Understanding the sensitivity of North Atlantic tropical cyclone limiting intensity to sea surface temperature: Observations versus climate simulations

Previous studies have examined relationships between sea surface temperature (SST) and tropical cyclone (TC) intensity that bear potentially important implications for the impact of climate change on TCs. We have previously shown that the sensitivity of TC limiting intensity to SST is 7.9 m/s/K for observed TCs over the North Atlantic basin, where limiting intensity is a statistically defined upper limit on hurricane intensity and sensitivity is estimated as the slope of a regression of limiting intensity onto SST. Subsequent research implemented the same methods and arrived at significantly lower sensitivity values for TCs generated by two atmospheric global climate models (GCMs). Here, we move toward attaining a deeper understanding of this sensitivity of limiting intensity to SST as well as the large discrepancies between sensitivities for observed and GCM-generated TCs. As a first step, we examine the sensitivity for five different models with horizontal resolutions ranging from ~20 km to 1°. Preliminary results suggest that while the lowest resolution models are associated with the lowest sensitivity of limiting intensity to SST, relatively high model resolution is not always associated with high sensitivity.

Additionally, the spatial and temporal structure of North Atlantic tropopause temperature from the NCEP CFS Reanalysis is compared with output from five different atmospheric GCMs. We hypothesize that warmer simulated tropopause temperatures may inhibit the thermodynamic efficiency of GCM-generated TCs, thus reducing limiting intensity values and the sensitivity of limiting intensity to SST. We next use a multiple regression of limiting intensity onto SST and tropopause temperature to gain insight into how much variance in limiting intensity can be accounted for by SST and tropopause temperatures. This is implemented for both observed/reanalyzed and GCM-generated data sets. Results focus on differences between observed and modeled sensitivities, which may indicate critical deficiencies in the models' physical representation of TCs.