Midwest/Northeast Severe Cold Events - Climatology, Causality, and Predictability

SPHEAR partners login here for the Shared Research Portal

Participating Partners: Chesapeake Energy, Citadel Investment Group, Susquehanna International Group
Lead Researcher: Alexander (Sasha) Gershunov
Project Coordinator: Stephen Bennett

Executive Summary

Extreme cold spikes in wintertime temperature spike demand for heating and natural gas. Short-term temperature extremes, both hot and cold, are highly sensitive to climate time scales as climate variability and change effect both the mean and variance structure of daily temperatures as they evolve over a season. We consider wintertime cold snaps over a large region of the Northeastern and Midwestern United States according to how local cold temperature thresholds (5th percentiles of local wintertime temperature recorded at each of almost 500 stations) are exceeded on daily timescales. A regional magnitude index reflecting the temperature intensity, duration and spatial extent of extreme cold spells is computed for 61 winters from 1948/9 to 2008/9 and for each day of each event. Observed variability of regional cold spells is then examined on timescales ranging from daily to interdecadal and scrutinized with respect to the climatic controls on their synoptic causes.  Relationships with known climate modes (ENSO, NAO, PDO, PSV, etc.) as well as other relevant objectively derived circulation and land-surface patterns are then used to develop sophisticated models and simple rule-of-thumb techniques for seasonal and improved medium-range probabilistic prediction of cold snap magnitude.  The main components of cold spells, i.e. intensity, duration and spatial extent, are explicitly considered. These forecasting tools are designed for straightforward operational application by practicing meteorologists working in energy load forecasting.

Project Summary

We first define regional cold temperature extreme magnitude in a way that reflects the events’ main components related to energy demand: intensity, duration and spatial extent. We suggest a definition based on local threshold exceedance, as Gershunov et al. (2009) used to define heat waves. For example, at a set of pre-selected weather stations with long and high quality daily temperature records (Tmax and Tmin) adequately representing a region, cold wave magnitude can be defined as follows.

Local and daily (or nightly) magnitudes are the Tmax (or Tmin) threshold exceedances (in a negative sense for cold waves) recorded at a specific station on a specific day or night, given a minimum duration. These can be aggregated (summed) over space and time into seasonal regional magnitude; daily regional magnitude summed over all n stations representing the region on a particular day or night; local cumulative cold wave magnitude at a single station summed over the duration of an event. Regional duration can be defined as the number of consecutive days or nights when local thresholds are exceeded. Spatial extent is defined as the percentage of representative stations where local thresholds are exceeded. Peak seasonal (or event) magnitude and spatial extent are defined as the maximum daily value over a season or over the duration of a particular cold wave. Regional duration, as well as spatial extent and magnitude, certainly depend on local duration, but this dependence is complicated by the synoptic dynamics of individual events that we propose to examine having defined the events of interest.

This definition is similar to the commonly used “degree days”, but we suggest using locally specific thresholds, defined, for example as the 1st of 5th percentile of the climatological seasonal (e.g. wintertime) distribution of daily Tmax or Tmin observed over a reference period (e.g. 1950 – 1999) at a specific station. This will filter out locally-specific climate information (that can always be added in later) and provide a clean focus on the regional nature of extreme events. We furthermore propose to examine Tmax and Tmin separately, to better understand the behavior of cold outbreaks during day and night.

Having defined the regional magnitude and salient components of cold outbreaks, we describe their observed variability over at least the last six decades in the context of large-scale climate features, e.g. sea level pressure, geopotential height and temperature fields, as well as outgoing longwave radiation and global angular momentum. This investigation can be performed at various lead times via canonical correlation analysis (CCA) as done for precipitation by Gershunov and Cayan (2003) and for summertime temperatures by Alfaro et al. (2006). The main benefits of a CCA-type approach include the (1) diagnostic analyses whereby, without having to make limiting a priori assumptions, the data are allowed to point out the relevant climate influences or modes such as ENSO, NAO, PDO, and/or others, even if yet unnamed; and (2) an objective framework for building these influences, including long-term trends if any, into prognostic models. Such statistical prognostic models are comprehensive and comprehensible (as opposed to various “black boxes” in use) in the sense that predictability can be tracked to specific physical influences; they are computationally reasonable so that optimal model complexity, lead time and prediction skill can be accurately assessed via cross-validation techniques (e.g. Gershunov and Cayan 2003, Alfaro et al. 2006). Such models can be used to examine and forecast any feature of regional cold outbreaks, aggregated seasonally for seasonal-to-interannual prediction.

We also examine the synoptic causes and precursors to individual cold extremes. Cold outbreaks are typically caused by extensive anticyclonically rotating polar masses of cold dense air whose trajectories over land are steered by topography. These anticyclones interact in complex ways with storm tracks and with cyclonic activity all in the context of large-scale climate variability and teleconnections (Favre and Gershunov 2006, 2008). Their behavior is certainly regionally specific, and statistical models similar those described above for seasonal-to-interannual prediction can and will also be used to enhance medium-range weather forecasts of individual extremes by utilizing all relevant antecedent weather and contemporary climate information.

In summary: In this pilot project, we (1) use a comprehensive definition to examine the variability of regional cold extremes; (2) use powerful statistical tools to investigate causality and demonstrate skillful seasonal-to-interannual predictability of the seasonal probability and other features of regional cold outbreaks; (3) examine synoptic causes and precursors of individual regional cold events; and (4) touse this information to improve the lead-time and skill of their extended range weather forecasts by tapping into relevant precursor weather information and conditioning the weather forecast on contemporaneous large-scale climate information.