Global Biogeochemical Cycles

Global coastal segmentation and its river catchment contributors: A new look at land-ocean linkage



[1] Here we present the COSCATs global database of 151 catchments in exorheic areas. The catchments connect to oceans through coastal segments according to three sets of criteria: natural limits (continents, oceans, regional seas, major capes, and bays), continental shelf topography (sills, basins, island chains), and geophysical dynamics (climate, ocean currents and tectonics). The COSCATs segmentation scheme is designed to improve Earth System analysis and to harmonize reporting of global riverine transfers from land to oceans. Each COSCAT is characterized by its coastal segment limits and length (median 2 400 km), by its catchment characteristics, including area (median 0.45 M km2), width, latitudinal range, runoff average value and direction, including its related physiographic units (n = 500). We apply the COSCAT segmentation to all 151 basins to estimate water discharge and total nitrogen impacts to oceans and find that the average runoff (mm/yr) and N yields (YN in kg km−2 yr−1) range over more than 3 orders of magnitude at this coarse resolution, and that their average population density ranges over 2 orders of magnitude. Hyperactive regions, defined as segments with 5 to 10 times the world average yield (river transfers per unit area of land), are differentially placed for water runoff and total contemporary nitrogen. COSCATs have been designed to facilitate the budget reporting and the analysis of global scale heterogeneity for riverine fluxes and can be applied to other material, such as suspended solids, carbon species or other nutrients, particularly for areas draining into regional seas.

1. Introduction

[2] The land-ocean connection through river catchments is facing multiple issues. Most world atlases, either general or oceanographic, have a somewhat schizophrenic representation of global physical geography. General atlases feature major river basin networks and continental relief but completely mask submarine morphology, while ocean atlases generally leave continents devoid of any information, particularly with regards to the river network. It is therefore difficult to visualize the linkage between land and ocean through rivers except in rare instances. Earlier physical geographers [Berghaus, 1891; De Martonne, 1955] attempted to differentiate the regions of the world that drain internally toward “interior seas” (endorheism), such as the Caspian and Aral, and lakes as Lake Chad from those facing the world oceans and their regional seas (exorheism).

[3] The importance of river material inputs to oceans has been recognized by many authors following Garrels and Mackenzie [1971] and Garrels et al. [1973] in their pioneering work on the cycling of sedimentary rocks and on global geochemical cycling. Their followers constructed the riverine inputs of water, sediment, carbon, nutrients and metals and gradually subdivided these fluxes for each ocean [Martin and Meybeck, 1979; Meybeck, 1979, 1982, 1987; Milliman and Meade, 1983; Walling and Webb, 1983; Degens et al., 1991; Milliman and Syvitski, 1992; Milliman et al., 1995; Probst, 1992; Caraco, 1995; Howarth et al., 1996; Caraco and Cole, 1999; Milliman, 2001; Oki et al., 2001; Fekete et al., 2002; Seitzinger et al., 2002; Amiotte-Suchet et al., 2003; Smith et al., 2003; Vörösmarty et al., 2003; Syvitski et al., 2005]. Some of these authors have pointed out the coarse-scale heterogeneity of riverine inputs to oceans where fluxes were aggregated at regional scales, such as Southeast Asia and Pacific islands for suspended solids [Milliman and Syvitski, 1992], humid tropics for water, carbon and silica [Meybeck, 1979, 1993; Walling and Webb, 1983; Probst, 1992], western Europe, North America, Southeast Asia for all nutrient estimates, and North Atlantic for nitrogen [Howarth et al., 1996]. However, these budgets have generally been reported with different spatial boundaries.

[4] In the last decade, the land-to-ocean linkage has been one focus of two international programmes, IGBP-BAHC [Vörösmarty and Meybeck, 2004; Meybeck et al., 2004] and IGBP-LOICZ ( Advance in mapping have now made in possible to differentiate endorheic from exorheic regions at the global scale, and to identify the drainage area of regional seas (e.g., South China Sea) and local seas (e.g., Adriatic). The estimation of global fluxes has also evolved. In the past, land-to-ocean fluxes have been established on the basis of a few documented major world rivers, through simple extrapolation [Martin and Meybeck, 1979; Meybeck, 1979, 1982], or through multiregression analyses [e.g., Jansen and Painter, 1974; Milliman and Syvitski, 1992; Probst, 1992; Howarth et al., 1996; Seitzinger et al., 2002]. Recent studies have estimated these fluxes using elaborated transfer models [Green et al., 2004] (see also papers in this special section), which take into account different natural and/or anthropogenic drivers of river transport from large-scale filters [Meybeck and Vörösmarty, 2005] to fine-scale processes [Sferratore et al., 2005]. The spatial resolution of these models, which started at 2° × 2°, has now reached 30′ × 30′ or even finer resolutions.

[5] The IGBP-LOICZ programme promotes local to global approaches that combine riverine nutrient and pollutant inputs and their impact on the coast at the 30′ × 30′ resolution, that include more than 6000 coastal cells, or finer ( [Smith et al., 2003]).

[6] Here we delineate and characterize a new set of ocean coastline segments (hereafter coastal segments) with their corresponding river catchments (COSCATs), based upon the global digitization of the continental landmass developed at the University of New Hampshire and the University of Paris VI within the IGBP-BAHC programme [Vörösmarty et al., 2000a, 2000b; Vörösmarty and Meybeck, 2004], with the following objectives: (1) delineation of the coastal zone into a set of coastal segments that are manageable for identification and characterization purposes; (2) linkage of the coastal segments to their related mega-catchments (COSCATs) for the whole ocean-facing side of continents; (3) characterization of these segments and catchments by geomorphic and hydrologic attributes; and (4) analysis of the coastal segments for water runoff, total N input and population pressure on the coastal zone at the global scale.

[7] First we review the major existing segmentations of the global coastline, and then, on the basis of a set of physical geographic criteria, we define and describe 151 main segments and catchments, which will be suited for use in various disciplines of Earth System Science as coastal geomorphology, geochemistry, hydrology, sedimentology, and ecology.

[8] The geographic distribution of river inputs to oceans has first been considered by global scale soviet geographers in their pioneering Physico-Geographic Atlas of the World [Gerasimov et al., 1964] and by hydrologists [Baumgartner and Reichel, 1975; Korzoun et al., 1978].

