N. Chapelon ® H. Douville ® P. Kosuth ® T. Oki Off-line simulation of the Amazon water balance: a sensitivity study with implications for GSWP Received: 28 March 2001 /Accepted: 29 October 2001 / Published online: 2 March 2002 Ó Springer-Verlag 2002 Abstract As a first step of the international Global Soil Wetness Project (GSWP), several state-of-the-art land surface models were recently forced by 1°· 1° atmo- spheric analyses and observations to obtain global soil moisture climatologies over the 1987–1988 period. Be- sides the 6-hourly atmospheric forcing, the models also used common boundary conditions (soil and vegetation parameters), also drawn from the ISLSCP Initiative I dataset.Whileveryfewinsituobservationsareavailable to validate the soil moisture fields, the simulated runoff can be compared with river discharge measurements. This strategy is employed here to assess the quality of GSWP simulations based on the ISBA land surface model. The ISBA runoff is transformed into 1°· 1° gridded stream flows by using the TRIP river routing model. The focus is on the Amazon basin where all models participating in GSWP showed a strong under- estimation in the annual runoff and a significant lag in the annual cycle of the runoff. A sample of 31 gauging stations is selected to validate the gridded runoff simu- lated by ISBA. Sensitivity tests have been performed, that suggest that deficiencies in both the precipitation forcing and the boundary conditions provided by ISLSCP contribute to the poor simulation of the Amazon water balance. The use of alternative precipi- tation, soil and vegetation datasets allows ISBA to produce a more realistic annual runoff, although theamplitudeoftheannualcycleremainsexaggeratedat thedownstreamgaugingstationofObidos.Amongthese experiments, the simulation leading to the best annual runoff has been used as a reference to test simple modi- fications in the TRIP river routing model. Tuning the parameters of TRIP or increasing the resolution of theriverchannelnetworkisnotsufficienttoimprovethe annual cycle of the simulated discharge. New develop- ments are necessary to deal more explicitly with the floodplain inundation that occurs during the rainy season over the Amazon basin, and to derive a geo- graphically variable and time-evolving stream flow velocity. In future, considerable efforts are needed to provide more reliable global land surface and forcing datasets in the continuation of GSWP, in order to val- idatemoreefficientlythelandsurfacemodelsandthereby to produce more realistic soil moisture climatologies. 1 Introduction The Global Soil Wetness Project (GSWP) is an ongoing modeling activity of the International Satellite Land Surface Climatology Project (ISLSCP), a contributing project of the Global Energy and Water Cycle Experi- ment(GEWEX).Itwaschargedwithproducinga2-year global data set of soil moisture (SM) on a 1°· 1° mesh by integrating state-of-the-art land surface models (LSMs) using identically prescribed 6-hourly atmo- spheric forcings and monthly (or fixed) soil and veget- ationboundaryconditions,allderivedfromtheISLSCP Initiative I dataset (Meeson et al. 1995). At the same time, GSWP provides an ideal means of testing and comparing LSMs on the global scale, since the off-line technique allows us to eliminate the strong systematic biases that are generally found in the atmospheric models. The first stage of GSWP was launched in 1996. Me´- te´o-France, with the ISBA (Interaction between Soil, Biosphere and Atmosphere) land surface model, was Climate Dynamics (2002) 19: 141–154 DOI 10.1007/s00382-001-0213-9 N. Chapelon ® H. Douville (&) Me´te´o-France, 42 Avenue Coriolis 31057 Toulouse Cedex, France E-mail: Herve.Douville@meteo.fr P. Kosuth Institut de Recherche pour le De´veloppement CP 7091 Lago Sul, 71619-970 Brasilia DF, Brazil T. Oki Institute of Industrial Science, University of Tokyo 4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, Japan