DAI ET AL.Changes in Continental Freshwater Discharge from 1948 to 2004AIGUo DAI, TAOTAO QIAN, AND KEVIN E. TRENBERTHNational Center for Atmospheric Research,* Boulder, ColoradoJOHN D. MILLIMANSchool of Marine Science, College of William and Mary, Gloucester Point, Virginia(Manuscript received 16 April 2008, in final form 17 November 2008)ABSTRACTA new dataset of historical monthly streamflow at the farthest downstream stations for the world's 925 largestocean-reaching rivers has been created for community use. Available new gauge records are added to a networkof gauges that covers -80 X 106 km2 or -80% of global ocean-draining land areas and accounts for about 73%of global total runoff. For most of the large rivers, the record for 1948-2004 is fairly complete. Data gaps in therecords are filled through linear regression using streamflow simulated by a land surface model [CommunityLand Model, version 3 (CLM3)] forced with observed precipitation and other atmospheric forcings that aresignificantly (and often strongly) correlated with the observed streamflow for most rivers. Compared withprevious studies, the new dataset has improved homogeneity and enables more reliable assessments of decadaland long-term changes in continental freshwater discharge into the oceans. The model-simulated runoff ratioover drainage areas with and without gauge records is used to estimate the contribution from the areas notmonitored by the gauges in deriving the total discharge into the global oceans.Results reveal large variations in yearly streamflow for most of the world's large rivers and for continentaldischarge, but only about one-third of the top 200 rivers (including the Congo, Mississippi, Yenisey, Parana,Ganges, Columbia, Uruguay, and Niger) show statistically significant trends during 1948-2004, with the rivershaving downward trends (45) outnumbering those with upward trends (19). The interannual variations arecorrelated with the El Nifio-Southern Oscillation (ENSO) events for discharge into the Atlantic, Pacific, Indian,and global ocean as a whole. For ocean basins other than the Arctic, and for the global ocean as a whole, thedischarge data show small or downward trends, which are statistically significant for the Pacific (- 9.4 kM3 Vr 1).Precipitation is a major driver for the discharge trends and large interannual-to-decadal variations. Compari-sons with the CLM3 simulation suggest that direct human influence on annual streamflow is likely smallcompared with climatic forcing during 1948-2004 for most of the world's major rivers. For the Arctic drainageareas, upward trends in streamflow are not accompanied by increasing precipitation, especially over Siberia,based on available data, although recent surface warming and associated downward trends in snow cover andsoil ice content over the northern high latitudes contribute to increased runoff in these regions. The results arequalitatively consistent with climate model projections but contradict an earlier report of increasing continentalrunoff during the recent decades based on limited records.1. IntroductionContinental freshwater runoff or discharge is an im-portant part of the global water cycle (Trenberth et al.* The National Center for Atmospheric Research is sponsoredby the National Science Foundation.Corresponding author address: Dr. Aiguo Dai, National Centerfor Atmospheric Research, P.O. Box 3000, Boulder, CO 80307-3000.E-mail: [email protected]). Precipitation over continents partly comes fromwater evaporated from the oceans, and streamflow re-turns this water back to the seas, thereby maintaining along-term balance of freshwater in the oceans. Thedischarge from rivers also brings large amounts of par-ticulate and dissolved minerals and nutrients to theoceans (e.g., Boyer et al. 2006); thus it also plays a keyrole in the global biogeochemical cycles. Unlike oceanicevaporation, continental discharge occurs mainly at themouths of the world's major rivers. Therefore, it pro-vides significant freshwater inflow locally and forcesocean circulations regionally through changes in densityDOI: 10.1 175/2008JCLI2592.1© 2009 American Meteorological Society277315 MAY 2009JOURNAL OF CLIMATE(Carton 1991). Continental runoff also represents amajor portion of freshwater resources available to ter-restrial inhabitants. As the world's population growsalong with increasing demands for freshwater, inter-annual variability and long-term changes in continentalrunoff are of great concern to water managers, espe-cially under a changing climate (Vor6smarty et al. 2000b;Oki and Kanae 2006).There are a large number of analyses of streamflowover individual river basins (e.g., Krepper et al, 2006;Qian et al. 2006; Ye et al. 2003; Yang et al. 2004a,b;Xiong and Guo 2004), countries (e.g., Birsan et al. 2005;Groisman et al. 2001; Guetter and Georgakakos 1993;Hyvarinen 2003; Lettenmaier et al. 1994; Lindstrom andBergstrom 2004; Lins and Slack 1999; Robson 2002;Shiklomanov et al. 2006: Zhang et al. 2001), and regions(Genta et al. 1998; Dettinger and Diaz 2000; Cluis andLaberge 2001; Lammers et al. 2001; Pasquini and Depetris2007). Streamflow records for the world's major riversshow large decadal to multidecadal variations, but oftenwith small secular trends (Cluis and Laberge 2001;Lammers et al. 2001; PekdrovA et al. 2003; Dai et al.2004b; Huntington 2006). However, increased stream-flow during the latter half of the twentieth century hasbeen reported over regions with increased precipitation,such as many parts of the United States (Lins and Slack1999; Groisman et al. 2001) and southeastern SouthAmerica (Genta et al. 1998; Pasquini and Depetris2007). Decreased streamflow, in contrast, has been re-ported over many Canadian river basins during the last30-50 yr (Zhang et al. 2001) in response to decreasedprecipitation. Because large dams and reservoirs builtalong many of the world's major rivers during the last100 years dramatically change the seasonal flow rates(e.g., by increasing winter low flow and reducing spring/summer peak flow; Cowell and Stoudt 2002; Ye et al.2003; Yang et al. 2004a,b), trends in seasonal streamflowrates (e.g., Lammers et al. 2001) can be affected greatlyby these human activities. Also, there is evidence thatthe rapid warming since the 1970s has caused an earlieronset of spring that induces earlier snowmelt and asso-ciated peak streamflow in the western United States(Cayan et al. 2001) and New England (Hodgkins et