How is freezing rain changing in a warming world?

This was the overarching question being addressed in our NSF-funded research, conducted between 2021-5 and entitled “Thermodynamic Modification of Mixed-Phase Winter Storms in a Warmer Climate” (AGS-211-0243). This project had the following goals:

(1) Extracting information on freezing rain from the European Center’s ERA5 reanalysis data, and building a long-term (40 + year) events database

(2) Using the above reanalysis data, and machine learning, to explore characteristic large-scale weather systems that instigate freezing rain across Eastern North America

(3) Identify past and future spatial and temporal statistics of winter precipitation and associated storm types at regional scales

(4) Identify storm-specific changes in the timing, distribution, phase partitioning, and intensity of winter precipitation phase types in the future climate, using an existing dataset that simulates past events in a warmer late 21st century climate (so-called ‘pseudo global warming‘ technique).

Our study domain and sub-regions

What is Freezing rain, and where does it tend to occur?

Click on the link below to read about freezing rain (By Kara Rydberg, Undergraduate intern).

Introduction to Freezing Rain

When freezing rain accumulates to 0.25 inches in a single event, then this is defined by the National Weather service as an Ice Storm. These are the events we are mainly concerned about in this work. Since the year 2000, winter storms in the United States have contributed between over $10 billion in damage. Even in the southern U.S., a region where such events are less frequent, the southern plains states experienced nearly $1 billion in damage, and sustained power outages for days to weeks following several high-impact ice storms in the decade 2000-2010^[1]. For a winter storm to produce freezing precipitation, particularly to the extent as to be classified as an ‘ice storm’, requires a complex and precise series of thermal, dynamic, and moisture ingredients^[2-5]. Perhaps because of this, and the relative lack of long-duration high-spatial and temporal resolution observations^[6-7], freezing precipitation is generally understudied with respect to both rain and snow, which is one of the reasons we focused on it in this project, now that we have datasets that can resolve freezing rain and its variations in space and time.

Our research activities, and key results

1. EVENT DATABASE and WEATHER PATTERN ANALYSIS

The data, methods, and key results for this aspect of the work are reported in peer-reviewed work; Britton and Mullens (2025).

Event Database:

The database consists of freezing rain events from 1979-2020, identified using precipitation type flag in the ERA5 reanalysis dataset, and compiled using GIS tools. We located all grid points in which freezing rain was flagged by the model, integrating additional surface-based metrics into the database, including temperature, precipitation rates, and temperature. Following this, data was grouped into specific events Each data point was compared and checked for any neighbors in direction and time. These neighbors were marked with a cluster number. This dataset can be employed using GIS to visualize freezing rain event ‘swaths’ in time, as shown below.

Archetypal Weather Patterns

The technique used to identify these large-scale weather patterns is called ‘Self Organizing Maps’. It is a univariate or multivariate ‘deep learning’ statistical method which maps multidimensional data onto lower-dimensional spaces that capture fundamental patterns in the data. In everyday terms, imagine that you have in front of you dozens of photographs of a certain set of variables (e.g., pressure, temperature) in the atmosphere at the time that a freezing rain event was just beginning. How would you cluster theses together to come up with a series of common patterns? This is what the SOM technique does. It provides us with distinct atmospheric states that are physically realistic and allows us to group all events into X number of distinct archetypes. In our case, we selected 16 initial archetypes, which was further reduced to 4 using an additional clustering process (known as ‘K-means clustering’).

A general public summary of our results can be found in the linked document below. This summarizes the whole domain patterns (i.e., those produced from considering freezing rain events over the entire eastern and central North America.

Graphics displaying regional weather regimes and meteorological characteristics are provided in the document below for each of the seven subdomains shown above.

Key Outcome

The weather regimes reveal overarching patterns linked to freezing rain, and where those patterns tend to produce freezing rain, and we have computed these for the full domain space, and for freezing rain events in each subdomain. Therefore, we can get the ‘big picture’ view for the whole region, and a more specific, granular view for each subdomain. Therefore, these regimes can be used by forecasters for situational awareness regarding freezing rain weather patterns and how past events within regimes evolve to aid future prediction. Moreover, we have been able to characterize some meteorological features of each regime – such as distributions of temperature, wind speed, ice accretion, sea level pressure, and duration.

