A decade of German heatwaves evident in social media data

doi:10.21203/rs.3.rs-4497852/v1

Download PDF

Research Article

A decade of German heatwaves evident in social media data

https://doi.org/10.21203/rs.3.rs-4497852/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Heat is a global public health risk which kills many people every year. This is true even in temperate climates like that of Europe which has experienced an increasing number of intense heatwaves in the last decade. This study uses a unique dataset of more than 57,000 X (formerly Twitter) conversations from users in Germany about heatwaves over the period 2013 to 2022. Structural topic modelling revealed nine topics, with an increase in the number and diversity of topics since 2018, a year of unprecedented drought and heatwaves in Europe, including Germany. Before 2018, German tweets related to heatwaves primarily expressed concern about heatwaves on a global scale, such about wildfires in Australia. In 2015 there was brief spike tweets in which people began to comment on the local impacts of heat on their lives. Topics in 2015 included complaints about heat, air-conditioning shortages and heat-related train delays. From 2018, however, conversations have proliferated about local heat impacts and personal adaptations along with frequent heat warnings and advice on how to cope with and relieve heat. European droughts and wildfires also gained prominence from 2018 onwards, with concern about heat and drought shifting from being a problem outside Europe to one affecting many European countries, including those frequently visited by Germans as tourists.

drought

European summer

heat relief

temperature records

sentiment analysis

text mining

tweets

Average temperatures are predicted to be at least 2.4 degrees higher than pre-industrial levels by 2100, and heatwaves are set to become longer and more frequent (IPCC 2021). As it is, heat is already a major public health concern (Capon et al. 2019; Ebi et al. 2021; Jay et al. 2021). Extreme temperatures cause illness directly or exacerbate existing illnesses (Bi et al. 2011; Jay et al. 2021) and have led to the deaths of about 30 million people in the last thirty years (Vicedo-Cabrera et al. 2021). Heat also reduces labour productivity (Zander et al. 2015), increases violence (Heilmann et al. 2021) and compromises general well-being (Zander et al. 2017). People living in cities, currently 56% of the world population and rising (UN-Habitat 2022), are particularly vulnerable to heat because of the urban heat island effect (Wouters et al. 2017). There is therefore increasing recognition of the need for research on adaptation to the direct effects of heat on human society. Such adaptation to heat can occur at a personal level (e.g. drinking, resting, shifting activities to cooler times of the day), at a household level and at an institutional level in terms of policy and infrastructure (Turek-Hankins et al. 2021).

While research on how people are learning to cope with increased heat (Bakhsh et al. 2018; Jay et al. 2021; Zander et al. 2021; Osberghaus and Abeling 2022) is increasing, far less is known about the extent of heat stress in the wider society and societal views on longer and increasingly frequent heatwaves. Climate change is not the only reason for increased heat exposure - urban population growth has increased heat exposure by nearly 200% in recent decades (Tuholske et al. 2021). Under the worst-case climate scenario, half of the world’s population could be exposed to unprecedented heat by 2080 (Lenton et al. 2023). However, exposure to heat is not the same as heat stress because many people are acclimatised to heat, and/or take appropriate heat relief measures (Guardaro et al. 2022). There is also evidence that the consequences of heatwaves are felt more strongly in regions where extremely hot days are rare than in regions that are generally warm or hot, such as those with a tropical climate (Zander et al. 2023d).

Most scientific evidence on the societal burden from heat is limited to heat health impacts (Faurie et al. 2022), particularly hospital data. However, only extreme cases require hospitalisation. Data on people affected by heat who do not require hospitalisation, as well as on heat adaptation, has mainly been obtained through location-specific survey-based studies. Such data is costly and time consuming to collect, provides a snapshot from one point in time which may not coincide with the highest temperatures and is questionably extrapolated to the wider population (Hausmann et al. 2020). As a result, no country has national data on heat impacts and adaptation behaviour over time because there are neither longitudinal datasets nor social surveys. One exception is longitudinal panel data on the availability of cooling devices (fans and air-conditioning; e.g. Osberghaus and Abeling 2022).

A relatively new source of large datasets is social media. Increasingly, the public turns to social media during crises to share the latest information while agencies (governmental, non-governmental or non-institutional) use social media for formal communication of updated information, warnings and specific advice, or to give directions to the public and stakeholders such as volunteers (Wang et al. 2016; Abedin and Babar 2018). Therefore, social media data have become a useful tool in disaster risk mitigation, communication and recovery, mainly for sudden and severe hazards such as hurricanes, floods and earthquakes (Luna and Pennock 2018; Ogie et al. 2022). Very few studies have examined social media use to explore information seeking behaviour, impacts and responses to slow onset hazards.

Germany has experienced record high temperatures and droughts in the last few years (Winklmayr et al. 2022; Kabisch et al. 2023) with an estimated 48,000 heat-related deaths occurring during the last decade (Wang et al. 2024). Heatwaves in 2018 and 2019 are estimated to have cost the German economy 40 billion Euros (Trenczek et al. 2022). The severe heatwaves in Europe would have been extremely unlikely in the absence of human-induced climate change (Scott et al. 2004; Yiou et al. 2020). The aims of this study were to understand 1) trends in heatwave perceptions in Germany and 2) what people communicate about during heatwaves. For this we applied machine learning based methods to analyse the text and identify topics emerging in a unique dataset of German language social media messages.

Our paper extends the applications of disaster-related social media studies by assessing a slow onset hazard - heatwaves, though often deadly, are rarely perceived as severe and destructive as are some rapid-onset natural hazards like floods or wildfire. To the best of our knowledge, social media analysis has been employed to explore how people are impacted by heatwaves in only a handful of country-specific case studies (Japan: Murakami et al. 2020; India: Cecinati et al. 2019; Italy: Grasso et al. 2017; and the US: Jung and Uejio 2017; Moore et al. 2019) and one global study (Zander et al. 2023b).

Our study also contributes to a growing body of literature examining the social and health effects of increasing heat and heatwaves in Germany (e.g. Mücke and Litvinovitch 2020; Beckmann et al. 2021; Sandholz et al. 2021; Osberghaus and Abeling 2022; Hannemann et al. 2023; Zanocco and Sousa-Silva 2023; Heidenreich and Thieken 2024). This recent increase in studies underpins the urgency with which social heat studies from Europe are thought to be needed. The topic analysis can provide near real-time analysis of where and when the society suffers from heat which can foster rapid decision-making and timely resource allocation during heat waves (Lyu et al. 2024) as well as for extremely hot days that are not necessarily labelled as heat waves and for which no warnings are issued.

2.1. Data collection

Our data set consists of tweets, which are messages with a constraint on maximum length (140 characters to November 2017, 280 characters thereafter) sent using the social media platform X (formerly), written in the German language. There were five reasons we decided to use German X data. First, by early 2023 there were eight million X (then Twitter) users in Germany (Blog2Social 2023). Although this is not representative of German society as a whole, conversations about heatwaves among users still provide a far broader reflection of the views of German society on heatwaves than any survey could have done. Second, the German language has a narrow spatial footprint, so social media messages in German were most likely to have been from users living in the small number of countries in which German is the dominant language. Third, the German search term we used, ‘Hitzewelle’ (heatwave) has had fewer applications unrelated to hot weather than in English, where the term has been applied to a popular film, a popular song and a disco-funk band, so that the tweets retrieved are likely to be more focussed on the subject topic. Fourth, the platform was frequently used by news outlets, government officials and emergency services for disaster communication and management (Kusumasari and Prabowo 2020). Lastly, users of X agreed to a privacy agreement that allowed the company to make data publicly available while data from other social media, such as Facebook, is considered private (Mahoney et al. 2022). Furthermore, until recently, X provided a certain percentage of all messages sent on the platform, formerly referred to as tweets, including associated meta-data, to be downloaded and used for non-commercial research via a free API. For academics with a Twitter research account, 20% of all tweets could be downloaded for free, making the platform the social media source used most frequently by academics for research. Although free and open access to tweets for research purposes was suspended when Twitter came under new ownership in 2023, we were able to obtain the data on heatwaves that was freely available for the ten years up to the end of 2022.