[9] Parallel to these approaches the ocean ecologists have divided the world oceans and their regional seas into a set of relatively homogeneous entities: Large Marine Ecosystems (LMEs), established by NOAA and IUCN (, and their linked watersheds. In turn, LMEs are the basis of another global scale classification of the coastal zone and related drainage basins made by the Global International Water Assessment [GIWA Core Team, 2001; Hempel and Daler, 2004] established by UNEP Global Environment Facility (GEF). The global LME delineations are primarily targeted at facilitating the assessment of the ecological state of the marine environment including the coastal zone while the GIWA classification is aimed to facilitate the assessment of human pressures and impacts, on water uses within drainage basins and on coastal zone management. Although LME and GIWA classifications are generally similar, they are not identical. Further their suitability for use in Earth System Science, for example, in global scale models of land to ocean fluxes, may be limited because: (1) the size and geometry of these coastal entities has a large variability (some entities only include few islands, e.g., Feroe Islands, while others are very large, such as the Russian Arctic Basin), (2) some are not defined by river catchment boundaries but by river courses, and others are delineated by political boundaries, and (3) some entities are shared by several oceans or continents.

[10] Robinson and Brink [1998] also reviewed the dynamics of 32 coastal entities. Previous attempts of coastal segmentation are detailed and discussed in auxiliary material (Annex A).

[11] Both the GIWA and LME approaches have important limitations, and combine different segmentation rules. These two approaches do not allow for the assessment of river budgets for entire continents or by individual marine sedimentary basins.

[12] In contrast, the LOICZ database ( is flexible and can support multiple segmentation schemes according to the study objectives, including global river inputs to oceans [Smith et al., 2003]. It is targeted to the finest possible description of the coast and its functioning including for local management. However, it does not explicitly take into account the geomorphological boundaries of the coast, either on land or on the continental shelf, and is therefore not suitable for the estimation of global river material budgets at coarse resolutions, nor to the analysis of the distribution of river fluxes on land, although this objective can be easily met from the LOICZ database.

2. COSCAT Segmentation Approach

2.1. Segmentation Criteria

[13] The coastal segmentation and its related catchment data set (COSCATs data set) proposed here is an attempt to reduce some of the shortcomings of previous types of delineation of exorheic areas and/or to increase potential applications in Earth System analysis both from an ocean and continental perspective. It also aims to characterize the coast and its related catchments with a manageable set of 151 segments including five Greenland segments and eight hook-up segments, discussed later, and aims to allow eventual clusters, for local and regional seas, ocean basins and continents, biomes and climate zones, and human pressures. Seven major criteria (C1 to C7) are used to define how coastal zones are segmented in the COSCAT data set as follows. Additional minor criteria are used (Figure 1, Annex B).

Figure 1.

Schematic representation of land-ocean linkage at the global scale. Limits of COSCATs: A, coastal morphology (e.g., cape); B, runoff gradient; C, coastal morphology (Island Chain); D, coastal morphology (widening shelf); E, ocean current divergence; F, plate tectonics; G1, G2, G3, limits of hook-up segments.

[14] 1. In C1, coastal segments are not delimited by major rivers. The assignment of major river inputs to a segment needs to be clear. For the largest rivers (Amazon, Congo, Chang Jiang, Mississippi, etc.), segment boundaries have been placed as far as possible from the river mouth in order to better distribute the influence of these mega-inputs to the whole coastal segment.

[15] 2. In C2, segments are first limited by natural continent and ocean boundaries.

[16] 3. In C3, major capes and the regional sea limits are defined as natural boundaries, for example, A and D on Figure 1.

[17] 4. In C4, shelf morphology and ocean floor relief are considered to relate the continental inputs to major marine sedimentary basins [Lisitzin, 1996]. Coastal segment endmembers are delineated when undersea sills reach the coast and/or limiting regional seas (Okhotsk, Andaman, Caribbean seas, Gulf of Guinea, Bering Sea, Western Africa segments, etc.) (e.g., C, Figure 1).

[18] 5. In C5, the segmentation is preferentially made in the maximum climate gradient area in regions of high climate variability (e.g., B, Figure 1). This criterion is particularly applied to gradients in coastal precipitation in order that resulting segments are more homogeneous (e.g., the Guinea–Sierra Leone–Liberia coasts are within one segment).

[19] 6. In C6, coastal currents are considered in cases where major diverging coastal currents are hitting the coast (e.g., E, Figure 1). For example, a coastal segment boundary is placed between the Amazon mouth and Capo de Sao Roque, assuming that most of the Amazon waters are diverted into the North Atlantic Ocean. If the material budget for a given marine sedimentary basin requires the combination of both land and ocean inputs, then segments separated by currents can be clustered.

[20] 7. In C7, former sea level at the Last Glacial Maximum is considered. The reconstruction of past land-ocean linkages during the Last Glacial Maximum may be required for Earth System analysis. Therefore we have checked that the coastal catchment delineation remains coherent when sea level was lower. This criterion was particularly applied to the South China Sea, the Siberian platform, the Patagonia coast, the North Sea and for the Arafura Sea/Carpentaria Gulf, the catchment of which currently combines two COSCAT entities (1416, North Arafura Sea and part of 1415, South Arafura). The boundaries of continents and oceans are detailed in auxiliary material (Annex B).

2.2. Data Sources, Procedures, and COSCAT Gazetteer

[21] First, we applied the criteria to delineate coastal segments through visual analysis of atlases, including the Physico-Geographic Atlas of the World [Gerasimov et al., 1964] (probably the best of its kind so far), the Physical Relief Globe of the National Geographic Society (1974), and the Times Atlas [Times, 1983, 1989]. We also made use of several works on continental morphology, including that of Snead [1980], Bridges [1990], and Summerfield [1991], and on ocean morphology, including Vanney [1991], Pernetta [1994], and Leier [2001]. The Encyclopaedia of Oceanography and Encyclopaedia of Geomorphology [Fairbridge, 1966, 1968] have been widely used as well. In particular, we relied upon Bridges [1990] to define most of the world's physiographic units in the coastal catchments (Annex tables), and upon Lisitzin [1996] for a description of oceanic sedimentary basins.

[22] The application of segmentation criteria on a global digitized map at a 30′ × 30′ resolution has generated the main coastal segments (n = 151, Antarctica excluded), which have been identified by their boundary cells using the GTOPO30 continental and oceanic topography databases [U.S. Geological Survey EROS Data Center, 1996]. Both databases have been used in previous works describing the potential river network [Vörösmarty et al., 2000a, 2000b; Fekete et al., 2002].

