1. Introduction
Accurately predicting the onset of spring has long been a source of speculation in human culture and continues today, primarily for agricultural purposes. Having both cue and clue of when snow or rains will recede to open fields for seeding can mean the difference between a boom or bust for farmers and the societies and economies they support. In lieu of highly complex weather models and plant phenology, Indigenous and settler cultures across the globe often used faunal biological indicators as oracles of the onset of spring (Turner and Clifton 2009; Pareek and Trivedi 2011; Risiro et al. 2012). For example, clams are said to predict inclement weather by producing more bubbles than usual, and cows licking their forefeet are one of 11 different bovine-related signs of rain (Wallisch 1999). Similarly, it has been observed that sugar gliders induce torpor in response to inclement weather (Nowack et al. 2015). Despite this diversity of weather pattern prophets, the groundhog Marmota monax appears to have claimed the throne as the appointed predictor of spring onset in the United States and Canada.
For over 100 years, a growing collection of North American villages, towns, and cities have used groundhogs to predict the early or late arrival of spring. The tradition states that if, on removal from its burrow, a groundhog “sees its shadow,” spring will come late; alternatively, if the groundhog does not see its shadow, spring will come early. Groundhog Day is an exclusively North American holiday, introduced by European immigrants to Pennsylvania as a variant of an ancient Christian celebration of Candlemas, which marked the midway point between the winter solstice and the spring equinox (Yoder 2003). The original Candlemas tradition portended 40 more days of winter if the weather on this midway point was sunny. On the basis, presumably, of an apparent overabundance of groundhogs in the region at the time, these newly minted Pennsylvanians altered the traditional celebration by including the use of this large rodent and the shadow it casts on the assigned day as the harbinger of spring. The first celebrated instance of Groundhog Day was in Punxsutawney, Pennsylvania, in 1887. In regions where this holiday has been celebrated for decades, these weather-predicting groundhogs have become well known celebrities, expanding beyond the original Punxsutawney Phil (Pennsylvania) to include other notable members of the genus Marmota such as Wiarton Willie (Ontario, Canada), Buckeye Chuck (Ohio), and Shubanacadie Sam (Nova Scotia, Canada; Yoder 2003).
Notwithstanding, the accuracy of groundhog forecasts has rarely been questioned and has never once been comprehensively assessed, even with the multitude of groundhogs now purportedly forecasting weather across North America. Possibly because of his big-hog status (and long-term data record), Punxsutawney Phil has had his predictions most heavily evaluated. Using daily minimum and maximum temperatures and snowfall depth from the nearest weather station, Aaron et al. (2001) analyzed the accuracy of Phil’s forecasting abilities from 1950 until 1999. For 6 weeks following 2 February, if the average daily minimum and maximum temperatures were significantly greater than, and the average daily snowfall was significantly less than, those for the 6 weeks before 2 February, the authors considered it to be an early spring (Aaron et al. 2001). With this method, the authors found that Phil made accurate predictions 70% of the time (Aaron et al. 2001). Other sources, which use different and largely ambiguous thresholds of prognosticating success, tell a different tale; the U.S. National Oceanic and Atmospheric Administration (NOAA) indicates that Punxsutawney Phil was correct only 40% of the time (NOAA 2020), and the Old Farmer’s Almanac reports that groundhogs are 50% accurate when forecasting the seasons (Thomas 1997), although no quantitative analyses are provided to substantiate these claims. Some groundhogs like General Beauregard Lee (Jackson, Georgia) self-report accuracy rates of 99% (Yoder 2003).
Predictions by Aaron et al. (2001), which form the most comprehensive assessment of groundhog predictions to date, fall short in two critical ways: 1) predictive success was measured for only a single groundhog, Punxsutawney Phil; and 2) the methods that the authors used do not permit evaluation of groundhogs across latitudinal and longitudinal scales because of climatic variance that groundhog forecasters encounter across their “natural” range (i.e., snow accumulation is unlikely to be a relevant metric at the southern end of groundhog distributions).