The data in this research were collected using snscrape, a pre-built scraper for social networking services in Python. The TwitterSearchScraper method of snscrape’s Twitter module (JustAnotherArchivist 2020) allows users to retrieve historical tweets based on a set of keywords. Here we searched for tweets posted between 1 January 2013 and 31 December 2022 (10 years) and including the keywords “Hitzewelle” (German for heatwave). Apart from the text of the tweets, we collected meta information about the user location (city and state) and the date. From the date we categorised the tweets into years and into season (spring: March, April, May; summer: June, July, August; autumn: September, October, November; winter: December, January, February).

2.2. Data processing

Because we were interested in tweets in German and from Germany only, we deleted all tweets that were in a different language (n=2,888) and those that were in German but posted from outside Germany (location clearly indicating place outside Germany such as Austria; n=1,245). We also deleted duplicates (identical text posted by the same user; n=32,004). From the 93,282 tweets over the ten years that we initially downloaded, 57,145 tweets were kept. To further clean the text, we applied common standards for processing short social media text data which included steps such as removing special characters, punctation and numbers, converting texts to lower case, and removing common German stopwords (see Xue et al. 2020). Stopwords are words that do not add meaning to sentences.

2.3. Structural topic modelling and coherence score

Topic modelling is a popular and important unsupervised machine learning and natural language processing (NLP) algorithm (Ref.). It is a tool designed to analyse and extract hidden topics from large datasets, in this case Twitter (X) tweets, which cannot easily be read and analysed objectively without a machine. Topic models classify articles according to their key topics and word frequencies and thereby reveal hidden clusters (Chang et al. 2009). Articles within each topic have similar words that occur together frequently (Blei et al. 2003). However, the researchers’ knowledge and experience are also needed to interpret the model outcome into meaningful themes (Karami et al. 2018).

There are different approaches to topic modelling. The most common topic modelling approach is Latent Dirichlet Allocation (LDA), an unsupervised machine learning technique that generates topics based on the pattern of occurrence or co-occurrence of words in a document (Jacobi et al. 2015). LDA, however, performs poorly on very short texts, as is the case with tweets, due to the lack of information on co-occurrence of words (Qiang et al. 2022) and does not allow consideration of covariance (i.e. the relationship between topics and meta data such as the year of the tweets r user location). We therefore analysed the data using structural topic modelling, an extension of LDA topic models allowing correlated structures (Blei and Lafferty 2007) and the inclusion of non-textual covariates into the model (Roberts et al. 2016).

To do so we first had to assess how many topics best fit the data. This was done by combining quantitative and qualitative approaches. First, to determine a likely number of possible topics, we assumed, based on other studies (Ref, Ref) that a model with between 5 and 20 topics would make a sensible and manageable number of topics to interpret. We then applied the function ‘ksearch’ of the stm R package (Roberts et al. 2019 to evaluate the performance of models with 5 to 20 topics by maximising two functions will minimising the number of residuals. The two functions were the semantic coherence of the topics (Ref), a measure of semantic similarity between the top words within each topic, and the held-out likelihood (Ref), a measure of how well the model predicts unseen data that indicates model skill in capturing meaningful patterns in the data. We plotted the results of this function to visualise the relationship between the number of topics and the model fit indices. The optimal number of topics was then chosen based on a qualitative interpretation of this plot with the object of balancing coherence, held-out likelihood and residual number.

To account for variables that may explain topic prevalence, we then applied the 'estimateEffect' function of the stm R package to perform an ordinary least squares (OLS) regression of the covariates of topic prevalence. Correlations between topics were assessed using the 'topicCorr' function and visualised to depict the relationship between topics (Figure S3 in the Supplementary Materials). Topics connected by lines indicate a high probability of co-occurrence, with a correlation coefficient threshold set at 0.16 (Tonidandel et al. 2022). A qualitative evaluation of the results was derived from the plot. This visualisation enabled the linking of related topics without relying solely on keyword overlap.

2.4. Sentiment analysis

The sentiment analysis was done to complement the content analysis and can be particularly useful for disaggregation of texts with similar keywords and topics. The emotions which underpin the sentiments in the language analysed can be very strong during disasters such as hurricanes (Neppalli et al. 2017), wildfires (Zander et al. 2023c) and earthquakes (Wang and Taylor 2018) and can help planners identify when and what type of relief is most urgent. We know of no other sentiment analyses of heat-related social media conversations, although heat is known to increase irritability, violence, and conflict (Thompson et al. 2018; Otrachshenko et al. 2021; Li et al. 2023). A study from the US in which more than two million tweets about climate change, not heat, were collected over two years found that individuals were more likely to express a positive sentiment towards climate change when temperatures exceeded historic expectations, relative to when temperatures are below historic averages (Effrosynidis et al. 2022). We therefore analysed text to distinguish positive sentiments, which have been found to relate to adaptive capacity and support (Kušen and Strembeck 2021; Zander et al. 2023a), from negative sentiments associated with damage and distress (Kryvasheyeu et al. 2016).

While there are many methods to automatically assess sentiments in the English language, including freely available lists of negative, positive and neutral words, there is less choice for the German language. Here we used Rauh’s German political sentiment dictionary (Rauh 2017), available as an object in the quanteda R package. This dictionary has been found to show strong correlation with German dictionaries for automated sentiment analysis (Eberl et al. 2020). The sentiment analysis provides an augmented sentiment score which ranges from -1 (all words in a text are negative) to +1 (all words positive). This score is calculated by counting the frequency of definitive negative and positive keywords and by dividing it by the total number of words in each text (Eberl et al. 2020). A score of zero indicates a neutral sentiment. Using this score, we assigned each tweet to one of three sentiment categories: neural (score = 0), positive (score > 0) and negative (score <0).

3.1 Spatial and temporal distribution of tweets

We could assign 85% of the 57,145 tweets (n= 48,475) to one of the 15 German states. Most states were from the highly populated and urbanised states, which over-represents the city states of Berlin (16.2% of tweets, 4.5% of the German population) and Hamburg (6.2% and 2.2%; Figure 1 and Table S1 in the Supplementary Materials).

[Figure 1 here]

Most tweets (88% in total) were posted in the European summer but for some years had an unusually large number of relevant tweets in either spring (2014 and 2017) or autumn (2016 and 2020) (Figure 2).

[Figure 2 here]

A spike in the number of tweets occurred in 2015 then had a sustained increase from 2018 (Figure 3). The tweets activity was closely associated with major heatwave events. The 2015 spike in tweets related to heatwaves corresponded to a severe heatwave across India and Pakistan which killed more than 2,000 people followed by Germany’s most severe heatwave in Germany since 2003, with estimated attributable 6,200 deaths (an der Heiden 2019a). The year 2015 was the hottest year on record since modern record keeping began in 1880 (NASA and NOAA 2016). Although, globally, 2016 was hotter than 2015 (NASA and NOAA 2017), and 2017 was among the hottest, Germany was not directly affected and the number of heatwave related tweets posted in Germany was low (Figure 3).

Tweet activity increased from 1,451 in 2017 to 11,452 in 2018 (20% of all tweets) and remained high in 2019 (10,216 tweets; 18%) were posted. Although the year 2018 ranked behind years 2015-2017 at a global scale, the summer of 2018 was then the hottest in Germany since temperature data were recorded with 7,500 people dying as a direct consequence in the 2018 and 2019 heatwaves combined (Winklmayr et al. 2022), a number exceeded only in the summer of 2003 summer (70,000 died; Robine et al. 2008).

The year 2020 was Europe's warmest year on record, tying with 2016 at a global level (NASA 2021; 2023 after this study, was hotter still). Heat waves occurred in France, the Netherlands and UK (Boyer et al. 2021) but it was the winter season that was warmer than usual while the summer was about normal for most regions (Boyer et al. 2021). The European summer record was broken again in 2021 with heat and drought linked to wildfires in Italy, Greece and Turkey (Copernicus Climate Change Service 2021) while Germany and Belgium experienced extreme rainfall and devasting floods in July 2021. However, tweet numbers declined (2020: 8.7%; 2021: 7.6%). The final year, 2022, had the most tweets (25%) but the summer was again Europe’s hottest on record with an estimated 61,000 people dying because of heat (Ballester et al. 2023).

