Twitterati on COVID-19 pandemic-environment linkage: Insights from mining one year of tweets

Abstract

The COVID-19 pandemic seems to have had positive (although short-lived, e.g., reduction in pollution due to lockdown) as well as negative (e.g., increasing plastic pollution due to use of disposable masks, etc.) impacts on the environment. The pandemic-environment linkage also includes circumstances when regions experienced extreme weather events, such as floods and cyclones, and disaster management became challenging. This study aims to examine the trends in public discourses on Twitter on these interactions between the pandemic and environment. The present study follows the most recent literature on understanding public perceptions — which acknowledges Twitter to be an abundant source of information on public discussions on any global issue, including the pandemic. A Python-based code is developed to extract Twitter data spanning over a year, and analyze the presence of covid-environment related keywords and other attributes. It is found that the Twitterati aggressively viewed the impacts (such as economic slowdown and high mortality) of the pandemic as miniatures of the results of future climate change. The community was also highly concerned about the varying air and plastic pollution levels with the change in lockdown and covid prevention policies. Extreme weather events were a high-frequency topic when they impacted countries such as India, the USA, Australia, the Philippines and Vietnam. This study makes a novel attempt to provide an overview of public discourses on the pandemic-environment linkage and; can be a crucial addition to the literature on assessing public perception of environmental threats through Twitter data mining.

Graphical abstract

1. Introduction

The COVID-19 pandemic has adversely affected most countries around the world. Along with its direct impacts on the health of populations, it has had indirect repercussions for our social, economic, political and cultural setups. The pandemic has influenced or interacted with our environmental system as well (Zambrano-Monserrate et al., 2020). For example, in the short-term, the pandemic had led to cleaner air, water and thriving biodiversity because of the restrictions (e.g., lockdown) in human/industrial activities (Sarkar et al., 2021). However, as the pandemic progressed, there have been concerns about rising plastic pollution due to littering from disposable masks and other safety equipment (Patrício Silva et al., 2021). Thus, it seems that the pandemic has had both positive and negative links with the environment. As in the case of any other global subject of concern, there have been numerous dialogues about this pandemic-environment association among the general public on different social media platforms. Currently, the world is interconnected through various social media hubs and information dissemination is quick. It is seen that, since the pandemic has affected majority of the global population directly or indirectly, the public has actively participated in opinionating about the different aspects of the pandemic. This includes the link or impact of the pandemic on the environment. An analysis of the public discussions about COVID-19 and the environment can reveal the variety of perceptions that have developed regarding their linkage. Such studies can add to the vast literature on examining public perceptions of global issues of concern, such as climate change and the environment (Jang and Hart, 2015; Kirilenko and Stepchenkova, 2014), and can have implications for policies initiating public action.

Among the many social media platforms, Twitter is one of the most widely used to share snippets/texts of information and hence, can be an important source of data on public discussions. Earlier, public perception could be examined solely through surveys. However, with the advent of Twitter (the microblogging platform), ‘big data’ on public opinion could be accessed and assessed without designing a primary survey. Twitter is now a crucial and comprehensive source for understanding public perceptions. Hence, the recent literature has abundance of studies on examining public perceptions of a variety of issues, such as vaccine (Keim-Malpass et al., 2017), swine flu pandemic (Chew and Eysenbach, 2010), cancer (Modave et al., 2019), presidential election (Morris, 2018), smoking (Myslín et al., 2013), eateries (Park et al., 2016) and consumer engagement (Read et al., 2019), using Twitter data. In the context specific to the environment, there are studies that have assessed the importance or frequencies of the topic of climate change on Twitter (Kirilenko and Stepchenkova, 2014), tweeting activity during extreme weather events such as, warm spells and hurricanes (Kirilenko et al., 2015), Twitter discussion on air pollution (Chen et al., 2017), marine plastic pollution (Otero et al., 2021), and biodiversity threats (Barrios-O’Neill, 2021).

During the ongoing COVID-19 pandemic, numerous studies have used Tweets (microblogs written on Twitter) to examine the spread of misinformation (Kouzy et al., 2020; Shahi et al., 2021), and the impact of the pandemic on people's emotions (Xue et al., 2020), mental health (Guntuku et al., 2020), human mobility (Huang et al., 2020), etc. It is seen that after the start of the pandemic, a large volume of the literature has used Twitter data to assess the linkage of COVID with various concerns, but not with the environment. There seems to be limited studies on examining opinions on Twitter about environmental stresses, such as pollution and climate change during the pandemic. This is in exception to a few works published recently, such as (Loureiro and Alló, 2021; Smirnov and Hsieh, 2022), that have examined the impact of the pandemic on climate change discussions. However, there has been no aggregate assessment of Twitter discussions on the multiple aspects of the environment (apart from climate change) during the pandemic.

