New strategies to estimate future changes in tropical cyclone maximum wind speeds

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 4 February 2014: 9:30 AM
Room C205 (The Georgia World Congress Center )
Mari Jones, NCAR, Boulder, CO; and B. D. Youngman, G. J. Holland, D. B. Stephenson, and R. W. Katz

Reliable estimates of future changes in extreme weather phenomena, such as tropical cyclone (TC) or extra-tropical cyclone (ETC) maximum wind speeds are critical for climate change impact assessments and the development of appropriate adaptation strategies. However, General Circulation and Regional Climate Model outputs are often too coarse for direct use in these applications, resulting in very different probability distributions for observed and model simulated maximum wind speeds. This poses two problems: how can we best adjust model simulated wind speeds to make them more realistic; and can we use these relationships to make more reliable predictions of future maximum wind speeds?

Observed (1950-2010) and model simulated (12km and 36km grid for 1995-2005 and 2045-2055) North Atlantic Tropical cyclone maximum wind speeds (vmax) are used to examine parametric and semi-parametric approaches and identify relationships between the modelled and observed maximum wind speeds. All the maximum wind speeds are well described by left-truncated Weibull distributions, which allows us to exploit a power transformation to calibrate between the model and observed TC vmax. This transformation leads to two calibration strategies to estimate future TC vmax: bias correction or change factor. While the results are broadly consistent across different model resolutions, we conclude that there are differences in future estimates of maximum wind speeds from the two calibration strategies arising from the different treatment in distributional variance.

A further calibration strategy is proposed, which is not fully dependent on the selection of an appropriate statistical distribution. Quantile-quantile plots of the log-transformed observation and simulation data display an interesting linear property facilitating a location-scale semi-parametric downscaling. Sensitivity testing demonstrates that this method is robust for storm durations >48hours for all time periods used for the current climate, as well as for different model resolutions.