As a more objective and broadly applicable approach to evaluating the onset of spring, we propose the use of peak flowering timing in a widely distributed spring-blooming plant. The phenological link between ambient temperature, plant development, and seasons is well established (Piao et al. 2019) and has been used to assess the timing of spring and summer across continental Europe (Menzel et al. 2006). In the current study, we chose to use the climate-predicted phenology of one of North America’s earliest spring ephemerals, the Carolina spring beauty, or Claytonia caroliniana, as a phenological proxy of the arrival of spring and therefore as a means of evaluating the predictive capacity of groundhogs. The geographic distribution of C. caroliniana overlaps substantially with the distribution of groundhogs, including those involved in predicting the onset of spring (Fig. 1).
Native habitat distribution of groundhogs (brown shading) as well as the spring ephemeral Carolina spring beauty (green shading) in North America. Groundhog and Carolina spring beauty co-occurrence is indicated in yellow shading, and gray triangles indicate the locations of prognosticating groundhogs.
Citation: Weather, Climate, and Society 13, 3; 10.1175/WCAS-D-20-0171.1
Our goal for this study was to conduct a phenological assessment that would provide a definitive and comprehensive assessment of the spring prognostication abilities of groundhogs distributed across North America using currently available data. Our specific objectives were 1) to evaluate whether groundhog predictions were different from random chance; 2) to determine whether latitude or country of origin influenced prediction success rate; and 3) to determine whether, for the groundhogs with sufficient long-term records, there was any evidence that predictions of late springs were more common under climate warming currently observed in both North America and globally since the middle of the last century (Pachauri et al. 2014).
2. Methods
a. Natural history of groundhogs and Carolina spring beauty
Groundhogs (Marmota monax) are among North America’s largest rodents (2–5 kg) and are found broadly across Canada and throughout the northeastern and southeastern United States (Burt and Grossenheider 1976). They are solitary burrowing mammals whose home ranges consist of open areas such as fields, clearings, open forests, or rocky slopes. Similar to other hibernating squirrels, groundhogs undergo bouts of torpor through winter months where they lower their heart rate and body temperature to conserve energy (Ruf and Geiser 2015). Whereas Groundhog Day occurs annually on 2 February, groundhogs typically emerge from hibernation and their burrows in March, or even later in northern latitudes, where mating is often the first order of business (Zervanos et al. 2010). The onset of hibernation in groundhogs is caused by a combination of cool temperatures and lack of food availability (Davis 1967), and emergence from burrows appears to be related to circadian rhythms and aboveground temperature cues (Davis 1977).
The Carolina spring beauty (C. caroliniana) is among the earliest spring ephemerals in North America (Lehmberg et al. 2020, manuscript submitted to Can. Field Nat.). It is widely distributed across eastern North America, from the southern states of Alabama and Georgia through to the northern Canadian provinces of Ontario, Quebec, and the Maritimes (Miller and Chambers 2006), highly similar to the distribution of groundhogs (Fig. 1). The required annual cumulative growing degree-days (beginning 1 January) above 0°C (GDD0) value for C. caroliniana to bloom was recently calculated and reported to be 129° ± 17°C (Lehmberg et al. 2020, manuscript submitted to Can. Field Nat.).
b. Groundhog data acquisition
A list of historical and current prognosticating groundhogs was compiled through a comprehensive internet search. In total, n = 45 groundhogs from Canada and the United States were identified. Of these 45 groundhogs, the final list was pruned to n = 33 groundhogs (Table 1) because our criteria were 1) the animal was a true groundhog (M. monax) and not another member of the Sciuridae family; 2) the animal was a living, breathing groundhog and not a taxidermized groundhog or stuffed puppet; and 3) that predictions were made by a reported visual inspection of a shadow and not by the groundhog “whispering” its prediction to a handler or choosing between labeled food bowls. For all remaining groundhogs in our analysis, annual predictions of “early” or “late” spring for a given groundhog were allegedly recorded by handlers as the presence or absence of a shadow when the groundhog exited its burrow. These records were accessed from publicly available news reports, fan page sites, conservation authorities, email contact with persons running twitter accounts for various groundhogs, and government agencies.
