Flash droughts and their impacts—using newspaper articles to assess the perceived consequences of rapidly emerging droughts

Flash droughts (FDs) have attracted increasing attention in the past decade. They are characterised by a rapid depletion of soil moisture resulting from interactions between the soil and atmospheric conditions. To date, there is a lack of consistent FD definitions and an understanding of their socio-economic impacts. Here, we explore the relationship between biophysical FD conditions and the perceived impacts of FDs in Germany between 2000 and 2022. We measured perceived impacts by analysing consequences reported in newspaper articles (2000–2022) and online search behaviour using Google trends data (2004–2022). To characterise the physical process, we considered root zone soil moisture data. Our results show that FDs are becoming increasingly frequent in Germany, occurring once every two years on average. Despite the lack of knowledge from the general public regarding the phenomenon of FDs, the peaks of interest in drought impacts correspond to the physical occurrence of FDs across the country. We identified an average time gap of four weeks between FD onset and the reporting of perceived impacts. This gap is longer than the average duration of FDs’ onset. Consequently, our findings highlight that consistent monitoring of FD conditions and drivers is necessary to guarantee effective preparedness. As impact perception is too slow to allow the adoption of mitigation measures, FDs require new schemes for response measures compared with slowly emerging (conventional) drought events. The novel method also allows the consistent and impact-based validation of FD identification methods.


Introduction
Recently, the scientific community has shown an increased interest in studying flash droughts (FDs).In particular, the 2012 FD that started in late May and by July affected over 75% of croplands in the USA Midwest.This disastrous event devastated central areas in the USA and produced increased attention from scholars on a term that was rarely used before (Basara et al 2019, Lisonbee et al 2021).The 2012 USA FD led to over US$30 billion in losses across the country because the rapid change from wet-to-dry soil moisture hindered timely prevention and mitigation (Christian et al 2019).The 2012 FD made the public aware of a new threat to the water-food nexus posed by rapid onset drought conditions (Pendergrass et al 2020), which is expected to occur more frequently worldwide under climate change (Yuan et al 2023).
FDs are a particular type of drought characterised by the rapid onset, in contrast to the slow-evolving attribute of conventional droughts.Multiple methods to define and identify FD have been proposed in the academic literature (Alencar et al 2022).An initial debate focused on whether FDs were shortlived droughts (Mo and Lettenmaier 2016), or intense evolving droughts (Ford and Lobosier 2017).The current consensus supports the latter (Christian et al 2024).A leading understanding describes FDs as emerging rapidly by transitioning from average to high soil moisture to dry conditions (Yuan et al 2023).However, seminal questions remain concerning what should be considered an FD.Although clear thresholds have been proposed (Christian et al 2019, Li et al 2020, Osman et al 2021, Yuan et al 2023), their transferability across case studies is limited (Alencar et al 2022).The lack of well-established criteria has led to a growing diversity of definitions and monitoring tools for FDs.However, these definitions and tools still lack validation and transferability (Lisonbee et al 2021, Alencar et al 2022).Hence, FD research would benefit from studies applying multiscale and multivariable frameworks (Liang and Yuan 2021).
Here, we propose a different perspective for FD identification.Similar to their slower counterparts (i.e.longer and slow-evolving droughts; henceforth referred to as conventional droughts), FDs affect multiple sectors of society, such as agriculture, animal husbandry, energy and water supply, and social activities (Christian et al 2020, Walker et al 2024).Although various existing FD indices are centred on hazards (Alencar et al 2022), they may not consistently translate into socioeconomic impacts.Therefore, the conditions leading to impactful FDs are still uncertain.Here, we suggest that understanding society's perception of the impacts of FDs can assist in diminishing these uncertainties.
We present a novel approach to studying FDs by considering the link between the biophysical occurrence of FDs and their perceived impacts.We assess the impacts based on impacts reported in newspaper articles and the internet search behaviour of local populations which represent two popular approaches for measuring local drought awareness.
Newspapers often serve as real-time recorders of societal responses and reactions to extreme weather events (e.g.Sodoge et al 2024, Bogdanovich et al 2024)), offering a unique perspective on the social impacts of FDs.Similarly, internet searches can approximate public awareness of a specific topic (Kam et al 2019, Kim andKim 2023).This approach can help researchers gain a more comprehensive understanding of the socio-economic implications of FDs, enhancing their ability to develop effective strategies for preparedness, mitigation, and adaptation.
To test whether digital trace data (e.g.news articles and Google searches) can support the identification of FD occurrence, we considered the case of Germany from 2000 to 2022.Germany was chosen as a case study because of its recent history of major conventional droughts (2003, 2015, and 2018-2022) that had widespread impacts on sectors such as agriculture, forestry, livestock, and waterways navigation, which have gained public attention (Tijdeman and Menzel 2021, Rakovec et al 2022, Sodoge et al 2024).We obtained daily time series for biophysical FD occurrence, reported impacts in newspapers, and online search behaviour on a district level in Germany.Thus, we investigate three hypotheses: -H1 -FD is a common phenomenon in Germany, and its occurrence has increased over the last few decades.-H2 -Impacts reported in the media and searched by internet users are a source of information for risk assessment that can be used to identify when FDs impact society -H3 -There is a delay between FD onset and impact reporting due to latency time for water/plant/soil/society response to drier conditions.

