A study demonstrates that Twitter can serve as a sensor for the onset of spring.
Technology is advancing at a lightning fast pace, and society has become accustomed to information at its fingertips. With 77 percent of all Americans owning a smartphone and 78 percent owning a computer, people can find and share information or connect to others with ease. The advent of social media platforms over the last decade has changed our communication patterns. Sharing life events, news stories, job opportunities, and even political perspectives with millions of socially connected listeners has become commonplace. Additionally, social media platforms are allowing participants to contribute environmental observations to a global conversation. I chose to explore whether such observations shared on the social media platform, Twitter, could support seasonal change science. The findings of this study indicate that Twitter observations can offer valuable environmental insights when they are examined collectively.
Commonly, environmental observations found on social platforms offer an indication of how the seasons are changing. Individuals may share a note on flowers blooming in their yard, identifying the first robin of the year, or changing leaves in the fall. The science of documenting seasonal changes in the environment is termed phenology. The USA National Phenology Network (USA-NPN) refers to the science as nature’s calendar. Helmut Lieth more thoroughly described the science as an activity of observing “life cycle phases or activities of plants and animals in their temporal occurrence throughout the year” [1, p. 4]. Central to the study is the concept of looking at nature to understand when seasonal events in plants and animals occur.
Over the course of the last decade, social media usage has exploded; the percent of all adult Americans with an account grew from 7 percent in 2005 to 65 percent in 2015. Because of social media’s ubiquity in the lives of a diverse American populous, understanding the contribution that social media users can make to science is extremely important, particularly because of the potential these posts offer to augment more formal collection of observations of natural events and timing.
Twitter, a microblogging site, allows users to share short messages, tweets, to other network users. Using the advertising slogan “What’s Happening?”, Twitter asks its user to share and discover what is being talked about or trending. Twitter restricts users to 280-character messages; a 140-character limit was recently doubled. Users can follow one another, respond to each other’s content, or share each other’s content by retweeting.
Twitter has been used as a sensor in a variety of contexts from earthquake detection to stock market and political election outcome predictions. However, its use as a social sensor for detecting the onset of seasonal change has not yet been broadly explored. With it serving as a platform where users commonly share observations of seasonal change, it merits consideration as tool that can help to inform phenology.
I sought to examine the intersection of social media and seasonal change observations. In this study, I investigated how Twitter might function as a forum for sharing seasonal change observations and discerned the importance and applicability of Twitter observations for phenological insights. Specifically, I asked, can Twitter be used as a platform to detect the onset of spring?
To determine whether Twitter can be used to detect seasonal change, six phrases/words with significant relevance to the onset of the spring season were tracked over the spring of 2016 and 2017. Both data sets collected tweets posted between January 1 and March 31. Tweets that were relevant to phenological observation based on the keyword criteria were incorporated into the analysis. Tweets were excluded when they contained the keyword but did not apply to spring emergence. For example, tweets that spoke of “early spring break” were removed.
The USA-NPN estimates the biological start of spring using Spring Indices (synthetic measures of spring season activity in plants) based on spring temperature conditions. These indices show that the start of spring was weeks earlier across much of the U.S. in 2017 compared to 2016. In particular, the start of spring was much earlier in 2017 in the Southeastern U.S. – the region from which the majority of spring-related tweet activity might be originating in the months evaluated in this study.
Twitter observations that noted an early spring initially followed a similar trend in 2016 and 2017. However, around February 20 a noticeable deviation from the 2016 trend line began to emerge. An obvious increase in tweets that addressed an early spring were being posted.
Between February 20 and March 14, 2017, the largest percentage of early spring tweets occurred. Interestingly, a similar increase can be recognized around the same time frame in 2016 but with far fewer tweets. It is also worth mentioning that a spike in tweets was observed in both years around Groundhog Day. Such tweets may have been referencing the outcome of the event without using any key identifying terms in the tweet. By Feb. 26, 2017, there had been more tweets addressing early spring than in the entire 2016 sample.
Comparing USA-NPN spring onset findings with the combined voice of Twitter users in 2016 and 2017 offered insights into the effectiveness of the platform as an environmental sensor. By examining tweets that occurred between Jan. 1 and March 31 for each year across various keywords, it was apparent that social observations on Twitter, when examined collectively, could detect a difference between the two spring seasons. Each of the six spring phrases that were considered resulted in more tweets in 2017 than in 2016, as well as a pattern of tweeting that reflected the emergence of spring coinciding with USA-NPN results for the respective year.
The 2017 data set revealed a much greater awareness of spring and spring indicators than the previous year. It would be reasonable to believe that such a noteworthy difference in tweets year over year would be the result of a gross difference in what would be considered normal. Perhaps Twitter users are more inclined to document abnormalities in the onset of spring. If an observer recognized that an event was taking place earlier than they anticipated, they may have felt it was worth sharing. While this speaks to the large difference in the number of tweets, it may or may not address the varied pattern of observations between years. For example, many of the 2016 trend lines saw a steady increase in tweets, possibly indicating a more standard progression of spring, whereas 2017 trend lines frequently assumed a much steeper slope after the middle of February. The sharp spike, resulting from a dramatic increase in tweets per day, could potentially indicate a major weather event, such as high temperatures or a shift in precipitation, which brought about changing phenophases for spring indicator species. The U.S. National Oceanic and Atmospheric Administration (NOAA) reported that the average temperature for the contiguous United States from January through March 2017 was 40.29 degrees F, an increase over the 2016 average of 39.73 degrees F. The three-month stretch represented the second warmest January, February, and March since 1895 and included an extremely warm February that was 7.36 degrees warmer than the 20th century average.
Cumulative tweets in 2017 for first robin and crocus both lagged 2016 until they surged past the 2016 trend around the third week of February. Lilac tweets in 2017 also lagged 2016, however the lag was much more dramatic than the other indicators that were measured. In addition, the 2017 spike in lilac tweets didn’t occur until the middle of March. Such a slope increase could potentially be attributed to the emergence of the flower suddenly as opposed to a more tapered, gradual emergence. If this is the case it may speak to the level of sensitivity that such analysis can generate.
This research explored how useful Twitter might be as an indicator for seasonal change. The results of this study revealed that Twitter can serve as a sensor for the onset of the spring season. Tweeting trends corresponding with NPN spring emergence trends were observed when comparing the onset of the 2016 spring with 2017. While individual tweets may or may not accurately represent the seasonal change, this study’s results suggest that when tweets are looked at collectively across the entire media platform they can offer an indication as to how nature is behaving.
The ability to leverage social tools to garner insights into nature’s seasonal calendar has the potential to support phenological research. Given the prevalence of social platforms, learning how to harness the observations that users offer for the support of scientific research can help to advance research. Twitter and other social media platforms may have the ability to help assess nature’s behaviors over large temporal and spatial dimensions. Expanding this research to study other elements of seasonal change could help science further understand how a changing climate may impact individual species as well as the dynamic interactions between species. Furthermore, understanding how nature behaves can offer clues regarding how humans should alter their patterns in the face of a warming climate. Given society’s propensity and desire to share observations on social platforms the data, or tweets, should be used in a collective manner to further understand the natural world.
This research offers only a glimpse into the potential that the social space may offer observational environmental research. While similar analysis could be completed to monitor a multitude of environmental factors and species, there is still much more to consider from a phenological perspective.
Philip Halliwell is a professor of environmental science at Colorado Mountain College. In addition, he is completing a doctoral degree in sustainability education at Prescott College. His research explores the role of citizen science in environmental research.
 H. Lieth, Ed., Phenology and seasonality modeling. Springer Science & Business Media, 1974. [E-book] Available: Google Books.