J. D. and R. H. Meade (1983)
World-wide delivery of river sediment to the oceans, J. Geology, 91, 1-21.
Mocko, D. M., G. K. Walker, and Y. C. Sud (1997).
Sensitivity of surface fluxes, soil energy, and hydrology to parameterization of evapotranspiration - a study based on the ISLSCP Initiative I data, In Proc. 13th Conf. on Hydrology, 331-334.
National Astronomical Observatory (1993)
Rika Nenpyo (Scientific Charts), National Astronomical Observatory, Maruzen.
Oki, T. and Y. C. Sud (1997)
Design of the global river channel network for Total Runoff Integrating Pathways (TRIP), Earth Interactions, 1, submitted.
Wessel, P. and W. H. F. Smith (1995)
New version of the generic mapping tools released, EOS Trans. Amer. Geophys. U., 76, 329.



[Fig.1] Figure 1: One degree mesh Total Runoff Integrating Pathways (TRIP) of North America. River mouths are plotted by (star). The thickness of line corresponds to the river order of streams. For identified rivers, river mouth, farthest grid from the river mouth, and gauging stations are marked by (circles), (triangles), and (inverse triangles), respectively. Lake grids are marked by (rectangles). [PostScript file is here.]


[Fig.2] Figure 2: Comparing whole area sizes of identified major river basins. [PostScript file is here.]


[Fig.3] Figure 3: Comparing drainage area sizes of selected gauging stations. [PostScript file is here.]


[Fig.4] Figure 4: Comparing river lengths(km) in GRCN with published numbers. [PostScript file is here.]


[Fig.5] Figure 5: Comparison of annual runoff simulated by SSiB and observation for calendar years of 1987 and 1988. [PostScript file is here.]


[Fig.6] Figure 6: Comparing monthly runoff at Vicksburg in the Mississippi. [PostScript file is here.]


There were accesses to this page in , .

(Last updated at Monday, 29-Jun-1998 12:01:07 JST, by Taikan OKI)