Flash drought drives rapid vegetation stress in arid regions in Europe

Flash droughts are characterised by rapid onset, combined with the potential to severely impact agriculture and ecosystems. However, assessments of the ecological impacts of flash droughts, especially in Europe, are largely lacking. Here we investigate ecosystem responses to flash droughts in Europe between 2001 and 2019 using diverse observational data, including gross primary production (GPP) and leaf area index (LAI). We find that in arid regions an abrupt transition to water-stressed conditions occurs within a few weeks, which negatively affects vegetation status and reduces carbon uptake in the initial stages of drought; normalised anomalies of LAI and GPP decrease to about −0.5. By contrast, vegetation in humid regions is not as severely affected, given that soil moisture recovers relatively quickly. We also show that soil moisture status before the onset of drought significantly impacts the timing (1–5 pentads) and degree (−0.33 to −0.71 of normalised LAI and GPP anomalies) of drought-induced vegetation stress, particularly in arid regions. Our results highlight the regional characteristics of flash drought impacts for more informative monitoring and early warning systems.


Introduction
Drought is one of the most destructive disasters with wide-ranging and far-reaching impacts [1,2]. While drought is typically considered a slow onset event that develops over many months or even years [3], during recent years, increasing attention has been devoted to droughts with a rapid onset or intensification, which are referred to as flash droughts [4][5][6][7][8]. The occurrence of flash droughts and the underlying mechanisms behind their rapid onset have been extensively examined in many regions worldwide, including Australia, China, India, and the US [6,[9][10][11][12][13][14][15][16]. In contrast to slower-developing conventional droughts, which are mainly driven by a precipitation shortage, flash droughts are triggered by a combination of strong precipitation deficits and anomalous increases in atmospheric evaporative demand [5,8], although the lead driver of flash drought development varies across regions [6].
Excessive evaporative demand at the land surface can be caused by drought-associated clear sky conditions (i.e. increased incoming solar radiation), amplification of air temperature, strong winds, or increased vapour pressure deficit, often under persistent atmospheric conditions [5,6,8,17].
Due to their sudden onset-often emerging within a few weeks-and complex contributing factors, flash droughts are difficult to predict [5,18]. The lack of early warning can lead to unusually extensive damage to agriculture and ecosystems. Earlier studies have recorded severe drought-induced vegetation stress and mortality during flash drought events, such as the 2012 central US flash drought, by analysing changes in vegetation conditions through satellite observations [19][20][21]. Significant changes in vegetation conditions following flash droughts have also been witnessed in India and China [22,23]. However, despite the clear evidence of the severity of flash drought impacts on ecosystems, our understanding of the ecological impacts of flash droughts remains incomplete. Most impact studies are limited to a few extreme events, or are very tightly geographically focused. Moreover, the increased likelihood of simultaneous dry and hot extremes under a changing climate may intensify flash drought conditions and associated impacts [7,13,[24][25][26], emphasising the necessity of systematically studying ecosystem responses to flash droughts using long-term records across different regions.
Here, we assess the impact of flash droughts on terrestrial ecosystems in Europe between 2001 and 2019 using diverse observational gridded datasets derived from remote sensing and in-situ measurements. While a recent study showed the widespread occurrence of flash droughts across Europe [27], an assessment of the impact of these events is still lacking. We identify flash drought events at a 0.25 • resolution based on soil moisture percentiles, and then compare the temporal evolution of ecosystem variables between different climate regimes (i.e. humid and arid regions). We also examine how prior soil moisture conditions influence the timing and magnitude of drought-induced damage. Our study aims to shed light on the regional characteristics of ecosystem responses during flash droughts in Europe, providing observational evidence of the ecological impacts of flash droughts and contributing to more effective drought monitoring and early warning.