Data and methods
Our research design comprises multiple steps that involve three distinct datasets: soil moisture, online search behaviour measured using Google trends, and reported drought impacts in newspaper articles (figure 1).For each, we obtained daily time series on a district-level using the statistical regions of Germany defined by the Nomenclature of Territorial Units for Statistics (NUTS) levels 1, 2, and 3. We used the most fine-grained NUTS level available for each dataset for the main analysis and provided the results for all NUTS levels in the supplementary material.To explore the relationship between hazard and impactrelated indicators, we employed time-series crosscorrelation analysis.The following sections provide a detailed description of each step in the research design.

Soil moisture data and FD identification
For identifying FDs, we used the method suggested by Ford and Labosier (2017) which is based on root-zone soil moisture and was successfully applied in multiple case studies (Otkin et al 2018, Lesinger andTian 2022).This method was chosen due to its use of soil moisture in shallow compartments, which has been pointed as the most important variable for FD occurrence and that best relates to impacts  if (1) the soil moisture dropped from at or above the 40th percentile to at or below the 20th percentile within four pentads, (2) the soil moisture remained below the 30th percentile for at least three pentads after the onset to avoid classifying very short events, and (3) the event ended after soil moisture reached the 40th percentile.Percentiles were computed based on a grouping of pentad-years, wherein each pentad (1 to 73) was juxtaposed with its corresponding ith pentad for each year within the 1980-2022 period.This means that the percentile distributions were calculated separately for each pentad across all years in the dataset.For instance, all data points corresponding to the 1st pentad of each year were used to compute the percentile distribution for the 1st pentads, all data points for the 2nd pentad were used to compute the percentile distribution for the 2nd pentads, and this process was repeated for all 73 pentads.These criteria were computed using the detailed soil moisture dataset by Boeing et al (2022) and the method of Ford and Lobosier (2017) implemented in the R-package fdClassify (Alencar et al 2022).Criterion 1 defines the intensification phase, which ends at the FD onset.Criterion 2 defines a minimum persistence after onset to avoid overidentification of events (Pendergrass et al 2020), and criterion 3 establishes the end of the event, indicating a return to normal conditions.Although the spatial resolution of the FD dataset corresponded to 4 km, we aggregated the occurrences at NUTS-1, NUTS-2, and NUTS-3 levels.For each NUTS level, the percentage of the area under FD conditions was computed at each pentad.At the NUTS-1 level, the three German city-states (Berlin, Bremen, and Hamburg) joined the surrounding states (Brandenburg, Niedersachsen, and Schleswig-Holstein, respectively) because of their relatively small size compared to the other states.