Therefore, the present study aims to analyze the public discourses that have taken place on Twitter since the beginning of the pandemic regarding the environment — particularly tweets that link the pandemic to various environmental threats and disasters. The aim of this manuscript is in continuation to previous studies (published especially before the pandemic) focusing on understanding Twitter discussions about environmental threats and concerns (as previously mentioned such as, Kirilenko et al., 2015; Kirilenko and Stepchenkova, 2014). This study develops a Python-based algorithm to download and analyze information or “big data” from Twitter; and hence can also be an addition to the developing domain of computational social science. Further, the study attempts to incorporate a long-term dataset that spans more than a year, which again has not been done in previous studies that have used Twitter dataset to examine aspects related to the pandemic.

2. Methodology

The present study accesses tweets or microblogs on Twitter on the COVID-19 pandemic and analyzes the presence of various keywords related to the environment. The methodology is divided into three steps: data download, data processing, and data analysis (described in the following sub-sections). A pictorial representation of the methodology is given in Fig. 1 .

Fig. 1.

Methodology followed in the study.

2.1. Data download

Tweets are the atomic blocks of Twitter. Each tweet is represented by a tweet ID which can be used for downloading the tweet text and related attributes. It is to be noted that Twitter's policy restricts public distribution of Twitter data, and only tweet IDs are available in the public domain. A developer account is required to access the tweets through these IDs. The Twitter streaming API (application programming interface) allows to mine/download tweets (Twitter data) of any user through the Twitter developer account. Each tweet is a JavaScript Object Notation (JSON) object in Twitter. The Twitter API returns tweet objects in JSON format as a collection of key-value pairs with named attributes and respective values. This study uses six main attributes: tweet_id, full_text, retweet_count, entities-hashtags, user-verified, and user-location (Table 1 ) (Twitter, 2021).

Table 1.

Description of the several attributes of the tweet JSON object.

Attribute	Type	Description
tweet_id	Integer	Unique identifier of a tweet object.
full_text	String	Regular expression of tweet text in UCS Transformation Format 8 (UTF-8).
retweet_count	Integer	Number of times a tweet is retweeted
entities-hashtags	An array of hashtag objects	Hashtags are metadata used to classify/label tweets. The hashtag is an attribute of object name ‘entities’ (a child of the tweet object) that provides parsing conveniences for tweet objects.
user-location	String	The location is an attribute of the object name ‘user’ (a child of the tweet object). It identifies the location of the user. The location is available if a user shares it while tweeting. Otherwise, by default, the location is the home/account location provided by the user. Please note that many tweets may not have any specified user location. Furthermore, the search device may fuzzily interpret the location of the user.
user-verified	Boolean (True/False)	Indicates whether the user account is verified (true) or unverified (false).

In this study, a Python library, namely Twarc, is used to access the Twitter API for collecting and storing tweet objects in JSON format while using the dehydrated tweet IDs of COV19TweetsDataset (Lamsal, 2020). The COV19TweetsDataset has comprehensively documented daily scale tweet IDs related to the COVID-19 pandemic from around the world. In the present study, the dehydrated tweet IDs for 409 days (March 21, 2020, to May 02, 2021) are used to archive tweet objects.

Archiving tweet objects in JSON format from the tweet IDs using Twarc is known as the hydration of tweets and is a computationally expensive process. The tweets produced each day are hydrated from tweet IDs and saved in 409 text files in JSON format for further processing and analysis. The total daily tweet IDs in COV19TweetsDataset vary from 288,277 to 4,774,322. However, archiving tweet objects in JSON format and storing them in a text file with a very high number of tweet IDs is computationally very expensive. For example, hydration of 100,000 tweet objects and saving them in a text file may take 2 hours of computational time and 3 gigabytes (GB) of memory space. Hence, to reduce the computational burden, an upper limit of total downloadable tweet objects is set to 600,000.

Thus, from the COV19TweetsDataset, 600,000 tweet objects per day are hydrated if the total tweet IDs exceed the upper limit. Otherwise, the exact number of tweet objects in the COV19TweetsDataset is hydrated. While hydrating tweets, each tweet_id is randomly sampled from the collection of tweet IDs without repetition to avoid time locality. Since every country has a different time zone, Twitter users from various regions will certainly tweet at their own comfortable time. The random selection of tweet_ids thus aims to avoid biases towards any country while archiving tweet objects.