2. 80-YEAR CLIMATOLOGY AND TRENDS/VARIABILITY

The data, methods, and key results for this aspect of the work are reported in peer-reviewed work by Mullens (2025). A conference presentation is also available to view: Mullens and Bonilla (2024).

The available ERA5 data, and the fact that it provided a reasonable reproduction of monthly and annual statistics of freezing rain (shown in the work linked above), allowed us to pursue questions related to why freezing rain trends in the past several decades have not been declining in a warming climate, and show very high inter-annual variability. We hypothesized that the role of natural variability is primarily responsible for the lack of trends, and the variability, but sought to determine which modes of variability that influence the North American region, had the greatest connection to freezing rain occurrence.

A few experiments were conducted to elucidate the role of various modes of natural variability. We selected modes based on wavelet analysis of important periodicities in each subdomain’s time series (using monthly data), and evaluation of the literature. These included The El Nino Southern Oscillation (Oceanic Nino index), Pacific North American pattern (PNA), East Pacific North Pacific pattern (EP-NP), North Pacific Index (NPI), Arctic Oscillation (AO), Pacific Decadal Oscillation (PDO), and the Atlantic Multidecadal Oscillation (AMO). Data sources and the specific statistical experiments used are described in the above works.

Key findings included (a) PNA continued to be the most important mode for many sub-domains, followed by year (indicative of interannual dynamical variations), while other modes varied more widely in importance across subdomains; (b) there was often a split between the U.S-based domains (south of 49N), and the Canadian domains. For example, the role of the EP-NP index suggested that many domains had more freezing rain when this mode was in its positive phase, whereas the southeastern Canada region showed the opposite finding. Another interesting finding was that most domains show higher freezing rain occurrence during a positive Arctic Oscillation, whereas cold air outbreaks are typically associated with a negative AO phase. We suggest that this result reflects the need to pay attention to vertical temperature structure during a cold air outbreak. Those that most commonly produce freezing rain tend to have a northeastward axis of cold air to warm air, with rising air being promoted by a western upper-level trough and eastern ridge over the continent. If the southeastern ridge is weak or absent, then cold air tends to take on another orientation with warm air to the southwest and cold air to the north and east, which is less conductive to freezing rain. Furthermore, we identified through regression of synoptic parameters onto the freezing rain time series that many subdomains showed a strong Arctic/Alaskan ridge pattern^[8], which is often seen during a negative PNA, positive EP-NP, and positive AO. This ridging allows cold air to funnel southeastwards from western Canada into the Plains states, creating the optimal orientation of the thermal boundary. The only regions in which this pattern is largely absent, are the Northern Plains, south central and southeastern Canada, helping to explain how these regions often show different relationships to the modes of natural variability used. We can see these results reflected in the types of weather systems that result in freezing rain in each subdomain from our previous regimes analysis.

Using these results, can we predict whether a given winter/cool season will have more or less freezing rain? Our climatological analysis indicates that there are potentially predictable features on monthly and seasonal scales. We identify that a negative PNA pattern has a particularly strong connection to an increase in freezing rain activity in the southern and eastern U.S subdomains. Using random-forest regression (a machine-learning technique) we can statistically model freezing rain occurrence to a high degree of fidelity when using all available data, however to this point, the intent of the work was to explore physical relationships with predictors, and so there was no temporal lag applied to the data (i.e., lead time between predictors and resulting freezing rain), nor was there distinct training and testing datasets. Relationships with natural variability vary by subdomain and are also non-linear. Our key finding, however, was that months dominated by anomalous Alaskan/North Pacific Ridging were linked to higher freezing rain activity in many subdomains in the U.S.. However, Canadian regions showed weaker connections to natural variability.

3. FREEZING RAIN EVENTS IN A LATE 21ST CENTURY CLIMATE

The output for this section of the project is discussed in Britton, A., and E.D Mullens, 2025. A Weather Regime-Based Analysis of Freezing Rain Trends in a Warmer Climate for Eastern North America. Int. J. Clim. https://doi.org/10.1002/joc.70233

This activity uses the CONUS1 dataset by Liu et al. (2017), which includes 13-year time slices of a control (CTRL) climate, or the actual conditions between October 2000 and September 2013, and a perturbed global warming (PGW) scenario, which takes the mean temperature change projection for a late 21st century climate from the coupled model inter-comparison project version 5 models, and re-runs this 13-year time slice with the modified thermal conditions. We therefore have simulations of actual weather systems that occurred during this time (CTRL), and those same actual weather systems, in the sequence they occurred, but now modified to exist as if they were in a late 21st century climate. 