Trend analysis
To explore the temporal patterns in the occurrence of FDs in Germany (H1), we assessed the trends in their intensity, duration, and frequency from 1980-2022.Here, we define FD intensity as the mean drop in soil moisture percentiles over the intensification phase, measured in percentile.day−1 .Duration refers to the period from the start of intensification until the return to normal conditions.Frequency is considered as the expected number of years with an occurrence of at least one FD, ranging from 0-1.
We used the regional Mann-Kendall test to perform this trend analysis and to compute and identify possible changes in FD intensity, duration, and frequency (Mann 1945, Helsel et al 2020).This test evaluates the monotonic trend in the time series data to identify the significance of increasing or decreasing trends.We used the data on identified FDs (section 2.1) at the NUTS-2 level to assess whether the frequency, duration, and intensity of FDs in Germany increased considerably during the study period (H1).Furthermore, we computed the regional Theil-Sen estimator (Sen 1968, Theil 1992) to estimate the slope of these trends (i.e. the trend direction and magnitude).The Theil-Sen estimator, known for its robustness in assessing rates of changes in time-series, is less sensitive to outliers, and does not rely on assumptions regarding the distribution of data and residuals (Wang and Yu 2005).It is computed as the median of the slope of all lines that connect all pairs of points in the data series (Helsel et al 2020).

Digital trace data on perceived impacts and public awareness 2.2.1. Socio-economic impact data
To assess the perceived impacts of FDs on different socio-economic sectors, we used information on reported socio-economic impacts in newspaper articles.Specifically, we used the approach and data by Sodoge et al (2024), which consider the frequency of newspaper articles reporting on a specific drought impact using text-mining approach as follows.We examined approximately 50 000 newspaper articles that included terms related to conventional drought (e.g.Dürre * and Trockenheit in German).These were published between 2000 and 2022 in 250 national, regional, and local German newspaper outlets included in the genios.denewspaper aggregator database.Articles were automatically reviewed using text-mining techniques to detect reported impacts in each article and its geographic scope using NUTS levels 1, 2, and 3. First, duplicate articles were automatically removed by considering the Jaccard similarity score (Mullen 2020).To this end, a threshold of 95% was defined, where 100% indicates full similarity.We then categorized the types of impacts using lasso logistic regression models.These models were trained and assessed using a set of 1800 manually annotated newspaper articles (Madruga de Brito et al 2020).Hyperparameter tuning was conducted using cross-validation within the training sample.A median accuracy of 89% was obtained for these models compared to the manually annotated data.The following impact types were covered: agriculture (crop yield losses), energy (impacts on reduced energy production in nuclear power plants), fire (forest wildfires or fires in other areas due to drought conditions), livestock (impacts on livestock farming and animal populations), and social (impacts on places and activities used for recreation, tourism, and leisure).To locate impact areas, we identified the cities, regions, and districts mentioned in each article using named-entity recognition.We then selected the regional cluster with the highest frequency of detected locations.A previous study by Sodoge et al (2024) provides a detailed description of this procedure, including extensive validation with external indicators and observed impacts such as crop yield losses, forest fire statistics, and precipitation deficit.
The resulting dataset contains daily timeseries data for each NUTS unit (levels 1,2, and 3) of the number of reported impacts.We further aggregated the time series into pentads to compare the data with the FD measurement.At the NUTS-1 level, the three German city-states (Berlin, Bremen, and Hamburg) joined the surrounding states (Brandenburg, Niedersachsen, and Schleswig-Holstein, respectively) because of their small size compared to the other states.

Public awareness data
To determine public awareness of FDs, we collected information on spatio-temporal patterns of internet search frequency for the topic 'drought' using Google trends data.This service provides an index of how many Google users on a particular day and region search for a specific term, where a higher value (maximum 100) corresponds to the maximum interest for a term.Previous studies have shown the capabilities of Google trends data to monitor the awareness of droughts from a spatio-temporal perspective (Kam et al 2019).However, Google does not reveal details and changes in the underlying algorithms.We extracted the daily time series for 2004-2022 for each NUTS-1 (only scale available) unit using the PyTrends Python package (Hogue and DeWilde 2023).We then aggregated the time series into pentads to compare the different datasets in subsequent analyses.