2.2. Data processing

The collection of tweet objects in text files in the form of JSON is not ideal for the statistical analysis of the attributes. Hence, it is required to extract the value of those attributes from the JSON objects and store them in separate text files. Doing so will make it convenient to carry out the statistical analysis of each attribute individually. Thus, the values of five attributes (full_text, retweet_count, entities-hashtags, user-verified, and user-location) from each JSON text file are extracted and stored into five separate text files. For example, the text files extracted from JSON text file “Covid_JSON_Text_ID.txt” are named as “full_text_ID.txt”, “retweet_count_ID.txt”, “entities_hashtags_ID.txt”, “user_verified_ID.txt”, and “user_location_ID.txt”. In the file names, the extension “ID” is used to designate a particular day of the daily-basis dataset, which ranges between [1, 409]. While storing the values of an attribute in the text file, the respective tweet_id is also stored in the same file in a separate column. The tweet_id is a unique identifier of a tweet object which is beneficial to cross-reference/link five other attributes during statistical analysis.

2.3. Data analysis

In the first step, the text of “full_text_ID.txt” and “entities_hashtags_ID.txt” files are cleaned up by removing punctuations and symbols and converting upper case letters to lower case. This study considers 17 keywords/phrases (Table 2 ) representing various environmental threats, dialogues around climate risk (and response) and extreme weather events. These have been filtered from almost 40 keywords which were compiled based on authors’ subjective judgement formed by going through generic news reports and social media posts about the impacts of the pandemic on the different environmental aspects. Many of these keywords were related and later, it was found that they were not at all present (or present in negligible numbers) in the tweets concerning the present study. Hence, they were dropped. We believe that this process of having a large pool of keywords from which less than half were finally considered has enabled the study to be inclusive of the significant keywords.

Table 2.

Keywords representing various environmental threats, climate change, and extreme weather events.

S.No.	Category	Fulltext	Hashtag
1.	Environmental threats	air pollution	airpollution
2.		water pollution	waterpollution
3.		plastic pollution	plasticpollution
4.		deforestation	deforestation
5.		biodiversity loss	biodiversityloss
6.		wildlife trade	wildlifetrade
7.	Related to climate change	climate change	climatechange
8.		climate emergency	climateemergency
9.		climate risk	climaterisk
10.		climate resilience	climateresilience
11.		climate adaptation	climateadaptation
12.		climate mitigation	climatemitigation
13.		carbon emission	carbonemission
14.	Extreme weather events	cyclone, typhoon, hurricane	cyclone, typhoon, hurricane
15.		flood	flood
16.		bushfire, bush fire	bushfire
17.		drought	drought

The frequency of occurrences of each keyword is counted in both files using computer programs. It is to be noted that there should be no space in a two-worded key phrase while searching it in the hashtag file. Also, in the case of the keyword cyclone, synonyms such as typhoon and hurricane are also considered. Moreover, for each keyword occurrence, the respective retweet counts, the user location, and user account status (verified or unverified) are also assessed. The exact steps associated with data analysis are explained as follows.

2.3.1. Analysis of full text

Start with a single keyword mentioned in Table 2. In the “full_text_ID.txt” file, count the number of times the keyword is present and save the total count in a CSV file. Also, archive the respective tweet_ids whenever the keyword is found in the “full_text_ID.txt” file. Repeat the process for all the keywords in Table 2.

2.3.2. Analysis of hashtags

Like the previous step, find the occurrences of a keyword (Table 2) in “entities_hashtags_ID.txt” file and save the total counts in a CSV file. Archive the corresponding tweet_ids in a separate text file. Repeat the process for all the keywords in Table 2.

2.3.3. Analysis of retweet counts

Search for archived tweet_ids (found while analyzing full text or hashtags) and corresponding retweet counts in the “retweet_count_ID.txt” file.

2.3.4. Analysis of user whether verified or unverified

Search for archived tweet_ids in the “user_verified_ID.txt.” file and find whether they are from verified or unverified accounts.

2.3.5. Analysis of user location

The analysis of user location is tricky and explained below in two steps. First, search for archived tweet_ids in the “user_location_ID.txt” file and store the corresponding locations in a text file named “user_location_ID_Step1.txt”.

The locations are usually the countries, cities, or districts/provinces/counties in the “user_location_ID.txt” file. Hence, determining the country to which a tweet_id belongs is not straightforward. Thus, we created a location dictionary containing the name of countries, major cities, and provinces/districts. Each element in the dictionary is a country whose sub-elements are its major cities/provinces/districts. This study considers 32 countries, as listed in Table 3 . The countries are selected if they have some of the highest numbers of Twitter users and if English is one of the major languages of the population (Statista, 2021).

Table 3.

List of countries considered in the study.

