Identifying Frostquakes in Central Canada and Neighbouring Regions in the United States with Social Media

Abstract

Following the ice storm of December 2013 in southern Ontario, the general public heard noises that resembled falling trees and reported these occurrences on social media. These were identified as a rare phenomenon called cryoseism, or more commonly known as frostquakes. These occurrences became the first large-scale documented frostquakes in Canada. Using meteorological metrics, we were able to forecast two subsequent frostquake events in January 2014 that coincided with reports on social media. In total, six more episodes of frostquakes as well as their locations were identified in January and February of 2014. Results showed that in central Canada, frostquake occurrences ranged from Windsor, Ontario to the west to Montreal, Quebec to the east and from Niagara Falls, Ontario to the south to North Bay, Ontario to the north. In the United States, the reports came from states bordering the Great Lakes and the New England areas. Two frostquake clusters were identified, one in and around the Greater Toronto Area and the other in eastern Wisconsin. Frostquakes were most frequently heard at nighttime. We critically assess the use of social media as an observation network including the possibility of false positives and population bias. This study demonstrates that rare phenomena such as frostquakes can be identified and assessed using data gathered through social media.

Full text

Identifying Frostquakes in Central Canada and
Neighbouring Regions in the United States with
Social Media
Andrew C.W. Leung*
1
, William A. Gough1, Yehong Shi1
Abstract
Following the ice storm of December 2013 in southern Ontario, the general public
heard noises that resembled falling trees and reported these occurrences on social
media. These were identified as a rare phenomenon called cryoseism, or more
commonly known as frostquakes. These occurrences became the first large-scale
documented frostquakes in Canada. Using meteorological metrics, we were able to
forecast two subsequent frostquake events in January 2014 that coincided with
reports on social media. In total, six more episodes of frostquakes as well as their
locations were identified in January and February of 2014. Results showed that in
central Canada, frostquake occurrences ranged from Windsor, Ontario to the west
to Montreal, Quebec to the east and from Niagara Falls, Ontario to the south to
North Bay, Ontario to the north. In the United States, the reports came from states
bordering the Great Lakes and the New England areas. Two frostquake clusters
were identified, one in and around the Greater Toronto Area and the other in
eastern Wisconsin. Frostquakes were most frequently heard at nighttime. We
critically assess the use of social media as an observation network including the
possibility of false positives and population bias. This study demonstrates that rare
phenomena such as frostquakes can be identified and assessed using data gathered
through social media.
1
Andrew C.W. Leung (), William A. Gough, Yehong Shi
Department of Physical & Environmental Sciences, University of Toronto Scarborough
1265 Military Trail, Scarborough, Ontario, Canada
andrewc.leung@mail.utoronto.ca

2
Keywords: frostquake; cryoseism; social media; crowdsourcing, collaborative
mapping
1 Introduction
Frostquakes, also known as cryoseism, are relatively rare weather phenomenon.
They occur after sudden freezing of the ground under specialized conditions and
are characterized by a “boom” or “cracking” noise that resembles falling trees.
Sometimes, a small tremor is also reported (Lacroix 1980). Occurrences are
infrequent (Nibonov 2010; Barosh 2000) and recurrences can be delayed by
decades or longer.
In southern Ontario, frostquakes were first heard on the night of December 24,
2013, just after the ice storm of 2013 (December 20–22, 2013). The general public
reported these noises on various social media platforms such as Twitter,
Facebook, and online discussion boards. Some described the noise as similar to
someone banging their fist against the wall or a gunshot (Allen 1993). Many
individuals reported on social media that they were asleep and were woken up by
the noise. A number of them mentioned that their pets became startled when the
noise began. According to media outlets, some people called the police believing
that someone was firing a shotgun or that their house was being broken into.
Meteorologists from local media in Toronto, Ontario identified that the noises
were likely the result of frostquakes and elaborated on the antecedent processes
that give rise to them. Seismic events were quickly ruled out as seismic stations in
Canada did not find any seismic waves in this area that night. Meteors were also
ruled out.
After reading the term online or hearing it in the news, the public appeared to be
looking for more information and turned to Wikipedia (Fig. 1). Prior to December
25, 2013, the page on cryoseism received about 300 views on average per day
(searching for the terms “ice quake” and “frostquake” on Wikipedia are redirected
to the cryoseism page). Just after the first wave of frostquakes in southern Ontario,
the number of views on cryoseism spiked up to 2772 views on December 25 and
then 6363 views on December 26. After that, visits to this page subsided but still
well above the average prior to the first wave. When frostquakes returned during
the night of January 2 to 3, 2014, the news media once again covered the event in
newspaper, on TV, and online. Because of the attention generated by these various
publications, the Wikipedia page drew over 21,000 views on January 3. The
public became conscious of the noise so that when frostquakes occurred once

