CHAT: The Colorado Hail Accumulation from Thunderstorms Project

Deep hail accumulations, sometimes up to 50 cm in depth, have occurred frequently enough to catch the attention of the National Weather Service (NWS), the general public, and social/digital media outlets. Motor vehicle accidents, road closures, airport delays, flooding, and swift-water rescues have all resulted from hail accumulations on the ground. 

Despite the extreme nature of these storms, adequate reports or measurements of accumulated hail depth are currently not collected or archived, and products to track or forecast these events do not exist, preventing guidance from being issued to emergency responders, transportation departments, and the general public.

To better identify and forecast hail accumulations from thunderstorms, forecasters from the NWS Forecast Office (NWSFO) in Boulder, Colorado, in collaboration with researchers from the University of Colorado Boulder, started the Colorado Hail Accumulation from Thunderstorms (CHAT) project in 2016, which aims to collect hail accumulation reports and study the behavior of hail-producing thunderstorms with dual-polarization weather radars and a lightning mapping array. The CHAT project has four main objectives: 

1. Building a database of reported hail depths, median hail sizes, and hail swath extent; 
2. Studying typical characteristics of thunderstorms that produce significant hail accumulations on the ground; 
3. Developing techniques to identify thunderstorms with hail accumulations on the ground using operational weather radar and lightning networks; and 
4. Developing techniques to nowcast and predict hail accumulation potential on the ground. 

So far, the CHAT project has focused primarily on storms occurring in eastern Colorado and southeast Wyoming using data from the NWS dual-polarization radar network [Weather Surveillance Radar-1988 Doppler (WSR-88D)] that became available in 2012 and the Colorado Lightning Mapping Array (COLMA) that was installed in northeastern Colorado in the spring of 2012 (Rison et al. 2012). Once results are more robust and algorithms are tested, the project can be expanded to the national level.

Identifying the conditions for hail accumulation

Previous research has focused on hail formation, growth, and decay as well as environmental conditions favoring hail production in convection. But this work has mainly analyzed processes that lead to the growth of hailstones rather than depth of accumulation. 

Maximum hail production may occur when hailstones travel through optimal growth environments within thunderstorms. That is, regions between −10° and −25°C with appropriate updraft strength and width, as well as sufficient supercooled water. Additionally, the research shows that the availability of hail embryos of appropriate size and concentration in locations where they can be advected into the hail growth zones is also important. Storm environmental conditions (e.g., vertical wind shear, buoyancy, vertical profile of humidity, aerosol concentration) have also been identified as important factors for changing storm structure, dynamics, and microphysics, specifically, hail formation and growth.

Much of the research on identifying and forecasting the growth of large hailstones has been implemented into algorithms and procedures used by the NWS. Currently, the NWS issues a severe thunderstorm warning when a thunderstorm is expected to produce hail ≥2.5 cm (1 in.) in diameter. Over the last decade, NWS has sought to increase the number of maximum hail size reports through social media, NWS storm reports (also referred to as Storm Data), or multiyear field campaigns. It has also evaluated the quality of these reports. Even so, reliable and detailed reports of accumulated hail depth, median hail size, and hail swath extent remain rare.

Kalina et al. (2016) performed one of the first comprehensive studies that analyzed synoptic conditions and radar and lighting signatures of four thunderstorms along the Front Range, each with >15 cm of hail accumulation at the surface. Though these events were associated with slow storm motion (6–9 m s–1), the radar and lightning signatures of these hail events were not substantively different from what has been observed in other severe hailstorms without hail accumulations. Nevertheless, Kalina et al. (2016) found that hail accumulations are associated with large hail production or presence in the cloud and slow storm propagation speeds or a combination of these. 

Continuing the work by Kalina et al. (2016), Wallace et al. (2019) used 20 reliable hail depth reports along the Front Range to refine a radar-based hail accumulation algorithm that was the basis of the Kalina et al. (2016) study. Wallace et al. (2019) validated this revised algorithm with a larger dataset of 32 thunderstorms and showed that the ratio between reported and radar-based hail accumulations at the time and location of the report ranged between 0.6 and 1.5 for 80% of the reports where >3 cm of hail accumulations was observed on the ground. Other NWSFOs such as Amarillo, Texas, have also started to use dual-polarization weather radar information to identify thunderstorms with hail accumulations (Ward et al. 2018).