The identity, location, and surrounding human population size during the most recent census for the 33 prognosticating groundhogs involved in this meta-analysis, sorted by latitude.
c. Method for determination of early and late spring
Climate records corresponding to each groundhog under evaluation were accessed using the nearest nationally catalogued weather station to the city, town, or village hosting a particular groundhog. Mean daily temperatures [(maximum temperature + minimum temperature)/2] were obtained and used to estimate GDD0. A minimum of 30 years of climate data were acquired for each groundhog location to ensure a robust calculation of mean date of spring onset; only weather stations with nearly continuous data coverage were chosen, but where necessary multiple weather stations within the same county were used to create a full dataset (see Table A1 in the appendix). Climate data were accessed for Canada and the United States from Environment and Climate Change Canada (ECCC; https://climate-change.canada.ca/climate-data) and the NOAA (https://www.ncdc.noaa.gov/cdo-web) weather repositories, respectively.
Early and late are the only options a forecasting groundhog has at its disposal, so an objective determination of the onset of spring for each groundhog location was required. As North American groundhogs involved in spring predictions span a gradient of 16° of latitude, it was also important that our assessment of spring captured local discrepancies in spring phenology that are based on latitudinal clines. At each groundhog location, the estimated Carolina spring beauty bloom date was calculated (as the nearest day when accumulated GDD0 surpassed 129 degree-days with starting date of 1 January of each year) and was assigned as the onset of spring for each year that a groundhog prediction was made. We then averaged these estimated Carolina spring beauty bloom dates across years to provide a mean onset of spring for each groundhog location. A locally objective early or late measure of spring for each groundhog-year prediction was determined if the estimated onset of spring (based on the estimated Carolina spring beauty bloom date from that year’s GDD0 estimate) occurred before or after this locally specific, time-averaged value. Because the estimated spring bloom date in a given year and location was reported as an integer-based yearday and average estimated bloom date for that location was left as a numeric value, we did not have any cases in which spring estimates ever fell on the estimated average date of spring.
To provide context for potential “error” in our estimated date of spring onset, given that the mean accumulated GDD0 for the estimated bloom date of C. caroliniana includes error of ± 17°C days, this translates into a potential error around our estimated onset of spring of ±1 day at 17°C, which is a mean daily temperature that is not unusual to encounter during spring. However, because this onset date was subsequently turned into a bimodal variable (either early or late), given its distance relative to the mean onset over a minimum 30-yr record, this only introduces uncertainty in those dates within 1 day of the mean spring onset, representing very few (9%) of our observations, and is therefore unlikely to influence our results or conclusions.
d. Data analysis
Each location- and year-specific groundhog prediction was compared with our locally objective binary early or late spring estimate (described above). If a groundhog saw its shadow (predicting a late spring) and spring onset of that year was estimated to occur after the mean spring date for that location (observed late spring, based on the estimated bloom dates of Carolina spring beauty), then that groundhog made a successful prediction. Similarly, if the groundhog failed to see its shadow (predicting an early spring), but spring onset occurred after the mean date for that location (late spring), it would be an incorrect prediction. Similarly, all other possible combinations of groundhog predictions with our objective measure of spring onset were considered to make these determinations.
Logistic regression models were fit to each set of groundhog prediction and spring-onset data to determine the ability of groundhogs to successfully predict the onset of spring. To evaluate whether groundhog guesses collectively were any better than random chance, data from all 33 groundhogs were included in a single logistic regression to evaluate the rate of successful predictions across all groundhog locations. Logistic regression models were also run for each individual groundhog, although predictive power for certain models was low because of low sample sizes.
Additional analyses were performed to understand whether regional or landscape factors influenced the success of spring groundhog predictions. We examined the degree to which the latitude at which groundhogs resided might explain prediction success, and to settle a long-standing debate among the authors that emerged after a long afternoon and evening at the campus pub, a Welch’s two-sample t test was conducted to evaluate the predictive success between Canadian and American groundhogs. Significance in all analyses was assessed at α = 0.05.