Correlation analysis
To identify whether the reported impacts in the news media and internet search behaviour were influenced by the occurrence of FDs (H2), we performed a cross-correlation analysis between the datasets.This allowed us to assess the temporal delays between the occurrence of FDs and impact perception (H3).Specifically, we employed cross-correlation functions (CCF-Shumway and Stoffer 2019) that can identify delays between two variables, offering insights into whether one variable can predict the other.In this study, these delays represent the asynchronous behaviours expected between the peaks of FD onset and impact reporting (hypothesis H3).CCFs are commonly used to assess the similarity between two series (waveforms) as a function of their relative delay (equation 1) where x and y are time series, γ is the co-variance between x and y, ρ x and ρ y are the standard deviations of x and y, respectively, and i and j are time indexes.
It is important to notice that i and j are related, being j = i + h, where h is the delay between the time series.Using equation (1), we can identify the delay between both time series x and y (e.g.FD onset and impact reporting) as being the h that maximizes ρ x,y (i, i + h).

H1-FD is a frequent and intensifying phenomenon in Germany
Using a high-resolution soil moisture dataset (Boeing et al 2022) and the method described by Ford and Lobosier (2017), we assessed the occurrence of FDs in Germany from 1980-2022 (see figure 2).The results showed that FDs are a frequent phenomenon in Germany, with an average frequency of one event every 1.5 years.Particularly, the southern and southwestern regions experience the highest frequency of events, with FDs occurring almost every year (frequency of approximately 0.87).Conversely, the northeast, including the federal states of Saxony-Anhalt, Brandenburg, and Berlin, has a lower rate of FD occurrence.Here, FDs occurred in approximately 55% of the analysed years.
Temporal trend analysis revealed that FDs have increased in frequency and affected areas over the last few decades (see figure 3).From 1980-2022, Germany experienced FD events every year, at least in parts of its territory, with the minimum values of affected areas ranging from 10% (e.g.1987 and 2002) to a notable maximum of approximately 99% (in 2020).The affected area has consistently increased at a rate >9% per decade.Hence, the average annual affected area by FDs in Germany increased from approximately 40% at the beginning of the 1980 s to approximately 80% in the 2020 s.In particular, 2007In particular, , 2012In particular, , 2020In particular, , and 2022 had major FD occurrences in Germany.
Using the regional Mann-Kendall test, we assessed the spatial distribution of trends in FD intensity, duration, and frequency (figure 4).Only positive trends were identified across the NUTS-2 regions in Germany for all three variables.Furthermore, most trends were statistically significant, as indicated by the hatched areas.These findings indicate that FD is not only a frequent phenomenon (occurring almost yearly in some regions) in Germany (H1), but also an intensifying one.Figure 4(a) shows that FDs intensified across southern Germany, with rates increasing up to 1 percentile.day−1 .decade−1 .Meanwhile, figure 4(b) illustrates changes in the duration of FDs, showing particularly major trends in the southern and western regions.Here, FDs have become notably longer over the 2000 s and 2010 s, with durations increasing up to 30 days.decade−1 .These two results underscore the heightened hazard posed by FDs in the south and southwest of the country, where soil is rapidly drying at a faster rate and persisting for longer durations.
Figure 4(c) presents the changes in FD occurrence in terms of events per year.These events have become more frequent across all regions, with an average frequency of 5%.decade −1 .The region with the largest increase was Berlin (11%.decade−1 ).In contrast to the patterns observed for the other two variables analysed, NUTS 2 regions in the south presented a slightly lower trend in FD occurrence than their northern counterparts.

H2-The blind perception of FDs
The term 'Blitzdürre' (FD, in German) is not widespread in the news or is common knowledge in German society.In fact, of the over 260 million articles in the genios.denewspaper aggregator database, only 21 mentioned this term after 2023.Comparing the interest in searches using the keywords Blitzdürre and Dürre (the term for conventional drought, in German), we find that interest in FDs is, on average, 14 times lower (figure 5).During peak periods of interest in FDs in the early 2000 s, this accounted for only 16% of the searches using the keyword drought.Over the last five years, searches focusing on FDs have decreased to 0.5%.The keywords used to assess impact perception from news articles and public awareness from Google searches focused solely on drought-related terms because of the absence of the term FD in public discourse.Nevertheless, there is a common pattern of FD emergence and an increase in drought-related report impact on the media illustrated in figure 6.This consistent pattern of acknowledging FD occurence via the impact perception and their reporting as conventional droughts (i.e.Dürre) is what we call here as a 'blind perception' .

