The impact of Argo on ECMWF operational ocean analyses has been assessed by conducting a set of observing system experiments spanning the period 2001–2006. The experiments evaluate the information content of Argo temperature and salinity data, gauged in terms of influence on the ocean state and the skill of seasonal forecasts. The salinity data from Argo is instrumental in correcting the salinity of the ECMWF analysis on a basin-wide scale. The effect of Argo temperature is noticeable in the Indian, Atlantic and Southern Oceans. In the Pacific, the impact is modest, being confined to the South American coast. The information content of Argo combines well with the altimeter information in most regions. The impact of Argo on seasonal forecasts of sea surface temperature has been assessed for the period 2001–2006. The assimilation of Argo improves seasonal forecast skill in most areas, although the reduction of the error is small.
 The ocean observing system is always changing but the last few years have seen major developments both in the observing platforms and in the way the observations are used. A key development has been the continued expansion of the Argo float network and continuation of high quality altimetry. The relative importance of different components of the observing system should always be kept under review. There is unlikely to be a unique ideal observing system as different applications will have different needs and emphases. Although the authors have a particular interest in the potential contribution to seasonal forecasting, a wider brief than just observing the tropical oceans is considered. Vidard et al.  assessed the relative importance of the tropical in situ mooring arrays (TAO, TRITON, PIRATA), XBTs and the developing Argo float network, showing a major role for the moorings and in some ways a smaller contribution from Argo than might have been expected. There were three limitations in that paper: firstly Argo was assessed at a relatively early stage of development (2001–2003), secondly the assimilation system did not use salinity data directly and thirdly altimeter data were not used. In this paper we use the latest version of the ECMWF ocean data assimilation system (S3) in which both salinity and altimetry are used. Results from a set of Observing System Experiments (OSEs) carried out to assess the impact of Argo temperature and salinity will be shown. First a brief description of the S3 ocean analysis system is presented.
2. S3 Ocean Analysis System
 The ocean data assimilation system for S3 is based on the HOPE-OI scheme. The first guess is obtained by forcing the ocean model with daily fluxes of momentum, heat, and fresh water from the ERA40 reanalyses for the period January 1959 to June 2002 and NWP operational analyses thereafter; the fresh water flux in the ERA40 has been corrected according to Troccoli and Kallberg . The ocean model has a horizontal resolution of 1° × 1° with equatorial refinement, i.e., the meridional resolution increases gradually towards the equator, where it is 0.3°. There are 29 levels in the vertical, with a typical vertical thickness of 10 m in the upper ocean. The barotopic mode is resolved explicitly. The observations come from the quality controlled data set prepared for the ENACT and ENSEMBLES projects until 2004 [Ingleby and Huddleston, 2006], and from the GTS thereafter. The model SSTs are strongly relaxed to analyzed daily SST maps from the OIv2 SST product [Reynolds et al., 2002] from 1982 onwards. Prior to that date, the same SST product as in the ERA40 reanalysis was used. The altimeter data used are global gridded weekly maps Le Traon et al.  from 1993 onwards.
 The different data streams are assimilated sequentially. First, the model background is bias-corrected [Balmaseda et al., 2007a], and the detrended altimeter-derived sea level anomalies combined with the bias-corrected model first-guess to produce a first analysis. This analysis is then used as a first guess for a second assimilation step, in which only subsurface temperature data are assimilated, and salinity is updated by imposing conservation of the model temperature/salinity (T/S) relationship, while the sea level and velocity field remain unchanged. In a third step, salinity observations are used to modify the model T/S relationship. In this step, the T/S information is spread along isotherms following the scheme of Haines et al. . Only salinity is modified in this step. Velocity updates are then derived from the temperature and salinity increments by imposing geostrophic balance. Finally, the trend in global (area averaged) sea level is assimilated. The analysis is performed every 10 days. Analysis increments in temperature, salinity and velocity are added slowly over the subsequent 10 days, after which a new background field is available, and the cycle repeated. Details about the data assimilation system are given by Balmaseda et al. [2007b].
3. Observing System Experiments
3.1. Experimental Setup
 The S3 system described above assimilates all data available (altimeter, Argo, moorings, XBTs,…). This analysis, denoted ALL, started 1/1/1959. Three experiments withdrawing the Argo, altimeter, and both altimeter and Argo, were conducted, called NOARGO, NOALTI and NOARAL respectively. A further experiment which uses all the altimeter and thermal data but does not use salinity data has been performed (NOSAL). Although no salinity data are used in NOSAL, salinity is still adjusted preserving the model T-S relationship. The four supplementary experiments all start from the long-term reanalysis (ALL) on 1/1/2001 and so can be compared directly to it.