With the notoriety, abundance of guesses, and national pride attached to predictions made by Punxsutawney Phil (Team USA; n = 107) and Wiarton Willie (Team Canada; n = 54), we performed an additional comparison for the predictions of these hogs alone. A chi-square test was conducted to evaluate the accuracy of these two hogs predicting the onset of spring. Significance was again assessed as α = 0.05. In addition, data from these groundhogs (being the longest datasets) were used to determine whether groundhog predictions of late springs were increasing in frequency over time (which might be expected under global climate warming).
3. Results and discussion
Using a novel method of assessing the onset of spring across all regions reporting groundhog predictions, our analyses revealed that the likelihood of North American groundhogs correctly predicting the arrival of spring was quite literally no better than a coin toss. Using data among all 33 prognosticating groundhogs, there was no significant relationship between groundhog prediction and the estimated actual onset of spring (logistic regression; P = 0.28), and the overall percentage of correct guesses was 50% (n = 530 across all groundhog location/years). Considering individual groundhogs, not a single groundhog we evaluated had a significant relationship between their predictions and the estimated true onset of spring (logistic regression; P value range = 0.061–1 across all 33 groundhogs).
Both from their relative aptitude—or consistent ineptitude—in predicting the onset of spring, certain groundhogs appeared to be better at portending the coming spring season than others. Essex Ed (Essex, Connecticut), Stonewall Jackson (Wantage, New Jersey), and Chuckles (Manchester, Connecticut) all had greater than a 70% success rate of correctly predicting spring onset (groups with n > 9 observations; Fig. 2). By contrast, Buckeye Chuck (Marion, Ohio), Dunkirk Dave (Dunkirk, New York), and Holland Huckleberry (Holland, Ohio) had similar precision to this first group, but with opposite accuracy, demonstrating correct guesses across all three groundhogs less than 30% of the time (Fig. 2); in other words, it is better to bet against this latter group of groundhogs than with them.
Proportion of correct guesses for each groundhog; Canadian groundhogs are shown in red, and American groundhogs are shown in blue. Data points outside the vertical red dotted lines indicate groundhogs that were reasonably good at predicting the correct timing of spring (>0.75) or reasonably good at predicting the opposite of spring onset (<0.25). The number of observations for which a groundhog offered a prediction of spring are shown within each data point.
Citation: Weather, Climate, and Society 13, 3; 10.1175/WCAS-D-20-0171.1
Whereas the group of Essex Ed, Stonewall Jackson, and Chuckles may represent a farmer’s best bet for knowing the onset of spring, their guesses should not be weighted equally. When observing the mean difference in yeardays between predictions of a late versus early spring, the guesses by both Essex Ed and Chuckles were categorically precise, with the onset of spring in a late prediction year arriving at least 17 days after the average estimated onset of spring in an early predicted year for their respective locations (Fig. 3). However, Stonewall Jackson was a good example of being lucky rather than accurate; despite having predicted spring successfully 73% of the time (Fig. 2), the difference in observed spring onset between his predictions of early and late springs was only a razor thin margin of 0.5 yeardays (Fig. 3).On the far end of the spectrum, it appears best to bet heavily against Winnipeg Willow’s predictions; while Willow prognosticated only 4 times, she was correct just 25% of the time, and on her prediction of a late spring, it actually arrived a full 38 days earlier than the mean onset of spring in the region. It is unfortunately clear that even if certain groundhogs can occasionally predict the onset of spring better than others, there appears to be no clear prophet among the group evaluated here.
The difference between mean observed spring dates for early and late predictions of each groundhog; positive values indicate a closer match between spring onset and groundhog predictions (i.e., spring occurs later on average when they predict late spring as compared with when they predict early spring), negative values indicate an opposite groundhog prediction to observed spring onset (i.e., spring occurs earlier on average when they predict late spring as compared with when they predict early spring), and values at or near zero represent no difference in spring onset between early and late groundhog predictions. Only groundhogs that have prognosticated at least twice were included in the figure.