H3-Relationship between FD occurrence, perceived impacts, and public awareness
Comparing the time series of FD indices and reported impacts in newspapers, we found positive correlations (r = 0.65) and that following a rapid increase in areas affected by FD, there is a wave of impact reporting.Using CCF, we assess time lags between the occurrence of FDs and the reporting of their impacts on the news media (figure 7(a)).
The results for Germany as a whole show a uniform baseline awareness of droughts (see the blue line in figure 7(a)).We interpret this as reflecting the cultural and social perceptions of droughts, encompassing the expectation of future events and the recollection of past events.The baseline value was derived from the CCF computation, where the horizontal axis can be arbitrarily long in the positive and negative directions.To assess the media's baseline awareness of drought impacts, we used CCF values for long delays (above 15 weeks, a period longer than usual FDs) and negative delays under two weeks (to exclude potential forecast reports).The concept of negative delay results from the mathematical nature of CCF, which allows the identification of synchronicity independent of causality.
When examining the relationship between media-reported impacts and FD occurrences, we found that there was a consistent delay of four weeks before the impacts peaked.This suggests persistence of interest and reported impacts long after the onset of FDs.The four-week delay can be interpreted as the average time between FD onset and the perception and reporting of socioeconomic impacts.However, there were spatial differences between the changes in impact reporting under FD occurrence.In NUTS-2 and NUTS-3 regions, there was a notably higher frequency of impact reporting within the first two weeks compared to the NUTS-1 region (supplementary material).This highlights the effectiveness of localized news sources in monitoring and reporting the local conditions and impacts of droughts and FDs.
Comparing FD occurrence and online search behaviour data from Google Trends, we found similar correlations (r = 0.61) and trends in public awareness in response to the occurrence of FDs (see figure 7(b)).Overall, the value of the baseline awareness aligns  with our previous observation (figure 7(a)).However, the peak delay was shorter (three weeks) compared to the response of reported impacts in newspaper articles (four weeks).This shorter timeframe indicates that the general community engaging in online searches has a quicker perception of the impacts of FDs.One possible explanation for this is the time required between perception and reporting of the news, as there is an intrinsic interval necessary to collect, verify, and write news compared to online search behaviour.

Discussion
FDs are becoming increasingly frequent under climate change, and require different adaptation schemes owing to their rapid emergence (Yuan et al 2023, Christian et al 2024).In this regard, we investigated the occurrence of FDs in Germany, as well as societal awareness (Kam et al 2019) and perceived impacts (Sodoge et al 2024).Using these digital trace data, our study shows the delayed response of social perception to FDs as reported in the media and according to online search behaviour.Expanding on the results of our study, we contextualize them within the structure of our hypothesis presented in the Introduction.