3
again on January 6 to 7 the cryoseism entry on Wikipedia also drew over 40,000
views, the highest daily visit in that page’s history. These page views are
considered to be a conservative number since the traffic statistics software did not
include mobile views into the total, which accounts for roughly 30% of all page
views (Heilman and West 2015).
Fig. 1 Number of non-mobile page views on cryoseism article on Wikipedia. *
indicates dates with missing page view data
Prior to 2013, the only frostquake officially reported in Canada occurred near a
seismic station in Sadowa, Ontario (44.8oN, 79.2oW) on January 18, 2000 and the
occurrence was recorded on a fortuitously closely located seismometer (Natural
Resources Canada 2016). Eight probable frostquakes from 1870 to 1898 in New
Brunswick were identified by Burke (2004).
A search of the scientific literature produced scant results. Lacroix (1980)
examined frostquakes and their intensity in the New England area up to 1979. He
also identified that frostquakes were frequent in January. Barosh (2000) reported
additional frostquakes in New England area, including the damage caused by
related cracking. Fujita and Sleep (1991) confirmed three frostquakes and four
probable from 1872 to 1922 in Michigan. Allen (1993) monitored and recorded
frostquake activities with seismographs in Sebago Lake region of Maine during
the winter of 1990 to 1991. Burke (2004) found three likely frostquake events in

4
eastern Maine on top of the eight events in New Brunswick. Nikonov (2010)
examined events spanning 1803 to 1908 in Eastern Europe. He identified three
critical factors were required for the formation of frostquakes: moist soil, low to
no snow cover, and a sudden drop in temperature that exceeded -20oC.
A major reason why frostquakes appear so infrequently in the scientific literature
is their relative infrequency and difficulty in detection, it is a largely unstudied
phenomenon. While networks have been set up to detect earthquakes, frostquakes
are too localized and infrequent to be effectively monitored in a similar fashion.
Past occurrences rely on anecdotal information such as journals or newspaper
reports (Burke 2004), the social media of the time. With the advent of the internet
and the contemporary social media, frostquakes have the potential to be reported
more readily and thus researched. We note that gathering data through social
media has been used in natural science and earth science research in the past
(Hyvärinen and Saltikoff 2010; Ogden 2013). For example, US Geological Survey
used Twitter to improve its earthquake monitoring response time (Earle et al.
2011). In Europe, forest fires are usually detected by remote sensing but also
augmented by citizens contributing volunteered geographic information (VGI) in
the forms of blogs, tweets, and photos (De Longueville et al. 2010). Other
climatology-related observations that benefitted from VGI include assessing the
availability of outdoor skating rinks due to warmer winters in Canada (Robertson
et al. 2015), flooding and storm surges caused by Hurricane Sandy in New York in
2012 as well as tornado damage in Oklahoma in 2013 (Middleton et al. 2014).
However, this is the first time that frostquake data are gathered using VGI.
In this work, the questions we seek to answer are as follows:
What were the climate conditions of the January 18, 2000 Sadowa frostquake?
What was the geographical range and associated climate conditions of frostquakes
in Canada and US during the winter of 2013–2014?
What role can social media play in detecting frostquakes?
2 Methods
Two approaches were taken to investigate the frostquakes in Ontario during the
winter of 2013–14, climate data analysis and social media reporting. We have
limited our analysis to begin with the frostquakes that occurred on January 2/3,
2014. While frostquakes were first heard on the night of December 24, 2013, we

5
are not including reports from that night for several reasons. First, in southern
Ontario, over 1 million houses lost electricity as a result of the ice storm that was a
precursor to the formation of frostquakes. Power was not restored to some homes
until a week later. There would be inherent bias of under-reporting or no reporting
towards those who lost power since they could not go online to report their
observations. Second, because of the ice storm, many trees and branches had
fallen. These noises that appeared to be coming from frostquake could actually be
trees falling down under the weight of the ice and could have been mistakenly
identified as frostquakes or vice versa. This could lead to false positive reports.
Third and finally, the term frostquake was not familiar to Canadians. Only a few
Canadian TV media outlets in the Toronto area ran online stories on December
25–26 about the booming sound by mentioning the term frostquake. The term was
not publicized until another round of frostquakes on January 2, 2014 as noted
above.
2.1 Climate Data Analysis
We analyze local weather data using the criteria established by Nikonov (2010).
These metrics included saturated soils, low to no snow cover and a rapid drop of
temperature to below -20oC. We did this first for the January 18, 2000 frostquake
reported at Sadowa, Ontario and then for the frostquakes that occurred during the
winter of 2013–14.
For the January 18, 2000 frostquake event we used Muskoka Airport (44°58' N,
79°18' W) weather station data for climatological analysis. The weather station is
approximately 27 km away from Sadowa. We examined the daily temperature
(minimum, mean, and maximum), precipitation (rain and snow), and snow on
ground.
For the winter of 2013–14 we used data from fifteen Canadian weather stations
from Environment Canada’s Climate Archives (Fig. 2a) and data from six
American weather stations from National Ocean and Atmospheric
Administration’s Climate Data Online (Fig. 2b). Station selection was based on
the spatial range of frostquake reports and the number of reports mentioned in that
area. Similar to the event in 2000, we used daily temperature, precipitation, and
snow on ground.
Precipitation is reported using “trace” as a measure and we sought to quantify this.
For Canadian stations, trace amount of daily rainfall (< 0.2 mm/day) was given a