 The effect of Argo data can be gauged by comparing the different experiments. The differences between experiment ALL and NOARGO show the impact of withdrawing Argo temperature and salinity when everything else is assimilated. The difference NOALTI minus NOARAL measures the impact of Argo when altimeter data are not assimilated. The impact of Argo temperature only is given by the differences between NOSAL and NOARGO. Finally, the difference between ALL and NOSAL gives a measure of the impact of salinity data from Argo, although in the Western Pacific this difference also includes the impact of TRITON salinity data. The statistics presented in following sections are for the period 2001–2006; for much of this period the Argo array was not complete [Roemmich et al., 2004; Gould, 2005], and therefore the impact of Argo might be underestimated.
 Recently, a problem has been detected with data from the SOLO/FSI Argo floats http://www.arago.ucsd.edu). An additional experiment, similar to ALL but with the SOLO/FSI floats removed was then conducted for the period 2001–2006. Results show that the impact of the SOLO/FSI data, noticeable at intermediate depths (500 m–1000 m) in the tropical Atlantic, does not alter the conclusions presented here.
3.2. Impact of Argo on the State of the Ocean
Figure 1 shows how the salinity averaged over the upper 300 m (S300) changes when the Argo data are withdrawn. The salinity information of Argo plays a dominant role in determining the salinity of most of the global ocean, by increasing the value of salinity values over most of the Indian ocean, the tropical Atlantic, the North Atlantic and North Pacific, and freshening the waters in the Indian ocean west of Australia, in the latitude band around 25 N in the Central Pacific, and in the Antarctic Circumpolar Current (ACC). Argo temperature has a modest effect on S300, freshening parts of the north Atlantic and southern Indian ocean. In the tropics, the effect of Argo on salinity is larger if no altimeter data are assimilated, suggesting that both Argo and altimeter are acting in the same direction (not shown). Most of the changes in S300 are due directly to the assimilation of salinity data, but there are areas where the effect of T on S is contrary to the direct effect of assimilating S; for example in the southern Indian Ocean and the Western Pacific/Indonesian Throughflow. In the latter, the assimilation of Argo temperature produces an increase of salinity, while the direct assimilation of salinity data produces fresher water, probably associated with the barrier layer. It is likely that Argo is correcting for errors in the ocean model vertical mixing and the fresh-water flux.
 The impact of Argo on the heat content of the upper ocean can be measured by its impact on the averaged temperature in the upper 300 m (T300), shown in Figure 2. For T300, the impact of Argo, although global, occurs at smaller spatial scales than for salinity, probably because there are other sources of temperature information. Argo has a substantial effect in reducing the heat content of the subtropical Indian ocean. It also increases the latitudinal gradient of heat content of the ACC region, by reducing T300 in the southern ocean and increasing it in the northern edge of the ACC, especially in the confluence with the Brazil/Falklands current, the Agulhas current and the New Zealand Subtropical Front. There are also large changes in the vicinity of the Kuroshio and Gulf Stream. In the tropical Atlantic, Argo cools the waters off the Caribbean area, and increases the heat content in the subtropical Atlantic around 15N; the east–west temperature gradient along the equatorial Atlantic is more pronounced with Argo, with colder waters in the Gulf of Guinea. The heat content in the eastern part of the South Atlantic and South Pacific gyres is reduced. In the Pacific, Argo has a large effect on the meridional temperature gradient associated with the North Equatorial Current system, but the latitude band 10N/10S is little affected by Argo due to the TAO/TRITON array, with the exception of the area close to the South American coast, where Argo reduces the heat content. Most of the Argo impact on the upper ocean heat content is due directly to the temperature data, although the salinity from Argo also influences the heat content, for example increasing the heat content in the tropical Atlantic. The interaction between Argo and altimeter seems to be constructive, in the sense that the impact of Argo in the absence of altimeter has similar structure to its impact when altimeter data are used, though the amplitude is weaker (not shown).
Figure 3 shows a small selection of vertical profiles of the RMS error in temperature and salinity of the different experiments compared to a fixed set of observations. These come from the new version of the ENSEMBLES observational data set (EN3 v1c on http://www.hadobs.org), which has been extended to 2006 and does not contain the SOLO/FSI floats.