Countries selected based on leading Twitter users	Countries selected if English is one of the major languages of the population
United States, Japan, India, Brazil, United Kingdom, Indonesia, Turkey, Saudi Arabia, Mexico, Thailand, France, Philippines, Spain, Canada, Germany, South Korea, Argentina, Egypt, Malaysia, Colombia,	Australia, Austria, Bangladesh, Belgium, Italy, Netherlands, Nigeria, Poland, South Africa, Switzerland, Uruguay, Pakistan

In the second step, count the number of times an element (country) and its sub-elements are present in the “user_location_ID_Step1.txt” file. Save the results in a CSV file. Then, repeat the steps for all the elements (countries). Please note that the process of counting user locations corresponding to archived tweet_ids is not 100% precise. It may be possible that two cities located in two different countries have the same name. For example, India and Pakistan have a city with the same name as Hyderabad. Moreover, the directory will miss a count if the location set by the user in the “user_location_ID.txt” file is a small town or village. However, these issues with regard to location are limited and hence, the usage of this attribute is continued.

3. Results and discussion

3.1. Frequency of keywords

3.1.1. In tweet texts

The top three occurring keywords are flood, cyclone and climate change (Fig. 2 ). Their high occurrence signifies that the interaction of these topics with the pandemic was of concern among the Twitterati. For example, flood/cyclone events during the pandemic raised concerns regarding the evacuation and social distancing of victims, and the lockdown created difficulties in transporting aid. Further, the peaks in Fig. 3 (a) and (b) show that the use of the keywords flood and cyclone/typhoon/hurricane may have increased when they struck any country. The peaks in the month of August in Fig. 3(a) are because of massive floods affecting different states in India. For cyclone, the increase in frequency in August is because of Hurricane Laura (of category 4) that affected the USA (Fig. 3 (b)). The highest peak can be observed on 13^th November 2020 when typhoon Vamco (which affected the Philippines and Vietnam) laddered up into a category 4-equivalent typhoon, indicating that the Vamco event received the most attention among all cyclones on Twitter. The pandemic also initiated people to draw parallels with the climate change crisis. Hence, climate change seems to be a top-occurring word in the tweets. The peak in November in Fig. 3(c) is primarily because of tweets criticizing the president of the Philippines for the devastating impacts of cyclones (which is seen to be a result of climate change) in the country. The peak in climate change frequency in January is because of the numerous retweets of the White House's statement that President Biden is taking action on the climate crisis (along with COVID-19, the resulting economic crisis and racial inequity).

Fig. 2.

Total frequencies of different key words/phrases in the daily tweets on COVID-19 from March 20, 2020, to May 02, 2021.

Fig. 3.

Frequency of the eight highest occurring keywords/phrases, that is, (a) flood, (b) cyclone, (c) climate change, (d) air pollution, (e) bushfire, (f) drought, (g) deforestation, and (h) carbon emission in the daily tweets on COVID-19 from March 20, 2020, to May 02, 2021.

Once the COVID-19 infections started to climb, countries went under lockdown. The closing of businesses, offices and industrial activities during the lockdown significantly lowered pollution, specifically that of air (Singh et al., 2020; Venter et al., 2020). The improved air quality across the world must have been intensely discussed on Twitter, resulting in ‘air pollution’ being one of the most mentioned phrases, especially a few weeks into the lockdown in April 2020 (Fig. 3(d)). Air pollution was also linked to greater vulnerability to COVID-19 infections in many of the tweets (Wang et al., 2020). In September, after lifting of the lockdown in many countries, air pollution levels increased, which again has been tweeted vigorously. The peak in the conversations on pollution in November was because of deteriorating air quality in Delhi, India, due to the bursting of crackers for the festival of Diwali. Around this time, Delhi had also seen an increase in the number of COVID-19 cases.

Next, the extreme events of bushfire and drought are the most mentioned. In the first half (January–May) of 2020, Australia was severely impacted by bushfire. Bushfire again affected parts of Australia for a week in February 2021. The numerous peaks in Fig. 3(e) show that the keyword bushfire has been used in many tweets across the study period. Some of the important discussions on Twitter were about the bushfire and the pandemic resulting in an economic recession in Australia, the government's inefficiency and the need for action towards managing the multi-disasters of bushfire, flood, drought and rodent plague. The various peaks (Fig. 3(f)) in tweet frequency on drought were majorly with regard to those experienced by USA and Australia.

The other frequently occurring keywords are the following in decreasing order: deforestation, carbon emission, climate emergency, wildlife trade, plastic pollution, biodiversity loss, water pollution and climate risk.

The many peaks regarding deforestation (Fig. 3(g)) show that this topic has been a consistent part of tweets during the pandemic. It is said that the lockdown catalyzed illegal deforestation. The highest peak (on 17–18 May 2020) is mostly because of tweets condemning illegal deforestation in Brazil and clearing of deforestation projects in India.