Data
Multiple approaches have been suggested to identify flash drought based on various variables reflecting the rapid depletion of soil moisture [28]. Soil moisture quantile is one of the widely-used indices to identify flash droughts [8,9,12,13,27]. Soil moisture measurements are, however, often sparsely available in space. Due to this reason, modelled soil moisture (reanalysis) or alternative variables such as evaporative stress index have been used in previous studies [9,12,27]. Here we overcome this limitation by using gap-free multi-layer soil moisture data, SoMo.ml, derived from machine learning trained with in-situ soil moisture [29]. The dataset provides daily soil moisture at a 0.25 • resolution from 2000 to 2019.
In addition, we obtain meteorological variables such as precipitation and temperature from the E-OBS gridded dataset [30]-an ensemble of interpolated ground station data driven by multiple stochastic techniques. The E-OBS data, especially precipitation, often include missing values due to low station density, and we fill the gaps using climatological values of the same nth day. We also obtain net radiation from the European Centre for Medium-Range Weather Forecast Reanalysis v5 (ERA5) [31], which is relatively well constrained with satellite radiance observation through data assimilation. Both E-OBS and ERA5 data are available at daily and 0.25 • resolutions.
Lastly, we employ multiple observation-based datasets derived from remote sensing or in-situ data to obtain gross primary productivity (GPP) and leaf area index (LAI). GPP, defined as the total influx of carbon into an ecosystem through photosynthetic fixation of carbon, is obtained from FLUXCOM [32] generated through upscaling of energy and carbon flux measurements from FLUXNET eddy covariance towers using machine learning algorithms; the data generated by random forest using remote sensingbased inputs (RS product) is used in this study. GPP generated from a revised two-leaf light use efficiency model driven by satellite and reanalysis data is additionally considered [33]. LAI quantifies the amount of leaf area in a vegetation canopy, and the data are obtained from the gap-filled satellite data, including Moderate Resolution Imaging Spectroradiometer (MODIS)/Terra [34] and Global LAnd Surface Satellite (GLASS) [35,36]. GLASS LAI is developed using a machine learning model trained with MODIS surface reflectance data. All the data are interpolated to the 0.25 • resolution to match the spatial resolution of soil moisture. All the data are available at an 8-daily scale, and thereby, we linearly interpolate to a daily scale.

Flash drought identification
We define flash droughts based on the rapid change of soil moisture, mainly following the approach of [27], who recently studied flash droughts in Europe. We employ soil moisture of the top 30 cm depth computed as a depth-weighted average in the top two soil layers (0-10 cm and 10-30 cm depths) of SoMo.ml. Soil moisture is aggregated into 5-daily averages (pentad) to reduce noise in daily variability and then converted to percentile through the Gringorten plotting position approach [37]. The percentiles are determined at each grid pixel for the same pentad throughout the study period. Finally, the following conditions are considered to identify flash drought: (a) flash drought starts when soil moisture declines from at least 40th percentile to below 20 percentile within two following pentads; (b) during the initiation of drought, soil moisture percentiles should consistently decrease with the mean rate of at least 0.1 per pentad; (c) the flash drought event terminates when soil moisture increases back to above the 20th percentile threshold and stays for at least two pentads; and (d) a drought event with a duration of 6-18 pentads is chosen to reflect drought intensity adequately and to distinguish it from a longduration traditional drought [8,27]. See figure S1 in supplementary for the visualised description of flash drought. We consider flash drought events only during the growing months from April to September, i.e. from the 19th to 55th pentads of each year, when a significant impact of drought on vegetation is expected.

