Using localized Twitter activity to assess harmful algal bloom impacts of Karenia brevis in Florida, USA

Highlights

• Twitter activity strongly correlates with local red tide conditions over time.

• Per-capita Twitter metrics yield stronger correlations than total metrics.

• Explicitly geo-tagged tweets prove more reliable than user geoprofile matches.

• Tweet term frequency analysis helps indicate issues of public’s utmost concern.

• Twitter sentiment analysis helps understand public’s awareness of red tide.

Abstract

Red tide blooms of the dinoflagellate Karenia brevis (K. brevis) produce toxic coastal conditions that can impact marine organisms and human health, while also affecting local economies. During the extreme Florida red tide event of 2017–2019, residents and visitors turned to social media platforms to both receive disaster-related information and communicate their own sentiments and experiences. This was the first major red tide event since the ubiquitous use of social media, thus providing unique crowd-sourced reporting of red tide impacts. We evaluated the spatial and temporal accuracy of red tide topic activity on Twitter, taking tweet sentiments and user types (e.g. media, citizens) into consideration, and compared tweet activity with reported red tide conditions, such as K. brevis cell counts, levels of dead fish and respiratory irritation on local beaches. The analysis was done on multiple levels with respect to both locality (e.g., entire Gulf coast, county-level, city-level, zip code tabulation areas) and temporal frequencies (e.g. daily, every three days, weekly), resulting in strong correlations between local per-capita Twitter activity and the actual red tide conditions observed in the area. Moreover, an association was observed between proximity to the affected coastal areas and per-capita counts for relevant tweets. Results show that Twitter presents a trustworthy reflection of the red tide’s local impacts and development over time, and can potentially augment the already existing tools for efficient assessment and a more coordinated response to the disaster.

Introduction

Red tide algal blooms associated with marine dinoflagellates are regular occurrences in coastal environments where oceanographic conditions, nutrient availability, and planktonic community interactions create conditions favorable for growth. Increases in the frequency, duration and severity of these events have occurred over the years (Tester, Steidinger, 1997, Tester, Steidinger, 1997, Kirkpatrick, Fleming, Squicciarini, Backer, Clark, Abraham, Benson, Cheng, Johnson, Pierce, et al., 2004, Kirkpatrick, Fleming, Squicciarini, Backer, Clark, Abraham, Benson, Cheng, Johnson, Pierce, et al., 2004, Glibert, Anderson, Gentien, Granéli, Sellner, Glibert, Anderson, Gentien, Granéli, Sellner), stimulated by multiple factors related to coastal land development and climate change (Glibert, 2020, Glibert, 2020). Toxins (Lin, Risk, Ray, Van Engen, Clardy, Golik, James, Nakanishi, 1981, Lin, Risk, Ray, Van Engen, Clardy, Golik, James, Nakanishi, 1981, Poli, Mende, Baden, 1986, Poli, Mende, Baden, 1986) produced by these species or degradation of water quality conditions with cell decomposition can have multiple negative impacts on marine environments and human communities in coastal areas. Large die-off events of marine organisms are commonly observed with red tides (Sievers, 1969, Sievers, 1969, Steidinger, 1996, Steidinger, 1996), including fish, marine associated birds, turtles, and large mammals (e.g., cetaceans, sirenians). Bioaccumulation of neurotoxins can further alter ecosystem trophic dynamics (Pierce, Kirkpatrick, 2001, Pierce, Kirkpatrick, 2001). Coastal conditions for humans during ride tides are also negatively impacted, as toxins are aerosolized out of the water column contributing to respiratory or skin irritation that can lead to hospitalizations (Backer, Kirkpatrick, Fleming, Cheng, Pierce, Bean, Clark, Johnson, Wanner, Tamer, et al., 2005, Backer, Kirkpatrick, Fleming, Cheng, Pierce, Bean, Clark, Johnson, Wanner, Tamer, et al., 2005, Fleming, Kirkpatrick, Backer, Bean, Wanner, Dalpra, Tamer, Zaias, Cheng, Pierce, et al., 2005, Fleming, Kirkpatrick, Backer, Bean, Wanner, Dalpra, Tamer, Zaias, Cheng, Pierce, et al., 2005, Fleming, Kirkpatrick, Backer, Bean, Wanner, Reich, Zaias, Cheng, Pierce, Naar, et al., 2007, Fleming, Kirkpatrick, Backer, Bean, Wanner, Reich, Zaias, Cheng, Pierce, Naar, et al., 2007, Milian, Nierenberg, Fleming, Bean, Wanner, Reich, Backer, Jayroe, Kirkpatrick, 2007, Milian, Nierenberg, Fleming, Bean, Wanner, Reich, Backer, Jayroe, Kirkpatrick, 2007). Economic impacts are also observed, as tourism or waterfront business revenues are reduced in areas affected by red tide.

