Hurricane Irene (2011) worst-case estimates of wind damage to property from exigent analysis of ECMWF ensemble forecasts

Author: Daniel Gombos and Ross N. Hoffman
January 10, 2013
Symposium on the Role of Statistical Methods in Weather and Climate Prediction

Daniel Gombos, and R. Hoffman (2013) Hurricane Irene (2011) worst-case estimates of wind damage to property from exigent analysis of ECMWF ensemble forecasts. Symposium on the Role of Statistical Methods in Weather and Climate Prediction, San Francisco, CA.

Ensemble-based exigent analysis uses the covariance estimated from an ensemble forecast of a damaging weather event to determine the worst-case scenario (WCS). The WCS (called the exigent damage state (ExDS)) is determined as the sum of the ensemble mean damage and the exigent damage perturbation (ExDP). The ExDP is the forecast damage ensemble perturbation that maximizes the weighted sum of the grid-point damage values (called a damage functional) at a user-prescribed uncertainty level (called a Mahalanobis distance quantile (MDQ)). The damage at each grid-point is estimated using a chosen damage function, which can range in complexity from simplified parameterizations to comprehensive commercial catastrophe (CAT) models.

The ExDS is potentially valuable because, for multivariate Gaussian ensembles, it is the unique scenario expected to be the most probable for a given amount of area-integrated damage and the most damaging for a given probability. It provides a compact view of what might (or might have) happened – a map depicting the multivariate uncertainty bound of the integrated damage for a forecast weather event that is consistent with the dynamics and statistics of the forecast ensemble.

Exigent analysis is applied to ECMWF ensemble forecasts to estimate WCS wind damage to property during Hurricane Irene, a Category 1 hurricane that made landfall in North Carolina on August 27, 2011. The ExDS for forecasts initialized 12 UTC 24 August has anomalously high damage along the New Jersey Atlantic coast, around New York Harbor centered on Jamaica Bay, Narragansett Bay, and along Cape Cod and Massachusetts Bay and high inland penetration of damage, especially in the Delmarva peninsula, New Jersey, and eastern Massachusetts. The ExDS for forecasts initialized 12 UTC 25 August tracks Irene more to the west, with damage concentrated in South Carolina, the mouth of the Chesapeake Bay, the New Jersey coast and both ends of Long Island, sparing much of eastern Massachusetts. For forecasts initialized on 12 UTC 26 August, the forecast track and damage patterns are intermediate between those with initial times of 12 UTC 24 and 25 August, but the level of damage is greater at the mouth of the Chesapeake and around New York Harbor. Luckily, in the case of Irene, the WCS did not occur; exigent total wind damage estimates were found to be as high as $3.52B at the 90th MDQ, compared to $0.59B as estimated from the analyses.

Exigent analysis has direct applications to emergency situations affecting life and property. The ExDS can be used as extreme, but definitely possible, scenarios by emergency responders or others planning for an impending severe weather event. Upper-bound estimate of the numbers of people and/or housing units expected to be affected could be used to preposition ambulances, food and shelter resources, insurance adjusters, construction and restoration supplies, and utility crews. Resources could be prepositioned in a Goldilocks zone - locations that are close, but to avoid dangerous conditions, not too close - to where they would be needed in a WCS.

In addition, extended exigent analysis (ExEA) can be used to determine the antecedent conditions associated with the ExDS. These antecedent conditions might be used to understand the atmospheric dynamics that may lead to the WCS or, when real-time observations align with the exigent antecedent conditions, to alert emergency responders before the next model run. Exigent analysis also has potential to guide weather modification experiments by estimating optimal timing and location of cloud seeding.