Main

First, weighting by inverse-fourth-power of separation distance in our analysis effectively eliminates contributions near to the 2,284-km radius of influence if other stations are closer. Only between the Antarctic Peninsula and the Ross Sea does the gap between points approach the radius of influence. Second, the trend maps in our Fig. 2 have features that are finer than the quarter-image radius of influence, indicating that that local data prevail in our analysis. Third, the interpolated fields are consistent with the summaries for 1976–2000 in Fig. 2.10 of the recent IPCC WG-I report1 and with trends based on a different data set2 for 1965–1999, an interval that was chosen because it was one of global warming. A recent depiction3 of combined summer and autumn Antarctic surface-temperature trends from 1969 to 2000 is similar to ours, although our data suggest that cooling is more pronounced during autumn periods than in summer.

The key question that arises is this: is our interpolation better than arithmetic averaging? We contend that, although the interpolations involve uncertainty, they highlight the fact that a full assessment of Antarctic temperature trends requires more than the averaging schemes that have so far been used to imply that Antarctica has been warming at a rate that is faster than the global average4,5.

In estimates of hemispheric anomalies or trends (for example, see the IPCC's Figure 2.7; ref. 1), regions that are devoid of data are effectively assigned anomalies (or trends) that are equal to the hemispheric means of anomalies (or trends) for areas for which data exist. This type of assignment ignores information from even nearest-neighbour grid cells. We maintain that our approach represents an improvement over arithmetic means of station values or grid cells, provided that the nature of the interpolation procedure is clearly stated.

Only by interpolation can one hope to determine the area fraction and spatial pattern of continental warming or cooling, regardless of the magnitudes of such trends — which are irrelevant to our conclusions. As Antarctic trends obviously vary spatially, seasonally and interdecadally, interpolation will ultimately be required to optimize the information contained in historical data.