minimal subsurface data coverage may be sufficient for initialising a model of this simplicity. Clearly, the surface wind forcing provides the bulk, but not all, of the information needed for initialisation (at which point we should note that the VOS and, in particular, TAO are also important sources of surface wind information). For this model at least it would seem that this extra information could be provided by just some of the TAO stations and VOS XBT lines.
6. Conclusions
This study examined the importance of subsurface thermal data for monitoring and predicting seasonal-to-interannual climate variations in the tropical Pacific Ocean. Such data are not the exclusive source of information on such variations, surface wind and SST being two, perhaps more important, alternative sources. The study takes advantage of analysis and coupled model hindcast results over the period 1970-94 to evaluate the importance of subsurface data and to evaluate the individual contributions from the VOS XBT network and the TAO buoy array.
The main conclusions to be drawn from the 400m depth-averaged temperature analyses, and from cross-reference to the study of Smith and Meyers [1996], were
(1)The effective information available for the equatorial and tropical regions of the Pacific Ocean grew by around a factor of 4 from the 1970's through to the late 1980's and early 1990's;
(2)The TAO system now provides the bulk of useful information in the equatorial region, and the largest proportion summed over the topical region as a whole;
(3)The VOS XBT network makes its most significant contribution in the off-equatorial region through resolution of westward propagating Rossby waves;
(4)There is little evidence of unacceptable levels of redundancy between TAO and the VOS XBT network, though within each of these systems, and under the terms of this study, the number of units of effective information are considerably smaller than the total number of original observation (around a factor of 4).
The principal results and conclusions from the observing system sensitivity hindcasts are perhaps less clear cut than one might have expected, though each is consistent with the conclusions above. In addition to examining hindcasts using TAO and VOS XBT alone, a series of hindcasts were also carried out with a severely degraded total data set (70% of the platforms removed). The TAO and VOS XBT experiments (for the period 1986-1994) showed that
(5)Neither the TAO-alone nor VOS XBT-alone hindcast performances were significantly poorer than those using the full data set.
There was a suggestion, consistent with conclusions (2) and (3), that
(6)The TAO-alone hindcasts performed better at short lead times (order 4 months) while the VOS XBT hindcasts performed better at longer lead times (6 to 12 months).
The model hindcast performance was less impressive in the 1970's and, to some extent, the 1990's, compared with the skill over the period 1982-91. The impact of subsurface data through the 1970's was positive, though small, at lead times out to around 10 months,
