Reply to comment by R. N. Maue and R. E. Hart on “Low frequency variability in globally integrated tropical cyclone power dissipation”

[1] The field of tropical cyclone (TC)-climate variability is relatively young. Differences remain about the interpretations of methods and terminology and the implications of results. We welcome the opportunity to clarify some of the concepts and results from Sriver and Huber [2006] (hereinafter referred to as SH06) by responding to the Maue and Hart [2007] (hereinafter referred to as MH07) critique. We thank the authors for verifying the methods of SH06, and we are pleased with the reproducibility of our essential results. The main criticism raised in MH07 is that SH06’s results are not independent of previous studies that employ the ‘Best Track’ (BT) wind data set. In this reply, we refute their main criticism and expand upon MH07’s results. [2] SH06 calculated the annually integrated TC power dissipation (PD) on a global scale. As an integrated measure, PD is the convolution of storm duration, frequency and intensity. SH06 built upon the results of Emanuel [2005] (hereinafter referred to as E05) that approximated PD as the power dissipation index (PDI) using BT maximum sustained wind estimates. SH06 investigated how utilizing a wind intensity data set independent of E05 might change PD and made no pretense of using different storm tracks. We considered the winds employed in SH06 independently derived and free from the controversial, ‘ad hoc and subjective adjustments’ used by E05. [3] SH06 showed close agreement between ERA40derived PD and PDI, and all ERA40-derived quantities were highly correlated with E05’s PDI for the combined Atlantic and northwestern Pacific regions post-1978. These results showed that post-1978, ERA40 robustly reproduced trends in PD observed in BT winds, PDI was an adequate approximation of PD, and combined trends in integrated intensity in the Atlantic and northwestern Pacific regions were reasonable indicators of low frequency variability in globally integrated TC activity. [4] MH07’s main criticism is that the results of SH06 are not independent of E05 because both studies utilized the same track data—‘‘the relationship between ERA40 PD and BT PDI is a result of dataset interdependence on frequency and lifecycle with less than 10% of the correlation arising from intensity’’. In other words, their hypothesis is that SH06’s results could not stray far from those of E05 because the same track data was used in both studies. [5] However, as shown in SH06, ERA40-derived PD and E05 PDI prior to 1978 do not agree well (Figure 1 in SH06 and Figure 1 in this paper using unfiltered data). We conjectured that this might be a consequence of less reliable winds in ERA40 during the pre-satellite period, but our analysis never attempted to prove this. Regardless, whichever data set is more correct, the fact that the two time series can differ strongly is sufficient grounds to reject the MH07 hypothesis that the two time series are trivially related. This answers MH07’s main criticism, which we thought was clear from Figure 1 of SH06. [6] MH07 further propose that wind field variations are not an important contributor to integrated TC intensity variability. They claim that the track-sensitive term (frequency and duration) governs PD/PDI. We note that this pattern must be repeated in both ERA40 and BT records for the main MH07 criticism to be valid. MH07’s Figure 1b shows that replacing TC winds from ERA40 with a constant value reproduces the variability of Emanuel’s PDI time series derived from BT winds. MH07 show that ERA-40 PD and E05’s PDI are 90% correlated, but then inaccurately suggest that 80% of the variance is explained by the track-sensitive term alone, suggesting that TC wind changes are relatively unimportant compared to track length and storm frequency changes when describing trends in integrated intensity. [7] If true, then the variance in E05’s time series is also dominated by variability in the tracks, with little role for wind speed variations, because of the strong correlation between the E05 and ERA-40-derived time series. This would be very surprising given the current debate on the importance of historical TC wind records [Landsea et al., 2006] and the fact that all studies so far have reiterated the importance of better knowledge of TC velocities [Chan, 2006; Hoyos et al., 2006; Webster et al., 2006]. Here we show that the MH07 conclusion is not true and is the byproduct of MH07’s flawed statistical methodology. [8] A fundamental aspect of statistical analysis is that the explainable variance of a time series is only separately attributable to uncorrelated components of the series. In this respect, MH07’s analysis contains a basic statistical flaw in their attempt to attribute explained variance in PD/ PDI. They attribute explainable variance in PD/PDI to the track-sensitive term while holding the wind-sensitive term constant. However, these terms are not independent of one another (i.e. not orthogonal), thus their technique fails to account for colinearity and covariance between the windsensitive and track-sensitive terms. For example, in the case of two perfectly co-linear time series, the track-sensitive term might appear to explain 100% of PD variance, which GEOPHYSICAL RESEARCH LETTERS, VOL. 34, L11704, doi:10.1029/2007GL029413, 2007 Click Here for Full Article