[23] Next we automatically generated the coastal catchments related to each coastal segment on the basis of a simulated topological network of 6000 world river basins at 30′ resolution (STN 30) as established by Vörösmarty et al. [2000a, 2000b] and Fekete et al. [2001] (see an example for the Western Europe/Western Mediterranean segments in Annex E). These basins are defining potential catchments that are strictly defined by land topography: Some have no active runoff in present-day climatic conditions, i.e., their drainage network is fossilized [Vörösmarty et al., 2000a, 2000b]. Each COSCAT is therefore aggregating dozens of those elementary basins defined at the 30′ × 30′ resolution. Finally, the general characteristics of the COSCAT catchments have then been determined from five databases using ESRI Arc Info at the 30′ × 30′ resolution (see Annex D for COSCAT data set and Annex F for the Western Mediterranean COSCAT example): potential hydrological network at 30′ × 30′ [Vörösmarty et al., 2000a, 2000b], NOAA World Vector Shoreline [Soluri and Woodson, 1990], population in 1995 at 30′ resolution [Vörösmarty et al., 2000c], runoff distribution at 30′ × 30′ [Fekete et al., 2002], total nitrogen flux from rivers (contemporary) at 30′ × 30′ [Green et al., 2004].

[24] COSCATs names are derived from their receiving ocean entities as mentioned in our selection of ocean atlases and encyclopedias [e.g., Vanney, 1991; Pernetta, 1994; Fairbridge, 1966]. COSCATs names are used for both the coastal catchments and their coastal segments.

2.3. Metrics Used in Coastal Segment, Coastal Catchment, and Coastal Ribbon Description, and Segment Coding

[25] We associate each coastal segment to a coastal ribbon of land so that comparisons with the rest of the catchment may be made (e.g., characteristics of coastal runoff, coastal population density, coastal lithology). Coastal ribbons are defined here as the 50-km-wide portion of the lands at the ocean interface as defined at a 30′ resolution (see an example in Annex F, Figure fs02). The coastal ribbon can eventually be characterized by the same descriptors as an entire catchment. The area of a coastal ribbon is defined here as the coastal segment length (in km) multiplied by 50 km. With such an operational definition the coastal ribbon area is no longer strictly defined at the 30′ resolution. We define the coastal population as the percentage of the 30′ resolution cells lying within 50 km of the coast as defined by the 30′ coastline (e.g., 100% of the population of a given cell is counted as coastal population if the entire cell lies within 50 km of the coast). As such coastal cells can be connected by their corners, their side or not be connected at all, i.e., the coastal catchment is so thin that the coastal ribbon is disconnected (see examples for Ligurian and Algerian coasts in Annex F, Figure fs02). Since cell size varies with latitude, depending on projection, the number of inland ribbon cells also varies (see example Figure fs01 in Annex E). In rare coastal features where the STN30 cell does not discharge directly to the coast, but to inland drainage, for example the Tyrrhenian coast East of Genoa (see Annex F) that drains into the Po catchment, the population of these cells is not counted as coastal population.

[26] We characterize each coastal segment and each catchment using 15 metrics, which, for this first version of COSCAT data set, concentrate on morphologic characteristics, as follows. Here, subscript “c” refers to the coastal ribbon, subscript “t” to the whole coastal catchment, “C” to the coastal segment center and “T” to the centroid of the catchment.

[27] 1. Lc denotes coastal segment length (km) at the 30′ × 30′ resolution, as determined by the trajectory linking all successive adjacent cells of the segment. This length is different from the “real” length as defined by Bartley et al. [2001] on the basis of the NOAA World Vector Shoreline [Soluri and Woodson, 1990].

[28] 2. At denotes coastal catchment area (M km2), defined at the 30′ × 30′ resolution.

[29] 3. Ac denotes coastal ribbon area (M km2), defined as Lc × 50 km width.

[30] 4. Wt denotes mean width of coastal catchment (km), defined as At/Lc.

[31] 5. Ltm denotes maximum possible length of river network (km), defined at 30′ × 30′ resolution within each coastal catchment on the basis of the potential river network, whether presently discharging to the ocean or not.

[32] 6. Ltm/Wt denotes relative length of river network (km).

[33] 7. Notation (x, y)T denotes position of centroid (latitude, longitude) for each costal catchment. It grossly defines the catchment position and allows for the worldwide analysis by latitude. For strangely shaped segments where this point fell in water it has been put by hand on land cells.

[34] 8. Notation (x, y)C denotes position of coastal segment center, which is the cell that has an equal number between it and the two most distal cells. It defines the segment position and allows for the worldwide analysis by latitude. The relative positions of catchment centroid and coastal segment center are used to define the overall direction of water flow within each catchment (α).

[35] 9. Notation (Δlat)TC denotes latitudinal difference (in degree) between coastal segment center (T) and coastal catchment centroid (C). It is a coarse indicator of the intracatchment climatic heterogeneity. In Annex C, northern and southern latitudes are respectively counted as positive and negative; eastern and western longitudes are respectively counted positive and negative. Catchments flowing north have a positive difference, those flowing south a negative one.

[36] 10. Ac/At denotes relative size of coastal ribbon, defined as the ratio of the coastal ribbon area over the catchment area. The coastal ribbon has a uniform width of 50 km.

[37] 11. Notation α denotes flow direction in degrees, counted from 0° (north) clockwise through to 359°, as determined by the angle between catchment centroid and the coastal segment center. A northeast flowing river will have a 45° vector angle, a northwest flowing river will have a 315° angle, etc.

[38] 12. Notation qt denotes average runoff (mm/yr) of the whole coastal catchment, defined as the sum of total water discharge to each coastal cell Vt divided by the total catchment area At.

[39] 13. Notation dtpop denotes average population density (p/km2), defined as the total population of whole catchment divided by the total catchment area.

[40] 14. Yt totN denotes yield of total nitrogen as carried by rivers to coast (kg km−2 yr−1).

[41] 15. Vt denotes annual volume of water discharged to the coastal segments (km3 yr−1). COSCATs coding has four digits; the first two are for the continents, the last two for the catchments within the continents (e.g., 04XX indicates Europe, and 0406 is the code for Gulf of Finland).

[42] Annex C tables provide the complete list of coastal segments together with the related major river basins (n = 375) as listed by Meybeck and Ragu [1995] and the principal physiographic regions of the oceans. The metrics of each coastal segment, ribbon, and catchment are presented in Annex D.

[43] We have made a preliminary assessment of the correspondence between COSCAT entities and Large Marine Ecosystems coastal (LMEs) and Global International Water Assessment continental and coastal (GIWA) entities (Annex C tables).

3. COSCAT Segments and Catchments

3.1. COSCAT Global Puzzle

[44] The coastal catchments generated by the coastal segments, look like a giant-puzzle of the continental area (Figure 2). The puzzle elements can be clustered into different configurations, which allow the budgeting at local (e.g., Adriatic Sea), regional (e.g., SW Africa, Caribbean Sea), continental or oceanic scales of all water borne fluxes as dissolved salts, sediments, nutrients, carbon and contaminants.

Figure 2.