Data processing
Daily data of meteorological and ecological variables are aggregated into pentad-means, as done for soil moisture, and expressed as normalised anomalies. Anomalies are defined as the difference from the long-term average, which are then normalised by the respective pentad-of-year standard deviation. For the main analysis, we exclude grid pixels with a longterm average temperature lower than 5 • C and aridity higher than 8, as we assume that regions with such relatively extreme conditions show higher uncertainty in the observational data of soil moisture and ecosystem variables [29,38] and have relatively sparse vegetation. This filtering excludes grid pixels mainly in the Scandinavian countries and around the Alps (figure 1). Aridity is defined as a ratio of long-term equivalent evaporation to precipitation; the former is converted from net radiation by multiplying the inverse of the latent heat of vaporization, and it is expressed in mm. The aridity map can be found in figure S2. Figure 1 depicts the aspects of flash drought occurrences and their associated environmental conditions. Over the last 20 years, widespread flash droughts occurred across Europe, with nearly 57% of grid pixels considered in the study domain experiencing four or more flash drought events (figure 1(a)). A long-term analysis (1950-2019) conducted by [27] found a substantial increase in the frequency of flash droughts over the last decades, particularly in Central and Mediterranean Europe. Recently, more grid pixels of Europe have been affected by flash droughts (see the bar plot of figure 1(a)), particularly in 2018 and 2019, when Europe was affected by exceptionally dry and hot summers [39,40]. Maps of flash drought frequency for each year can be found in figure S3. We find that a great extent of areas in Europe is affected by flash droughts, e.g. in 2007, 2015, 2018, and 2019, which is consistent with the findings from [8]. Most flash droughts occurred in drier and warmer conditions or in drier and colder conditions (approximately 79% and 15% of total flash drought events, respectively), indicating a crucial role of precipitation deficits in flash drought initiations [6,27] (see figure 1(b)). To address uncertainty in soil moisture data, we repeat the analysis using ERA5 soil moisture and find the overall spatial patterns and annual variability of flash drought occurrences remain similar ( figure S4). Figure 2 shows the temporal evolution of soil moisture, ecosystem variables, and drought-relevant hydrometeorological variables. We use the data from 2001 to 2019 based on the overlapping period for all datasets. To facilitate the understanding of regional characteristics of flash droughts and their impact, we divide the entire grid pixels, using aridity, into humid (aridity ⩽ 1) and arid (aridity > 1) regions. Lower-than-usual precipitation and higher-thanusual temperature are observed during flash drought (top plots), as expected in figure 1(b). In humid regions, the anomalies of precipitation and evapotranspiration (i.e. latent heat flux) are relatively strong, and soil moisture decreases more rapidly and its negative peak is more pronounced than in arid regions. Although there are no significant differences within the temporal pattern of net radiation anomalies between the regions, evapotranspiration in arid regions is relatively weak and decreases more rapidly, probably due to relatively low and limited absolute soil water content. Accordingly, the negative peak of soil moisture is smaller in arid regions than in humid regions. However, it takes longer to return to normal soil moisture conditions in arid regions.