The peaks in carbon emission (Fig. 3(h)) are mostly because of tweets highlighting scientific papers/reports on calculating carbon emissions during the pandemic. On 19 May 2020, a study (Le Quéré et al., 2020) that claimed a 17% reduction in global CO₂ emission (in April 2020 compared to values in 2019) due to the lockdown garnered a lot of attention on Twitter. This resulted in the peaking of the keyword carbon emission (Fig. 3(h)). Similarly, on 12 December, another study (GCP, 2020) showing a decline in carbon emissions by 6.7% in 2020 gained popularity on Twitter. Again, on 3 March 2021, the IEA's (International Energy Agency) report, stating that carbon emissions have increased (2% in December 2020 compared to that in 2019) after the global lockdown (IEA, 2021), was highly shared on Twitter resulting in the peak on the day in Fig. 3(h).

Figures showing the daily frequency of the rest of the key words/phrases are included in the Supplementary Material (Fig. S1). Fig. S1 (a) shows that climate emergency also peaked in a few instances. A peak on 18 October 2020 for climate emergency was observed as there were many retweets stating there was no climate emergency. On the other hand, the peak of climate emergency on 24 October 2020 was initiated by the USA's presidential debate, wherein climate action plans by both candidates were seen to be unsatisfactory. On 22 January 2021, an informative tweet explaining the Paris Agreement by the UN (United Nations) was largely shared, which again resulted in numerous tweets containing the phrase 'climate emergency.'

The phrase wildlife trade was frequently used at the beginning of the pandemic (Fig. S1(b)) because it was the alleged source of the pandemic. For example, on 23 April and 7 May 2020, there were many tweets discussing the link of the pandemic to wildlife trade and its need to be banned. On 22 May 2020, there was a peak in its usage because China put restrictions on wildlife trade, and EU (European Union) planned to take action against illegal wildlife trade. Similarly, on 6 June 2020, there was a campaign to end illegal wildlife trade. On 25 July 2020, there was a peak because Vietnam banned wildlife trade. Tweets on banning the trade were again making rounds the following year on 5 March 2021. Further, on 30 March 2021, the WHO (World Health Organization) raised the discussion on the link of the pandemic to wildlife trade, thus again shooting up the use of the phrase in the tweets.

As the pandemic progressed into a few months after March 2020, concerns about plastic pollution resulting from safety gear used to prevent infection started to rise. This is evident from the peak in the usage of ‘plastic pollution’ on 25 June 2020 (Fig. S1(c)). The peak on 9 August was because of the same concerns. Additionally, there were some tweets highlighting a scientific paper (Prata et al., 2020) regarding the proliferated use of plastics during the pandemic and mismanagement of its disposal. Again, on 19 March 2021, there were several retweets of UNEP's Executive Director's statement that marine plastic pollution has risen by ten times since 1980 (IPBES, 2021).

On a few days, there were peaks about biodiversity loss (Fig. S1(d)). On 6 June 2020, many netizens tweeted that the pandemic has shown that we need to take action on climate change and biodiversity loss. Nature is interlinked and harming any of its aspects can lead to alarming repercussions. On 31 October 2020, the Twitterati expressed that biodiversity loss can lead to a pandemic, and activities that lead to climate change and biodiversity loss can also lead to a pandemic.

An investigation of the peaks in ‘water pollution’ shows that the Twitterati has not directly linked it to the pandemic. The tweets merely talk about generic water pollution. For example, there are tweets slamming the US President and Delhi government (in India) for water pollution in their respective regions.

‘Climate risk’ is synonymous with ‘climate change’; however, it has not been used highly by the Twitterati. Climate risk peaked once when Greenpeace tweeted (on 7 September 2020) that economic recovery post-COVID cannot be resilient without considering climate risk. It peaked next on 16 January 2021 when the Pope's scientific advisor stated that the pandemic has ‘highlighted existential climate risk.’

It seems that technical combination words, which are mostly used by the academic/scientific community, such as climate-resilience, climate-adaptation and climate-mitigation are comparatively low in frequency. The peaks in these limited tweets speak about covid emphasizing the need to improve resilience to climate change, covid as a challenge for disaster relief and climate adaptation, and including climate mitigation and adaptation plans (and funding) within covid recovery.

3.1.2. In tweet hashtags

Fig. 4 shows the frequency of usage of the various hashtags. The rankings of the highest occurring hashtags are not precisely similar to that of keywords/phrases in the text (discussed in section 3.1.1). For example, ‘Climatechange’, which was the third highest occurring keyword in case of text, is the highest tweeted hashtag. However, these slight changes in the rankings do not seem to have any severe implications for this study. The generic reasons behind the usage of these keywords in the tweet text and hashtags remain almost the same. Hence, the reasonings already described in section 3.1.1 are not repeated in this section.

Fig. 4.

Total frequencies of different hashtags in the daily tweets on COVID-19 from March 20, 2020, to May 2, 2021.