[Figure 3 here]

3.2. Tweet content and related sentiments

The stm resulted in nine topics (Figure 4). The topic meanings were characterised by the authors based on the 50 most important words in each topic (see Figure S2 in the Supplementary Materials). The average sentiment score was also positive (0.09) and more tweets had positive sentiments (37%) than negative (19.5%) with the largest proportion being neutral (43.5%; Table S2 in the Supplementary Materials). The share of positive tweets increased from 30% to 38% after the year 2018 but so, too, did the share of negative tweets, from 17% to 20%. In summer (average sentiment score 0.102), tweets were more likely to be positive (38%) and less likely to be negative (19%) than in the other seasons. In winter (average sentiment score 0.011) the tweets were least likely to be positive (26%) and most likely to be neutral (50%) with tweets in spring (average sentiment score 0.102) more positive than in autumn (average sentiment score 0.054).

[Figure 4 here]

Topic 6. Global heatwaves and mortality (22% of tweets; average sentiment score: 0.057): Most tweets belonged to this topic (Figure 4) for which important keywords included: ‘heatwaves’, ‘Australia’, ‘temperature records’, ‘excess mortality’, ‘death’ and ‘news’. Topic 6 was the most prevalent topic from 2013 to 2017; Figure 5 and Table S3 in the Supplementary Materials). It was also the only topic more likely to occur before 2018 (Table 1). Until 2018 this topic was confined to the summer season (Table S3) while since 2018 it was more likely to be talked about outside the summer season. It was therefore the only topic that was, overall, mostly tweeted about in autumn, indicated by the negative and significant coefficients for the other three seasons (Table 1). Conversations within this topic were about record temperatures in Australia and associated mortality, how people in different countries (Spain and India besides Australia) swelter in the heat and the high risk of wildfires in these extremely hot conditions. The tweets within this topic were not about personal experience in Europe or Germany. Many tweets were posted in European winter, which is summer in the southern hemisphere (Figure 2). One reason topic 6 dominated the early years was because many tweets were repeated in different forms, such as the news about a German woman who has died in the heatwaves and associated wildfires in Australia in 2013. Tweets in these early years were broadly similar, hence the dominance of a single topic. From 2018 the absolute number of tweets in topic 6 declined while the variety of tweet content within this topic increased.

[Figure 5 here]

Topic 5. Local impact of climate change (20% of tweets; average sentiment score: 0.126): Most tweets within this topic were posted in summer (Table 1; Table S3) with the label based on frequent keywords like ‘hot’, ‘water’, ‘climate change’, ‘shade’, ‘car’, ‘air-conditioning’, ‘electricity’ and ‘body’ (Figure S2). People tweeted about the impacts of heat in Germany and what they personally do or cannot do. Examples of impacts concerning people were trains in Germany being too hot or getting stuck on the tracks because of the heat, the heat impact on dependent people living in homes for the elderly, the lack of air-conditioning in hospitals and nursing homes, and the danger of leaving children and pets inside cars. Because of the severe heatwave in Germany in 2015, this topic first had a considerable number of tweets in the summer of 2015, declined in importance in 2016 and 2017 before gaining in importance again from 2018 on (Figure 5).

Topic 4. Heat health impacts (11% of tweets; average sentiment score: 0.045): Characterised by keywords including ‘heat’, ‘cooling’, ‘India’, ‘causes’, ‘Pakistan’, ‘death’, ‘danger’, ‘help’, ‘suffering’, ‘corona’ and ‘fans’, tweets in this topic contained advice on how to stay healthy in the heat, advised people to drink plenty of fluids, avoid alcohol, open windows to create a breeze at home, use fans and avoid doing sport during hottest time of the day. Most of this advice was related to coping at home and about looking after people’s health. In 2020, the keyword ‘corona’ came up frequently when people compared the heatwave to be as dangerous to health as the covid-19 virus. This topic first appeared in 2015 and was associated not only with the German heatwave but also with heat related deaths in India (in 2015) and in Pakistan (in 2018) and the health implications of heat during Ramadan, particularly the implications of fasting. Its highest share of tweets was in the three very hot years of 2018, 2019 and 2022 (Figure 5). Before 2018 this topic was most likely to be tweeted about during European summer time, while since then it was a likely topic in the months which are relatively hot elsewhere in the world, which is European spring time (March to May) (Table 1), with a particular high proportion of tweets within that topic being posted during spring in 2021.

Topic 7. Storm risk (10% of tweets; average sentiment score: 0.018): This topic, which had the keywords ‘storm’ and ‘severe weather’ among the most important in this topic, was largely about the link between heatwaves and heavy storms, such as heavy storms that end heatwaves and cause cooler days and nights. Very few tweets of this topic occurred before 2015 and only some more in 2015 before increasing in 2018 again. In 2018 and 2021 this topic also covered fire risk from lack of rain in Germany. People tweeted about damages from storms, hail and hurricanes and warnings about potential risk of those. As such this was the most negative topic (Table S2). The topic occurred in all seasons but was least prevalent during the European winter time (December, January, February; Table 1).

Topic 8. Temperature records and warnings (10% of tweets; average sentiment score: 0.130): Tweets about temperature records and climate modelling forecasts considered not only Germany but also Asian countries, particularly China in 2022. Often the temperature records were phrased as warnings, mostly experts and meteorologists expressing concern about these records and heatwaves. The topic was most dominant in summer and least so in winter (Table 1). The topic first arose during the heatwave of 2015. There were then a few tweets in 2016 and 2017 before the topic gained momentum in 2018 with peaks in 2019 and 2022, in line with the severe heatwaves in Germany (Figure 3).

Topic 2. Heat and fire risk (9% of tweets; average sentiment score: 0.139): During 2018 this was the most dominant topic with numerous tweets about fire fighters in action and risking their lives. Many tweets expressed gratitude to the fire fighters 44% of the tweets within this topic were categorised as have a positive sentiment (Table S2). Wildfires across Europe, such as in Greece and Sweden, were mentioned as was the link between drought and climate change. Some warnings were also conveyed in these tweets, for example urging people not to have barbeques or throw away cigarette butts in dry forests. Tweets within this topic were significantly more likely to occur in summer than in any of the other three seasons (Table 1). Although mostly focusing on Europe, the topic also included tweets about fire risk in dry and hot weather in Australia and Asia (e.g. Korea) at the end of 2019 and the beginning of 2020 and, in 2022, in California (USA). In the US context, people also tweeted about water shortage for the fire fighters and old power networks that cannot withstand the heat.

Topic 3. Heat as a nuisance (8% of tweets; average sentiment score: 0.128): Based on keywords such as ‘today’, ‘tomorrow’, ‘summer’, ‘August’ and ‘June’ as well as keywords about how people feel, such as ‘suffering’ and ‘afraid’, 96% of the tweets within this topic were posted in the European summer and concerned Germany. Like most other topics, it only gained momentum in 2018. This topic was correlated with topic 9 (see below; Figure S3) but tended to treat negatively judging from its lower average sentiment score (Table S2). This topic included some texts about health effects (feeling dizzy and having headaches and suffering in the office) and, like topic 9, advice what to do, including cooling through taking showers and not to forget to drink plenty. This topic was checked manually to understand the context of the keywords. The way in which the keyword ‘holidays’ was used in this topic, for example, differed from how it was used in topic 1 (see below). Unlike topic 1, which considered heat in tourist destinations, use of holidays word in topic 8 was about Germany being hotter than tourist destinations; tweets complained about returning from Italy only to find it hotter at home in Germany. These thoughts were also linked to being afraid of the heat in Germany when coming back and that such a situation was not ‘normal’. The use of the keyword ‘fear’ was also ambiguous. The context was partly about being afraid of the unusual high temperatures in Germany and elsewhere, but also about the fear of heatwaves being unnecessary. Some people expressed their annoyance with the weather forecasts and warnings, arguing that the temperatures were that of a normal summer.

