Hurricane wind fields needed to assess risk to offshore wind farms

In their paper in PNAS, Rose et al. (1) applied a statistical model to estimate hurricane wind losses to wind turbines over a 20-y typical wind farm lifetime. They combined a county annual landfall frequency probability density function with a generalized extreme value (GEV) fit of maximum wind speeds to model the expected 20-y losses attributable to hurricane activity at four hypothetical offshore wind farm sites.

We found one error and three major flaws in this approach, which lead to an order of magnitude overestimation of risk:

• i) The GEV fit parameters used by Rose et al. (1) can only be reproduced if applied to historical maximum 1-min wind speed estimates (the paper incorrectly states 10-min wind estimates). Maximum 1-min winds are typically 10% higher than a 10-min mean wind measured over the same period (2).

• ii) The historical data used for the GEV fit are dominated by years (1851–1977) with high uncertainty for tropical cyclone intensity. Post-1977 records are more accurate because of improved reconnaissance aircraft measurements, and pre-1900 records are generally not considered reliable for hurricane risk modeling (3).

• iii) Hurricanes do not maintain their maximum “regional box” (box coordinates shown in table 1 of ref. 1) intensity all of the way to landfall (e.g., 4).

• iv) Hurricanes have spatially varying wind fields. Typically, the highest winds affect a very small area within the eye wall (e.g., H*Wind; www.aoml.noaa.gov/hrd/data_sub/wind.html).

Applying the Galveston Poisson landfall rate and a GEV fit of landfall maximum winds (table 1 of ref. 4) (winds converted to 10 min) over a 200,000-y period and applying figure 1 of ref. 1 (for unyawed turbines) to each storm, we find an 8% chance of losing 25 or more turbines over a 20-y period, compared with 30% as reported by Rose et al. (1). However, this is still an overestimation of risk, because it is unlikely that the maximum wind from each storm would have an impact on a relatively small wind farm.

Fig. 1.

Maximum 10-min average, 10-m level wind speeds at hypothetical wind farm locations offshore Galveston, TX (red: 29.091° N, 94.9° W; green: 29.091° N, 95.085° W; blue: 29.3° N, 94.747° W), and maximum in-box northwest Gulf of Mexico (25.5°–30° N, 92°–99° W) (black) for all historical hurricanes passing within 200 km of Galveston since 1900.

To account for spatial variation in hurricane winds, we used a wind field model similar to those used in the risk industry [e.g., Florida public hurricane loss model (5)] on all historical hurricane tracks passing within 200 km of Galveston from 1900 to 2010. Maximum winds were recorded at three wind farm locations as well as the peak (regional box) HURDAT (Best track: http://www.nhc.noaa.gov/pastall.shtml#hurdat) wind for each storm while traversing the northwest Gulf of Mexico (Fig. 1). We find that, on average, the wind farm locations experience winds that are 65% (Fig. 2) of the peak in-box wind speeds used in the study by Rose et al. (1), resulting in a loss of 2 turbines per 20-y period compared with 24 turbines if their unrealistic peak in-box winds are used. We recommend that future work toward estimation of hurricane risk use best practices already prevalent in the insurance industry, which include the use of physically based wind field models that account for spatial variation (www.sbafla.com/methodology).

Fig. 2.

Scatterplot of maximum landfall winds at wind farm locations (y axis) compared with the peak life cycle wind speed for the same hurricane while in the northwest Gulf of Mexico.