A recent article by Wang and Lee  (henceforth WL2008) claims to show observational evidence for a relationship between trends in land-falling US hurricanes and global warming of sea surface temperatures (SST). The detection of trends in hurricane activity, and the attribution of hurricane trends to specific SST patterns are problems that are fraught with methodological difficulties. In this comment I am concerned that the analysis by WL2008 does not properly take into account the presence of natural interannual variability in the SST record. In addition the technique does not differentiate between natural and anthropogenic longer-term variability, which may have different spatial patterns and therefore different effects on hurricane activity. Therefore I am concerned that WL2008 may be falsely attributing decreases of landfalling hurricanes to trends in “global warming of sea surface temperature”.
 It is well known that there is a relationship between El Nino/Southern Oscillation (ENSO) and the seasonal statistics of Atlantic hurricanes. This relationship is largely the result of ENSO's effects on vertical wind shear through atmospheric teleconnections. In fact, this relationship is one of the bases for the success of seasonal forecasts of hurricane activity [e.g., Gray, 1984]. WL2008 also rely on this relationship as the main causal link in their analysis of global warming:
“A secular warming of sea surface temperature occurs almost everywhere over the global ocean. Here we use observational data to show that global warming of the sea surface is associated with a secular increase of tropospheric vertical wind shear in the main development region (MDR) for Atlantic hurricanes.”
 However, the SST pattern that the authors call the “global warming of the sea surface” is defined as the first Empirical Orthogonal Function (EOF) of annual mean SST, and therein lies the problem. The first EOF is the spatial pattern whose time series, determined by orthogonal projection, explains the most variance in the data. This pattern contains a mixture of natural and anthropogenically forced SST trends with natural variability in which ENSO is very prominent. If this weren't clear enough from the similarity of the SST pattern to a composite El Nino SST pattern (WL2008, Figure 1a. We have repeated the analysis here in Figure 1a), then the time series of this pattern (WL2008, Figure 1b) with its peaks and valleys corresponding to prominent El Nino and La Nina events makes this even more evident.
 Compare, for example, the pattern of WL2008 to the same EOF analysis done when an 11-year running mean filter has been applied to the SST data (Figure 1b). This procedure is a first-cut at removing ENSO-related interannual variability, while retaining most natural and forced multidecadal variability and trends (M. Newman, personal communication, 2008; see also Newman  for more discussion of the timescales involved). The filtered patterns show much less emphasis on SSTs in the central and eastern Tropical Pacific – the core region of ENSO variability. In addition, the recent 50-year linear trend pattern shown in Figure 1c shows little trend in the core ENSO region. Individual climate model projections of SST for the 21st century also indicate that the spatial patterns of anthropogenically forced SST trends may differ considerably from interannual variability (not shown) [see Barsugli et al., 2006, Figures 1 and 2].
 While the authors allude to the presence of ENSO in their “global warming” pattern I am concerned that they do not grasp its significance. The fact that global mean temperature is positively correlated with El Nino along with the fact that El Nino is positively correlated with MDR vertical wind shear does not lead to the conclusion that “global warming”, as it is usually understood, is associated with wind shear (and hence landfalling hurricanes). Simply put, the warming may be happening in different places and have different impacts on the atmospheric circulation. The large ENSO signal may be masking, or even enhancing the true global warming signal on hurricanes. It is impossible to know the answer without a proper analysis that explicitly filters out the influence of ENSO variability.
 Unfortunately, filtering ENSO is not as simple as it sounds because of several factors [Penland and Sardeshmukh, 1995; Penland and Matrosova, 1998, 2006]. ENSO is fundamentally a multivariate phenomenon; filtering with a single index of ENSO will not remove all of the ENSO effects, and may in fact remove the trend you are seeking to study. ENSO, through its teleconnections, also involves SST variations in all tropical ocean basins, complicating the analysis. Finally, ENSO has a “low frequency tail” that contributes to natural decadal variability. Exactly how the ENSO filter incorporates these phenomena – particularly the ENSO-related SST variations in the Atlantic – will have a large effect on the attribution of the hurricane timeseries to “global warming of the sea surface”.
 ENSO is not the only natural factor that needs to be considered when attempting to isolate the global warming signal in observational data; it is singled out here because of its significant contribution to the variations in the global mean SST and to the modulation of hurricane statistics. Other natural phenomena including the Atlantic Multidecadal Oscillation (AMO) and other “modes” would need to be filtered (or explicitly included in a statistical model) as well, but these may prove even more difficult to treat than the ENSO signal.
 The only hope of understanding what the climate record has to teach us about the future of US landfalling hurricane activity is to combine statistically sound analysis of observed data, carefully constructed modeling studies, and theoretical reasoning. Carefully crafted statistical analyses can address the methodological issue raised here and contribute to an integrated understanding of this phenomenon. These studies need to be done.