We use a freezing rain algorithm by Bourgouin (2000) to extract freezing rain from the CONUS1 data for all identified events from our previous database. We then explore various facets of these events in a warmer climate, and separating them by regime type (using our four full area regimes). By conducting a short validation between CONUS1 and ERA5, we found that freezing rain events were well simulated by the former in time and space.

Key Outcomes

Freezing rain shifts northward in a warming climate – but generally does not decrease in amount as much as snowfall does.

The typical freezing rain line shift was estimated as 258 km for all events, contrasted with a northward push for snow of around 108 km, although the latter showed a notable westward shift of nearly 100 km when averaged across all events. There was large spread in these values between individual events. in many cases, freezing rain events show some decline in total amounts in a warmer climate, but that the snow phase declines far outweigh those of freezing rain. When calculating the mean percentage change in freezing rain LWE across all 82 events, we find a -1.7% mean decrease in amount, but with snow, this rises to -9.7%. This agrees with visual interpretation of the figure above, where all domain amounts for snow show a substantial decrease (nearly 6x that of freezing rain), whereas freezing rain shows more of a geographical shift. Amongst the regimes, R3 shows the largest decline in amounts, with ice accretion declining by nearly 30%. R1-3 all show lower amounts of total freezing rain in a warmer climate. However, R4 shows higher amounts, and an accretion increase of nearly 52%.

Sub-regions show different magnitudes of change

The largest decline in freezing rain accretion is within the Midwest region, whereas southeastern Canada and other northern domains shows an increase in freezing rain, at the notable expense of snowfall. Cold and warm layer energies are consistent with the location shifts from past to future. Both layers move northward, and the warm layer intensifies whereas the cool near-surface layer tends to shrink, except for the southeastern Canada domain. We might have expected that the southernmost domain would show the largest decreases in freezing rain and snow, but this is not the case. We hypothesize that the warming of the Great Lakes, which sits at the northern extent of the Midwest may limit freezing rain formation. On the other hand, the southeast events are dominated by potent cold air outbreaks, and driven by a blocking high or Alaskan ridge pattern. Outside of this pattern, freezing rain events can occur due to cold air damming east of the Appalachians when an anticyclone is present to the north, and often under the approach of a Gulf low pressure. While there is a notable decrease in non-damming related freezing rain activity, the topographic influence does promote freezing rain in a warmer climate still – just further north.

What about duration and precipitation rates?

There is little change in total amounts for the northeast and south-central Canada, but a discernable increase for the Northern Plains and southeastern Canada. Freezing precipitation rates show a decrease in some domains, and increases in others. Domains favorable for more freezing rain activity also show higher freezing precipitation rates during events, in general. However, overall there is insignificant changes compared with the historical climate. Likewise for duration, which shows some evidence of increase, particularly for the U.S southeast, but this is also not statistically distinct.

The role of regime

Freezing rain constitutes a much smaller proportion of total precipitation in a winter weather system – typically being confined to a narrow swath close to the surface freezing line, under an environment with a distinct thermal inversion in the lower atmosphere (below freezing near surface and above freezing between 0.5-2km above the ground). As such, the proportional declines in freezing rain are smaller, and often there is a northward shift in the freezing rain axis as opposed to a substantial decrease. However, 67% of all events do show less freezing rain in the future than in the past. The amount of change in freezing rain amounts and occurrence does change with regime, and typically regimes that were cooler and more marginal for freezing rain (Regimes 1 and 4) in the historical climate become optimal for freezing rain in the warmer climate. Regime 4 in particular shows an increase, with 11 of 17 events showing more freezing rain in a warmer climate.

Student Training and Participation

PhD – Austin Britton (2022-6)

Mr. Austin Britton has conducted the bulk of the research for this project, in support of his PhD dissertation, and is lead author on two published manuscripts. He has also presented his work at two American Meteorological Society Annual Meetings (2024, 5).

Undergraduates – AG Cornell and Kara Rydberg (2025)

Two undergraduate students participated in this work from January-May 2025. Kara Rydberg (junior, meteorology) conducted a public-communication work, creating two factsheets on the key messages from the project research. The factsheets include “What is freezing rain” – which includes climatological information derived from this project; “What Atmospheric Patterns result in Freezing Rain in Eastern North America?” – including key messages from paper 1, both of which are available to view above. Anna Grace Cornell (junior, Geography) evaluated ERA5 freezing rain against control simulations in the climate simulation dataset (WRF-CONUS1) and create narratives to explain the findings of the sub-domain self-organizing maps analysis for the seven regions used throughout the project, which is available above.

