We began by assuming that the topographic/tectonic character of a river basin plays the major role in determining its sediment load/yield, and that sediment yield was partly determined by basin area. Rather than using mean basin elevation as the topographic parameter, we used maximum headwater elevation, because in many rivers much of the sediment load comes from mountains where the river originates. The Amazon is a widely cited example, in which >80% of the sediment load is derived from the Andes, which constitute only about 10% of the river basin area (Gibbs, 1965; Meade et al., 1985). Also, maximum elevations can be estimated quickly from a topographic map. Ahnert (1970) pointed out the strong correlation between local relief and denudation (see review by Summerfield, 1991), but such a calculation becomes difficult when dealing with the number and diversity of rivers cited here.
We subdivided river basins into five categories based on the maximum elevation within the hinterland: high mountain (headwaters at elevations >3000 m), mountain (1000-3000 m), highland (500-1000 m), lowland (100-500 m) and coastal plain (<100 m). Based on a preliminary analysis of the yields, mountainous rivers, comprising the largest data set, were subdivided into three categories: Asia and Oceania (generally with very high sediment loads/yields); the high Arctic and non-alpine Europe (with low sediment loads/yields); and the rest of the world (i.e., North and South America, Africa, the Alps, and Asia Minor, Australia, etc.) Clearly this classification is not without problems. For example, in terms of relief, a small island with elevations of 800-900 m probably should be considered mountainous, not upland. Still, as seen in the following analysis, our elevation-based classification seems valid.
Geomorphologists and hydrologists often use the terms "yield," "sediment yield," or "specific yield" to compare sediment loads between disparate river basins by normalizing sediment load relative to size of the river basin (t/km2/yr). Waythomas and Williams (1988) argue, however, that statistically the comparison of yield versus basin area can give spurious results, since area is common to both axes; they propose the comparison of sediment load and basin area instead. In this paper, data are presented in terms of both yield and load.
Our data base consists of the loads and yields for 280 rivers (Table 1, Milliman and Syvitski, 1992). Collectively these rivers account for >62 x 106 km2, or about two-thirds of the land surface draining into the ocean (Milliman and Meade, 1983). Basin sizes range from <200 km2 to >6,000,000 km2, and loads vary from <0.02 to >1000 mt/yr. Where discharge values are available, we have converted them to runoff (discharge/basin area). The data come from many sources and from a wide variety of techniques, and therefore the quality is variable. Moreover, many of the data are recycled: for example, some of the data used by Lisitzin (1972) are from Strakov (1961), some of which came from Lopatin (1950) and early IAHS/Unesco compilations.
Modern river sediment loads seldom represent natural loads. Sediment discharge changes as erosion levels change or sediment is stored (i.e., river diversion projects). With the exception of Arctic rivers, where human civilization has had minimal impact, most rivers reflect the results of human activity on the erosional capacity of the rivers, both through deforestation and poor soil conservation (see Milliman et al., 1987) and urbanization (Meade, 1982). In contrast, the increased diversion and damming of many rivers has decreased sediment discharge dramatically. The Nile and Colorado deliver no sediment to the ocean, and many other rivers, such as the Mississippi, Zambesi, and Indus, have experienced markedly decreased sediment discharges in recent years. Sediment loads of other rivers have decreased because of other human activities; for example, present-day bed loads of some northeast Italian rivers are 1.5 to 20 times lower than they were in the early 1950s because of legal and illegal riverbed dredging (Idrosser, 1983; I.N. McCave written communication, 1991). Often these human impacts work in conflicting ways: dams on the Ganges have decreased sediment discharge, whereas increased erosion in the mountains of Nepal (from deforestation) has increased the load of the confluent Brahmaputra (Hossain, 1991). In this paper we cite sediment loads of rivers prior to river diversion (at least, where data are available). However, the values given in this paper still reflect increased soil erosion and thus probably are higher than they would be in natural conditions.
Plots of runoff versus basin area, load versus runoff, and load/yield versus basin area (Figure 5.2) show a variety of trends. Runoff decreases with increased basin area (Figure 5.2a), probably because larger river basins tend to include a greater proportion of "lowland," with reduced precipitation and increased evapotranspiration (D. Walling, written communication, 1991). Also, our data for smaller rivers are biased toward rivers with high runoff, as small rivers with low runoff are seldom gauged. With respect to sediment load versus runoff, we find the same random relationship noted by Walling and Webb (1985) for load versus precipitation (Figure 5.2b). In contrast, load/yield vary directly/indirectly with basin area, although the scatter is considerable (Figure 5.2c, d).
When we divide the rivers into the seven topographic categories, a number of trends show much better correlation.