Topic 1. European drought (6% of tweets; average sentiment score: 0.088): This topic first appeared in 2015 and was then prominent if 2018 and 2022. In 2018 the tweets related to Topic 1 were mostly about the unprecedented drought with particular reference to northern Germany. The topic also included similar keywords to topic 2 about fire risk and fire fighters. In 2020, 2021 and 2022 the focus had shifted from the German drought that featured heavily in 2018 to droughts in southern Europe, particularly Italy and France. Users tweeted about the heat in some of the tourist destination countries (Italy, Greece, Turkey) and the danger of fires for people living there and tourists, as well as water shortages in these countries (in particular in 2021). Conversations about heat and drought at tourist destinations continued in 2022 with reference to harvest failure, dead wildlife because of water shortages and wildfires and avoidance of these destinations to reduce strain on local fire fighting capacities. More subtle issues related to heat and drought included the costs of power (in Germany in 2018), of drinking water (bottles) and ice cream in Italy (in 2021).

Topic 9. Local heat adaptation (4% of tweets; average sentiment score: 0.158): This small, diverse topic was, like Topic 3, largely focused on Germany and what people in Germany do to cope with heat. It featured key words such as ‘drinking’, air-conditioning’, ‘cooling’, ‘bath’, ‘sweating’, ‘house’, ‘window’, ‘pets’, and ‘work’. These keywords indicate that the conversations were about how to cope with heat at home and at work and had the most tweets in 2018, 2019 and 2022, the hottest years. This topic was most likely to be posted in summer, and less likely in winter than in the other two seasons (Table 1) and was the topic with the most positive language use, i.e. with highest average sentiments core (Table S2). Many of the tweets included practical suggestions about what to do, such as avoiding lengthy sun exposure, drinking plenty of water, drinking warm rather than cold drinks, having cold showers, using fans, consuming light food and wearing airy clothing. Tweets within this topic also included entreated people to look out for older people and pets during heatwaves and to be careful when swimming in lakes and ponds. Like topic 5, topic 9 also included ‘air-conditioning’, but here it was more about how people were using it, how much heat relief it provided and that buying one was a good decision, while in topic 5, the sentiment around air-conditioning was more negative (i.e., the lack of air-conditioning).

Table 1. Results of a structural topic model with covariance

Variable

Topic 1

Topic 2

Topic 3

Topic 4

Topic 5

Topic 6

Topic 7

Topic 8

Topic 9

Before 2018

-0.004*

(0.004)

-0.013**

(0.006)

-0.022***

(0.005)

-0.033***

(0.005)

-0.019**

(0.008)

0.072***

(0.006)

-0.012**

(0.005)

-0.197*

(0.004)

-0.022***

(0.004)

Spring

-0.002

(0.002)

0.005

(0.004)

-0.012***

(0.003)

0.016***

(0.003)

-0.009*

(0.005)

-0.015***

(0.004)

0.014***

(0.003)

0.001

(0.002)

0.004*

(0.002)

Summer

0.029***

(0.002)

0.011***

(0.003)

0.011***

(0.002)

-0.006**

(0.002)

0.019***

(0.004)

-0.033***

(0.003)

0.022***

(0.002)

0.065***

(0.001)

0.011***

(0.002)

Winter

-0.015***

(0.003)

0.006

(0.004)

-0.005

(0.003)

-0.021***

(0.003)

-0.023***

(0.005)

-0.029***

(0.005)

-0.007**

(0.003)

-0.103***

(0.003)

-0.010***

(0.003)

Before 2018 * Spring

0.002

(0.005)

0.001

(0.008)

-0.001

(0.006)

0.004

(0.007)

0.002

(0.010)

0.002

(0.007)

-0.004

(0.006)

-0.005

(0.005)

-0.001

(0.005)

Before 2018 * Summer

0.007*

(0.004)

0.005

(0.007)

0.002

(0.005)

0.017***

(0.005)

0.009

(0.008)

0.022***

(0.006)

0.005

(0.005)

0.072***

(0.004)

0.006

(0.004)

Before 2018 * Winter

-0.008*

(0.005)

-0.029***

(0.009)

-0.017**

(0.006)

-0.006

(0.007)

-0.013

(0.010)

-0.007

(0.008)

-0.014**

(0.006)

-0.102***

(0.006)

-0.007

(0.005)

Levels of significance: 1% = ‘***’ , 5% = ‘**’ , 10% = ‘*’

Topic 1: European droughts; Topic 2: Heat and fire risk; Topic 3: Heat as a nuisance; Topic 4: Heat health impacts; Topic 5: Local impact of climate change; Topic 6: Global heat waves and mortality; Topic 7: Storm risk; Topic 8: Temperature records and warnings; Topic 9: Local heat adaptation and advice

[Table 1 here]

4.1. Tweet activities and sentiments over time

Previous studies have noted that social media is a good indicator of extreme weather events with increases in social media activity immediately before and during fast onset natural hazards such as tornados and hurricanes (e.g. Silver and Andrey 2019; Torricelli et al. 2023). Less evidence exists for heat related studies, or slow onset hazards in general. As study from Spain in which Spanish tweets were analysed over the course of two years showed an increase in weekly temperature volatility predicted public attention toward climate change (Mumenthaler et al. 2021). Zander et al. (2023b), analysing English tweets on a global scale, found that the number of tweets corresponded well with major global heat wave events, suggesting that communities all over the world are aware of the threat from heat and its social and health impacts. Our study echoes these results with increasing tweet activities over the last decade whenever Germany experienced extreme temperatures and heat waves. Not surprisingly, 88% of the tweets were posted in European summer. However, in some years, large number of tweets were also posted in either autumn or spring which relates well to global heat events in the southern hemisphere such as extreme heat and heat-related fires in Australia (see Figure 3). Showing high awareness of local heat events was also indicted by the highest shares of summer tweets occurring in the hottest years (2015, 2018,2019 and 2022; see Figure 2).

Our results also show that tweet activity decreased in years that were cold relative to other recent years, suggesting that tweet activities are driven by recent experience rather than historical periods when recent ‘cold’ years might have been considered warm (Moore et al. 2019). However, we expect increasing number of heat related tweets in the future as Europe’s temperatures and heat wave events are set to rise more quickly than the global average (IPCC 2021; Rousi et al. 2022), putting the health and productivity of European’s people at risk.

Sentiments have become more positive when heat conversations were linked to climate change, and also over time in general, as shown elsewhere. Europeans have been found to be among those who tweet most positively about climate change (Effrosynidis et al. 2022). However, while Effrosynidis et al. (2022) found that positive sentiment towards climate change were more likely when temperatures exceeded historic expectations, relative to when temperatures were below historic averages, sentiments in our study were a mix of positive and negative from the year 2018 on. This suggests that rising awareness and sentiments still predominated in the knowledge of climate change may be dividing opinions about hot weather. Nevertheless, positive sentiments predominated in the German summer and it was in winter when negative sentiments were more common, a time when deadly heat waves were more likely in the southern hemisphere.

4.2. The Year 2018

The year 2018 in Germany was climatically extraordinary (Zscheischler and Fischer 2020), even in the light of subsequent hot years, and has since been the focus of several studies of heat-related mortality (e.g. an der Heiden et al. 2019b; Winklmayr et al. 2023; Wang et al. 2024). Although the United Nations Climate Change Conference COP21 at the end of 2015 has been described as “one of the triggering moments that led people to update their mental template of climate change and its manifestations” (Mumenthaler et al. 2021), our analyses suggest it was the weather in the summer of 2018 that initiated a long-term shift in perceptions of heatwaves in German society. The proliferation of topics heat-wave related topics meant that local (German and European) experience of heat replaced global heat-related issues as the dominant feature of Twitter/X conversations (Figure 5). In the five years before 2018, most heatwave-related tweets related to death and suffering outside Europe. From 2018 on, the most important topics have been impacts of heat on health (Topic 4), local impacts of climate change (Topic 5), local heat adaptation and advice (Topic 9) and associated weather extremes including droughts in Europe (Topic 1), fire risk (Topic 2) and storms (Topic 7). Many of these topics were barely mentioned before 2018 other than during a short spike in the hot year of 2015. In 2018 the spike was not only much larger but, unlike after 2015, they persisted; while climatic conditions in many of the years since 2018 have also been exceptionally warm, the topics persisted even through the slightly cooler years of 2021 and 2022. This suggests that 2018 sensitised people to climate change and heat in Germany and Europe.