‘Climatechange’ being the top used hashtag, suggests that the Twitterati overwhelmingly linked the COVID crisis with climate change. There have been many discussions where it is urged that the effects (such as the meltdown of the health and economic systems) of the pandemic should be taken as a sample or warning against the more devastating impacts of climate change that will be experienced shortly in the future. Further, Fig. 5 (a) shows that the hashtag ‘climatechange’ spiked up numerous times in the year.

Fig. 5.

Frequency of the eight highest occurring hashtags, that is, (a) climatechange, (b) cyclone, (c) climateemergency, (d) airpollution, (e)flood, (f) bushfire, (g) plasticpollution, and (h) wildlifetrade in the daily tweets on COVID-19 from March 20, 2020, to May 02, 2021.

The next highest occurring hashtags are cyclone, climateemergency, airpollution, flood, bushfire and plastic pollution. The usage of the hashtag cyclone spiked up when some of the biggest cyclones affected countries (Fig. 5(b)). For example, cyclone Amphan in eastern India and Bangladesh on 22 May 2020, cyclone Nisarga on 4 June 2020 in western India, and Hurricane Laura on 28 Aug in parts of Central America and the USA. Next, the hashtag ‘climateemergency’ aggressively peaked when the Fijian Prime Minister congratulated Biden (on 7 Nov 2020) on becoming the US President and highlighted that US is required to fight climate emergency (Fig. 5(c)).

As mentioned earlier, there were many tweets emphasizing the increased vulnerability to COVID infections because of air pollution. The use of the hashtag ‘airpollution’ peaked in October–November 2020 when air pollution started rising after the lockdown (Fig. 5(d)). Some tweets also mentioned scientific studies (Pozzer et al., 2020) providing evidence of the same.

The usage of the hashtag ‘flood’ peaked mostly when a flooding event took place in India (Fig. 5(e)). For example, Assam floods on 20 July, Hyderabad floods on 15 October, and Chennai floods on 27 November 2020. The hashtag bushfire, which mostly relates to those that happened in Australia, has been used in different frequencies across the study period (Fig. 5(f), similar to insights discussed in section 3.1.1). The hashtag plasticpollution was highly used (Fig. 5(g)) when there was an incidence of a penguin dying after swallowing a plastic facemask (Srikanth, 2020), amplifying the evils of plastic pollution.

After plasticpollution, the hashtags wildlifetrade, deforestation, drought, carbonemission and biodiversityloss are the most used. The hashtag wildlife trade spiked (Fig. 5(h)) on the World Wildlife Day observed on 3 March 2021. The daily frequencies of the rest of the hashtags are shown in Fig. S2 in Supplementary Material. The peaks in deforestation are because of the widespread and illegal cutting down of the Amazon forests in Brazil. The usage of drought peaked significantly on the Desertification and Drought Day on 17 June 2020. Some of the peaks in carbonemission usage are because of tweets sharing results of scientific studies calculating the increase/decrease in emissions (as mentioned in section 3.1.1). Biodiversityloss peaks were because of general concern of the Twitterati regarding it. The rest of the hashtags, that is, climaterisk, climateresilience, climateadaptation, waterpollution and climatemitigation are very few in number. However, there was a small peak in climaterisk when climate activist Greta Thunberg resumed her strikes (against climate inaction) maintaining covid social distancing norms on 25 September 2020. Further, climateadaptation peaked on 22 January 2020 because of retweets of news reporting that the AstraZeneca vaccine factory needs UK emergency services to prevent its flooding.

3.2. Retweet count

Fig. 6 shows the average retweets of those containing the different key words/phrases. The tweets with the keywords cyclone, climate change and flood have been retweeted the most on average. Fig. 6 shows that the average retweets of other keywords are not that high compared to the top three keywords. It is also seen that the order (highest to lowest) of average retweets of the keywords is somewhat similar to the order of the most frequently used keywords in tweets (Fig. 2). That is, as expected, large volume of retweets has led to peaks in the corresponding keyword frequency.

Fig. 6.

Average retweets of those containing the keywords from March 20, 2020, to May 02, 2021.

3.3. Tweets from verified accounts

Although climate resilience, climate adaptation and climate risk are some of the least tweeted key words/phrases, Fig. 7 shows that they have been tweeted the most (12.39%, 11.85% and 11.19%, respectively) by verified accounts. These accounts seem to belong to climate awareness and action groups. The next highest tweeted keywords by verified accounts are biodiversity loss, plastic pollution, carbon emission, drought, air pollution, climate emergency, deforestation, bushfire and wildlife trade. The least tweeted keywords by verified accounts are climate change, cyclone, flood, climate mitigation and water pollution.

Fig. 7.

Percentage tweets (containing the keywords) by verified Twitter accounts.