Red tides from K. brevis have been observed on the southwest coast of Florida for over a century. Similar to global trends in harmful algal bloom (HAB) occurrences, these events have increased in severity, frequency, and geographic extent in recent years. While blooms originate offshore on the West Florida Shelf, onshore winds and currents can bring red tide blooms inshore to beaches and estuaries along Florida’s Gulf Coast, where they may be exacerbated by coastal nutrient sources. Extensive hydrologic modification of southwest Florida has altered the quantity, location, and timing of nutrient delivery from background conditions. Land use conversion of wetlands to agriculture and urbanization of coastal areas has further altered drainage patterns. The impacts of these changes on coastal water quality in southwest Florida have been exacerbated by climate change, as seasonal temperature and precipitation patterns have deviated from historical averages. Although no single cause has been identified, the synergistic effects of multiple stressors have likely contributed to significant red tide events. One of the longest and most severe events to date persisted on the coast of southwest Florida for 16 months from Fall of 2017 to early 2019.

Given the potential for large negative effects of red tide events on the environment, economy and public health, additional datasets that can augment conventional monitoring data could improve how local governments plan and respond to these events. Local governments and regional planning councils rely on regularly updated and routine monitoring data to respond appropriately to environmental disasters. These data inform how future response is managed to anticipated events, as well as how real-time response to ongoing events changes relative to where impacts are most observed. For red tide events, local governments may require data on where blooms are most severe to inform clean-up events (e.g., in response to fish die-offs) or to issue public safety announcements regarding beach closures or air quality advisories. Real-time rapid assessment of in situ conditions can inform these decisions, whereas more detailed information about oceanographic conditions and phytoplankton cell counts that require lab processing can be used for forecasting where bloom events are likely to occur in the future (National Oceanic and Atmospheric Administration (NOAA) (2018), University of South Florida Ocean Circulation Group (2010)). Moreover, conducting keyword analysis of social media messages could help in determining the issues of public’s utmost concern (e.g. is it economic impacts? health consequences? general concern for the environment?), hence informing decisions on the optimal distribution of the resources and funding (e.g. assisting waterfront businesses, distributing protective equipment, organizing marine debris cleanup, etc).