6
value of 0.1 mm/day and trace amount of daily snowfall (< 0.2 cm/day) was given
a value of 0.07 mm/day (Mekis and Vincent 2011). The adjustment for trace
amounts of precipitation in United States stations was more problematic due to
different measuring equipment and different definition of trace precipitation that
stems from its use of imperial units. Unlike Canada, US weather stations use
standard rain gauges to measure the amount of snowfall (Doesken and Judson
1997). In Yang et al. (1998), trace amount of daily rainfall (< 0.01 in/day; 0.254
mm/day) was assigned the same value (0.1 mm/day) as Canadian weather stations.
For trace amount of daily snowfall (< 0.1 in/day; 0.254 cm/day), Sugiura et al.
(2003) suggested an assigned value equal to a quarter of its measuring limit, which
is 0.025 in/day or 0.0635 mm/day. This assigned trace daily snowfall value for
American weather stations was almost identical to the value given to Canadian
weather stations.
Handling trace snow depth was more problematic. For the US, substantive
inconsistencies exist in terms of how trace amounts of snow depth was interpreted
and recorded among airport weather stations and volunteer stations (Doesken and
Judson 1997). In addition, while all Canadian airport and volunteer weather
observers record snow depth at or around 6 am daily, some US volunteer stations
record snow depth in the early evening while US airport weather stations report
snow depth at midnight. In both Canada and US, snow depth is rounded to the
nearest whole unit of measurement (cm for Canada, inch for US). Thus, a snow
depth of 0.5 cm to 1.0 cm is reported as 1 cm. For Canadian stations snow depth
below 0.5 cm is described as “trace”. Since snow depths of 0.1 cm to 0.4 cm were
considered equally likely to occur, the average value of 0.25 cm was assigned.
This approach is identical in principle with that used by Mekis and Vincent
(2011). Similarly, for US stations, trace snow depth between 0.1 in and 0.4 in was
given a value of 0.25 in, which is equal to 0.635 cm.
2.2 Social Media
The first approach was to analyze social media reports, particularly from Twitter
and produce maps of frostquake reports for each frostquake episode during the
winter of 2013–14. To identify frostquake events on Twitter, the search terms
“frostquake”, “cryoseism”, and “ice quake” were used. All frostquake reports
from January 2 to February 28, 2014 were examined. Location was identified as
the city mentioned in an individual’s tweet or post if given. Otherwise, the city
location that an individual associated with on their user profile was assumed to be
the location where the frostquake occurred. For date and time, t he timestamp of

7
the tweet or post was assumed to be the time of occurrence if the user mentioned
that they just heard the noise just prior to posting. If this was not case and the
individual specified the approximate time of the noise, then that time was used as
a proxy for the actual frostquake occurrence. Finally, if the individual did not
specify the time the frostquake was heard, only the date was assigned to the
location of the report. Those without specifying location were not included in the
study. Additional locations were obtained from a user-generated online Google
Map
(https://www.google.com/maps/d/viewer?mid=zId7WwTT0PPk.kmYXHjIndA-
w), which solicited social media users to collaboratively mark when and where
they heard the frostquake. This online map’s URL was also linked in multiple
news media’s online version of the story and encouraged the readers to add their
reports. Results from all crowdsourced information were sorted by date and
grouped by individual towns and cities. The reports were cleaned by examining
obvious plotting errors on the user-generated Google Map. Locations were
removed if the points were plotted in the middle of a large waterbody (e.g. Lake
Ontario).

8

9
Fig. 2 Weather stations in (a) Canada and (b) US chosen for temperature and
snow depth analysis

10
3 Results
3.1 Climate Data Analysis
3.1.1 January 18, 2000
The only frostquake officially confirmed by seismic record was the one from
Sadowa, Ontario on January 18, 2000 at 6:55pm. According to Natural Resources
Canada (2016), that night was very cold and 12 frostquakes were recorded within
a two-hour period. Coincidentally, individuals from Skowhegan, Maine also
reported frostquakes around the same time, on January 14 to 15, 2000 (Maine
Geological Survey 2016). Upon examining the weather conditions for both
locations at that time, both had above 0oC temperature two days prior to a quick
drop in temperature. The temperature drop on January 16, 2000 was quite large,
from 0oC to -25oC in one day. On the day of the frostquake, Sadowa had 8 cm of
snow cover on the day of frostquake while Skohegan had none. The way that
water entered the soil was also different between these locations. At Skowhegan,
rain was recorded two days prior to the frostquake. But at Sadowa, rain event
occurred seven days before the frostquake occurred. We believed that the increase
in soil moisture was caused by melting of the snow cover on the ground, as the
snow cover reduced from 11 cm to 8 cm. Therefore, it appeared that saturated soil
that caused frostquake could be the result of either rainfall or melting snow on the
ground. Since the events in Sadowa and Skowhegan happened within the same
week, we believe that the spatial variability is mainly caused by the particular
temperature, rainfall, and snow depth at the respective locations.
3.1.2 Winter of 2013–14
Temperature graphs for Canadian and American weather stations are shown in
Fig. 3. Thawing followed by a quick drop in temperature was observed in the
following periods at Canadian stations: December 19–22, December 26–29,
January 3–6, January 9–14, January 16–17, and February 18–23. Similar
observations at the American stations were found on December 26–29, January 9–
15, and February 17–23.
Snow depths are presented in Fig. 4. Most stations in southern Ontario had less
than 30 cm of snow on the ground from December 2013 to January 2014. All of
the American stations saw a decrease in snow cover after January 6–12. On
average, almost all of these American stations had less than 20 cm of snow depth
on average.

