The conclusions of any OSE are inevitably linked to the "terms-of-reference" chosen for the study; that is, the method by which we judge whether a suite of analyses or forecasts is better or worse. In this study we focus on forecasts of the so-called NINO3 index (the mean SST over the box 505-50N, 1500W-900W), yet we know that what we are really interested in are, for example, forecasts of precipitation over the continents which come under the influence of El Niflo. Variability of the NINO3 index is only a crude proxy for variability in precipitation from which it follows that improvements in NINO3 forecast/hindcast skill are probably no more than a guide to improved climate prediction skill in the more general situation. In the future, if not now, the demands will be for more detailed (spatial, temporal and physical) predictions and so any results and conclusions from this study are inevitably preliminary and not final.
In summary then, we are noting that the results of this study should be regarded as suggestive and not definitive, and caution against over-interpretation. Having done this, it should also be recognised and acknowledged that the alternatives to this approach are few, the most common being "scientific subjective assessment"; that is, relying on the subjective opinions of scientists who are knowledgable in the field. The recent report of the Ocean Observing System Development Panel [OOSDP 1995] provides a good example. However, even such assessments benefit from objective studies which explore various dependencies of the model analyses and predictions on the input data stream. The fact that decisions regarding long-term support for elements of the in situ and/or remote observing networks will have to be made at some time in the near future is without dispute, so we should endeavour to explore the subtleties of the system dependencies as far as possible in order to allow those decisions to be as well-informed as possible. That is, we wish to moderate, though not eliminate, the degree of subjectivity in such decisions.
2. Information interpretation and the methodology
Prior to presenting experimental results we wish to discuss a little further the concept of (observed) information, how it is used to provide an analysis or initial state for a forecast model, and present a little of the philosophy behind our approach. Analysis systems and ocean models are in effect tools for interpreting, interpolating and extrapolating information in space and time and, perhaps, into different regions of variable space [Smith 1993; Woods 1995]. If these tools were perfect, and the space/time chaotic nature of the systems was not an issue, then the quality of analyses (spatial interpolation and extrapolation) and forecasts (temporal extrapolation) would only depend on the quality of ingested data. Clearly this is not the case and we are faced with the predicament of choosing (for observing system sensitivity and evaluation tests) between a method which is complex, but in theory more complete, such as a coupled ocean-atmosphere general circulation model, and simpler tools that, while flawed through their simplicity, are easier to use and offer greater hope of understanding fundamental sensitivities. In this study we are adopting the latter approach.
The two methods we use to interpret the subsuiface thermal data are (a) the operational ocean analysis system of Smith [1995] and (b) the Kleeman et al. [1995] intermediate coupled model. In method (a) we are attempting to combine data spread in