Impact of flash drought on ecosystems in humid versus arid regions
The temporal evolution of GPP and LAI show similar behaviours in each region. Vegetation shows positive responses (increased photosynthesis and healthy vegetation) during the early stage of drought onset, in line with the increased evapotranspiration, possibly contributing further to the rapid onset of drought. This enhancement of vegetation conditions during drought was often reported in humid regions [41,42]. This is likely because the vegetation in such energy-limited environments benefits from increased radiation brought on by the clear-sky conditions associated with drought, alongside the relatively larger amount of soil moisture available to plants. Interestingly, we also find a weaker but positive vegetation response in the arid regions. Unlike in conventional droughts where the onset occurs slowly, when a flash drought occurs it is likely that favourable hydrometeorological conditions, i.e. increased energy and adequate amount of soil moisture, can occur simultaneously during the onset stages. This initial positive response of vegetation can further accelerate soil moisture depletion and induce an abrupt shift into water-stressed conditions. In both regions, the transition from increasing to decreasing pattern of ecosystem functioning occurs within only one to two weeks. As droughts progress, arid regions show a faster decline and longer duration of negative anomalies for both GPP and LAI than in humid regions. On average, flash droughts do not cause significant vegetation stress in humid regions, but it should be noted that there is substantial variability across the individual grid pixels. It is noteworthy that the same ecosystem variables obtained from different datasets do not exhibit exactly the same temporal patterns; for instance, LAI from the GLASS data shows an initial positive response of much smaller magnitude and, consequently, a negative impact of flash drought on ecosystems is detected earlier.
The contrasting patterns of drought-induced impact on ecosystems between humid and arid regimes can be clearly observed in figure 3. While the overall pattern of soil moisture, including the timing of negative peaks, shows similarity between the regimes, the temporal change of GPP and LAI throughout the development and recovery of flash drought events differs along gradations within these climate regimes. An abrupt change in GPP and LAI from positive, or near zero, to negative anomalies is more pronounced in arid regions with lower absolute soil water contents (aridity > 2), indicating higher drought vulnerability in those regions (e.g. Spain). Accordingly, drought-induced vegetation stress (i.e. negative anomalies) lasts longer and becomes more intense in arid regions. In the very humid regions with aridity < 0.8 (e.g. UK), vegetation water stress is negligible. GPP and LAI from the other datasets show a similar spatiotemporal pattern (not shown). Figure 4 summarises the relationship between the timing and intensity of drought-induced vegetation stress across different climate regimes and the soil moisture status (anomaly) of the affected region prior to the onset of drought. We further split the flash drought events into five subgroups based on soil moisture percentile thresholds (namely 0%-20%, 20%-40%, 40%-60%, 60%-80%, 80%-100%) for both humid and arid regions, and recompute the temporal evolution of figure 2 separately for each subgroup. The timing is determined as the pentad when the anomaly of ecosystem variables becomes negative for the first time ( figure 4(a)). Plants are subject to drought stress earlier in arid regions, particularly when the pre-drought soil moisture content is lower (drier); water deficits could occur e.g. due to multiple droughts (flash drought can develop during an ongoing drought [43]). Similarly, the intensity of drought impact, defined as the peak value of the negative anomaly of ecosystem variables during a flash drought, is more significant in arid regions, where the effect of pre-drought soil moisture content is greater ( figure 4(b)). For instance, vegetation in arid regions quickly responds to soil moisture deficiency [42,44,45]. Soil moisture thereby could be depleted already at the very early stage of drought, leading to faster and greater drought stress to vegetation. Relationship between soil water conditions prior to the onset of drought and ecosystem impact in terms of (a) timing and (b) intensity. Inset plots represent the definition of timing and intensity, which are defined as the pentad when the anomaly of ecosystem variables becomes negative and as the minimum value of the anomaly, respectively. We group the flash events by climatic regimes (humid versus arid) and then, for each regime, subgroup the events into quintiles based on normalised soil moisture anomalies before the drought onset. Box plots show medians and interquartile ranges, and dots show individual values obtained from the GPP and LAI data.

Timing and intensity of drought-induced impact
Moreover, drought stress is a major limiting factor of vegetation functioning in such water-limited environments. On the other hand, in humid regions, overall drought-induced vegetation stress is weaker, and pre-drought soil moisture conditions do not significantly affect the timing or intensity of drought impact.