11
Fig. 3 Temperature at weather stations in (a) Canada and (b) US during the winter
of 2013–14

12
Fig. 4 Snow depth at weather stations in (a) Canada and (b) US
3.2 Social Media
Overall, there were 2301 frostquake reports recorded through social media (Table
1). We have generated maps of frostquake reports for three events with the highest
number of reports: January 2/3, January 6/7 and January 20–22 (Fig. 5). Over
2100 public reports were recorded from these three events alone and the majority
were from Canada (Table 1). Spatial analysis showed that regions with high
population density (Greater Toronto Area) also experienced the highest number of

13
reports. For the January 2/3 event, most reports came from Toronto and Brampton,
Ontario (Fig. 5a). Virtually all of the reports were from Ontario, though there were
two reports from Wisconsin and one each from Indiana and New York State. For
January 6/7 event, the highest number of reports came from Toronto and around
the Green Lake area in Wisconsin (Fig. 5b). There were also reports from
Montreal, Quebec (n=3), Montague, Prince Edward Island (n=1), and St. John’s,
Newfoundland and Labrador (n=1). From the United States, we received multiple
reports from Indiana, Ohio, Michigan, Vermont and Maine. We also had
Colorado, Iowa and Virginia that did not experience frostquake prior to this event.
For the January 20–22 event, Toronto and Newmarket, Ontario had the highest
number of reports (Fig. 5c). We also one frostquake reported in Minnesota, a new
state for reporting. There were 55 reports not classified because the public only
specified the location and did not include the time or date of the event.

14

15
Fig. 5 Plots of reported frostquake locations on (a) January 2/3, (b) January 6/7,
and (c) January 20–22 of 2014. Dot sizes and colours are scaled to the number of
reports in each community. Larger dots represent more reports from a particular
town or city
Using all of the gathered reports from Fig. 5, we combined all the counts from
each location on various dates to create a density report map (Fig. 6). The density
report map identifies actual cluster of reports by taking in higher population in
urban areas into consideration. The density report values in each community is
calculated by adding the total number of frostquakes reported on Jan. 2/3, 6/7 and
20–22 of 2014 then divided by the population of the community. The population
of each community is based on Statistics Canada’s 2011 Census and U.S. Census
Bureau’s 2010 Census data. In total, there were 236 communities that experienced
frostquakes during those periods. We found two clusters of reports around the
Toronto region and eastern Wisconsin region.

16
Fig. 6 Density report map for frostquakes. Individual values are classified by
number of reports per 10,000 individuals in the community
Based on Google Map reports, Twitter and Facebook posts, we created a temporal
distribution of the timing of the frostquakes (Fig. 7). We found that the most
common time when frostquakes were reported was at night. In Canada, most
reported hearing frostquake during the overnight period, especially between 1am
to 3am (Fig. 7a). In the US, most reported that they heard frostquakes between
7pm and 11pm (Fig. 7b).

17
Fig. 7 The local time in (a) Canada and (b) US at which the public reported to
have heard a frostquake.
Table 1 Breakdown of date ranges with frostquake and the number of reports
within each range
Dates with frostquake
Number of reports
Canada
US
Total
January 2–3
878
4
882

18
January 6–7
824
158
982
January 13–15
5
0
5
January 20–22
261
10
271
January 23–25
7
11
18
January 26–29
9
16
25
February 1–3
7
18
25
February 5–7
4
32
36
February 8–12
6
6
12
February 17
0
1
1
February 22–23
4
2
6
February 25–28
27
11
38
Total
2032
269
2301
3.3 Coincidence of Frostquake Reporting and Weather
Conditions
For the winter of 2013–14 the climate data analysis indicated for southern Ontario,
dramatic temperature drops for the following dates: December 19–22, December
26–29, January 3–6, January 9–14, January 17–20, January 24–27, February 15–
19 and February 26–28 (Table 2). Frostquakes are reported in Table 1 for days
where there was a large difference between maximum temperature and minimum
temperature and often after temperature passed through the melting point of ice.
Since Toronto had the most reports in the overall period, we used the weather
station located in downtown Toronto (Toronto City) as the representative for the
region. For January’s frostquakes, the dates that had most public reports (January
2/3, January 6/7, January 20–22) all had a large drop in temperature and the
minimum temperature was below -20oC. Other dates with frostquake events also
had a considerable drop in temperature but the minimum temperature did not drop
below -20oC.
Table 2 Daily maximum, minimum and the difference between maximum and
minimum temperature for the Toronto City weather station in downtown Toronto.
Dates in bold indicate that over 10 reports were recorded in those periods
Date
Maximum
Temperature (oC)
Minimum
Temperature (oC)
∆Temperature (oC)
1/1/2014
-8.4
-14.5
6.1
1/2/2014
-14.3
-19.2
4.9
1/3/2014
-7.1
-22.3
15.2
1/4/2014
0.3
-7.2
7.5