Delineation of COSCATs catchment and coastal segments and limits of continents/ocean drainage basins (Africa, Asia, Australasia, Europe, and North and South America; numbers refer to Annex C tables and Annex D). Major endorheic regions are as follows: A, Central Asia endorheic plus Aral basin, Caspian basin; B, Lake Eyre basin; C, Way-Carnegie basin; D, Dead Sea; E, Umm As Samim basin; F, East African Rift; G, Chad Basin; H, Okawango; I, Altiplano and Mar Chiquita basin; J, Great Basin.

[45] In each piece of the puzzle, the transfer of river material originating from any physiographic entity (Annex C tables) follows the river network, sometimes over thousands of kilometers, and connects to a receiving ocean basin. At the catchment intersections (triple points in the topography), the neighboring upstream cells can be connected to very distant coastal cells, sometimes located in different oceans. Some of the major exorheic river basins of the world (n = 375 out of 6000 catchments defined at the 30′ × 30′ resolution) are attributed to a specific coastal segment and to a catchment (Annex C tables).

[46] The application of the seven selection criteria has generated a few catchments with odd and unexpected shapes: (1) the Rio de la Plata COSCAT (code 1108) includes Parana and Uruguay Rivers catchments plus some local catchments located on a narrow strip of land in Uruguay, (2) the Newfoundland COSCAT (code 0826), which has been set up to separate the Labrador Coast catchment from the New England COSCAT (code 0827) catchment and to individualize the Saint Lawrence COSCAT (code 0825). Also, there are eight hook-up segments such as South Timor COSCAT (code 1333) and the North East Black Sea COSCAT (code 0413), which have only a few cells. The number of segments analyzed in our statistical analysis of COSCAT metrics (Tables 1 to 7) varies slightly from 138 to 141 depending on the metrics; some hook-up segments and the five Greenland COSCATs are discarded in this analysis.

Table 1. Distribution of COSCATs Segment Length for 139 COSCATsa
Lc, km0–10001000–20002000–30003000–40004000–50005000–60006000–70007000–80008000–90009000–10,000Total Length (Lc)i, kmTotal Exorheic Drainage (At ex)i, M km2Average Catchment Width (Wt)i, km
  • a

    For each continent: COSCATs segment length, Lc; total coastal length (30′ resolution), (Lc)i; total exorheic drainage, (At ex)i; average width of catchments, (Wt)i. Few hook-up segments are omitted.

  • b

    Caspian and Aral basins not considered.

  • c


  • d

    Canadian Archipelago (0816).

  • e

    Sunda (1330), Sulu-Celebes (1331).

North America210103511d102,69021.86213
South America268..31,56017.39551
Australasia7532 40,1106.72168
Global exorheic7484714118211380,970114.46295

3.2. Morphological Characteristics of Coastal Segments and Their Catchments

3.2.1. Segment Length (Lc)

[47] As defined, most coastal segments have a length exceeding 1000 km (Table 1) with a maximum reaching 6000 km. The longest segments correspond with major archipelagos: the Canadian Archipelago COSCAT (code 0816), Sunda COSCAT (code 1330), and Sulu/Celebes COSCAT (code 1331). The Barents COSCAT (code 0408) is the longest segment (Lc = 9 311 km) that does not contain an archipelago. The median coastal length of COSCAT segments is 2400 km and the average is 2700 km.

[48] The total length of the nonglaciated coast, i.e., without Greenland and Antarctica, is estimated to be 380,000 km when defined at the 30′ × 30′ resolution. It is interesting to note that total coastal lengths of Africa (39,200 km) and Australasia (i.e., Australia, New Guinea, New Zealand, see Figure 2) (40,100 km) are almost identical and slightly greater than South America's length (31,600 km). Owing to its many regional seas, Europe's coastal length (48,600 km) is higher than for these continents. We have already mentioned that these lengths are scale dependent, as coastline morphology is in essence a fractal-like variable. Thus care should be exercised in using the reported lengths, which are specific to the 30′ × 30′ (latitude × longitude) resolution.

3.2.2. Coastal Catchment Area (At)

[49] The median coastal catchment area is 0.45 M km2 and the average is 0.82 M km2 (Table 2). Twenty percent of catchments are smaller than 0.2 M km2 and 27% greater than 1 M km2 (Table 2). The South Guiana–Amazon COSCAT (code 1104) is the largest catchment (7.1 M km2), for it includes the Amazon River catchment that cannot be split. The second and third largest catchments are notably smaller: 4.38 M km2 for the Angola COSCAT (code 0014) which includes the Congo River catchment, and 4.52 M km2 for the East Mediterranean COSCAT (code 0003) which includes the Nile River catchment.

Table 2. Distribution of COSCATs Catchment Areaa
Area, M km20–0.20.2–0.50.5–1.01.0–1.51.5–2.02.0–3.03.0–4.04.0–6.0>6.0Average
  • a

    Catchment area, At (M km2). Hook-up segments are omitted.

North America51353120.75
South America3623111.09
Global exorheic26492117684210.83

3.2.3. Mean Catchment Width (Wt)

[50] The mean calculated width of catchments is an important metric of the land-ocean connection, which ranges between 30 km for several archipelago segments and some very narrow coastal catchments to 3286 km for the East Mediterranean catchment (code 0003) (Table 3). When considering the group of catchments, for which Wt > 1000 km, two of these large catchments appear to drain the Sahara and therefore are not active at present in that, their river network is essentially fossilized (arheism) (code 0002, South Ionian Sea, and code 0020, South Canary Basin). Some catchments also combine active and fossil river networks, i.e., rheic and arheic area, and most of them are located in the Sahara, Arabia, Middle East and Central Australia.

Table 3. Coastal Catchment Basins Exceeding an Average Width (Wt) of 1000 km
Average Depth, kmBasin NumberBasinMain Rivers
32860003East MediterraneanNile
24611104South Guiana–AmazonAmazon
23701108Rio de la PlataParaná
22190016Niger DeltaNiger
15730834North Gulf of MexicoMississippi
14450020South CanaryTamanrasetta
13811340Indus DeltaIndus
13571103North GuianaOrinoco
11221308East KaraYenisei
11221309West LaptevLena
10460002South IonianIrharhara

[51] According to the definition of coastal width, some coastal catchments may not be very wide despite a very long river network. For instance, the West Kara Sea catchment (code 1307, 3.66 M km2) has a medium mean width (619 km only), due to a very complex coastline (Lc = 5 900 km).

[52] The catchment width can be averaged over any spatial entity. At the continental scale, the average width is equivalent for Europe (168 km) and Australasia (168 km) but extends to 551 km for South America and 670 km for Africa (Table 1). The average width reflects both the coastal zone sinuosity and the continent's size: for Australia alone (exorheic parts), it is higher and reaches 250 km.