Hypothesis 1-FD is a common phenomenon, and its occurrence has increased over the last few decades in Germany
Our analysis revealed that FDs have become increasingly frequent in Germany, occurring once every two years on average (figure 2).Notably, they are more prevalent in the southern and western parts of the country.In contrast, northeast Germany Furthermore, the trend analysis showed that in addition to being a common phenomenon, FDs are becoming more intense, longer, and more frequent across Germany, with positive trends in all three variables observed in all regions (NUTS-2).However, these trends are not uniform and can be linked to different factors.Specifically, FD intensity increased in the southern region, which is consistent with previous reports (Brunner et al 2023, Yuan et al 2023), especially in mountainous regions.The results indicated that the western and southern regions of Germany experienced an increase in the duration of FDs.These trends might be a consequence of differences in the regional climate.-3) and B: searches on Google and FD occurrences (NUTS-1).We observe a baseline of correlation, which reflects noise and pre-existing awareness or discussions regarding drought impacts in the media.The red line indicates the peak in the delay of FDs and impact reporting.Graphs for all NUTS levels are provided in the supplementary material.in all three variables, possibly induced by urban anthropogenic causes, such as soil sealing and water multiple water abstractions (Kuhlemann et al 2020), as well as land use changes upstream, particularly in Lausitz, where Lignite mining which artificially added water to the Berlin system and has progressively being reduced, with projection to complete phase-out by 2030 (Gerwin et al 2023).Nevertheless, the exact cause of the different trends in FD features is beyond the scope of this study, and further investigation is required.
Hypothesis 2-The impacts reported in the media and searched by internet users are a source of risk assessment that can be leveraged to identify when FDs impact society While prior studies focused on hazard-centric indicators (Ford and Lobosier 2017, Pendergrass et al 2020, Noguera et al 2021, Ho et al 2023) to identify FDs, our study demonstrates the effectiveness of using digital trace data to track FDs and their societal impacts.
We found that the use of the term 'FD' is not widespread in the German public discourse.Consequently, terms associated with drought were used to evaluate impact perception and public awareness.Despite the distinct nature and drivers of droughts and FDs, they are often perceived as a single phenomenon, as evidenced by the positive cross-correlation values (figure 7).Additionally, negative correlations between event occurrence, impact perception, and public awareness may be attributed to the semantic blending of flash and conventional droughts in public discourse.This blending results from drought-related impacts that are examined for synchronicity with FD events, as well as FDs that eventually become longer (conventional) droughts (Basara et al 2019).Additionally, while previous studies used Google Trends and newspaper data to enhance the modeling of social awareness in response to droughts (Kam et al 2019, Kim and Kim 2023), this study is the first to model FDs using innovative digital trace data and relate the results to hazard events.Thus, while modeling for larger, less-fine-grained temporal resolutions of heatwaves has been shown in previous studies (Polt et al 2023), we highlight how shorter drought hazards, such as FDs, can be monitored for shortterm responses.These results underscore the value of digital trace data in comprehending the impacts of FDs, unraveling the distinct characteristics that differentiate them from more gradually unfolding droughts.
Hypothesis 3-There is a delay between FD onset and impact reporting due to latency time for water/plant/soil/society response to drier conditions Using high-resolution soil moisture data and digital trace data at multiple levels of spatial aggregation (NUTS1, 2, and 3), we found a correlation between FD occurrence, impact reporting (r = 0.65), and Google searches (r = 0.61).The analysis of these digital trace data revealed a consistent delay between FD occurrence and the reporting of its impacts on the media (four weeks), as well as in Google searches (three weeks).Hence, these correlations and observed delays support H3 and H2.However, these delays are too long to be actionable because they are longer than the usual onset duration for FDs (up to 20 days-Ford and Lobosier 2017, Pendergrass et al 2020, Lisonbee et al 2021).There are two consequences of this finding.First, it reinforces the necessity of earlywarning and early-action systems to inform stakeholders about the occurrence of FDs (Sehgal et al 2021, Otkin et al 2022).Second, it provides a consistent metric for identifying impactful FDs that can be used to validate hazard identification methods, monitoring, and early warning systems (Alencar et al 2022).
Although recent studies have attempted to monitor and forecast FDs (Mukherjee and Mishra 2022, Ho et al 2023, Christian et al 2024), these efforts often fail to identify impactful events and validate their metrics and indices with observed societal consequences, leading to misidentification and misinformation (Alencar et al 2022).This study provides a novel technique to bridge this gap and allows the production of more reliable methods for FD identification and monitoring based on validated events.Currently, Germany does not monitor FDs, although German drought monitoring provides daily assessments of drought conditions (Zink et al 2016).Moreover, there is no tool in place for alert or early warning, despite the frequent occurrence of FDs and their major impact on multiple sectors, particularly agriculture.

Conclusions
Since 1980, we have observed an increase in the frequency and intensity of FDs.Different adaptation strategies are necessary to cope with their rapid emergence.We investigated the occurrence of FDs and their relationship to societal awareness in Germany.By analysing digital trace data on the perception of drought impacts, our study shows a delayed response of social perception to FDs in newspaper media and online searches.
We found that FD occurrence in Germany has been increasing over the past few decades, making FDs a common phenomenon.Synchronicities were observed between the reporting of drought-related impacts in the newspaper media and the incidence of FDs in Germany.However, societal awareness, measured by online search behaviour and reported impacts in newspapers, was dampened by a delay of, on average, four weeks compared to the emergence of the FD.These findings have two key implications: (1) digital trace data coupled with high-quality soil moisture data can successfully assist modelers in accurately validating designed metrics for FD identification, (2) although digital trace data serve as a rich source of information to determine when and where impactful FDs occur, it is insufficient to provide actionable information owing to delays in the publication of data.With the increasing threat posed by FDs in a changing climate, it is necessary to design information systems capable of integrating forecast assessments with potential impacts in vulnerable areas.