Building a hail depth database

For the period of study, 2012–17, we collected hail depth information from storm reports compiled by the NWS (Storm Data) and the Community Collaborative Rain, Hail and Snow (CoCoRaHS; see www.cocorahs.org/; Reges et al. 2016) network or reported in newspapers and by broadcast media. Since hail depth is currently not required for hail reporting, we asked amateur meteorologists and storm spotters to contribute texts, photos, video, and drone footage of hail depth, hail size distribution, and hail swath extent using Facebook, Twitter, telephone, or e-mail. 

A total of 91 hail depth reports were collected from 60 thunderstorms in the study area from 2012 through 2017 (Fig. 3); 64% of the reports were from 2016. So far, we have analyzed 32 storms (52 reports) that occurred within the COLMA and the range of the operational dual-polarization radars in Pueblo, Colorado; Denver; and Cheyenne, Wyoming, and that passed our quality control criteria. For a report to be included in the analysis, it had to meet the following requirements: 

1. The report had to include the precise location of the event 
2. The location had to be within 150 km of a Next Generation Weather Radar (NEXRAD) 
3. Precipitation had to be detected by the radar at the location of the report between the time of the report and 1 h prior
4. If the report was transmitted via social media, it had to be accompanied by a picture to verify the reported hail depth. 

Fig. 3 Archived reports of hail accumulations from thunderstorms along the Colorado Front Range between 2012 and 2017 (color coded). The number of reports for each year is listed in parentheses. Sources include CoCoRaHS reports, NWS storm reports (Storm Data), Twitter, Facebook, news outlets, and trained spotters. 

The quality of the reports varies greatly depending on the source. Unfortunately, out of the 60 thunderstorms (91 reports), 28 storms (59 reports) could not be analyzed because they did not satisfy the quality control criteria. Of the 32 analyzed thunderstorms, 14 had traces of hail or accumulations under 3 cm, 9 had accumulations between 3 and 10 cm (moderate), and 9 had more than 10 cm of hail accumulation (deep).

Which thunderstorms produce hail accumulations on the ground?

Though hail depth reports are crucial in determining which thunderstorms produce moderate-to-deep hail accumulations, more information is needed. To remedy this, Kalina et al. (2016) used radar reflectivity and a radar-based hydrometeor classification to estimate surface hail accumulations. Wallace et al. (2019) improved upon this by including information on maximum hail size from the radar-based maximum estimated size of hail (MESH) algorithm to derive maximum fall velocity using the diameter–fall velocity relationship for rimed particles from Heymsfield and Wright (2014). Validating this revised algorithm against 20 high-quality hail depth reports resulted in a correlation coefficient between radar-based and reported hail accumulations of 0.88, an improvement from the value of 0.69 obtained by Kalina et al. (see Fig. 8 in Wallace et al. 2019).

Two examples of radar-based hail accumulations using the validated algorithm in Wallace et al. (2019) are shown in Fig. 4. One example shows a series of multicell thunderstorms that occurred on 28–29 June 2016, which started to accumulate hail ∼20 km northwest of Denver, moving southeast at a speed of about 12 meters per second (Fig. 4a). We received three hail depth reports on that day in or close to the areas of deepest accumulations indicated by the radar. Differences between reported and radar-derived accumulations ranged between 0.6 and 3.8 cm around Denver and 5.7 cm at Arvada, Colorado. The second example shows a supercell thunderstorm first observed about 30 km east of Cheyenne. It moved east at 8 meters per second during the accumulation period. A 7-cm accumulation was reported along Interstate Highway 80 (I-80) close to Pines Bluff, Wyoming, while radar indicated a hail accumulation of about 11 cm. From Fig. 4, the ratios between the reported and radar-based hail accumulations are 0.49, 0.35, 0.87, and 0.64. This compares fairly well with the range of ratios from 0.6 to 1.5, quoted earlier from Wallace et al. (2019), for 80% of 32 hail depth reports.

Fig. 4. Radar-based total hail accumulations (a) between 2300 UTC 28 Jun and 0100 UTC 29 Jun and (b) between 2230 and 2359 UTC 27 Jul 2016 with report locations indicated by red arrows and small black squares. Radar data from WSR-88Ds at (a) Denver (KFTG) and (b) Cheyenne (KCYS) were used for this analysis. Hail accumulations between 1 and <3 cm are outlined by the magenta contours. Reported hail depths are listed with radar-based accumulations in parentheses. 