Some topics have not only increased in salience since 2018 but have also shifted seasonal focus. This was most notable for conversations about global heat waves and mortality (Topic 6). Before 2018, tweets about this topic tended to occur in summer, whereas since 2018 fewer than 1% of Topic 6 tweets were in summer with most occurring in autumn (Tables 1 and S3). Conversations about heat health impacts (Topic 4) also shifted from being mainly a summer topic before 2018 (with peak in 2015) to a subject discussed across spring, autumn and summer which suggests that very hot conditions increasingly bother German Twitter/X users outside European summer in terms of their health and well-being.

4.3. Majority topics

Of the nine topics, we define those five that had a proportion of at least 10% of the total share of tweets over the decade as major topics (see Figure 4). Of these, the most important were two with shares of around 20% each. The absolute number of tweets on global health and mortality (Topic 6; nearly 22% of all tweets over the decade) has remained reasonably constant even though the proportion declined after 2018. Its importance demonstrates concern for global issues, often in countries of the global South (e.g. India and Pakistan). The second topic with 20% of all tweets (Topic 5: Local impact of climate change) only emerged in 2015 and increased since 2018 as the link between uncomfortable heat and climate change became more obvious, particularly in the hottest years (2015, 2018, 2019 and 2022). A two-year study from the US which analysed more than two million tweets containing the terms “climate change” and “global warming” also found that the public recognise and connect extreme temperature anomalies to climate change (Kirilenko et al. 2015). Mumenthaler et al. (2021) found similar results for Spain, suggesting that a heightened awareness of climate change is being reflected in social media conservations because of increased media coverage.

Conversations about health impacts of heat (Topic 4; 11% of all tweets) were among the most frequent after the two dominant topics. This topic, like Topic 6, covers heat related mortality, including in other countries, but differed by inclusion of the term ‘corona’ with many conversations about the double burden of heat and corona.

In Germany, storms and floods are the rapid-onset weather events that affect most people and cause the highest number of fatalities (Butsch et al. 2023) with the flood in July 2021 in Western Europe accounting for one of the most costly disasters in Europe, causing EUR 32 billion worth of damage (Mohr et al. 2023) and killing more than 200 people, 184 in Germany alone (Kreienkamp et al. 2021). Heavier rainfall is associated with heatwaves because hot air can hold more water than cold air (Prein et al. 2017) and downpours from electrical storms often occur after extremely hot days. As a result, even though we did not search messages for terms related to flooding, storms or rainfall, storm risk still emerged as a topic on its own (Topic 7; 10%). This topic also refers to other extreme heat-related events such as fires, though these were generally outside Germany. The conversations suggested that people were linking different climate change related hazards and recognising that heat contributes to multiple hazards.

Another topic which also averaged about 10% concerned temperature records and warnings (Topic 8). Mücke and Litvinovitch (2020) argue for a strong need to implement heat health warning systems in Germany after the extremely hot summers of 2018 and 2019. The German early warning landscape is currently being updated and warnings through cell broadcast have been tested and implemented since February 2023. There are also warning apps that people can subscribe to and download in Germany (e.g., NINA and Katwarn). So far, these apps have mainly been used for fast onset hazards such as floods and storms. While specific local heat health warning systems remain rare in Germany (Hannemann et al. 2023), people have been increasingly turning to social media and official Government channels such as of national weather agencies to inform themselves about impending heat. This may be partly a consequence of the heat in 2018 – it was only in the hot years after that, particularly the summer months in 2019 and 2022 (see Table S3), that this topic emerged as important suggesting the heat warning systems were a response to the very hot and dry year of 2018.

4.4. Minority topics

Marginally below the threshold for major topics was that related to wildfire (Topic 2 Heat and fire risk; 9%). This topic encompassed fires in both Europe and other countries in Asia, Australia and north America where wildfire is common in very dry seasons and also increasing as a consequence of climate change, drought and heat (Mukherjee et al. 2023). The number of posts was particularly high in 2022, when it was the third most important topic (Table S3), reflecting the numerous fires in southern European countries that Germans visit on summer vacation. The topic has a relatively high sentiment score, reflecting the gratitude people felt towards those fighting fires. Potentially, the attractiveness of Mediterranean tourist destinations, particularly Portugal and the Greek and Spanish islands (Neger et al. 2024), may decline if they come to be considered too hot to enjoy and too risky in terms of wildfires.

Tweets in Topic 3 (Heat as a nuisance; 8%) complained about the heat, sometimes light-heartedly, and treated heat as an annoyance or inconvenience rather than as serious threat to health or property. This could imply that people those who posted the tweets suffered little from heat, were underestimating its impacts or are being pre-conditioned to seeing heat as a major problem in the future. Topic 3 was correlated to topic 9 (heat adaptation and advice), implying there is a fine line between annoyance at heat and providing helpful advice contributing to resilience and strategies for coping with heat (Topic 9).

The topic that related European droughts to heat (Topic 1, 6%) was prominent in only two years, 2018 and 2022. In 2018 very dry conditions in Germany caused cost >$50 million in reduced winter wheat yields (Schmitt et al. 2022) and forest canopy loss over half a million hectares, killing up to two-thirds of the conifers in some areas (Thonfeld et al. 2022). In 2022, droughts in southern Europe (Greece, Italy, France and Turkey) were associated with heatwaves and wildfires (Butsch et al. 2023), so the link between heat related tweets and drought was not surprising with many tweets mentioning of fire risk in tourist destinations.

The topic attracting the fewest tweets was on local heat adaptation and advice (Topic 9; 3.8%) and mostly covered personal coping strategies such as wearing airy clothing, drinking plenty of water and staying inside air-conditioned rooms. That communication about effective personal coping strategies through social media had fewest tweets suggests that although awareness of heat within Germany is high and increasing, adaptive capacity may be low, as concluded by Osberghaus and Abeling (2022). There may even be active resistance to adaptation; one reason why use of air-conditioning is rarely discussed during hot weather could be a reluctance among Germans to install air conditioning (Zanocco and Sousa-Silva 2023).

4.5. Study limitations

As with every data collection approach, using only data from one source, here from Twitter/X comes with limitations. First of all, those people who did not use Twitter/X between 2013 and 2022 were not included in our data analysis. This might include groups of people who are particularly vulnerable to heat, particularly elderly people; of the approximately 4% of all Germans who used Twitter/X in 2021, the largest cohort was of people aged 14 to 29 years (Smart Home Fox 2024).

A second study limitation could be that number of characters allowed in a message could constrain the amount of information that can be conveyed given that German words contain, on average, 20% more letters than English words (Carlson et al. 1985). However, the median Tweet length in English is only 38 characters, similar to the message length in books (48 characters) and films (25 characters; Alis and Lim 2012), well below the Twitter character limit. This also suggests that the increase characters limit from 140 to 280 in November 2017 is unlikely to have influenced the trends detected. While this could have contributed to the increase in topic number, given the increase did alter the tone, type and effectiveness of language used by commercial firms and politicians (Jaidka et al. 2022; Al Balawi et al. 2023), a comparison between 140 and 280 character-limit tweets found them to be indistinguishable semantically and syntactically (Gligoric et al. 2022) and there is no a priori reason to expect content bias related to heat waves based on character limits.

Our study shows that social media data could be used to monitor the views of society over time when faced with a slow onset hazard such as extreme heat. Unfortunately, the new management of X no longer allows free access to its data. While payment of these data may still be worthwhile under some circumstances, research of the type undertaken here will be constrained by budget limitations in many countries and institutions. The rebranding from Twitter to ‘X’ also compromises the comparison of recent messages with those from before the rebranding. While there are likely to have been gradual shifts in user characteristics over the ten years we studied the conversations, for which we have no information, the change in ownership is likely to have caused a step change in user type. Since the ownership change there has not only been a decline in use of the platform in general but many scientists and environmentally oriented users now refrain from using X to disseminate information (Chang et al. 2023). This can compromise the warning and hazard information landscape with real-time information now needed to be sources from other social media data. We should therefore learn has much as we can from the data while we can with probably dwindling number of studies using these data. Our study and previous studies also showed that the Twitter platform was a pivotal platform for the dissemination of real-time information and sharing informal advice how to deal with extreme situations.