19
1/5/2014
1.4
-1.7
3.1
1/6/2014
2.4
-15.8
18.2
1/7/2014
-15.8
-22.2
6.4
1/8/2014
-7.7
-16.2
8.5
1/9/2014
-2.9
-11.8
8.9
1/10/2014
4.1
-4.0
8.1
1/11/2014
7.3
2.9
4.4
1/12/2014
3.4
1.3
2.1
1/13/2014
7.4
1.9
5.5
1/14/2014
5.3
1.3
4.0
1/15/2014
2.4
-3.5
5.9
1/16/2014
-0.4
-3.6
3.2
1/17/2014
2.3
-1.2
3.5
1/18/2014
-0.8
-6.9
6.1
1/19/2014
-1.9
-7.2
5.3
1/20/2014
-1.8
-16.5
14.7
1/21/2014
-14.2
-19.8
5.6
1/22/2014
-12.1
-20.5
8.4
1/23/2014
-12.2
-17.4
5.2
1/24/2014
-6.8
-17.7
10.9
1/25/2014
-2.8
-13.8
11
1/26/2014
-3.8
-14.6
10.8
1/27/2014
-3.7
-16
12.3
1/28/2014
-11.6
-18.6
7.0
1/29/2014
-9.2
-16.5
7.3
1/30/2014
-0.4
-10.6
10.2
1/31/2014
1.5
-2.5
4.0
2/1/2014
1.4
-1.1
2.5
2/2/2014
0.9
-6.1
7.0
2/3/2014
-2.4
-9.7
7.3
2/4/2014
-3.9
-10.5
6.6
2/5/2014
-5.2
-11.7
6.5
2/6/2014
-6.5
-13.2
6.7
2/7/2014
-8.5
-13.9
5.4

20
2/8/2014
-8.6
-14.2
5.6
2/9/2014
-7.2
-12.1
4.9
2/10/2014
-5.7
-12.0
6.3
2/11/2014
-8.8
-13.6
4.8
2/12/2014
-5.2
-14.1
8.9
2/13/2014
-1.6
-9.0
7.4
2/14/2014
0.7
-4.4
5.1
2/15/2014
-2.4
-11.9
9.5
2/16/2014
-6.9
-12.8
5.9
2/17/2014
-2.2
-14.0
11.8
2/18/2014
0.7
-4.2
4.9
2/19/2014
7.8
-1.4
9.2
2/20/2014
2.6
-1.0
3.6
2/21/2014
5.1
0.7
4.4
2/22/2014
4.3
-0.6
4.9
2/23/2014
1.4
-4.7
6.1
2/24/2014
-4.3
-8.8
4.5
2/25/2014
-5.2
-10.4
5.2
2/26/2014
-8.2
-14.2
6.0
2/27/2014
-8.3
-16.0
7.7
2/28/2014
-6.9
-17.7
10.8
4. Discussion and Conclusions
4.1 Weather Conditions
Nikonov (2010) concluded that moist soil, low snow cover, and a sudden drop in
temperature were the variables required for frostquakes to occur. However there is
a paucity of observations of frostquakes to explore the nature of the precursors.
Using social media reported frostquakes and coincident climate data, we have an
opportunity to study frostquakes in more detail.
The weather station data showed that these conditions were met during our study
period. The snow depth was shallow when the temperature went from below 0oC
to above (Fig. 4). A few days later, thawing stopped when the temperature quickly