Discussion
Existing drought monitoring tools are primarily designed to capture slower-developing conventional droughts, and they may not be able to adequately respond to flash droughts [43]. Therefore, improving our understanding of the unique characteristics of ecosystem responses to flash droughts is crucial to establishing effective early warning and alerting systems. Previous studies have attempted to estimate the impacts of flash droughts based on the total reduction of GPP or crop production at a country level [21][22][23]; however, ecosystem dynamics during flash droughts are not well understood especially for Europe.
This study focuses on recent flash droughts in Europe and provides additional insight into the regional characteristics of their ecological impacts using large-scale datasets across different climate regimes. While meteorological drought conditions result in rapid and continuous soil moisture reduction, vegetation exhibits markedly non-uniform responses over time. During the early stages of flash droughts, where sufficient soil water is temporarily available, observations of vegetation have revealed positive responses (e.g. increased evapotranspiration) that likely contribute to further depletion of soil moisture. This is particularly true for humid regions, where enhanced vegetation conditions are observed across all of the ecosystem variables under study. However, as soil moisture dry-down continues, flash droughts lead to negative effects on vegetation, such as decreases in photosynthetic activity and canopy dieback in both humid and arid regions. This can occur within a few weeks, presenting a challenge for forecasting and early warning systems, which may be unable to detect relevant changes within that time. Overall, vegetation in humid regions is not severely affected, whereas water stress in arid regions persists longer even during the recovery of soil moisture. In addition, soil moisture conditions before a flash drought, which are closely related to preconditions of vegetation and related to the proximity to the soil moisture threshold of plant water stress (i.e. critical soil moisture [46,47]), are important in determining the timing and intensity of potential drought-induced impacts. This is particularly relevant in arid regions (e.g. Southern and Eastern Europe), where soil moisture content plays an essential role in maintaining vegetation and in the response of plants to drought conditions on short timescales.
A unique aspect of this study is the diversity of the observational data used. Satellite data, in particular, hold great potential for enhancing preparedness and minimising the possible impacts of drought by providing large-scale and near-real-time monitoring indicators [48]. Our study demonstrates the usefulness of satellite-based data for detecting rapid temporal changes in ecosystem status during flash droughts. However, satellite-based data are subject to a significant degree of uncertainty due to the indirect nature of satellite measurements [49,50]. To account for such uncertainty, we employ multiple ecosystem indices, such as latent heat flux, GPP, and LAI, and the regional characteristics of flash drought impacts on ecosystems are detected robustly in the data of all indices studied here. Further, we conduct additional data uncertainty analysis and confirm that our core results stay the same, regardless of possible uncertainties in the input data. As our results are based on the averaged composite of temporal evolution across many grid pixels, the overall results are not significantly affected by uncertainties in parts of the data. Also, it should be noted that the definition of flash drought is inconsistent across studies, even among those where soil moisture percentiles are used. We repeat our analysis using different conditions to identify flash droughts and confirm that although the frequency of flash droughts identified is largely affected, the key findings regarding the ecosystem responses remain similar. See supplementary for more details (Uncertainty analysis of data and figure S9).

Conclusions
Changes in the hydrological cycle and landatmosphere interactions under a warming climate may increase the risk of flash droughts across the world [7,51], necessitating a better understanding of terrestrial ecosystem responses to flash droughts. The computed values (e.g. anomaly magnitude) from our main analysis may not always be directly comparable with those obtained from other drought studies due to differences in time periods or data resolutions of study. Nonetheless, the overall temporal patterns of ecosystem responses and their regional differences reported here could serve as a basis for future studies. For instance, the sign of evapotranspiration changes during droughts is not realistically reproduced by Earth system models due to their misrepresentation of relevant soil-plant-atmosphere interactions [52]. In this context, our observation-based results help to identify target processes, for example, the timing of non-stressed to water-stressed vegetation conditions, that could facilitate physics model evaluation on subseasonal scales.
An increase in the frequency of occurrence and spatial extent of flash droughts has been reported across Europe [27], and areas which have experienced increases in the frequency of conventional droughts are also likely to experience more flash droughts, as observed in Spain [15]. Figure 1 also suggests that flash droughts have occurred more frequently during the recent hot and dry European summers. Although there have been extensive studies on major seasonal droughts-such as those that occurred in 2013, 2018 and 2019-and their ecological impacts [53][54][55], the potential interactions between flash and seasonal droughts and their combined impacts on ecosystems have been overlooked [56]. These interactions should be considered in future studies.
The relatively coarse resolution of the datasets used here cannot fully address the spatial heterogeneity of the land surface and ecosystem types. For instance, point-level measurements (e.g. GPP) have been used to show the response of ecosystems to flash drought across different vegetation types [57]. Future research on a global scale encompassing a wider variety of vegetation types and climate regimes could provide more significant insights into the spatial variations of flash drought impacts.

Data availability statement
All datasets used in this study are publicly available from the references indicated. FLUXCOM data can be requested from www.fluxcom.org. More information on MODIS and GLASS data can be found from www.earthdata.nasa.gov and www.glass.umd. edu, respectively. The data that support the findings of this study are available upon reasonable request from the authors.