Figure 1 .
Figure 1.Overview of the research design: based on the three datasets for soil moisture, reported drought impacts, and online search behaviour, daily time series are obtained on a district level in Germany.Cross-correlation analysis is then employed to analyse the relationship between hazard indicators and consequent impacts.

Figure 2 .
Figure 2. Ratio of years with FD occurrence in Germany from 1980-2022.Events were identified using the Ford and Lobosier (2017) approach, which is based on soil moisture percentile changes at a depth of 40 cm.Soil moisture was assessed using the Helmholtz Centre for Environmental Research (UFZ) drought monitor dataset (Boeing et al 2022).The soil moisture data is generated through the mesoscale hydrological model (mHM-Samaniego et al 2017) and validated with TERENO and FLUXNET stations (Pastorello et al 2020, Bogena et al 2022) at a resolution of 4 kilometers.

Figure 3 .
Figure 3. Annual percentage of areas affected by FDs in Germany.Using the Mann-Kendall test, we found strong evidence (p-value < 0.01) of an upward trend in the yearly affected area, with an increase of >9% per decade.Grey shaded area indicates the 95% confidence interval.

Figure 4 .
Figure 4. Regional Kendall tests to assess trends in (a): intensity (drop in soil moisture percentiles per day during the onset phase), (b): duration (total event duration in days), and (c): frequency (years with events) of occurrence of FDs in Germany at NUTS-2 level from 1980-2022.Most regions presented statistically significant trends for all variables.

Figure 5 .
Figure 5.Comparison of searches on Google for the terms 'Dürre' (Drought-light grey) and 'Blitzdürre' (Flash drought-dark grey) in Germany, from 2004 to 2023.Searches in Google Trends are reported as a percentage of the maximum observed in the monitored period.
(e.g.Berlin and Brandenburg), which has sandier soils and lower plant-available water (Kruse 2015, Gebauer et al 2022), exhibited faster variations in soil moisture in terms of percentiles (Alencar et al 2022).This characteristic leads to a lower rate of FD intensification, which may explain the lower occurrence of FDs in the region.This does not necessarily imply that regions like Brandenburg experience less rapid drying such as FDs (Shah et al 2022).Rather, this suggests a limitation in the method used to identify these local events (Alencar et al 2022).Impact-based and multivariate indices and analysis are possible solutions for overcoming the limitations of traditional methods of identifying FDs (Mukherjee and Mishra 2022, Zhang et al 2022, Shyrokaya et al 2023).
The western and southern regions have a temperate oceanic climate (Cfb-Köppen-Geiger classification; Beck et al 2018) whereas the eastern regions have a colder humid continental climate (Dfb).Moreover, east Germany has sandier soils, with lower organic matter content and plant-available water (Kruse 2015, Gebauer et al 2022).The city of Berlin shows large trends

Figure 6 .
Figure 6.Examples of synchronicity between FD occurrence and impact reporting.These examples show the patterns of synchronicity at multiple locations (a) Schleswig-Holstein; (b) Brandenburg; (c) Nordhein-Westfalen; (d) Sachsen; (e) Baden-Wüttemberg; (f) Bayern).Prevalence indicates the percentage of the area affected by FDs, and Perception is the reporting of drought-related impacts according to news articles.For more examples, visit the open application PerceptionFD, used to generate the multiple plots in this figure.