3.2.4. Relative Length of River Network (Lm/Wt Ratio)

[53] We hypothesize that the relative length is related to the catchment shape. This ratio is always greater than 1.5 (Table 4), except for the Rio de la Plata COSCAT in South America (code 1108), which has an exceptional relative length of 1.16, due to its very odd shape. The highest relative length values exceed 10 as for the Saint Lawrence COSCAT (code 0825), the Andaman COSCAT (code 1335, Irrawaddy/Salween Rivers) and the West South China Sea COSCAT (code 1329, Mekong River), which all have relatively elongated basins [Vörösmarty et al., 2000b] (Table 4). The median relative length is 4.2 and ranges from 2.5 for Africa to 4.8 for Asia. In archipelagos, this indicator is meaningless (17 for the Canadian Archipelago).

Table 4. Distribution of the Maximum River Length/Mean Width of Catchment Ratio in COSCATsa
  • a

    Length/mean width, Ltm/Wt.

North America174410414.1
South America1243513.2
Global exorheic11228252533944.2

3.2.5. Latitudinal Difference of Coastal Catchments ((Δlat)TC) and Orientation of Flow (α)

[54] The latitudinal difference of coastal catchments ((Δlat)TC) is presented on Table 5: 68% of them have a latitudinal difference between −2 to +2°. In this catchment cluster more flow south (36 catchments between −2 and −0.5°) than north (21 catchments between +0.5° to 2°). For the group of catchments with the largest range in (Δlat)TC (Table 6), exceeding 2°, more flow from north to south (n = 22) than from south to north (n = 13) (Table 5).

Table 5. Distribution of Latitudinal Range (Δ lat)TC Between Catchment Centroid (T) and Coastal Segment Center (C) in COSCATsa
 (Δ lat)TC, South Flowing, deg(Δ lat)TC, North Flowing, deg
  • a

    Latitudinal range, (Δ lat)TC; T, catchment centroid (T); coastal segment center, C.

N. America1226736122
S. America1112611111
Global exorheic31651018183751633743
Table 6. COSCATs With the Largest Latitudinal Ranges
Δ lat, degBasin NumberBasinMajor Rivers
South-Flowing Basins
−80016Niger DeltaNiger
−101108Rio de la PlataParaná
−4.11329West South China SeaMekong
−7.71340Indus DeltaIndus
North-Flowing Basins
+160003East MediterraneanNile
+8.50816Canadian ArchipelagoBack
+8.41104South Guiana–AmazonAmazon
+14.31307West KaraOb
+14.71308East KaraYenisei

[55] The East Mediterranean COSCAT (code 0003) that includes the Nile River basin, has the largest latitudinal difference between catchment centroid and coastal segment center (16°) (Table 6). Since this catchment is oriented south-north, the latitudinal range between the upper catchment in Tanzania and the Nile mouth is actually much more, from 4°S to 31°N. The East and West Kara Sea COSCATs (Yenissei and Ob basins), and the Rio de la Plata COSCAT (Parana River catchment) have also very wide ranges of latitudes, i.e., of climatic features. We find that, when all catchments are clustered by quadrants (0–90–180–270–360°), there is no marked bias in the direction of flow at the global scale (Annex D).

3.2.6. Relative Size of Coastal Ribbon (Ac/At Ratio)

[56] The ratio between coastal ribbon area (Ac), and catchment area are (At) ranges from more than 0.90 for 15% of COSCATs, typically with very narrow coasts, islands and/or archipelagos, to less than 0.10 for 25% of catchments exceeding 1 000 km mean width (Table 7). At the 30′ × 30′ resolution, the definition of coastal ribbon area is difficult. We are using here as an operational definition the coastal segment length (LC) times 50 km. Few noncoastal cells are left out if the coast is very narrow (see examples for the West Mediterranean COSCATs, Annex F, Figure fs02). In such conditions, Ac/At is poorly quantified. The median proportion of coastal area is 0.35, but since the widest catchments have also the greatest area, the global average ratio is much less: The average mean width of the exorheic part of continents (including arheic portions as most of the Sahara) as defined here is 295 km, the average coastal ribbon width/mean width is therefore 50/295 = 0.17.

Table 7. Distribution of Coastal Ribbon Area/Catchment Area Ratio in COSCATsa
Ac/At<0.050.05–0.10.1–0.20.2–0.30.3–0.50.5–0.70.7–0.9>0.9Average Ac/AtCoastal Area, M km2
  • a

    Coastal ribbon area/catchment area ration, Ac/At. Hook-up segments are excluded.

North America154275260.2355.13
South America3221530.0911.58
Global exorheic1317201626205210.16619.0

4. Discussion and Perspectives

4.1. Segmentation of Land-Ocean Interface

[57] The COSCAT approach focuses on the general land to ocean connection as required for Earth System Analysis for geochemistry, sedimentology, tectonics, now and in the Pleistocene, and does not focus on the coastal zone. As such it is much different from the “IGBP-LOICZ typology” [Land-Ocean Interactions in the Coastal Zone, 1995; Bartley et al., 2001; Maxwell and Buddemeier, 2002; Buddemeier et al., 2002; McLaughlin et al., 2003; Talaue-McManus et al., 2003], which is particularly targeted to biogeochemical budgets at the fine scale as for N and P and to coastal management issues. The LOICZ approach and data base ( is a very flexible one since the spatial aggregation is actually not fixed but depends on the question addressed (e.g., biodiversity, reef distribution, coastal management); both COSCAT reporting and LOICZ analysis share the river inputs to oceans [Smith et al., 2003].

[58] In the COSCAT data set, the boundaries of segments are fixed and aim to be permanent for present-day segments so that (1) the land-ocean connection can be easily described and mapped at a coarse resolution (the median coastal length is 2400 km) and (2) all land-to-oceans fluxes can be reported in the same format. The fact that the catchment boundaries for the Arctic Ocean in Canada and Alaska are somewhat different from those fixed by international treaties is not expected to be a problem. The COSCATs boundaries of the Bering Sea, the Foxe, Hudson, Ungawa and Baffin Bays, are almost identical to conventional boundaries [e.g., Arctic Monitoring and Assessment Programme (AMAP), 1998], therefore the AMAP conventional Arctic Basin can be reconstituted by adding those entities to the Arctic Ocean drainage. The Norway Sea is considered in the COSCAT data set as part of the Arctic Ocean, as for AMAP, and its boundaries are those of AMAP.

[59] The COSCAT data set is based solely on physico-geographic criteria. Its flexibility allows for the computation of land-to-ocean inputs through rivers and groundwater at different scales: continent, ocean, regional sea basins and marine sedimentary basins. Its medium to coarse spatial resolution also allows for gross distributions of inputs to oceans per latitudinal zone. Within COSCAT each continental cell of the exorheic domain (n = 51,400), at 30′ × 30′ resolution, is linked through river networks to a coastal segment and finally to a marine sedimentary basin. Conversely, each coastal cell (n = 6200) is linked to an upstream catchment and a river network. This structure is similar to that used by Vörösmarty et al. [2000a, 2000b] except that they did not cluster individual river basins (6200 potentially flowing networks at the 30′ × 30′ resolution) into rational entities targeted at land-ocean flux assessment and therefore limited the presentation and the analysis of fluxes at the global scale.