This allows us to analyze the temporal and spatial evolution of lightning and radar variables using the validated radar-based hail accumulations along the Colorado and southern Wyoming Front Range. Radar-based hail accumulations are also calculated in real time during the convective seasons for eastern Colorado, near Rapid City, South Dakota, and Amarillo (http://clouds.colorado.edu/Real-timeHailMaps). This preliminary nowcasting product is currently tested by the Boulder NWSFO and results are used for further research. We anticipate further validation of the radar-based hail depth algorithm as we receive more reports but also would like to test the algorithm in other areas first. Thus, we wish to solicit hail depth reports across the entire United States. For more information on how to submit reports, visit our website (http://clouds.colorado.edu/deephail) or reach us (@DeepHailCO) or our local weather forecast office on Twitter (e.g., @NWSBoulder, #deephail).

We also track several additional lightning variables associated with hailstorms. These include lightning flash rate and flash extent density. The former refers to the number of flash initiation points in over an area of 1 km × 1 km, and the latter is the number of flashes that cross a vertical column with a cross section of 1 km² in 1 min. These variables are both derived from Lightning Mapping Array measurements and are linked to storm updraft strength, updraft volume, and graupel mass. Numerous studies have shown that increases in lightning flash rate precede hailfall by 5–20 min. For the two examples shown in Fig. 4, enhanced flash extent density was observed in the vicinity of the deepest accumulations (Figs. 5a,b). On 28–29 June, east of Denver, flash extent density peaked at 2.5 flashes per km² per minute. And on 27 July, a maximum of 3.5–4.0 flashes per km² per minute was observed over the area of maximum observed hail accumulation, east and south of Pines Bluff. For flash extent density, we found that the changes are typically more important than the specific values for determining hail potential.

Fig. 5 Radar-based (top) total hail accumulation (color coded) as in Figs. 4a and 4b overlaid with flash extent density and (bottom) VII. All variables are accumulated (a),(c) between 2300 UTC 28 Jun and 0100 UTC 29 Jun and (b),(d) between 2230 and 2345 UTC 27 Jul 2016. Red or white arrows point to reports. Radar data from WSR-88Ds at (a),(c) Denver and (b),(d) Cheyenne were used for this analysis. In (a) and (b), the flash extent density contours start at 1 flash km–2 min–1. The contour interval is 0.5 flashes km–2 min–1. Enhanced areas of flash extent density are enclosed by thick black contours at 2.5 km–2 min–1 in (a) and (b).

As part of the real-time hail accumulation maps, we track vertically integrated ice (VII; Figs. 5c,d), which integrates radar reflectivity >35 dBZ at altitudes where the temperature ranges from −10° to −40°C and converts it into VII following the method described in Carey and Rutledge (2000), Gauthier et al. (2006), and Mosier et al. (2011). Enhanced VII is often observed upstream or in the area of moderate-to-deep hail accumulations. Figure 5 bears this out. 

After analyzing the radar and lightning variables for all 32 thunderstorms, vertically integrated ice stood out. A maximum in VII has been consistently observed 5–20 min prior to the first maximum of the radar-based accumulation rate associated with hailfall, while the temporal evolution of other lightning and radar variables varies from case to case. Because VII and hail accumulations were both derived from radar data, the variables are not necessarily independent. Many questions remain to be answered regarding lightning activity and hail accumulation potential. Automating the computation of these variables throughout the lifetime of the storm and comparing them with the two-dimensional maps will be a first step toward a nowcasting algorithm for better estimating the time and location of deep hailfall.

The goal of future research is to improve our basic knowledge about the evolution of radar and lightning characteristics of thunderstorms producing copious hail. We are working to include more cases from Colorado and elsewhere to provide more robust statistics and results that can be implemented into a nowcasting algorithm. We plan to investigate the role of terrain-induced boundaries and thunderstorm outflow boundaries on the rapid intensification of thunderstorms as well as the effect of melting on hailstones below the freezing level. Surface boundary interactions affected several thunderstorms in our dataset and may have influenced hail accumulations.

Finally, we want to take advantage of new measurement technologies on board Geostationary Operational Environmental Satellite-16 (GOES-16). Images at 30- and 60-s intervals and total lightning data will be a boon, especially in data-sparse areas, for revealing storm-scale boundaries, circulations, and the locations of hail swaths in real time. However, the ultimate goal is to predict hail accumulations from thunderstorms either through a nowcasting system or with numerical weather prediction models.

This article has been edited and adapted specifically for the AMS Weather Band. Any errors or omissions should be attributed to AMS Staff.