Figure 7 .
Figure 7. Results of the cross-correlation function (CCF) between A: Impact reporting in news articles and FD occurrences (NUTS-3) and B: searches on Google and FD occurrences (NUTS-1).We observe a baseline of correlation, which reflects noise and pre-existing awareness or discussions regarding drought impacts in the media.The red line indicates the peak in the delay of FDs and impact reporting.Graphs for all NUTS levels are provided in the supplementary material.
et al 2022 Getting ahead of flash drought: from early warning to early action Bull.Am.Meteorol.Soc.103 E2188-202 Otkin J A, Svoboda M, Hunt E D, Ford T W, Anderson M C, Hain C and Basara J B 2018 Flash droughts: a review and assessment of the challenges imposed by rapid-onset droughts in the united states Bull.Am.Meteorol.Soc.99 911-9 Pastorello G et al 2020 The FLUXNET2015 dataset and the ONEFlux processing pipeline for eddy covariance data Sci.Data 7 225 Pendergrass A G et al 2020 Flash droughts present a new challenge for subseasonal-to-seasonal prediction Nat.Clim.Change 10 191-9 Polt K D, Ward P J, de Ruiter M, Bogdanovich E, Reichstein M, Frank D and Orth R 2023 Quantifying impact-relevant heatwave durations Environ.Res.Lett.18 104005 Rakovec O, Samaniego L, Hari V, Markonis Y, Moravec V, Thober S, Hanel M and Kumar R 2022 The 2018-2020 multi year drought sets a new benchmark in europe Earth's Future 10 e2021EF002394 Samaniego L et al 2017 Toward seamless hydrologic predictions across spatial scales Hydrol.Earth Syst.Sci.21 4323-46 Sehgal V, Gaur N and Mohanty B P 2021 Global flash drought monitoring using surface soil moisture Water Resour.Res.57 e2021WR029901 Sen P K 1968 Estimates of the regression coefficient based on Kendall's tau J. Am.Stat.Assoc.63 1379-89 Shah J, Hari V, Rakovec O, Markonis Y, Samaniego L, Mishra V, Hanel M, Hinz C and Kumar R 2022 Increasing footprint of climate warming on flash droughts occurrence in europe Environ.Res.Lett.17 064017 Shumway R H and Stoffer D S 2019 Time Series: A Data Analysis Approach Using R (Chapman and Hall/CRC) Shyrokaya A, Pappenberger F, Pechlivanidis I, Messori G, Khatami S, Mazzoleni M and Di Baldassarre G 2023 Advances and gaps in the science and practice of impact based forecasting of droughts WIREs Water 11 e1698 Sodoge J, Kuhlicke C, Mahecha M D and de Brito M M 2024 Text mining uncovers the unique dynamics of socio-economic impacts of the 2018-2022 multi-year drought in Germany Nat.Hazards Earth Syst.Sci.24 1757-77 Alencar P H L and Paton E N 2022 How do we identify flash droughts?a case study in central european croplands Hydrol.Wang Z, Wu X, Chen J, Guo S and Zhang Z 2020 A new framework for tracking flash drought events in space and time CATENA 194 104763 Liang M and Yuan X 2021 Critical role of soil moisture memory in predicting 2012 central USA flash drought Front.Earth Sci. 9 46 Lisonbee J, Woloszyn M and Skumanich M 2021 Making sense of flash drought: definitions, indicators and where we go from here J. Appl.Service Climatol.2021 1-19 Madruga de Brito M, Kuhlicke C and Marx A 2020 Near-realtime drought impact assessment: a text mining approach on the 2018/19 drought in germany Environ.Res.Lett.15 1040a9 Mann H B 1945 Nonparametric tests against trend Econometrica 13 245 Mo K C and Lettenmaier D P 2016 Precipitation deficit flash droughts over the united states J. Hydrometeorol.17 1169-84 Mukherjee S and Mishra A K 2022 A multivariate flash drought indicator for identifying global hotspots and associated climate controls Geophys.Res.Lett.49 e2021GL096804 Mullen L 2020 textreuse: detect text reuse and document similarity (available at: https://docs.ropensci.org/textreuse)Noguera I, Domínguez-Castro F and Vicente-Serrano S M 2021 Flash drought response to precipitation and atmospheric evaporative demand in spain Atmosphere 12 165 Osman M, Zaitchik B F, Badr H S, Christian J I, Tadesse T, Otkin J A and Anderson M C 2021 Flash drought onset over the contiguous united states: sensitivity of inventories and trends to quantitative definitions Hydrol.Earth Syst.Sci.