The recent bloom in southwest Florida was the first major red tide event since broad public use of social media platforms, offering potentially unique opportunities to assess complementary sources of information that can aid management response to disaster events. During catastrophes like floods, earthquakes, and hurricanes people have been turning to platforms such as Twitter (Eismann, Posegga, Fischbach, Eismann, Posegga, Fischbach, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Mukkamala, Beck, 2017, Mukkamala, Beck, 2017, Zou, Lam, Shams, Cai, Meyer, Yang, Lee, Park, Reams, 2019, Zou, Lam, Shams, Cai, Meyer, Yang, Lee, Park, Reams, 2019), Facebook (Bird, Ling, Haynes, et al., 2012, Bird, Ling, Haynes, et al., 2012, Eismann, Posegga, Fischbach, Eismann, Posegga, Fischbach, Bhuvana, Aram, 2019, Bhuvana, Aram, 2019, Jayasekara, 2019, Jayasekara, 2019), Instagram (Sherchan, Pervin, Butler, Lai, Ghahremanlou, Han, 2017, Sherchan, Pervin, Butler, Lai, Ghahremanlou, Han, 2017), Flickr (Liu, Palen, Sutton, Hughes, Vieweg, et al., 2008, Liu, Palen, Sutton, Hughes, Vieweg, et al., 2008, Preis, Moat, Bishop, Treleaven, Stanley, 2013, Preis, Moat, Bishop, Treleaven, Stanley, 2013, Chien, Comber, Carver, 2017, Chien, Comber, Carver, 2017) to share their sentiments and experiences, making social media a valuable source of information. There’s been a variety of aspects studied in cases of utilizing social media within the context of disaster events, ranging from collective behavior analysis (Liu, Palen, Sutton, Hughes, Vieweg, et al., 2008, Liu, Palen, Sutton, Hughes, Vieweg, et al., 2008, Eismann, Posegga, Fischbach, Eismann, Posegga, Fischbach), to risk and crisis communication (Veil, Buehner, Palenchar, 2011, Veil, Buehner, Palenchar, 2011, Bird, Ling, Haynes, et al., 2012, Bird, Ling, Haynes, et al., 2012), to relief effort coordination (Gao, Barbier, Goolsby, 2011, Gao, Barbier, Goolsby, 2011, Purohit, Castillo, Diaz, Sheth, Meier, 2014, Purohit, Castillo, Diaz, Sheth, Meier, 2014, Bhuvana, Aram, 2019, Bhuvana, Aram, 2019), among others. Meanwhile, in this work we intend to focus on studying the spatio-temporal correlations between localized social media activity and disaster impacts to respective areas (Preis, Moat, Bishop, Treleaven, Stanley, 2013, Preis, Moat, Bishop, Treleaven, Stanley, 2013, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Chien, Comber, Carver, 2017, Chien, Comber, Carver, 2017, Chen, Wang, Ji, 2020, Chen, Wang, Ji, 2020), along with conducting keyword and sentiment analysis for gauging public’s situational awareness during the disaster (Verma, Vieweg, Corvey, Palen, Martin, Palmer, Schram, Anderson, 2011, Verma, Vieweg, Corvey, Palen, Martin, Palmer, Schram, Anderson, 2011, Yin, Karimi, Robinson, Cameron, 2012, Yin, Karimi, Robinson, Cameron, 2012, Mukkamala, Beck, 2017, Mukkamala, Beck, 2017). That would serve to provide an additional use case and partially confirm some of the research done in Preis, Moat, Bishop, Treleaven, Stanley, 2013, Preis, Moat, Bishop, Treleaven, Stanley, 2013, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, when user activity on Twitter and Flickr was shown to be a good proxy for evaluating local damages during Hurricane Sandy, or the study conducted in Chen, Wang, Ji, 2020, Chen, Wang, Ji, 2020, where tweets containing certain keywords were shown to correlate with road closures due to Hurricane Harvey flooding.

Our study focused on Twitter - a "microblogging" service used by over 330 million people worldwide to share short strings of text, or "tweets", to convey opinions or ideas on any topic. Besides its efficiency of communication (due to character restrictions imposed on tweet content), a key component of Twitter is the ability to share or "retweet" information from other users such that a single message can reach a much wider audience than the user’s immediate followers. Tweets can also be used to assess a response in time (i.e., how do people react after an event) that can be traced to a geographic location if, for example, a tweet is spatially referenced by the user. Thus, interpretation of tweet content across multiple users, relative numbers of retweets, spatial, and temporal information can provide an assessment of topic importance that can guide response and recovery efforts. The use of Twitter to both gauge the impacts and inform disaster response has received some attention in the recent literature. For example, links between Twitter activity and localized storm damages (Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016, Kryvasheyeu, Chen, Obradovich, Moro, Van Hentenryck, Fowler, Cebrian, 2016) and community response to storm events when power outages prevented use of conventional communication outlets (Pourebrahim, Sultana, Edwards, Gochanour, Mohanty, 2019, Pourebrahim, Sultana, Edwards, Gochanour, Mohanty, 2019) have both been demonstrated with an analysis of tweet data during Hurricane Sandy. While storm events unfold over days, red tide events can last several months and thus introduce a significant temporal as well as spatial component to the impacts.

Here, we present a multi-level analysis of Twitter activity as it relates to the Florida 2017–2019 red tide event. Our overall goal was to evaluate the use of Twitter as a potential complement for data on real-time conditions as measured by in situ observations (e.g., phytoplankton cell counts) or reported conditions relevant to public health (e.g., beach conditions). The intent is not to supplant existing methods for tracking status and trends of red tide conditions, but to understand social responses to these events that may have management implications for improving communication about red tide, both general education and real time conditions, and for prioritizing prevention or mitigation actions by local governments. The use of Twitter as a reflection of red tide conditions could also augment existing datasets by providing either confirmatory, or in some cases anticipatory, information in the absence of in situ data. For example, this may have value if regular monitoring programs for tracking conditions do not exist or lab processing times prevent rapid response to current conditions. Lastly, we analyzed sentiment and topical nature of the tweets for the purposes of gauging public’s situational awareness during the event, determining the issues of concern (environment, health, economy, government), and potentially improving the accuracy of Twitter metrics as proxies for local red tide conditions.