[60] The COSCAT data set separates endorheic from arheic areas and this constitutes a major difference with De Martonne's [1955] pioneer work and an improvement of Gerasimov et al.'s [1964] maps that had difficulties to differentiate these two domains. During wetter periods, as for the Green Sahara event some 6000 years ago, the potential river basins can be reconnected to the coast and their fluxes can be assessed [see Vörösmarty and Meybeck, 2004, Figure D.45].

[61] The COSCAT puzzle (Figure 2) allows for the visualization of the complex spatial distribution of land-to-ocean fluxes. This global vision of the continents by river hydrologists is well-known and popularized through continental divides. For example, the three neighboring African Great Lakes flow into three different coastal segments located some ten thousand kilometers apart: Lake Victoria is linked to the East Mediterranean through the Nile basin, Lake Tanganyika to the Mid-Atlantic Ocean through the Congo, and Lake Malawi to the Mozambique Coast on the Indian Ocean, through the Zambezi River network. This oriented and channeled circulation of riverine material across continents as can be assessed with COSCATs is fundamentally different from the circulation of atmospheric and ocean material.

4.2. Hot Spots of Land-to-Ocean Fluxes at a Coarse Resolution: Water Runoff and Total Nitrogen Examples

[62] The COSCAT data set establishes the global puzzle of coastal segments and catchments as corresponding to 138 major entities (Greenland and hook-up segments not counted) for which the average concentrations (mg/L) and yields (m−3 m−2 yr−1 or mm yr−1 for water; t km−2 yr−1, g m−2 yr−1 for river-borne material as suspended particulate matter, nitrogen, organic carbon) of riverine material can be calculated. In order to facilitate the comparison and the mapping of various yields, as for water runoff, suspended matter, total nitrogen or dissolved inorganic carbon, here we use a relative yield scale with a nonlinear progression to account for the huge variability of river-borne yields, that range over two to three orders of magnitude at the resolution of COSCATs (≈106 km2).

[63] Many authors have emphasized the importance of some regions in global river flux analysis using terms such as “key regions,” “hot spots,” or “major inputs.” In order to facilitate the visualization, mapping and comparison of river fluxes for any given material, here we normalize all yields (Yi) to their global average (Y*).

[64] We divide the Yi/Y* ratio into eight classes: (1) “hot spots for material transfers” where Yi/Y* > 10; (2) “hyperactive regions” where 5 < Yi/Y* < 10; (3) “eury-active regions” where 2 < Yi/Y* < 5; (4) “meso-active regions” where 0.5 < Yi/Y* < 2; (5) “hypo-active regions” where 0,2 < Yi/Y* < 0.5; (6) “oligo-active regions” where 0,1 < Yi/Y* < 0.2; (7) “steno-active regions” where 0,01 < Yi/Y* < 0.1; and (8) “nonactive regions” or “inactive regions” where Yi/Y* < 0.01.

[65] Here we apply the COSCAT delineation to estimate water runoff, which is the number one controlling factor of riverine fluxes [Vörösmarty and Meybeck, 2004] in coastal catchments (Table 8). Runoff is estimated for all catchments in the global database generated by Fekete et al. [2002]. The global average runoff q* of the exorheic regions is estimated as 340 mm/yr, not considering water consumption. Thus, using the above definitions, the meso-active catchments are those for which q is between 170 and 680 mm/yr, the runoff hot spots exceeds 3400 mm/yr, the inactive regions are those where q < 3.4 mm/yr.

Table 8. Distribution of COSCATs Ranked Per Average Runoff in Eight Classes of Runoff Normalized to the Global Exorheic Average With Corresponding River Discharge, Catchment Area, Population, Runoff, and Population Densitya
Classqi/q*tqi, mm/yrVt, km3/yrAt, M km2Pt, M Peopleqt, mm/yrdtpop, p/km2COSCAT Examples
  • a

    Average runoff, (qt)i global exorheic average, q*t, 340 mm/yr; river discharge, Vt; catchment area, At; population, Pt; runoff, qt; population density, dtpop. Greenland and hook-up segments are excluded.

Hot spots>10>3400000   
Hyper-active10–53400–1700739 1.9%0.39 0.34%40.1 0.76%1890103Challenger, E. South China Sea
Eury-active5–21700–68019,555 50.0%19.0 16.6%1244 23.7%103065.5Bengal, S. Guyana/Amazon, Philippines, Hutton Rockall, Galapagos
Meso-active2–0.5680–17016,635 42.6%55.3 48.2%2798 53.3%30150.6Coral, North Sea, N. South China Sea, W. Yellow, New England, Rio de la Plata, E. Yellow Sea, NW Okhotsk, W. Deccan, E. Deccan, Niger Delta
Hypo–active0.5–0.2170–681495 3.8%12.1 10.6%381 7.3%12331.4Cape Verde, Indus Delta, Peru, E. Chukchi, Beaufort, Azov, Algerian
Oligo-active0.2–0.168–34468 1.2%10,2 8.9%499 9.5%46.049.1Gulf, Pohai, Agulhas, Canadian Archipelago, Pampa
Steno-active0.1–0.0134–3.4187 0.5%11.7 10.2%219 4.2%16.018.8N. Canary, California Gulf, Somali, W. Red Sea, Cape
Inactive<0.01<3.43.0 0.01%6.1 5.3%71.5 1.4%0.511.7Oman, S. Arabian, S. Ionian, E. Red Sea, Baja California
Global  39,081 100%114.7 100%5252 100%34045.8 

[66] There is no runoff hot spot at the COSCATs resolution (∼1 M km2) (Figure 3 and Table 8), owing to the integrating effect of drainage basins in the global water cycle. At a finer resolution (30′ × 30′), however, cells and/or small patches of runoff hot spots (q > 3400 mm/yr) indeed can be found [Fekete et al., 2002].

Figure 3.

Relative runoff for COSCATs related to mean annual runoff for the exorheic realm. Runoff data are from Fekete et al. [2002]. Greenland is not considered. Runoff classes boundaries are defined in Table 8.