3.4. Country of origin of the tweets with the keywords

The highest number of tweets with the keywords considered in the study are from the USA and India (as they have some of the highest numbers of Twitter users) (Fig. 8 ). The hashtags flood, cyclone, air pollution, bushfire and drought are used by the countries that are most affected by these extreme events (Table S1 in Supplementary Material). For example, cyclone is most used in tweets from the USA and India, air pollution is used most in India and the UK, flood in India, bushfire in Australia, and drought in Australia and the USA. Tweets with keywords climate change, climate emergency, climate risk, climate resilience, climate adaptation and climate mitigation are the highest from the USA and UK (Table S1 in Supplementary Material), indicating that these countries may be most concerned about this wicked problem. The USA and UK seem to be the most concerned about other environmental threats as well; as they have the highest tweets using the keywords deforestation, carbon emission, wildlife trade, plastic pollution and biodiversity loss. Tweets regarding water pollution were the most from India.

Fig. 8.

Total tweets containing the keywords considered in the study from the countries.

4. Some observations and discussion

4.1. Pandemic-climate change parallels

The high occurrence of the keyword climate change in the tweets (in the text as well as in hashtag) suggests that the Twitterati drew parallels between the pandemic and climate change. There have been numerous tweets indicating that the impacts of climate change (on human health, economy, etc.) will be far more disastrous than the pandemic, and humankind should be taking serious action to mitigate climate change. Consequently, there have been tweets tagging political and authoritative figures urging action on climate change during the pandemic. On the other hand, there have also been a few opposing tweets that question the reality of climate change.

Studies (Loureiro and Alló, 2021; Smirnov and Hsieh, 2022) have showed that Twitterati's attention on climate change had declined during the pandemic. However, within this limited activity, in this study, it is found that the general perception (about the graveness of climate change) aligns with the comments of the scientific community and hence, highlights the scientific temperament of the Twitter masses. There have been studies that have opinionated that there are similarities between the impacts of pandemic and the climate crisis (Manzanedo and Manning, 2020), and consequences of climate change can be worse (Mattar et al., 2021). There have also been comments that climate change can bring about unforeseen spread of other infectious diseases (Bernstein, 2020). Hence, studies have suggested that climate adaptation plans should strategize the management of pandemics in the future (Phillips et al., 2020). However, it is to be noted that technical keywords, that are generally used by the scientific community, such as, climate mitigation and adaptation, were extremely low in frequency. This shows that the masses on Twitter may not be much aware of technical terms but understand that the future may be grim because of climate change.

4.2. Linking environmental pollution with the pandemic

Air pollution and carbon emissions were some of the most discussed environmental threats during the pandemic. Studies have shown that the lockdowns during the pandemic led to a drastic decline in air pollution (Singh et al., 2020; Venter et al., 2020) and carbon emissions (Le Quéré et al., 2020). However, they increased (IEA, 2021) once the lockdowns were lifted. Such radical lows and sudden highs in air pollution during and after the lockdown, respectively, must have garnered attention on Twitter. The high usage of the keywords air pollution and carbon emission suggests that the Twitterati was concerned and has closely monitored the air quality in their capacity as well as through scientific studies. Further, studies have also shown that air pollution can be linked to increased COVID-19 transmission (Wang et al., 2020), which again may have led to increased attention on air pollution. On the other hand, while air pollution and carbon emissions decreased, plastic pollution aggravated during the pandemic due to the increased use of personal protection equipment (such as face masks and surgical scrubs) (Patrício Silva et al., 2021). This, again, has been a highly discussed topic on Twitter.

In comparison to the number (and quality) of studies on assessing the impact of lockdown on air pollution, the scientific literature does not seem to provide global evidence regarding the decline in water pollution during the lockdown (except for a few regional studies, e.g., B. Chakraborty et al. (2021) and Chakraborty et al. (2021)). This may have resulted in the low attention on water pollution on Twitter.

4.3. Interest in scientific studies

The discussions in sections 4.1 and 4.2 also highlight that topics on which scientific literature is abundant have been popular on Twitter. For example, as discussed earlier, since many have quantified changes in air pollution and carbon emissions during the pandemic, Twitterati seems to have gained and popularized insights from these studies. On the other hand, as scientific assessments of water pollution were scarce (at the beginning of the pandemic, compared to air pollution), the keyword was not high in frequency in the tweets. This shows that the netizens were interested in scientific studies related to the pandemic-environment linkage.