4.6. Study applications

Understanding societal responses to heatwaves can be important for policy and adaptation (Wang et al. 2021). While analysing Twitter/X conversations has become more challenging since the change in ownership, other social media platforms could be explored to understand how responses and sentiments evolve in response to rising temperatures. Trends in social media can be used by advocates to influence policy responses to climate change as they draw on such a wide cross-section of society. At a personal level, emotional responses to climate change can also translate into action, though the connection between the two is only just starting to be understood (Brosch 2021). Social media analysis could help in understanding how heatwaves condition people to take action, or to accept policy change, which could in turn be important for strengthening national responses to increasing heatwave frequency. Social media analysis can also help in provision of social support for people emotionally affected by impending climate change. Although concern about climate change is currently lower in European countries than where climate change is having a stronger direct effect (Clayton et al. 2023), negative effects of climate change on mental health are greater in European countries than elsewhere (Ogunbode et al. 2023).

Social media analysis can also help understand the salience of different issues to the population. In the ten years of data we analysed, some concerns remained steady, others pulsed in response to immediate events while others were novel in the exceptional heat of 2018 and have remained so since. Continuing analysis will undoubtedly identify new issues of concern as well as changes in emphasis and emotional intensity towards existing topics. The proliferation and perpetuation of tweets and topics after 2018 suggest it marked a significant turn in domestic sentiment relating to the effects of climate change. Timely recognition of such events could help ensure that such changes are reflected in policy. At the same time, social media analysis could be used to test the impact of policy initiatives, such as the heat mitigation plans that have been introduced to some German cites (Hannemann et al. 2023).

More immediately, as with other severe natural hazards (Luna and Pennock 2018; Ogie et al. 2022), compiling social media data on heatwaves can not only allow authorities to transmit warnings and advice but also help them respond in a timely manner (see also Lyu et al. 2024). Although the geographic spread of heatwaves tends to be far broader than that of most sudden onset hazards like floods, earthquakes, tsunamis, the level of distress emerging from real-time analysis of social media could trigger warnings to people who are vulnerable to heat or those responsible for them. As it is, heat wave warnings often fail to anticipate heat-related distress in Germany, where some people start to react negatively to heat below current heatwave warning temperature thresholds (Bogdanovich et al. 2024). Real time social media analysis may therefore be a means to meet a need for targeted warnings.

Understanding how different populations are responding to heatwaves is especially timely for European countries since the region has been warming twice as fast as the global average, becoming the fastest-warming continent on Earth (Copernicus Climate Change Service 2024). Our analysis of social media conversations on Twitter/X on heatwaves is consistent with analyses of social media activity on sudden natural hazards, in particular the locations where the hazards occur and the affective and emotional responses of the affected population. As heatwaves increase in frequency and intensity, timely analysis of social media data can allow rapid responses to those most distressed. Judging from the topics, we can confirm earlier findings that heat sensitivity of households in Germany is high, adaptive capacity is low (Osberghaus and Abeling 2022). However, with the rising trend of topics about advice on what to do, capacity to adequately adapt to heat and heatwaves may be increasing, making the German society more resilient to future events. The dry and hot year in 2018 seems to have helped not only to raise awareness but also to increase coping capacity. The shift from global to local content from 2018 on suggests that German society is becoming more aware of the impending social and health impacts of heat and the need for adaptation. This may increase acceptance and adoption of heat health warnings and adaptive capacity, a change that could become apparent once social media is again readily available for analysis.

Funding: This work was funded by an internal grant from the Charles Darwin University. Kerstin Zander has received research support from the Alexander von Humboldt-Foundation.

Competing interests: The authors have no relevant financial or non-financial interests to disclose.

Data availability: Data (all Twitter conversations) are made available upon request. Please contact the corresponding author.

Author Contributions: Kerstin Zander: Project management, Funding acquisition, Supervision, Conceptualisation, Methodology, Data analysis and interpretation, Visualisation, Drafting the article; Duy Nguyen: Methodology, Data collection, Data analysis and interpretation; Stephen Garnett: Conceptualisation, Drafting the article.