 In all panels (including those not shown) the best fit to the subsurface data is achieved when Argo and altimeter are both assimilated. Panels in the upper row show the fit to T and the lower panels the fit to S. By comparing the black and grey dashed lines one can see the effect of the altimeter if there were no Argo data. In general black is better than grey, indicating the positive impact of altimetry. This difference is smaller when Argo is also present, as indicated by comparing the black and grey solid lines. However, even with Argo there is a positive impact from altimetry in the southern Indian ocean and the north subtropical Pacific, more marked in T than in S. The differences between solid and dashed black lines are especially noticeable indicating the large effect of withdrawing Argo. The differences are largest in the upper 100 m of salinity, though there are deeper effects in both T and S. The grey dashed line uses the least amount of data (no Argo and no altimeter) and as expected has the worst fit. In the Equatorial Pacific, temperature is well constrained by the TAO/TRITON array, and Argo has little impact except for the regions close to the South American coast (not shown).
 Poleward of 30°, where the altimeter data are given little weight or not assimilated at all, Argo is the main contributor to the observing system, and in general the best fit to the data is obtained when Argo data are assimilated (not shown). (The assimilation of altimeter data is disabled in areas with weak vertical stratification, where the performance of the assimilation scheme, based on the vertical displacement of the profiles, is not guaranteed).
3.3. Impact on the Skill of Seasonal Forecasts
 In order to assess the impact of the data assimilation on the skill of seasonal forecasts, two sets of coupled seasonal forecast experiments have been conducted, using initial conditions from ALL and NOARGO. Each set consists of ensemble forecasts initialised on 22 different dates, spanning the period 2001–2006 and sampled every three months (Jan, Apr, Jul and Oct). For each date, an ensemble of 5 coupled forecasts with perturbed initial conditions is integrated up to 6-months lead time. The coupled model is that used by S3 seasonal forecasting system [Anderson et al., 2007]. The forecast SST anomalies are computed with respect to the model climatology (which depends on the lead time). Figure 4 shows the scatter diagram of the Mean Absolute Error (MAE) in the forecast of SST anomalies over the 6 months of the forecasts for two regions: Niño4 in the Central Pacific the EQIND in the Indian Ocean.
 The assimilation of Argo improves the skill of the seasonal forecasts in most areas, although with the current sample size the improvement is hardly significant. Only in the Equatorial and Tropical Atlantic is the skill of the seasonal forecast degraded when Argo data are used (not shown). This fact seems contradictory with the results shown in Figure 3, where the experiment ALL has the best fit to the temperature and salinity data. This is a clear example of the ambiguity in the meaning of “best analysis”, and the difficulty of finding metrics for evaluation: a best fit to the data does not necessarily give the best forecast initial conditions.
4. Summary and Conclusions
 The impact of Argo data on the S3 ocean analyses has been evaluated by conducting a set of observing system experiments during the period 2001–2006. Results show that the salinity data from Argo changes the values of the S3 salinity analysis on a global scale. Argo data may be correcting for errors in the fresh-water fluxes, errors in the circulation, and errors in the water mass characteristics of the first guess. Temperature data from Argo also has an effect on the upper ocean heat content, being especially noticeable in the Indian, Atlantic and Southern Oceans. In the Equatorial Pacific, where the TAO/TRITON array is active, the impact of Argo temperature on upper ocean heat content is small, being confined to the regions close to the South American coast. The information content of Argo combines well with the altimeter information, and results show that the fit to the subsurface data is best when both altimeter and Argo are assimilated. But while the reduction in the RMS error of salinity is quite substantial, the profiles of the temperature error in the tropical regions are remarkably (disappointingly) similar for the different experiments. This suggests that further improvement in the temperature analysis needs the development of both the ocean model and the assimilation system.
 To gauge the impact of Argo on the initialization of seasonal forecasts, coupled forecasts using ocean initial conditions with and without Argo have been conducted. The assimilation of Argo improves the skill of seasonal forecasts in most areas, although the reduction of the error is modest. In the Equatorial Atlantic the skill of the seasonal forecasts is degraded with Argo, even when ocean initial conditions show a better fit to the temperature and salinity data. Better assimilation methods and forecast models are needed to make optimal use of Argo data in the initialization of seasonal forecasts in the Atlantic region.
 The Argo data were collected and made freely available by the International Argo Project and the national programs that contribute http://www.argo.ucsd.edu, http://argo.jcommops.org). Argo is a pilot program of the Global Ocean Observing System. Bruce Ingleby provided the extended version of the ENSEMBLES data set used in the verification.