Citation: Weather, Climate, and Society 13, 3; 10.1175/WCAS-D-20-0171.1
Two groundhogs in particular have received national fame in their respective countries but are equally poor at predicting spring. Punxsutawney Phil from Pennsylvania, whose record spans over 100 years and who is the only groundhog to have been objectively studied to date, only predicted at a 52% success rate according to our evaluation criteria (chi-square test;
Our analyses also indicated that any hope for geographic or national groundhog pride was unsupported. The proportion of correct spring predictions was not affected by the latitude where a groundhog forecast from (linear regression; F1,30 = 0.13, R2 = 0.002, and P = 0.72), and although the density of guessing groundhogs is much higher in the United States than Canada, there was no difference in the overall (in)ability of Canadian or American groundhogs to correctly predict spring (Welch’s two-sample t test; t = 0.20; degrees of freedom df = 5, 24; P = 0.85).
It appears that Punxsutawney Phil has a firm understanding of the warming world in which we now find ourselves. In the later part of his >100-yr climate record, Punxsutawney Phil has been predicting early springs (not seeing his shadow) with higher frequency (chi-square test;
The dearth of groundhogs’ predictive ability with regard to the onset of spring reported in our study may not solely rest on the haunches of the groundhogs themselves. With the exception of the experience of fictional character Phil Connors (Ramis 1993), Groundhog Day is celebrated annually on 2 February; however, in North America groundhogs typically emerge from their burrows during March at the earliest (Zervanos et al. 2010). Perhaps if the holiday was shifted closer to March to better coincide with the typical emergence date for this species, these groundhogs might have a larger number of natural environmental cues to inform their predictions and improve their accuracy (although potentially reducing the usefulness of predictions, given the reduced timeframe between the date of prediction and actual spring onset). In addition, much of the available data with regard to groundhog predictions is based on relatively few observations (94% of our groundhogs had n < 30 predictions). As such, a repeat of our study in 50 years might provide a more robust dataset for analysis. One last drawback to our analysis and potential future endeavors in evaluating groundhog predictions is that the average groundhog life span in captivity is 14 years (Patrick et al. 2016). Unfortunately, few organizations report the “replacement” dates of particular groundhogs when their predecessors die. As such, predictive capacities of groundhogs could potentially vary individually within a given location, but such information is currently not available.
Through the most comprehensive and biologically relevant assessment of groundhog predictive capacity to date, and with the use of a novel and arguably more accurate assessment of the onset of spring than has been used in previous studies, it seems clear that groundhog predictions are entirely unrelated to the onset of spring. While these activities provide good reason to celebrate the anticipation of spring’s arrival and a glimpse of hope for things to come from the doldrums of winter as February arrives each year, it is clear from our analysis that the predictions provided by these groundhogs are unlikely to replace the efforts of human meteorologists.
Acknowledgments
This study was the result of a truly laboratory-driven effort that started over speculation on a Friday at the campus pub and has led to a cumulative effort of students in the Community Ecology and Energetics Laboratory at Lakehead University over several years. Brendan Allan, Patrick Kennedy, Victoria Langen, Andrew Milling, and Stephen Slongo assisted with early components of compiling information for the paper. Thanks are given to all of the municipal, state, and provincial organizations that keep records of groundhog “predictions” that can be tracked and occasionally examined. We thank the constructive suggestions from three anonymous reviewers that were helpful in improving the paper. This work was supported by grants from the Canada Research Chair program, an NSERC Discovery grant and an Ontario Early Career Award to author Rennie, and an NSERC PGS-D grant to author Ross.
Data availability statement
All data are available in the public domain; however, the authors are happy to share the dataset that supports the analyses presented here upon request.
APPENDIX
Comprehensive List of Weather Stations
Table A1 provides a list of the full weather station identification codes that were used for estimating bloom date of Carolina spring beauty for all of our groundhog locations, along with the year range of data that were accessed.
Comprehensive list of weather stations used for estimating bloom date of Carolina spring beauty across groundhog locations. For each groundhog location, a minimum of 30 years of climate data were accessed. As a result of noncontinuous data collection for certain weather stations, there can be years for which multiple weather stations within the same county were used to calculate accumulated growing degree-days above 0°C (GDD0).
