Extreme winds from tropical cyclones (TCs) regularly threaten communities worldwide. In recent decades significant efforts have been put towards improving our understanding of the mechanisms involved. In particular detailed analysis of satellite imagery and observations from aircraft reconnaissance missions have allowed formulation of a well-accepted framework whereby asymmetries in the TC wind field structure are attributed to the forward motion of the system: stronger winds occur to the right (left) of a moving TC in the northern (southern) hemisphere with the magnitude of the left/right asymmetry increasing as the storm moves faster.
During the same period many simple parametric formulations have flourished to model the spatial distribution of winds around a TC based on this framework (e.g. Holland 1980, Willoughby et al. 2006). The simplicity of the approach allows for fast and efficient formulations that can be used in the context of risk assessment systems where millions of simulations are often required (Vickery et al. 2009).
Recent observations however have highlighted the limitations of this framework and many researchers now stress the role played by the environmental flow on determining the location of the surface wind maximum (Uhlhorn et al. 2014, Klotz and Jiang 2016). In particular TCs evolving in strongly sheared environments routinely produce devastating winds of similar magnitudes on both sides of their tracks. Such phenomenon is very common in the western North Pacific (Loridan et al. 2014, 2015) and is clear from analysis of Hurricane Sandy’s wind field at landfall: www.earthobservatory.nasa.gov/IOTD/view.php?id=79626
To account for such cases that strongly depart from the popular motion-induced asymmetry (MIA) framework discussed above, and as an alternative to current parametric wind modelling techniques, Loridan et al. (2017) recently investigated the potential of machine learning (ML) methods. ML algorithms can infer patterns from large amounts of data without any need for explicit instructions (self-taught) and therefore offer a very promising alternative to commonly used parametrization techniques (a.k.a. “knowledge engineering”).
The ML method developed is able to simulate a much wider range of TC wind field patterns than is possible with current wind field parametrizations as there is no need for explicit model rules to be prescribed. In particular the MIA assumption can be relaxed and atypical wind fields that do not exhibit stronger winds to the right (left) of the moving TC in the northern (southern) hemisphere can now be simulated. This includes wind fields from extratropical transitioning systems (see Evans et al. 2017) often characterized by strong winds on both sides of the TC track.
The new ML-based model simulates the full distribution of the TC surface wind field based on its track, intensity and wind field size parameters. By capturing the full distribution rather than a single estimate the method can directly characterize the uncertainty in model predictions. This has direct benefits if used in applications where a large number of model samples are needed, such as in risk assessment systems or probabilistic forecasting (see Fig. 1).
The flexibility of the modelling approach also allows easy visualization of the risk potential during a live event. Fig. 2 illustrates this with our live ML event simulator for Typhoon Noru 2017 (right of Fig. 2). The live simulator allows key wind parameters to be adjusted by the user in order to sample the full distribution of risk.
For more information, go to www.riskfrontiers.com
1 Evans, C and coauthors, 2017: The Extratropical Transition of Tropical Cyclones. Part I: Cyclone Evolution and Direct Impacts. Mon. Wea. Rev. https://doi.org/10.1175/MWR-D-17-0027.1
2 Holland, G. J., 1980: An analytic model of the wind and pressure proﬁles in hurricanes. Mon. Wea. Rev., 108, 1212–1218, doi:10.1175/1520-0493(1980)108,1212:AAMOTW.2.0.CO;2.
3 Loridan, T., E. Scherer, M. Dixon, E. Bellone, and S. Khare, 2014: Cyclone wind ﬁeld asymmetries during extratropical transition in the western North Paciﬁc. J. Appl. Meteor. Climatol., 53, 421–428, doi:10.1175/JAMC-D-13-0257.1.
4 Loridan, T., S. Khare, E. Scherer, M. Dixon, and E. Bellone, 2015: Parametric modelling of transitioning cyclone wind fields for risk assessment studies in the western North Pacific. J. Appl. Meteor. Climatol., 54, 624-642, DOI: 10.1175/JAMC-D-14-0095.1.
5 Loridan, T., R.P. Crompton, and E. Dubossarsky, 2017: A Machine Learning Approach to Modeling Tropical Cyclone Wind Field Uncertainty. Mon. Wea. Rev., 145, 3203–3221, https://doi.org/10.1175/MWR-D-16-0429.1
6 Uhlhorn E. W., B. W. Klotz, T. Vukicevic, P. D. Reasor, and R. F. Rogers, 2014: Observed hurricane wind speed asymmetries and relationships to motion and environmental shear. Mon. Wea. Rev., 142, 1290–1311, doi:10.1175/MWR-D-13-00249.1
7 Vickery, P. J., F. J. Masters, M. D. Powell, and D. Wadhera, 2009: Hurricane hazard modeling: The past, present, and future. J. Wind Eng. Ind. Aerodyn., 97, 392–405, doi:10.1016/j.jweia.2009.05.005.
8 Willoughby, H. E., R. W. R. Darling, and M. E. Rahn, 2006: Parametric representation of the primary hurricane vortex. Part II: A family of sectionally continuous proﬁles. Mon. Wea. Rev., 134, 1102– 1120, doi:10.1175/MWR3106.1.