[67] The meso-active coastal catchments correspond to a total area of 55.3 M km2, i.e., 48.2% of the exorheic domain, and to a total water volume discharged to the ocean of 16,635 km3/yr (42.6% of the global figure) (Table 8). At the other end of the runoff scale, the “inactive” regions are defined by the 3.4 mm/yr threshold which is similar to the one used by Vörösmarty et al. [2000a, 2000b] to operationally define the arheic regions. We find seven “inactive” coastal catchments covering altogether 6.1 M km2 and corresponding to 15,320 km of coastline: code 0002 South Ionian COSCAT; code 0005 South Aden COSCAT; code 0020 South Canary COSCAT; code 0806 Baja California COSCAT; code 1341 Oman Gulf COSCAT; code 1343 South Arabian COSCAT; code 1344 East Red Sea COSCAT. About 28,200 km of coastline have no permanent flow to the coast (3.4 < q < 34 mm/yr) and are mostly linked to seasonal or occasional river flows. Altogether about 11% of the worlds coastline has no perennial river flow, when the runoff is averaged over the coastal segments (i.e., at a resolution of about 0.5 M km2). This proportion is likely to be a minimum, since, at the 30′ × 30′ resolution, about 50 km × 50 km at the equator, the arheic regions that are potentially drained to the oceans are much more extensive (34.2 M km2). The relative yield scale used here depends upon the level of spatial aggregation and should be interpreted with caution.

[68] Next we use the COSCAT delineation plus an existing estimate of contemporary total nitrogen (totN = dissolved N + particulate N, including organic N) at 30′ × 30′ resolution [Green et al., 2004] to re-aggregate the modeled outputs from 6000 individual coastal catchments for the 138 coastal catchments. The average yield of exorheic areas (Y*t totN) is estimated by Green et al. [2004] to 355 kgN km−2 yr−1. This figure corresponds to 40.8 Tg totN/year with a dissolved inorganic nitrogen (DIN) flux estimate at 14.5 Tg/year, not very far from other estimates as 18.9 Tg N/year, for the DIN, found by Smith et al. [2003]. Both studies used a multiregression analysis, also used by many researchers working on nutrient fluxes [Caraco, 1995; Howarth et al., 1996; Seitzinger et al., 2002]. We establish, rank, and tabulate the total nitrogen yields for the hot spots (Yt totN > 2 550 kgN km−2 yr−1) to inactive regions (<3.55 kgN km−2 yr−1) (Figure 5). At the COSCAT scale, the hyperactive COSCATs for totN are found in Japan, the Indian subcontinent and Western Europe (Figure 4). The drivers of total N inputs are multiple and combine water runoff and climate with population density, cattle density and chemical N fertilizer use [Howarth et al., 1996; Smith et al., 2003; Green et al., 2004].

Figure 4.

Relative contemporary nitrogen yield for COSCATs related to the mean N yield for the exorheic realm. Total contemporary nitrogen data are from Green et al. [2004]. Greenland is not considered. Nitrogen yield classes boundaries are defined in Figure 5.

[69] The interpretation of COSCAT-derived flux pattern must take into account the level of aggregation of catchments. For instance Southern Iceland is aggregated with Ireland and western United Kingdom into the Hutton-Rockall catchment qualified here as a hyperactive N flux region. At the finer scale, field data for Iceland and Ireland do not validate this contention. This is an obvious and expected limit of COSCAT aggregation: Fixed catchments cannot be considered homogeneous for every river material, and care must be exercised in the interpretation.

[70] The relative weight of hyperactive regions, and others, can be represented in a comparative bar diagram for any type of riverborne material as proposed for total nitrogen in Figure 5. The hyperactive COSCATs regions for totN, i.e. with yields exceeding 5 times the global average, correspond to 1.9% of the global exorheic area and to 11.6% of the totN fluxes. The hyperactive regions also have a correspondingly higher runoff than average since their water volumes discharged to the sea is 4.5% of the global exorheic total; these regions are also much more populated than average as they correspond to 10.4% of the population. Another interpretation of the Figure 5 diagram is that 90% of world's totN is carried by 81% of river waters flowing from 50% of the exorheic area and populated by 70% of the exorheic domain population. For the lowest N yields, 25% of the area, occupied by 8.7% of the population, contributes only 1.6% of the river totN flux to oceans and the eury-active zone (40.1% of totN budget) corresponds to only 12.7% of the exorheic area and to 35.2% of the population.

Figure 5.

Proportions of total contemporary river nitrogen flux to ocean, and proportions of related drainage area, water volume discharged and population, per COSCATs classes relative to contemporary nitrogen flux (hook-up segments excluded). Total contemporary nitrogen data are from Green et al. [2004].

[71] Smith et al. [2003] and Green et al. [2004] have used another approach to present fluvial inputs to the coastal zone. They determined by modeling on approximately 6200 cells at 30′ × 30′ resolution. Both works are using, for river catchment delineation, the database developed by Vörösmarty et al. [2000b] and, for the riverine nitrogen concentration on which multiregressions are base, the database developed by Meybeck and Ragu [1995]. Both results are presented using a color code of the receiving flux for each cell, targeted to biogeochemical budgets at the finest scale possible (30′ × 30′) but the visual presentation of each 6200 cells is very difficult to examine at the global scale. Our approach is different: The very coarse scale (catchments with areas typically between 0.5 and 1 M km2) has several advantages in that it (1) facilitates the global vision of riverine production on land through the ∼140 piece puzzle, and (2) facilitates gross regional budgets, as for Regional Seas, portions of continents, biomes, climate zones, clusters of human pressures (e.g., North America and Europe; South Asia).

4.3. Population Pressures on Coastal Catchments

[72] We determine the total catchment population and the average population density of each class combining the global population database at 30′ × 30′ resolution as defined by Vörösmarty et al. [2000c] with the coastal catchments reordered into their eight runoff classes (Table 8). COSCATs with meso-runoff (170 < q < 680 mm/yr) are occupied by 2800 M people, i.e., 53.3% of the total population living in the exorheic regions, which corresponds to an average population density of 50.6 people per km2 (p/km2). COSCATs with hyperactive runoff (q > 1 700 mm/yr) have higher population density (103 p/km2). The steno-active and nonactive COSCATs for runoff (i.e., less than 34 mm/yr average runoff), are occupied by 219 and 71.5 M people, respectively, (5.6% of the population) that rely essentially on fossil water resources, a small number of headwater streams that do not reach the coast, and, in some catchments, on salt-water desalinization on the coast. This population is under marked water stress as mapped at the 30′ × 30′ resolution by Vörösmarty et al. [2000c].

