Extreme windstorms in the Northeastern USA in the contemporary and future climate

Abstract

Cold-season windstorms represent an important, and potentially changing, geophysical hazard in the Northeastern United States. Here we employ an integrated research methodology including both a storyline approach, where three intense windstorms from the current climate are subjected to pseudo-global warming (PGW) experiments, and a long-term transient simulation using the Weather Research and Forecasting (WRF) model. An ensemble of WRF simulations is built for each windstorm using different planetary boundary layer and microphysical parameterizations. The fidelity assessment suggests all ensemble members capture the windstorm evolution in contemporary climate. The configuration with highest fidelity is used in the PGW experiments performed with perturbed temperature fields, constant relative humidity, and deiced Great Lakes. These perturbation simulations indicate some evidence for a reduction of sea level pressure and increases in wind speed over and downwind of the Great Lakes and over the Atlantic Ocean plus an increase in precipitation accumulation but a reduction in snow coverage. These changes are spatially inhomogeneous in terms of magnitude and sign but are consistent with changes in potential vorticity. Alberta Clippers and Colorado Lows dominate the cyclones responsible for historical windstorms and thus are sampled in the PGW simulations. However, the transient simulation suggests an increasing role for tropical cyclones that undergo transition to extratropical cyclones. This reinforces the value of combining information from both PGW perturbation experiments within a storyline context and transient simulations when seeking to quantify the future risk associated with cold-season windstorms under changing climate.

Appendix: Evaluation framework and composite skill score

A composite skill score is calculated across multiple fidelity metrics and physical properties. As in Taylor diagrams, three aspects are considered when evaluating field statistics: root mean square error (RMSE), ratio of standard deviation ($\sigma$), and correlation ($r$) (Taylor 2001; Wilks 2011). The RMSE is normalized by the ensemble mean (overbar) to generate a (normalized) value that can be combined with other metrics:

$${S}_{RMSE}= RMSE/\widehat{RMSE},$$

(2)

For $\sigma$, the skill score is calculated as:

$${S}_{\sigma }= \frac{abs\left({\sigma }_{x}-{\sigma }_{o}\right)}{{\sigma }_{o}},$$

(3)

where subscripts x and o represent the simulation and observation respectively. For the correlation coefficient between x and o, the skill score is calculated as

$${S}_{r}=1-{r}_{xo},$$

(4)

For fields sampled in both space and time, the above metrics can be either applied to the time series of each grid to evaluate the temporal features or applied to the field at a given time step to evaluate the spatial patterns.

We also consider the location of the feature centroid and the spatial coverage of extreme values. For the centroid location, the field is smoothed before the maximum or minimum value is geolocated. Then the difference in locations of the max or min, ${d}_{c}$, is calculated between the simulation and observation. The skill score of centroid location, ${S}_{dc}$, is then calculated as the difference normalized by the ensemble mean difference:

$${S}_{dc}= {d}_{c}/\widehat{{d}_{c}},$$

(5)

To assess the spatial coverage of extreme conditions we count the number of grids with SLP < 1000 hPa, WS at 10-m > 10 ms⁻¹, WS at 100-m > 15 ms⁻¹, RR > 4 mm h⁻¹, cREF > 35 dBZ and P_tot (accumulated precipitation over the entire simulation) > 25 mm. While these thresholds are to some degree arbitrary they are selected to avoid small sample sizes generated by selection of more extreme threshold and in the case of cREF employs a threshold used to identify convective storms (Nisi et al. 2018). The grid cells that meet the threshold (N_hit) are compared between WRF and observation to evaluate the spatial coverage of hazardous features:

$${S}_{N\_hit}= \frac{abs({N}_{x\_hit}-{N}_{o\_hit})}{{N}_{o\_hit}},$$

(6)

in which ${N}_{x\_hit}$ and ${N}_{o\_hit}$ are the numbers of grids that meet the threshold in the simulation and observations, respectively. To weigh the scores of individual metrics/fields equally, all skill scores are normalized by the ensemble mean across different cases and model configurations,

$${S}_{MN}= {S}_{M}/\widehat{{S}_{M}}.$$

(7)

Table

Table 3 Properties and metrics used to evaluate simulated windstorms

3 shows a comprehensive list of the fields and metrics used for model evaluation. In total 47 individual skill scores are calculated for each simulation. The individual skill scores are then summed across all windstorm cases to produce a composite skill score. The precipitation related scores are given half the weight of those associated with wind conditions due to our focus on intense windstorms.

For comparison, an alternative metric-framework, the relative Euclidean Distance (D (Wu et al. 2012)), is also considered,

$$D= \sqrt{{\left(\frac{\overline{x }-\overline{o} }{\overline{o} }\right)}^{2}+{\left(\frac{{\sigma }_{x}-{\sigma }_{o}}{{\sigma }_{o}}\right)}^{2}+{\left({r}_{xo}-1\right)}^{2}},$$

(8)

where the overbar represents spatial or temporal average of a field. D considers all three aspects of a field at once in a normalized form, except the first term in D representing the bias in mean value.

When the assessed variable is averaged (or integrated) over both time and space, only one value is used for comparison (e.g., mean P_tot). In this case, the percent error (PE (Wilks 2011)) is used for evaluation,

$$PE= \frac{\overline{x }-\overline{o} }{\overline{o} },$$

(9)

All evaluations are done within the window ± 24 h from the maximum storm intensity (t_p).