Prolongation of Compound Dry–Hot Seasons Over Europe Under Climate Change Scenarios
Abstract
Compound effects of drought and heat are regarded as one of the greatest hazards in relation to climate change. We study characteristics of dry–hot seasons in Europe in an ensemble of CORDEX regional climate models (RCMs). Evaluation against the E-OBS gridded data set for 1976–2005 showed that the RCMs were able to reproduce the spatial pattern of the dry–hot season length but the simulated seasons tended to start later and interannual variability of their length was underestimated. Bias was larger (smaller) in the case of maximum (minimum) length over the 30-year period compared to the median length. Changes in the dry–hot seasons were then analyzed for three time slices (2006–2035, 2036–2065, and 2066–2095) and low and high greenhouse gas concentration pathways. Distinct prolongation (compared to the 1976–2005 simulated climate) was projected for 2036–2065 in the Mediterranean and Western Europe (10–30 days), regardless of the concentration pathway. The dry–hot seasons length was similar in the 2036–2065 and 2066–2095 time slices under the low concentration pathway but major extensions were found under the high concentration scenario over large parts of Europe (20–50 days). The projected spatial patterns of changes in the dry–hot seasons length depend primarily on the driving global climate model. Although the extensions are predominantly driven by increasing temperature, simulated precipitation changes modulate the resulting pattern by amplifying (reducing) the dry–hot seasons length in Southern (Northern) Europe.
Key Points
-
Distinct prolongation of future dry–hot seasons is projected across Europe, especially in the Mediterranean and Western Europe
-
Effects of earlier onset and/or later termination on the dry–hot seasons extensions vary across the domain
-
Uncertainties are primarily related to simulation of future spatio-temporal precipitation patterns in climate models
Plain Language Summary
As already observed in recent years, climate change is bringing more frequent and intense heat waves. In combination with drought, these have serious negative impacts on ecosystems and society. Using a group of climate models, we study characteristics of dry–hot seasons (several months long periods that are prone to heat waves and drought) in the European climate of the 21st century. We show that distinct prolongation of dry–hot seasons may occur already in the near future, especially over the Mediterranean and Western Europe. If global greenhouse gas emissions peak in the mid-21st century and begin to decline thereafter, further extensions of dry–hot seasons projected toward the end of the century remain relatively minor. On the other hand, the climate change scenario based on extensive use of fossil fuels through the 21st century suggests major changes in European hydroclimate due to rising temperatures, amplified by rainfall decline mainly in southern regions.
1 Introduction
Drought and heat waves are among the natural hazards with the most serious consequences for society and the environment. Their impacts tend to be stronger when effects of both dryness and heat combine, causing a major threat to food production (Zampieri et al., 2017), ecosystems (Sippel et al., 2018), water resources, and electricity production (van Vliet et al., 2016). The concurrent effect of two or more hazards is referred to as a multivariate compound event (Zscheischler et al., 2020). Compound dry–hot events have recently been observed across mid-latitudes worldwide: record-breaking heat combined with dry conditions was reported in 2022 over Europe, China, and South America, and in the previous summer of 2021, the Pacific coast of North America was struck by long-lasting and intense hot and dry conditions associated with the heat dome phenomenon (Osman et al., 2022), which caused widespread negative impacts on society and the environment (Bratu et al., 2022; Henderson et al., 2022).
Compound dry–hot events occurring in warm season are driven by precipitation deficits and excessively high temperatures during heat waves. In recent decades, heat waves have become more frequent, severe, and longer lasting, especially over extra-tropical Northern Hemisphere (Freychet et al., 2017; Schoof et al., 2017; Wehrli et al., 2019). In Europe, magnitude of these events increased substantially in the past two decades (Lhotka & Kyselý, 2022a) and the return periods of the major heat waves have been considerably shortened due to ongoing climate change (Dosio et al., 2018; Vogel et al., 2019). In contrast to these large shifts in heat waves' characteristics, changes in droughts have been less significant and with geographically variable patterns (Gudmundsson & Seneviratne, 2016).