4.4. Coinciding extreme weather events with the pandemic and biases toward events

Many regions have experienced extreme weather events and disasters during the pandemic. Social distancing mandates and travel restrictions had made disaster management quite challenging (Ishiwatari et al., 2020, Malakar and Lu, 2022), particularly at the beginning of the pandemic, around March 2020. The use of keywords depicting extreme events such as floods, cyclones and bushfires peaked when they affected parts of the world. This is especially the case when they hit regions that are populated or have a high number of English-speaking Twitter users, such as the USA, India and Australia. A similar trend, that is, heightened Twitter mentions, has been observed by Kirilenko and Stepchenkova (2014) in case of the extreme event, Hurricane Sandy, in the USA (before the pandemic). On the other hand, droughts (and consequently food insecurity) which have been ravaging many parts of the African continent (such as, Angola, Zimbabwe), have not been much part of Twitter discussions. The low attention to the African droughts may also suggest the disparity in accessing Twitter or internet services between African countries and the rest of the world.

4.5. Error analysis

In order to gauge the errors in frequencies of the different key words/phrases in the tweets reported in the study, the true frequencies are calculated — for the top six occurring keywords (that is, flood, cyclone, climate change, air pollution, bushfire and drought) on six randomly selected days. Thereafter, the percentage difference between the true and extrapolated frequencies, that is, the error, is computed. This error introduced in the analysis because of the use of extrapolated data is one of the limitations of the study. A box plot of the error percentages is shown in Fig. 9 . On average, the error is 4.23%; that is, it is less than 5% and seems to be statistically acceptable. Previous studies using extrapolated Twitter big data have not explicitly stated the error percentages, which makes comparison of the values with the literature difficult. However, it is suggested that when the sample size is large, as in this study using big data, sampling errors are small and sufficiently acceptable (Dillman et al., 2009).

Fig. 9.

Box plot representing errors between extrapolated and true frequencies.

5. Conclusion

The COVID-19 pandemic has affected almost all countries of the world and is a global concern. During this pandemic, countries have also been simultaneously affected by various environmental threats, such as air and plastic pollution, and extreme weather events, such as floods and cyclones. This study uses tweets to understand the trends and perceptions of the global public regarding these threats and extreme events during the first year of the pandemic (since its beginning in March 2020). The present study follows the approach of the literature of the past decade, wherein Twitter has become a popular data source for understanding public perceptions about issues concerning the global (or large) population. The contribution of the present study lies in its objective to provide an analysis of the public perceptions about the links between the pandemic and human-environment system — which has not been attempted previously in the literature.

The COVID-19 pandemic initially slowed down human activity across the globe, which consequently may have reduced pollution and somewhat reduced the burden on our environment. It also brought to the forefront discussions about the looming evil of climate change and ways for sustainable development. Later on, the COVID-pandemic has given rise to plastic pollution because of the use of disposable masks and other safety equipment. Air pollution also increased once the lockdown was lifted and it was business as usual. The present study reveals that the Twitterati was the most concerned about climate change, air and plastic pollution during the pandemic. They drew climate change-pandemic parallels and monitored the change in air and plastic pollution with the change in lockdown policies. This shows that there is a heightened level of awareness about climate change and environmental pollution among the Twitter masses. This is a positive indication that can perhaps direct more aggressive climate mitigation policies globally in the future. Further, when countries were impacted by floods, cyclones and bushfires, the chatter on Twitter regarding these extreme events increased. Many tweets recognized the challenges in evacuating and providing aid to disaster-hit regions during a pandemic. Next, wildlife trade, deforestation and biodiversity loss are some of the other issues that the Twitter masses were concerned about. Technical terms such as mitigation, adaptation and resilience to climate change were not popular. However, the Twitterati seemed to be interested in scientific studies which quantified the increase or decrease in carbon emissions in a few instances. Further, Twitter users of the USA and UK seemed to be most environmentally concerned as there was colossal tweet traffic (containing the selected key words/phrases in this study) from these countries.

Overall, the present study provides a glimpse of the trends in Twitter chatter regarding environmental and climate problems during the COVID-19 pandemic. The study focuses on the environment-pandemic linkage and the insights can be a valuable addition to the computational social science literature that mine Twitter/big data to explain public perceptions and behavior. This can consequently feed into global policy discussions and debates on consciousness about environmental threats. The insights provided in the study can be a starting point of discussion about public perceptions of different environmental risks, particularly during a global emergency, such as, the pandemic. The study highlights that data from social media platforms, which is generally publicly available and free, can be an eminent way to understand global (that is, large-scale) views about issues. Further, the results of the present study are part of social media analytics, and hence, social media handles associated with providing information and generating awareness about environmental change may benefit from them. Lastly, the present study can be complemented by sentiment analysis of the tweets constituting the different environment-related keywords. This can enable further understanding of the connotation (whether positive or negative) under the tweets and pulse of the Twitterati regarding the environment.

Credit author statement

Krishna Malakar: Conceptualization, Methodology, Formal analysis, Writing – original draft and revision, Partha Majumder: Methodology, Software, Data Curation, Coding, Formal Analysis, Writing– original draft, Chunhui Lu: Fund acquisition, Resources, Writing – review & editing, supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

The authors do not have permission to share data.