21
dropped and in some locations, by up to 20oC within 24 hours (Fig. 3). During and
shortly after the temperature plunged, the public started to hear the noise or feel
the shaking from the frostquakes.
In our study, we speculate that these steps need to have a specific sequence and we
have documented this using local weather data. For example, if sudden drop in
temperature that results in freezing occurred before the soil became moist or
before the snow cover was reduced to low levels, frostquake would not have
happened because the ground would have sufficient space for the soil to freeze
and not cause any cracking noise. Therefore, the only realistic sequence would be
a location first having low to no snow cover which allowed the soil to become
moist. After that, the temperature drop must greatly exceed the insulation effect of
the snow cover so that the temperature in soil will quickly drop and water
molecules inside the soil will freeze.
4.2 Social Media
It is a commonly used research method to utilize VGI data gathered from the
general public. Very often, these studies provide online forms for the public to fill
out in a structured manner. This ensures the completeness of the data often by
asking the user to select from a list of pre-determined options. In contrast, data
collection from social media often faces bigger challenges. The submitted
information are considered “free flowing” because the user chooses the level of
detail and in an unstructured manner that sometimes require follow-up prior to
analyzing the data. The main benefits of using social media for data gathering is a
larger sample size that does not require the public to fill out a web form to report
their findings. Instead of waiting for the public to engage with the scientists, social
media allows scientists to reach out and obtain data directly from the public. In
our study, the spatial data production of frostquakes can be classified as “bottom-
up, amateur, and asserted” (Cinnamon 2015) as almost all of the data points were
generated by the general public on social media or Google Map rather than by
authoritative experts.
Hyvärinen and Saltikoff (2010) listed the service provider’s terms and conditions,
data retention policy, privacy, and copyright as the biggest challenges for
collecting meteorological observations from social media. Their study analyzed
user-submitted photos that were uploaded to Flickr (an online photo depository)
and identified meteorological events at a specific time and location from the
photos. We believe that our study has fewer issues related to content removal
policy, privacy, and copyright. We collected location data in near real-life time,
which circumvented the issue with the deletion of older materials. Hyvärinen and
Saltikoff (2010) indicated that Twitter only stored one week of tweets before
deletion. We noted that Twitter has since modified its content removal policy to
keep tweets in perpetuity unless deleted by the user. In addition, there are third-

22
party providers who store tweets on a particular topic or keyword for future
retrieval. However, we did find that some frostquake tweets were deleted by the
user in rare instances (<0.1% of all reports) by comparing the differences between
the search results from Twitter with the third-party’s. Privacy was not a concern
for our study. We aggregated all reports from a town or city into one location thus
resulting in a coarse resolution of the locations. Even though some users enabled
their GPS locations while tweeting, the coordinates were imprecise or sometimes
obviously inaccurate and unreliable on Twitter. For example, on a few occasions,
the coordinates identify the user’s location to be in the middle of Lake Ontario,
500 m away from the shorelines of Toronto.
While this is a novel approach of using social media for scientific data collection,
there are some potential issues with using Twitter and Facebook to gather data. It
is difficult to pinpoint the exact time and location when the frostquake occurred.
20% of the reports we gathered did not specify approximate time of occurrence.
The number of reports is dependent on the population of a city and citizens who
use social media, thus skewing towards larger cities like Toronto. Another under
reporting problem arose from those who heard the noise but decided not to report
it on social media. Furthermore, there is an age disparity for social media users. A
survey conducted in 2009 in the US found that 75% of young adults from age 18
to 24 had a social media accounts whereas only 7% of those aged 65 and above
had an account (Lenhart 2009). Therefore, reports gathered from different social
media were more likely to be coming from teenagers and young adults rather than
older adults or seniors. Our study’s age bias was somewhat lessened when
mainstream media included the link to our Google Map reporting system in their
online news stories. Privacy setting on social media accounts also suppressed
some reports. While Twitter’s tweets were set to public by default, Facebook’s
posts were private by default. 60% of teenagers reported that they set their profiles
(along with their posts) to private and only visible to their friends (Madden et al.
2013). Hence, frostquake reports were less likely to be found on Facebook
because of the users’ privacy settings. On the other hand, false positives are not
uncommon on social media (Hyvärinen and Saltikoff 2010) and this was
particularly true for frostquake since the only identification was a banging noise
and light tremor but usually no physical observation can be made and in some
cases a report could be completely fictitious. Some obvious false positive
frostquake reports were identified. Some users reported hearing frostquake noises
from places such as San Diego which did not have frost on the ground at that time
of the year. We also had a number of well-intentioned reports from the Pickering,
Ontario area on the morning of January 21, 2014. The volunteers later corrected
themselves after discovering that the shaking and noise were the result of a nearby
waste water treatment plant explosion and not by frostquake. These reports were
removed from the analysis. A few plausible locations such as Atlanta, Georgia and
Denver, Colorado were kept after examining the climatological conditions on
January 6–7, 2014 and found the conditions to be possible for frostquake to occur
(Fig. 5b).