Pre-College students (2023)

During the second year of this project, two pre-college students joined our research group as part of a Summer Science training program facilitated by the University of Florida’s Center for Precollegiate Education and Training (CPET). Their participation in this program was funded through this project with both students working on elements of the research. The PI and one student worked on evaluating ERA5 against long-term first order and cooperative network station observations compiled by Prof. S Changnon^[9-10]. The second student worked with Austin to develop a framework for multi-variate self-organizing maps (SOM) and applied this to a sample dataset of freezing rain events in the Northeastern US. Both students successfully concluded their summer program with a poster and oral presentation. Their work has helped to tackle aspects of the project work and was utilized further. The students were included as co-authors on presentations at the American Meteorological Society annual meeting in 2024.

Conference Sessions inspired by this work

The PI developed and co-led three sessions from 2023-5 within the AMS Conference on Climate Variability and Change. These sessions, entitled ‘Winter Weather in a Warming World (4W)’ were designed to bring together researchers and professionals who are engaged in activities that seek to advance our understanding in these areas, to provide global and regional insights into changes in winter weather, and to connect these changes with their impacts on society, ecosystems, and infrastructure. Every session was very well attended.

For more information – see the links below

2025 – Winter Weather in a Warming World session 1 and session 2.

2024 – Winter Weather in a Warming World session 1 and session 2

2023 – Winter Weather in a Warming World session 1 and session 2

References

[1] Grout. T, H. Yang, J. Basara, B. Balasundaram, Z. Kong, and T. S. Bukkapatnam, 2012: Significant Winter Weather Events and associated Socioeconomic Impacts (Federal Aid Expenditures) across Oklahoma: 2000–10. Wea. Climate Soc., 4, 48–58.

[2] Britton, A., and E. D. Mullens. 2025: Temporal and Spatial Analysis of Freezing Rain over Eastern North America. J. Appl. Meteor. Clim. https://doi.org/10.1175/JAMC-D-240173.1

[3] Changnon, S. A., & T. R. Karl, 2003: Temporal and spatial variations of freezing rain in the contiguous United States, Journal of Applied Meteorology, 42(9), 1302-1316.

[4] Cortinas Jr., John V., Ben C. Bernstein, Christopher C. Robbins, J. Walter Strapp, 2004: An Analysis of Freezing Rain, Freezing Drizzle, and Ice Pellets across the United States and Canada: 1976–90. Wea. Forecasting, 19, 377–390

[5] McCray, C. D., E. H. Atallah, and J. R. Gyakum, 2019: Long-Duration Freezing Rain Events over North America: Regional Climatology and Thermodynamic Evolution. Wea. Forecasting, 34, 665–681, https://doi.org/10.1175/WAF-D-18-0154.1

[6] Mullens, E. D., L. M. Leslie, and P. J. Lamb, 2016: Impacts of Gulf of Mexico SST Anomalies on Southern Plains Freezing Precipitation: ARW Sensitivity Study of the 28–30 January 2010 Winter Storm. J. Appl. Meteor. Climatol., 55, 119–143, https://doi.org/10.1175/JAMC-D-14-0289.1.

[7] Mullens, E. D., and R. McPherson, 2017: A Multialgorithm Reanalysis-Based Freezing-Precipitation Dataset for Climate Studies in the South-Central United States. J. Appl. Meteor. Climatol., 56, 495–517, https://doi.org/10.1175/JAMC-D-16-0180.1.

Liu, C., Ikeda, K., Rasmussen, R., Barlage, M., Newman, A. J., Prein, A. F., Chen, F., Chen, L., Clark, M., Dai, A., Dudhia, J., Eidhammer, T., Gochis, D., Gutmann, E., Kurkute, S., Li, Y., Thompson, G., & Yates, D. (2017). Continental-scale convection-permitting modeling of the current and future climate of North America. Climate Dynamics, 49(1), 71–95. https://doi.org/10.1007/s00382-016-3327-9

Bourgouin, P., 2000: A method to determine precipitation types. Wea. Forecasting. 15, 583-592