Assessing and Modeling the Compound Hazard of Freezing Rain and Wind Gusts

Compound events with freezing rain (FZ) induced ice accumulation plus high magnitude wind gusts (G) pose a major risk to infrastructure and travel. The US Department of Homeland Security, National Protection and Programs Directorate identify freezing rain and icing events as a threat to critical energy and transport infrastructure if they lead to accumulation of 0.5 cm or more on exposed surfaces. For example, the event of 5-9 January 1998 caused damage of over $4 billion dollars and impacted much of New England and eastern Canada with accumulations of 10 cm of ice on some power lines and other electricity distribution infrastructure. Resulting damage to vegetation and energy infrastructure can cause power outages lasting several days while road and air travel disruptions hamper recovery efforts and economic activity. However, relatively few studies have sought to derive a robust climatology of such events or to diagnose geophysical causes. This is due in part to the lack of ice accumulation data, measurement of which is not commonly undertaken at in-situ meteorological stations. Further, the drivers and scales of such events are not fully captured by numerical models. The goals of this work are twofold: generate a record of these compound events from observations, as well as to assess their variability and how effectively largescale predictors can be used to predict their occurrence and magnitude.

In-situ observations are taken from the Automated Surface Observing Station (ASOS) records from sites across the Contiguous United States (CONUS) during 2005-2022. ASOS reports data at 1-minute and 5-minute intervals, with indicators for freezing rain occurrence in the 5-minute data which result from precipitation identification and changes in the 1-minute resonance frequency reported by an icing sensor. Precipitation type is determined by the ASOS temperature sensor and the Light Emitting Diode Weather Indicator (LEDWI), which differentiates between liquid precipitation and snowflake fall-speeds. Where the ice sensor shows a drop in resonance frequency (i.e., ice has built up on the plate) and liquid or unknown precipitation is measured, freezing rain is indicated. Ice sensor frequency changes can be converted to estimates of ice accumulation, though the sensors are prone to biases due to ‘ice-bridging’ and the sensor records are less complete than the rest of the ASOS data.

Ice accumulation is modeled using the ASOS data from 883 stations and the Cold Regions Research and Engineering Laboratory (CRREL) model, which uses precipitation rate, wind speed, air temperature, relative humidity and surface solar flux to simulate the fraction of impinging precipitation that freezes on contact with a surface. All of the predictors are measured at ASOS locations except the solar flux, which is estimated using hourly top of the atmosphere flux estimated using the solar constant and hour angle, and a local albedo from deep clouds of 0.82. The CRREL model is applied to hourly values as calculated from averages of the 5-minute data and the hourly liquid precipitation accumulation during all 5-minute periods when FZ is indicated within each hour. Consecutive hours of freezing rain result in more ice accretion, with accumulated ice decreasing in the subsequent hours without freezing rain at the e-folding rate. Validation of the CREEL model output are performed using the ice sensor data, as well as estimates from storm reports for events around CONUS taken from the National Weather Service (NWS) and the National Oceanic and Atmospheric Administration (NOAA).

Wind speeds and gusts are measured at ASOS sites using a 2-D sonic anemometer. Here the maximum of the 5-minute wind gusts reported in each hour are used in conjunction with the ice accumulations and the damage categories used by the NWS to calculate the risk category during each event, ranging from 1 (slight damage risk) to 5 (catastrophic damage risk). Relative frequencies of events and their associated impacts are assessed for the full period, with emphasis on major events in each of the 7 National Climate Assessment (NCA) regions across CONUS. The bulk of events occur in an arc from Texas to the upper Midwest and Great Lakes and across the Northeast, with secondary peaks in South along the Appalachian Mountains and in the Pacific Northwest.

Data from the ERA5 reanalysis are used to examine the meteorological context for FZG events identified from the ASOS observations and those observational estimates are also used to evaluate ERA5 estimates of precipitation type and wind gusts. With respect to the latter, our analyses clearly indicate significant biases in the spatial pattern of freezing rain, gust magnitude and event frequency and severity estimated from ERA5. Significant events with large ice accumulation are under-estimated in ERA5, potentially due to the sub-grid scale processes inherent in freezing rain and wind gust variability, indicating the clear need for use of the in-situ data sets in characterizing this hazard. Nevertheless, the ERA5 reanalysis has been shown to characterize large scale meteorological conditions with high fidelity and thus near-surface and tropospheric variables from ERA5 are used as predictors in machine learning algorithms (stepwise regression, decision trees and artificial neural networks) to predict ice accumulation and wind gust occurrence independently, which are then combined to obtain the compound events. Stepwise regression is used to identify the important predictors. Significant predictors identified in the stepwise regression are applied in the artificial neural networks (ANNs) to avoid overfitting. Predictors selected in this statistical analysis match physical expectations for freezing rain occurrence (e.g.., warmer air temperatures above the surface than at the surface, near or subfreezing temperatures at the surface) and wind gusts (e.g. higher mean wind speeds at 850 hPa and below). However, although the predictors are physically interpretable, ANNs improve on the results from stepwise regression, and both the spatial patterns and event timing of FZG events are relatively well captured, significant biases remain. The intensity of the events is underestimated. Both peak ice accumulation and wind gusts tend to be lower than observed, leading to underestimated risk categories.