23
We believe that the nature of the reports on social media between Canada and the
US were different. The reported frostquake time in the US tends to be in the
evening period where most people were still awake (Fig. 7b). They might have
heard the noise and decided to mention it on the internet. On the other hand, the
majority of Canadian reported frostquakes took place during the overnight period
after most people went to bed (Fig. 7a). It is likely in this case that the public was
woken up by the noise and decided to share their experience. In a number of
reports, the public said that their pets were woken up by the sound or felt the jolt
and the pets woke up their owners. A lot of Canadians mentioned in their tweets
that they were surprised by the sound and in some cases they were delighted to
finally hear it after their friends who shared similar experience with them. There
were several explanations which explained why more frostquakes were heard at
night than during the day. Night time temperature tends to be colder and a quick
temperature drop appears to be a requirement for frostquakes to occur. Another
reason to explain the temporal difference is that it is quieter at night and people
are less active, which makes it easier for the frostquake to be heard or felt. On a
greater temporal scale, we found much more frostquakes happening in January
than in February (Table 1) and this appeared to be in agreement with Lacroix
(1980).
On social media, the public stated that the sound appeared to be coming from the
roof even though the cause of the noise was the expansion of ice within the soil.
At the time of their reports, these individuals were in various types of buildings
(detached houses, apartments). It is unclear how the vibration sound resonated
through different building materials (e.g. wood, concrete) and propagate to upper -
level floors in apartment buildings. In addition, the frostquakes in 2013–2014
occurred in highly populated areas. This was noticeably different from reports
from New Brunswick and New England where most of the people who heard the
frostquakes were living in farm houses in rural communities (Allen 1993; Barosh
2000; Burke 2004).
4.3 Frostquake Clusters
Given that urban centres like Toronto and Montreal have large populations, their
total number of reports are not unexpectedly high when compared to suburban and
rural areas (Fig. 5). However, we did not observe a cluster of reports around
Montreal and that reports from Montreal area only appeared on January 6/7, 2014
(Fig. 5b). The density map showing a cluster in Toronto (Fig. 6) is not surprising
given that there were extensive media coverage and social media presence. It was
also the first area that received prominent attention given that many people
believed to have experienced it just after the ice storm of 2013. The spike in
Wikipedia traffic to cryoseism article (Fig. 1) on December 26, 2013 and January
3, 2014 were very likely to be coming from people in the Toronto cluster because
there were only 4 reports coming from the US up until January 3, 2014 (Table 1).

24
The Toronto cluster consists of four of the top ten largest municipalities in Canada
(Brampton, Hamilton, Mississauga, and Toronto) plus a number of suburban
towns and rural villages. While most people heard or felt the frostquake were
living in detached homes, some of them, especially in downtown Toronto, heard it
inside their apartments. Despite the large population size, Hamilton and
Mississauga were placed 28th and 30th percentile respectively on the density map
while Toronto was placed at 57th percentile and Brampton was placed at 73rd
percentile. There were a mix of low and high density reports from other
communities within the Toronto cluster. However, the density map showed that
Montreal had very low number of reports per 10,000 individuals.
The Wisconsin cluster, in contrast, is very different from the Toronto cluster. Most
of the communities in the Wisconsin clusters had very small population but high
density of reports. In fact, the top nine communities with the highest density
reports were all from the Wisconsin cluster and seven of the nine communities had
multiple reports of frostquakes. Yet, none of these nine communities had a
population over 1500 and practically everyone lives in detached housing in these
rural villages and towns. One thing in common between the Toronto and
Wisconsin clusters was that both areas were reported extensively by mainstream
media. In Wisconsin’s case, the attention was drawn to The Weather Network
(2014)’s video of a Wisconsin farmer discovering a crack about 30 m long and 20
cm deep after hearing booming noise which was attributed to frostquake. The
video caused heightened awareness of this phenomenon and was the subject of
local discourse. It led to a positive feedback loop of more awareness leading to
more reports, and that in turn led to even greater awareness. We found that similar
cracks were also present in two instances in Maine and Massachusetts (Allen,
1993; Barosh, 2000). In Allen (1993)’s study, the frostquakes were heard and felt
between 7–9 pm, which is consistent with our temporal analysis of when
frostquakes were most likely to be occur in United States (Fig. 7b). The timing
was quite similar to the event in Sadowa, Ontario at 6:55pm (Natural Resources
Canada 2016) and the event in Rothesay, New Brunswick in 1884 from 9:30pm–
10:30pm (Burke 2004). However, we found that the Canadian events in this study
took place much later in the night (Fig. 7a). Also, we were unable to explain why
the traffic to Wikipedia’s croyseism article did not show a noticeable spike after
the third major event on January 20–22, 2014 (Fig. 1).
Using Fig. 6 to compare with existing literature, we were able to get a snapshot of
which provinces and states were new to frostquakes. In Canada, there were no
frostquakes reported in Quebec, Prince Edward Island and Newfoundland and
Labrador prior to this study. Likewise, in United States, the states that did not
have reported frostquakes before January 2014 were Colorado, Georgia, Illinois,
Indiana, Iowa, Pennsylvania and Virginia. Almost all of the Canadian provinces
and US states which experienced the first frostquake took place on Jan. 6/7.
Our research demonstrated that yet another type of rare weather phenomenon like
frostquake can be monitored through social media. We linked the VGI reports of

