The effect of successive correction on variability estimates for climatological datasets

The effects on a dataset of smoothing by successive correction have been in vestigated. The resulting spatial resolution is estimated using a distribut ion of ship reports from a sample month. Although the smoothing uses the sa me characteristic radii over the whole globe, the resulting resolution is s patially variable and, in data-sparse regions, will show large month-to-mon th variability with changes in the distribution of the ship tracks. The cli matological dataset, which is gridded at 1 degrees, is shown to have a typi cal resolution of 3 degrees. In some regions the resolution is much coarser . Using sea surface temperature as an example, it is shown that the successiv e correction procedure as used, for example, in a recent climatological dat aset, is not successful in removing all of the noise in data-sparse regions . Additionally, the well-defined intermonthly variability in the main shipp ing lanes, where there are many observations, is degraded by the influence of poorer-quality data in the surrounding regions. This typically increases the intermonthly variability estimates in the shipping lanes by a factor o f 2. Further, the reduction of intermonthly variability, by up to a factor of 6, in highly variable regions such as the Gulf Stream, is greater than c an be accounted for by noise in the individual ship reports. This reduction is due to the removal of small-scale variability by the smoothing process. Removal of coherent and persistent small-scale variability has an effect o n the temporal and spatial characteristics of the data. It is suggested tha t smoothing by successive correction, although commonly used, is poorly sui ted to such spatially inhomogenous data as those from the merchant ships. However, the effect of successive correction on variability analysis using empirical orthogonal functions (EOFs) is shown to be small for the most sig nificant modes of variability identified in the Gulf Stream region. This is because the EOF analysis picks out the large-scale variability in the high est-order modes. However, too large a fraction of the total variance explai ned is ascribed to the large-scale modes of variability. Variability with s mall spatial scales is more likely to be significant if raw data are used i n the EOF analysis. Little significance should be given to EOF modes with s patial scales similar to the size of gaps between shipping lanes; this vari es from region to region.