REFERENCES
Aaron, M. A., B. B. Boyd, M. J. Curtis, and P. M. Sommers, 2001: Punxsutawney’s phenomenal phorecaster. Coll. Math. J., 32, 26–29, https://doi.org/10.1080/07468342.2001.11921847.
Burt, W. H., and R. P. Grossenheider, 1976: A Field Guide to the Mammals of America North of Mexico. 3rd ed. Houghton Mifflin Company, 289 pp.
Davis, D. E., 1967: The role of environmental factors in hibernation of woodchucks (Marmota monax). Ecology, 48, 683–689, https://doi.org/10.2307/1936520.
Davis, D. E., 1977: Role of ambient temperature in emergence of woodchucks (Marmota monax) from hibernation. Amer. Midl. Nat., 97, 224–229, https://doi.org/10.2307/2424700.
Menzel, A., and Coauthors, 2006: European phenological response to climate change matches the warming pattern. Global Change Biol., 12, 1969–1976, https://doi.org/10.1111/j.1365-2486.2006.01193.x.
Miller, J. M., and K. L. Chambers, 2006: Systematics of Claytonia (Portulacaceae). Systematic Botany Monogr., Vol. 78, American Society of Plant Taxonomists, 236 pp.
NOAA, 2020: Groundhog Day forecasts and climate history. NCEI, accessed 11 June 2020, https://www.ncei.noaa.gov/news/groundhog-day-forecasts-and-climate-history.
Nowack, J., A. D. Rojas, G. Körtner, and F. Geiser, 2015: Snoozing through the storm: Torpor use during a natural disaster. Sci. Rep., 5, 11243, https://doi.org/10.1038/srep11243.
Pachauri, R. K., and Coauthors, 2014: Climate Change 2014: Synthesis Report. Cambridge University Press, 151 pp., https://www.ipcc.ch/site/assets/uploads/2018/02/SYR_AR5_FINAL_full.pdf.
Pareek, A., and P. C. Trivedi, 2011: Cultural values and indigenous knowledge of climate change and disaster prediction in Rajasthan, India. Indian J. Tradit. Knowl., 10, 183–189.
Patrick, A., and Coauthors, 2016: Sensitivity of primary fibroblasts in culture to atmospheric oxygen does not correlate with species lifespan. Aging, 8, 841–847, https://doi.org/10.18632/aging.100958.
Piao, S., and Coauthors, 2019: Plant phenology and global climate change: Current progresses and challenges. Global Change Biol., 25, 1922–1940, https://doi.org/10.1111/gcb.14619.
Ramis, H., 1993: Groundhog Day. Columbia Pictures, 101 min.
Risiro, J., D. Mashoko, D. T. Tshumba, and E. Rurinda, 2012: Weather forecasting and Indigenous knowledge systems in Chimanimani District Manicaland, Zimbabwe. J. Emerging Trends Educ. Res. Policy Stud., 3, 561–566.
Ruf, T., and F. Geiser, 2015: Daily torpor and hibernation in birds and mammals. Biol. Rev. Cambridge Philos. Soc., 90, 891–926, https://doi.org/10.1111/brv.12137.
Thomas, R. B., 1997: Old Farmer’s Almanac. Yankee Publishing, 288 pp.
Turner, N. J., and H. Clifton, 2009: “It’s so different today”: Climate change and Indigenous lifeways in British Columbia, Canada. Global Environ. Change, 19, 180–190, https://doi.org/10.1016/j.gloenvcha.2009.01.005.
Wallisch, K., 1999: Animal behavior as a weather predictor. Ph.D. thesis, Oklahoma State University, 130 pp.
Yoder, D., 2003: Groundhog Day. Stackpole Books, 144 pp.
Zervanos, S. M., C. R. Maher, J. A. Waldvogel, and G. L. Florant, 2010: Latitudinal differences in the hibernation characteristics of Woodchucks (Marmota monax). Physiol. Biochem. Zool., 83, 135–141, https://doi.org/10.1086/648736.
