The paper compares ten different global precipitation data sets over the oceans and discusses their respective strengths and weaknesses in ocean regions where they are potentially important to the salinity and buoyancy budgets of surface waters. Data sets (acronyms of which are given in Section 2) are categorised according to their source of data, which are (1) in situ for Center for Climatic Research (Legates and Willmott, 1990; Archive of Precipitation Version 3.01, http://climate.geog.udel.edu/similar to climate), Southampton Oceanography Centre (SOC) (Josey et al., J Clim 12:2856-2880, 1999) and University of Wisconsin-Milwaukee (UWM) (Da Silva et al. 1994); (2) satellite for Microwave Sounding Unit (MSU) (Spencer, J Clim 6:1301-1326, 1993), TOPEX (Quartly et al., J Geophys Res 104:31489-31516, 1999), and Hamburg Ocean Atmosphere Parameters and Fluxes from Satellite (HOAPS) (Bauer and Schluessel, J Geophys Res 98:20737-20759, 1993); (3) atmospheric forecast model re-analyses for European Centre for Medium-range Weather Forecast (ECMWF) (Gibson et al. 1997) and National Center for Environmental Prediction (NCEP) (Kalnay et al., Bull Am Meteorol Soc 77:437-471, 1996); and (4) composite for Global Precipitation Climatology Project (GPCP) (satellites and rain gauges, Huffman et al., Bull Am Meteorol Soc 78(1):5-20, 1997) and Climate Prediction Center Merged Analysis of Precipitation (CMAP) (satellites, rain gauges and atmospheric forecast model, Xie and Arkin, Bull Am Meteorol Soc 78(11):2539-2558, 1997). Although there is no absolute field of reference, composite data sets are often considered as the best estimates. First, a qualitative comparison is carried out, which provides for each data set, a description of the geographical distribution of the climatological mean precipitation field. A more careful comparison between data sets is undertaken over periods they have in common. First, six among the ten data sets (SOC, UWM, ECMWF, NCEP, MSU and CMAP) are compared over their common period of 14 years, from 1980 to 1993. Then CMAP is compared to GPCP over the 1988-1995 period and to HOAPS over the 1992-1998 period. Usual diagnostics, like comparison of the precipitation patterns exhibited in the annual climatological means of zonal averages and global budget, are used to investigate differences between the various precipitation fields. In addition, precipitation rates are spatially integrated over 16 regional boxes, which are representative of the major ocean gyres or large-scale ocean circulation patterns. Seasonal and inter-annual variations are studied over these boxes in terms of time series anomalies or correlation coefficients. The analysis attempts to characterise differences and biases according to the original source of data (i.e. in situ or satellite, etc.). Qualitative agreement can be observed in all climatologies, which reproduce the major characteristics of the precipitation patterns over the oceans. However, great disagreements occur in terms of quantitative values and regional patterns, especially in regions of high precipitation. However, a better agreement is generally found in the northern hemisphere. The most significant differences, observed between data sets in the mean seasonal cycles and interannual variations, are discussed. A major result of the paper, which was not expected a priori, is that differences between data sets are much more dependent upon the ocean region that is considered than upon the origin of the data sets (in situ vs satellite vs model, etc.). Our analysis did not provide enough objective elements which would allow us to clearly recommend a given data set as reference or best estimate. However, composite data sets (GPCP, and especially CMAP), because they never appeared to be really "off" when compared to other data sets, may represent the best recent data set available. CMAP would certainly be our first choice to drive an ocean GCM.