Abedin B, Babar A. 2018. Institutional vs. non-institutional use of social media during emergency response: a case of Twitter in 2014 Australian bush fire. Inf Syst Front 20:729-740.
Al Balawi, R., Hu, Y., & Qiu, L. (2023). Get a Word in Edgewise: Post Character Limit and Social Media-Based Customer Service. Proceedings of the 56th Hawaii International Conference on System Sciences 5761-5770
an der Heiden M, Muthers S, Niemann H, et al. 2019a. Schätzung hitzebedingter Todesfälle in Deutschland zwischen 2001 und 2015. Bundesgesundheitsbl 62:571-579.
an der Heiden AM, Buchholz U, Uphoff H. 2019b. Schätzung der Zahl hitzebedingter Sterbefälle infolge derHitzewelle2018. Epid Bull 23:193-197.
Bakhsh K, Rauf S, Zulfiqar F. 2018. Adaptation strategies for minimizing heat wave induced morbidity and its determinants. Sustain Cities Soc 41:95-103.
Ballester J, Quijal-Zamorano M, Méndez Turrubiates RF, et al. 2023. Heat-related mortality in Europe during the summer of 2022. Nat Med 29:1857-1866.
Beckmann SK, Hiete M, Schneider M, Beck C. 2021. Heat adaptation measures in private households: an application and adaptation of the protective action decision model. Humanit Soc Sci Commun 8:227.
Bi P, Williams S, Loughnan M, Lloyd G, et al. 2011. The effects of extreme heat on human mortality and morbidity in Australia: implications for public health. Asia Pac J Public Health 23:27S-36S.
Blei DM, Ng AY, Jordan MI. 2003. Latent Dirichlet Allocation. J Mach Learn Res 3:993-1022.
Blei DM, Lafferty JD. 2007. A correlated topic model of science. Ann Appl Stat 1:17-35.
Blog2Social 2023. Social-Media-Nutzerzahlen 2023: Social-Media-Nutzer in Deutschland und weltweit. https://www.blog2social.com/de/blog/social-media-nutzer [accessed Feb 2024]
Bogdanovich E, Brenning A, Reichstein M, et al. 2024. Official heat warnings miss situations with a detectable societal heat response in European countries. Int J Disaster Risk Reduct 100:104206.
Boyer T, Blunden J, Dunn RJH. 2021. State of the Climate in 2020. Bull Amer Meteor Soc 102: S1-S10.
Brosch T. 2021. Affect and emotions as drivers of climate change perception and action: a review. Curr Opin Behav Sci 42:15-21.
Butsch C, Beckers LM, Nilson E, Frassl M, Brennholt N, Kwiatkowski R, Söder M. 2023. Health impacts of extreme weather events – Cascading risks in a changing climate. J Health Monit 8:33-56.
Capon A, Jay O, Ebi K, Lo S. 2019. Heat and health: a forthcoming Lancet Series. Lancet 394:551-552.
Carlson R, Elenius K, Granstrom B, Hunnicutt S. 1985. Phonetic and orthographic properties of the basic vocabulary of five European languages. Q Prog Stat Rep 1:63-94.
Cecinati F, Matthews T, Natarajan S, McCullen N, Coley D. 2019. Mining social media to identify heat waves. Int J Environ Res Public Health 16:762.
Chang J, Boyd-Graber J, Gerrish S, Wang C, Blei DM. 2009. Reading tea leaves: how humans interpret topic models. Neural Inf Process Syst 22:288-296.
Chang CH, Deshmukh NR, Armsworth PR, Masuda YJ. 2023. Environmental users abandoned Twitter after Musk takeover. Trends Ecol Evol 38:893-895.
Copernicus Climate Change Service. 2021. European State of the Climate 2021, Summary. Doi:10.21957/9d7g-hn83
Copernicus Climate Change Service. 2024. European State of the Climate 2023, Summary. Doi:10.24381/bs9v-8c66
Eberl JM, Tolochko P, Jost P, Heidenreich T, Boomgaarden HG. 2020. What’s in a post? How sentiment and issue salience affect users’ emotional reactions on Facebook. J Inf Technol Politics 17:48-65.
Ebi KL, et al. 2021. Hot weather and heat extremes: health risks. Lancet 398:698-708.
Effrosynidis D, Sylaios G, Arampatzis A. 2022. Exploring climate change on Twitter using seven aspects: Stance, sentiment, aggressiveness, temperature, gender, topics, and disasters. Plos One 17(9):e0274213.
Faurie C, Varghese BM, Liu J, Bi P. 2022. Association between high temperature and heatwaves with heat-related illnesses: A systematic review and meta-analysis. Sci Total Env 852:158332.
Gligoric K, Czestochowska J, Anderson A, West R. 2022. Anticipated versus actual effects of platform design change: a case study of twitter's character limit. Proc ACM Hum-Comput Interact 6:1-29.
Grasso V, Crisci A, Morabito M, Nesi P, Pantaleo G. 2017. Public crowdsensing of heat waves by social media data. Adv Sci Res 14:217-226.
Guardaro M, Hondula DM, Ortiz J, Redman CL. 2022. Adaptive capacity to extreme urban heat: the dynamics of differing narratives. Clim Risk Manag 35:100415.
Hannemann L, Janson D, Grewe HA, et al. 2023. Heat in German cities: a study on existing and planned measures to protect human health. J Public Health, ahead of print. doi:10.1007/s10389-023-01932-2
Hausmann A, Toivonen T, Fink C, et al. 2020. Understanding sentiment of national park visitors from social media data. People Nat 2:750-760.
Heidenreich A, Thieken AH. 2024. Individual heat adaptation: analyzing risk communication, warnings, heat risk perception, and protective behavior in three German cities. Risk Analysis, ahead of print. doi:10.1111/risa.14278
Heilmann K, Kahn ME, Tang CK. 2021. The urban crime and heat gradient in high and low poverty areas. J Public Econ 197:104408.
IPCC. 2021: Summary for Policymakers. In: Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, Masson-Delmotte V, Zhai P, Pirani A, et al. (eds), Cambridge University Press, Cambridge, UK, pp 3-32.
Jacobi C, van Atteveldt W, Welbers K. 2016. Quantitative analysis of large amounts of journalistic texts using topic modelling. Digital Journalism 4:89-106.
Jaidka K, Zhou A, Lelkes Y. 2019. Brevity is the soul of Twitter: the constraint affordance and political discussion. J Commun 69:345‑372.
Jay O. 2021. Reducing the health effects of hot weather and heat extremes: from personal cooling strategies to green cities. Lancet 398:709-724.
Jung J, Uejio CK. 2017. Social media responses to heat waves. Int J Biometeorol 61:1247-1260.
JustAnotherArchivist. 2020. Snscrape. GitHub. https://github.com/JustAnotherArchivist/snscrape [accessed June 2022]
Kabisch N, Remahne F, Ilsemann C, et al. 2023. The urban heat island under extreme heat conditions: a case study of Hannover, Germany. Sci Rep 13:23017.
Karami A, Gangopadhyay A, Zhou B, Kharrazi H. 2018. Fuzzy approach topic discovery in health and medical corpora. Int J Fuzzy Syst 20:1334-1345.
Kušen E, Strembeck M. 2021. “Evacuate everyone south of that line” Analyzing structural communication patterns during natural disasters. J Comput Soc Sc 4:531-565.
Kusumasari B, Prabowo NPA. 2020. Scraping social media data for disaster communication: how the pattern of Twitter users affects disasters in Asia and the Pacific. Nat Hazards 103:3415-3435.
Kirilenko AP, Molodtsova T, Stepchenkova SO. 2015. People as sensors: Mass media and local temperature influence climate change discussion on Twitter. Glob Environ Change 30:92-100.
Kreienkamp F, Philip SY, Tradowsky JS, et al. 2021. Rapid Attribution of Heavy Rainfall Events Leading to the Severe Flooding in Western Europe during July 2021, World Weather Attribution. https://www.worldweatherattribution.org/wp-content/uploads/Scientific-report-Western-Europe-floods-2021-attribution.pdf [accessed May 2024]
Kryvasheyeu Y, Chen H, Obradovich N, et al. 2016. Rapid assessment of disaster damage using social media activity. Sci Adv 2:e1500779.
Lenton TM, Xu C, Abrams JF et al 2023. Quantifying the human cost of global warming. Nat Sustain 6:1237-1247.
Li J, Feng C, Yang J. 2023. Climate attribution of interpersonal violence: International evidence. Environ Res 236:116836.
Luna S, Pennock MJ. 2018. Social media applications and emergency management: a literature review and research agenda. Int J Disaster Risk Reduct 28: 565-577.
Lyu F, Zhou L, Park P, Baig F, Wang S. 2024. Mapping dynamic human sentiments of heat exposure with location-based social media data. Int J Geogr Inf Sci, ahead of print, doi:10.1080/13658816.2024.2343063
Mahoney J, Le Louvier K, Lawson S, Bertel D, Ambrosetti E. 2022. Ethical considerations in social media analytics in the context of migration: lessons learned from a Horizon 2020 project. Res Ethics 18:226-240.
Mohr S, Ehret U, Kunz M, et al. 2023. A multi-disciplinary analysis of the exceptional flood event of July 2021 in central Europe–Part 1: event description and analysis. Nat Hazards Earth Syst Sci 23:525-551.
Moore F, Obradovich N, Lehner F, Baylis P. 2019. Rapidly declining remarkability of temperature anomalies may obscure public perception of climate change. Proc Natl Acad Sci USA 116:4905-4910.
Mücke H-G, Litvinovitch JM. 2020. Heat extremes, public health impacts, and adaptation policy in Germany. Int J Environ Res Public Health 17:7862.
Mukherjee S, Mishra AK, Zscheischler J, et al. 2023. Interaction between dry and hot extremes at a global scale using a cascade modeling framework. Nat Commun 14:277.