25
frostquakes to concurrent weather conditions. Through the use of social media, we
collected the greatest number of frostquake locations to date and identified two
frostquake clusters. We also found three Canadian provinces and seven US states
that experienced its first ever reported frostquake during our study period.
Acknowledgements
We thank Ashley King for creating the initial user-generated Google Map that
served as the collaborative VGI mapping platform for the public to report their
observations.
References
Allen RP (1993) A Study Of Cryoseisms (“Frostquakes”) In The Sebago Lake
Region, Maine. In: Symposium on the Application of Geophysics to
Engineering and Environmental Problems 1993:415–430.
doi:10.4133/1.2922017
Barosh PJ (2000) Frostquakes in New England. Eng Geol 56(2000):389–394.
doi:10.1016/S0013-7952(99)00092-7
Burke KBS (2004) Historical Seismicity in the Central Highlands,
Passamaquoddy Bay, and Moncton Regions of New Brunswick, Canada,
1817-1961. Seismol Res Lett 75(3):419–431. doi:10.1785/gssrl.75.3.419
Cinnamon J (2015) Deconstructing the binaries of spatial data production:
Towards hybridity. Can Geogr 59(1):35–51. doi:10.1111/cag.12119
De Longueville B, Annoni A, Schade S, Ostlaender N, Whitmore C (2010) Digital
Earth’s Nervous System for crisis events: real-time Sensor Web Enablement
of Volunteered Geographic Information. Int J Digit Earth 3(3):242–259.
doi:10.1080/17538947.2010.484869
Doesken NJ, Judson A (1997) A Guide to the Science, Climatology, and
Measurement of Snow in the United States. Fort Colins, CO: Colorado State
University Department of Atmospheric Science.
Earle PS, Bowden DC, Guy M (2011) Twitter earthquake detection: earthquake
monitoring in a social world. Ann Geophys 54(6):708–715. doi:10.4401/ag-
5364

26
Fujita K, Sleep NH (1991) A re-examination of the seismicity of Michigan.
Tectonophysics 186(1-2): 75-106. doi:10.1016/0040-1951(91)90386-7
Heilman JM, West AG (2015) Wikipedia and Medicine: Quantifying Readership,
Editors, and the Significance of Natural Language. J Med Internet Res
17(3):e62. doi:10.2196/jmir.4069
Hyvärinen O, Saltikoff E (2010) Social Media as a Source of Meteorological
Observations. Mon Weather Rev 138(8):3175–3184.
doi:10.1175/2010MWR3270.1
Lacroix AV (1980) A short note on cryoseisms. Earthquake Notes 51(1):15–20.
doi:10.1785/gssrl.51.1.15.
Lenhart A (2009) Adults and Social Networking Websites. Pew Research Internet
Project. Available via http://www.pewinternet.org/2009/01/14/adults-and-
social-network-websites/. Accessed 20 May 2016
Madden M, Lenhart A, Cortesi S, Gasser U, Duggan M, Smith A, Beaton M
(2013) Teens, Social Media, and Privacy. Pew Research Internet Project.
Available via http://www.pewinternet.org/2013/05/21/teens-social-media-
and-privacy/. Accessed 20 May 2016
Maine Geological Survey (2016) Reports of earth shaking in Maine possibly due
to cryoseisms. http://www.maine.gov/dacf/mgs/hazards/earthquakes/quake-
cryolist.htm. Accessed 20 May 2016
Mekis E, Vincent LA (2011) An Overview of the Second Generation Adjusted
Daily Precipitation Dataset for Trend Analysis in Canada. Atmos-Ocean
49(2):163–177. doi:10.1080/07055900.2011.583910
Middleton SE, Middleton L, Modafferi S (2014) Real-time Crisis Mapping of
Natural Disasters using Social Media. IEEE Intell Syst 29(2):9–17.
doi:10.1109/MIS.2013.126
Natural Resources Canada (2016) Frequently Asked Questions about Earthquakes
(FAQ). http://www.earthquakescanada.nrcan.gc.ca/info-gen/faq-eng.php.
Accessed 20 May 2016
Nikonov AA (2010) Frost Quakes as a Particular Class of Seismic Events:
Observations within the East-European Platform. Izv Phys Solid Earth,
46(3):257–273. doi:10.1134/S1069351310030079
Ogden LE (2013) Tags, Blogs, Tweets: Social Media as Science Tool?
Bioscience, 63(2):148. doi:10.1525/bio.2013.63.2.15

27
Robertson C, McLeman R, Lawrence H (2015) Winters too warm to skate?
Citizen-science reported variability in availability of outdoor skating in
Canada. Can Geogr 59(4):383–390. doi:10.1111/cag.1222
Sugiura K, Yang D, Ohata T (2003) Systematic error aspects of gauge-measured
solid precipitation in the Arctic, Barrow, Alaska. Geophys Res Lett
30(4):1192–1195. doi:10.1029/2002GL015547
The Weather Network (2014) Wisconsin: Ice quake causes property damage.
http://www.theweathernetwork.com/news/articles/wisconsin-ice-quake-
causes-property-damage/19445. Accessed 20 May 2016
Yang D, Goodison BE, Ishida S, Benson CS (1998) Adjustment of daily
precipitation data at 10 climate stations in Alaska: Application of World
Meteorological Organization intercomparison results. Water Resour Res
34(2):241–256. doi:10.1029/97WR02681.