[73] In this first analysis of the spatial distribution of population we compare the population density of the coastal ribbon to the overall population density for the whole catchment. For the COSCATs that have below-average catchment width (Wt < 200 km) this ratio is very close to 1.0 as expected. For the other COSCATs (Wt > 200 km, Figure fs03 in Annex F) this ratio is very variable. The coastal population density/whole catchment population density ratio ranges from less than 0.05, where the coast is depopulated relative to the whole basin, as in many Arctic catchments, with regards to the whole basin, to more than 5, where the coast is five times more populated than the whole catchment, as in the East Mediterranean COSCAT (code 0003), Niger Delta COSCAT (code 0016), or South East Australia COSCAT (code 1411) (see full data set of population densities in Annex D). Therefore we believe that common belief such as “the coastal zone is densely populated” should be more carefully checked at the regional level for each coastal catchment. Further analysis at finer scales is needed to confirm our first analysis.

4.4. Reconstruction of Past Coastal Catchments

[74] COSCAT segmentation is only valid for the present period and prior to any human impacts such as water diversions. The coastline and the related coastal segments can be markedly changed by lower sea level such as the one that occurred at the Last Glacial Maximum (LGM) some 20 ka ago. In internal regions, the river network has also been much affected during the last 20,000 years by changes of the hydrological balance. Also, ice caps have covered most of North America and part of Eurasia and acted as barriers to river runoff. A first review of these changes has been made by Meybeck and Vörösmarty [2005]. At longer geologic timescales (>1 M years), the land-to-ocean fluxes should also consider the location of continents, their uplift and continental volcanism. The future delineation of COSCATs for the LGM should include: (1) the ice cap extension, (2) the revision of river drainage in presently endorheic regions as in Central Asia, (3) the reconstitution of coastal catchments on extended continental platforms exposed at lower sea level (SE Asia, N Siberia, Patagonia, Persian Gulf, North Sea, etc.).

4.5. Future Changes in Land-to-Ocean Fluxes

[75] The great sensitivity of river systems to Holocene climate variability has been recently reviewed by Meybeck and Vörösmarty [2005], who also drew attention to the increasing modifications of global river hydrology by present water engineering (damming, diversion) and by water consumption, particularly by irrigation. These anthropogenic changes have occurred mostly over the last 50 years. Now they affect a global area of about 40 M km2, or on the same order than the extension of major natural changes over the last 10,000 years.

[76] This extremely fast acceleration of Earth System Change due to multiple human impacts [Steffen et al., 2004], leading to the term “Anthropocene” describing the present period [Crutzen and Stoermer, 2000; Crutzen, 2002], is also typical of riverine changes occurring at the global scale [Meybeck, 2002, 2003; Vörösmarty et al., 2004; Meybeck and Vörösmarty, 2005]. Most water diversions are presently occurring within one COSCAT entity. However, there is probably already a transfer of water between the James Bay catchment (code 0819) and some tributaries of the East Hudson Bay (code 0819). The Colorado River diversion from Arizona to Southern California is also well known, and most of this water is evaporated in irrigated fields, particularly in the Salton Sea area. Large-scale water transfers were once planned across the North American continent and are being studied between the Zambezi and Southern Africa [Golubev and Biswas, 1985]. If they ever occurred the related modification of water transfers will need to be taken into account in COSCAT budgets.

[77] There is presently drastic reduction of observed river flow, from 50% to more than 90%, in many dry and semiarid regions such as in the Colorado and Rio Grande [Meade and Parker, 1985], Nile, Orange, Indus, Murray, Huang He. The reduction in flow has now reached a global scale and has been termed “neoarheism” [Meybeck, 2003]. This impact is generally associated with the fragmentation and regulation of river flow and to sediment trapping [Dynesius and Nilsson, 1994; Vörösmarty et al., 2003; Nilsson et al., 2005; Syvitski et al., 2005], which are two other syndromes of global river change [Meybeck, 2003].

4.6. Perspectives

[78] The COSCAT scheme and database of coastal hydrologic segments and segmentation are intended to improve descriptions of land and relative riverine fluxes from an oceanic perspective, with standard reporting for approximately 140 major entities. The delineation of these segments is designed to facilitate river input budgeting from the subregional scale, as local seas, up to the full global scale, and is designed for reconstruction of past inputs. COSCAT can be used to make coarse-resolution descriptions and maps of multiple human impacts on river systems. The segmentation scheme can also be used to optimize the design of operational monitoring of changes in the land-coastal system within each segment [Salisbury et al., 2001, 2004], as well as to provide a template upon which dynamic constituent transport models can be constructed (W. M. Wollheim et al., The spatially distributed Framework for Aquatic Modelling of the Earth System (FrAMES), submitted to Global Biogeochemical Cycles, 2005).

[79] The COSCAT coastal segmentation is not aimed to establish biogeochemical budgets at the fine scale, as in local bays and estuaries, where the nested scale approach is necessary. It is therefore not adapted to local coastal management but can be used to assess and report regional issues, such as for GIWA entities. For fine-scale biogeochemical models, great care should be taken to scale issues, because for a given segment, the budgets for small local basins or the major river deltas and estuaries may be quite different [Crossland et al., 2005].

[80] Our first application to river runoff and population pressure reveals a global heterogeneity of the global catchment puzzle with average yields and population densities ranging over 2 to 3 orders of magnitude. At such a coarse resolution (approximately 106 km2), some coastal catchments may still exceed five times the global average yield while other yields are more than 10 times less than the global average. This global heterogeneity in land/ocean connections is also expected for other riverine material such as organic and inorganic carbon, nutrients, ions, sediments and contaminants, the study of which will be the next COSCAT application.

[81] Another foreseen COSCAT application is to recluster the 140 catchments into mega-ensembles, defining the Earth System. These ensembles will be of intermediate size between the six continents and the 140 catchments, and will individualize regional sea catchments (n ≈ 40), and open ocean segmentations (n ≈ 15). A third step will be to complement this first description of the 140 piece puzzle with other important descriptors such as climate zones, lithology and relief types.

[82] Finally, it is important to remember that coastal zone budgets should also consider that (1) net riverine fluxes to oceans are altered by biogeochemical processes and sediment trapping occurring downstream of river mouth in the estuarine zone, (2) other land-based inputs should be taken into account, such as aeolian transport of dust, atmospheric fallout of nutrients, direct urban wastes inputs, and (3) ocean upwelling can be dominant nutrient inputs in some coastal zones.


[83] The technical assistance of Séverine Roussennac (Sisyphe, Université Paris 6) and Pamela Green (University of New Hampshire) is greatly appreciated. This work has been partially completed during Hans Dürr's Post-Doctoral Studies (department of Ecologie des Systèmes Aquatiques (ESA), directed by C. Lancelot) at the University of Brussels (EU programme Si-WEBS, contract HPRN-CT-2002-000218). UNH work has been supported by the NASA EOS Interdisciplinary Science Team program (grant NNG04GH75G). The critical reviews of R. Buddemeier (Kansas Geological Survey) and of S. Sterling (Sisyphe, Université Paris 6) have been particularly appreciated.