Mumenthaler C, Renaud O, Gava R, Brosch T. 2021. The impact of local temperature volatility on attention to climate change: Evidence from Spanish tweets. Glob Environ Change 69:102286.
Murakami D, Yamagata Y. 2020. Spatial heat-wave assessments using geotagged Twitter data. In Spatial Analysis Using Big Data, Yamagata Y, Seya H (eds), Elsevier Academic Press, London, pp 227-238.
NASA and NOAA. 2016. NASA, NOAA Analyses Reveal Record-Shattering Global Warm Temperatures in 2015. https://www.nasa.gov/news-release/nasa-noaa-analyses-reveal-record-shattering-global-warm-temperatures-in-2015/ [accessed March 2024]
NASA and NOAA. 2017. NASA, NOAA data show 2016 warmest year on record globally. https://www.nasa.gov/news-release/nasa-noaa-data-show-2016-warmest-year-on-record-globally/ [accessed March 2024]
NASA. 2021. 2020 Tied for warmest year on record, NASA analysis shows. https://climate.nasa.gov/news/3061/2020-tied-for-warmest-year-on-record-nasa-analysis-shows/ [accessed March 2024]
Neger C, León-Cruz JF, Gössling S. 2024. The tourism fire exposure index for the European Union. Tour Manag 103:104901.
Neppalli VK, Caragea C, Squicciarini A, Tapia A, Stehe S. 2017. Sentiment analysis during Hurricane Sandy in emergency response. Int J Disaster Risk Reduct 21:213-222.
Ogie RI, James S, Moore A, Dilworth T, Amirghasemi M, Whittaker J. 2022. Social media use in disaster recovery: a systematic literature review. Int J Disaster Risk Reduct 70:102783.
Qiang J, Qian Z, Li Y, Yuan Y, Wu X. 2022. Short text topic modeling techniques, applications, and performance: a survey. IEEE Trans Knowl Data Eng 34:1427-1445.
Ogunbode CA, Pallesen S, Böhm G, et al. 2023. Negative emotions about climate change are related to insomnia symptoms and mental health: Cross-sectional evidence from 25 countries. Curr Psychol 42:845-854.
Osberghaus D, Abeling T. 2022. Heat vulnerability and adaptation of low-income households in Germany. Glob Environ Change 72:102446.
Otrachshenko V, Popova O, Tavares J. 2021. Extreme temperature and extreme violence: evidence from Russia. Econ Inq 59:243-262.
Prein AF, Rasmussen RM, Ikeda K, et al. 2017. The future intensification of hourly precipitation extremes. Nature Clim Change 7:48-52.
Rauh C. 2018. Validating a sentiment dictionary for German political language—a workbench note. J Inf Technol Politics 15:319-343.
Roberts ME, Stewart BM, Airoldi EM. 2016. A model of text for experimentation in the social sciences. J Am Stat Assoc 111:988-1003.
Roberts M, Stewart BM, Tingley D. 2019. Stm: an R Package for Structural Topic Models. J Stat Softw 91:1-40.
Robine J-M, Cheung SLK, Le Roy S, et al. 2008. Death toll exceeded 70,000 in Europe during the summer of 2003. C R Biol 331:171-178.
Rousi E, Kornhuber K, Beobide-Arsuaga G, et al. 2022. Accelerated western European heatwave trends linked to more-persistent double jets over Eurasia. Nat Commun 13:3851.
Sandholz S, Sett D, Greco A, Wannewitz M, Garschagen M. 2021. Rethinking urban heat stress: Assessing risk and adaptation options across socioeconomic groups in Bonn, Germany. Urban Clim 37:100857.
Schmitt J, Offermann F, Söder M, Frühauf C, Finger R. 2022. Extreme weather events cause significant crop yield losses at the farm level in German agriculture. Food Policy 112:102359.
Silver A, Andrey J. 2019. Public attention to extreme weather as reflected by social media activity. J Conting Crisis Manag 27:346-358.
Smart Home Fox. 2024. Twitter Statistiken. https://www.smart-home-fox.de/twitter-nutzer-statistiken#:~:text=Die%20meisten%20deutschen%20Nutzer%20sind%2014%20bis%2029,wird%20von%2043%20%25%20der%20Unternehmen%20weltweit%20verwendet [accessed May 2024]
Thompson R, Hornigold R, Page L, Waite T. 2018.Associations between high ambient temperatures and heat waves with mental health outcomes: a systematic review. Publ Health 161:171-191.
Thonfeld F, Gessner U, Holzwarth S, et al. 2022. A first assessment of canopy cover loss in Germany’s forests after the 2018–2020 drought years. Remote Sens 14(3):562.
Tonidandel S, Summerville KM, Gentry WA, Young SF. 2022. Using structural topic modeling to gain insight into challenges faced by leaders. Leadersh Q 33:101576.
Torricelli M, Falkenberg M, Galeazzi A, et al. 2023. How does extreme weather impact the climate change discourse? Insights from the Twitter discussion on hurricanes. PLOS Climate 2(11):e0000277.
Trenczek J, Lühr O, Eiserbeck L, Sandhövel M, Ibens D. 2022. Schäden der Dürre- und Hitzeextreme 2018 und 2019. Prognos AG. https://www.prognos.com/sites/default/files/2022-07/Prognos_KlimawandelfolgenDeutschland_Detailuntersuchung%20Hitzesommer%2018_19_AP2_3a_.pdf [accessed March 2024]
Tuholske C, Caylor K, Funk C, Verdin A, Sweeney S, Grace K, Peterson P, Evans T. 2021. Global urban population exposure to extreme heat. Proc Natl Acad Sci USA 118:e2024792118.
Turek-Hankins LL, Coughlan de Perez E, Scarpa EG, et al. 2021. Climate change adaptation to extreme heat: a global systematic review of implemented action. Oxford Open Climate Change 1:kgab005.
UN-Habitat. 2022. World Cities Report 2022. United Nations Human Settlements Programme (UN-Habitat), Nairobi, Kenya.
Vicedo-Cabrera AM, Scovronick N, Sera F, et al. 2021. The burden of heat-related mortality attributable to recent human-induced climate change. Nat Clim Change 11:492-500.
Wang Z, Ye X, Tsou MH. 2016. Spatial, temporal, and content analysis of Twitter for wildfire hazards. Nat Hazards 83:523-540.
Wang J, s Nikolaou N, an der Heiden M, Irrgang C. 2024. Past, present, and future heat-related mortality in Germany. Preprint 9 Feb 2024. doi:10.21203/rs.3.rs-3920808/v1
Wang Y, Taylor JE. 2018. Coupling sentiment and human mobility in natural disasters: a Twitter-based study of the 2014 South Napa Earthquake. Nat Hazards 92:907-925.
Winklmayr C, Muthers S, Niemann H, Mücke H-G, Heiden M. 2022. Heat related mortality in Germany from 1992 to 2021. Dtsch Arztebl Int 119:451-457.
Winklmayr C, Matthies-Wiesler F, et al. 2023. Heat in Germany: Health risks and preventive measures. J Health Monit 8:3-32.
Wouters H, De Ridder K, Poelmans L, et al. 2017. Heat stress increase under climate change twice as large in cities as in rural areas: a study for a densely populated midlatitude maritime region. Geophys Res Lett 44:8997-9007.
Xue J, Chen J, Chen C, Zheng C, Li S, Zhu T. 2020. Public discourse and sentiment during the COVID 19 pandemic: using Latent Dirichlet Allocation for topic modeling on Twitter. PloS One 15:e0239441.
Yiou P, Cattiaux J, Faranda D, et al. 2020. Analyses of the northern European summer heatwave of 2018. Bull Am Meteorol Soc 101:S35-S40.
Zander KK, Botzen WJW, Oppermann E, Kjellstrom T, Garnett ST. 2015. Heat stress causes substantial labour productivity loss in Australia. Nat Clim Change 5:647-651.
Zander KK, Moss SA, Garnett ST. 2017. Drivers of self-reported heat stress in the Australian labour force. Environ Res 152:272-279.
Zander KK, Shalley F, Tan G, Dyrting S, Taylor A. 2021. “Run air-conditioning all day”: adaptation pathways to increasing heat in the Northern Territory of Australia. Sustain Cities Soc 74:103194.
Zander KK, Baggen HS, Garnett ST. 2023a. Topic modelling the mobility response to heat and drought. Clim Change 176:42.
Zander KK, Rieskamp J, Mirabaie M, Alazab A, Nguyen D. 2023b. Responses to heat waves: what can Twitter data tell us? Nat Hazards 116:3547-3564.
Zander KK, Garnett ST, Ogie R, Alazab M, Nguyen D. 2023c. Trends in bushfire related tweets during the Australian ‘Black Summer’ of 2019/20. For Ecol Manag 545:121274.
Zander KK, van Hoof J, Carter S, Garnett ST. 2023d. Living comfortably with heat in Australia – preferred temperatures and climate zones. Sustain Cities Soc 96:104706.
Zanocco C, Sousa-Silva R. 2023. Extreme heat experience influences public support for local climate adaptation policies in Germany. Urban Clim 52:101759.
Zscheischler J, Fischer EM. 2020. The record-breaking compound hot and dry 2018 growing season in Germany. Weather and Clim Extremes 29:100270.

SupplementaryMaterials.docx

Download PDF

Reviewers agreed at journal
22 Jul, 2024
Reviewers invited by journal
16 Jul, 2024
Editor assigned by journal
03 Jun, 2024
First submitted to journal
30 May, 2024

You are reading this latest preprint version

A decade of German heatwaves evident in social media data

Status:

Version 1

Abstract

Figures

1. Introduction

2. Methods

2.1. Data collection

2.2. Data processing

2.3. Structural topic modelling and coherence score

2.4. Sentiment analysis

3. Results

3.1 Spatial and temporal distribution of tweets

3.2. Tweet content and related sentiments

4. Discussion

4.1. Tweet activities and sentiments over time

4.2. The Year 2018

4.3. Majority topics

4.4. Minority topics

4.5. Study limitations

4.6. Study applications

Conclusion

Declarations

References

Supplementary Files

Status:

Version 1