In general, droughts are associated with precipitation deficits and/or increased evapotranspiration due to high temperatures. In Europe, trends in drought differ among regions but a small increase of the area prone to drought has been found from the early 1980s onwards (Spinoni et al., 2015). Europe recently experienced an exceptionally dry period 2015–2018 (Hoy et al., 2017; Lhotka et al., 2020) associated with major negative effects on ecosystems (Buras et al., 2020). Moravec et al. (2021) showed that while negative precipitation anomaly in Central Europe of the 5-year 2014–2018 period was about 0.5–2.5 standard deviations, positive temperature anomalies were much higher (up to 4 standard deviations), indicating that high temperatures (and related amplified evapotranspiration) substantially exacerbated dry conditions. The growing importance of higher evapotranspiration for drought development is in accordance with Teuling (2018), who highlighted stronger coupling between drought and heat in a future warmer climate.
Climate change induced shifts in compound dry–hot events is a topic attracting growing interest, and several indices and metrics have been applied. Vogel et al. (2021) analyzed compound hot–dry events in the Mediterranean, using relative temperature thresholds (quantiles of temperature distributions) and drought indices (SPI and SPEI; Vicente-Serrano et al., 2010). A similar method was adopted by Jha et al. (2023), who assessed changes in compound hot–dry events worldwide, using an ensemble of CMIP6 global climate models (O'Neill et al., 2016). Alternative approaches make use of recently proposed standardized indices for compound events. Hao et al. (2018) employed the Standardized dry and hot index (SDHI; based on ratios of precipitation quantiles to temperature quantiles relative to the local climatology) in order to study trends of dry–hot events over global land areas. Other more complex indices are the Standardized compound event indicator (SCEI; Hao et al., 2019), which was shown to perform analogously to SDHI over China (Wu et al., 2020), and the Blended dry and hot index (BDHI), based on standardized precipitation and temperature indices (Wu et al., 2021).
For assessing rapidly emerging compound dry–hot events on the daily temporal scale, the Meteorological drought composite index (MCI; Yu & Zhai, 2020), the Standardized compound drought and heat index (SCDHI; Li et al., 2021), and the Daily-scale compound dry and hot index (DCDHI; Wang et al., 2023) have recently been developed. While these standardized metrics are useful for a relative comparison of dry–hot events' magnitude across various climatic zones with different precipitation and temperature regimes, in our study, by contrast, we aimed to provide projected changes in dry–hot events in absolute numbers which have straightforward interpretation and may be more easily communicated also to stakeholders and decision makers. Because heat waves in Europe usually persist from several days to a few weeks, while droughts are linked to much longer timescales (Mishra & Singh, 2010), we propose a novel alternative approach based on dry–hot seasons to analyze compound effects of these hazards on a joint timescale. Dry–hot seasons were defined based on the Climatic water balance index (CWB; difference between precipitation and potential evapotranspiration, Section 2.2), and represent periods prone to heat waves and droughts lasting from a couple of weeks to several months.
Due to climate change related changes in precipitation and temperature patterns, characteristics of dry–hot seasons are expected to be altered. The main aims of this study are to evaluate dry–hot seasons' characteristics across Europe in an ensemble of climate models in comparison to observed data, and quantify their possible shifts in the 21st century and uncertainties due to different climate models and greenhouse gas concentration pathways. In addition, contributions of temperature and precipitation changes to modified dry–hot seasons patterns in future European climate are investigated.
2 Materials and Methods
2.1 Regional Climate Models and Observed Data
Nine regional climate model (RCM) simulations from the CORDEX project (Jacob et al., 2020) driven by the CMIP5 global climate models (GCMs; Taylor et al., 2012) were analyzed for the historical period and under low (Representative Concentration Pathway [RCP] 4.5) and high (RCP 8.5) greenhouse gas concentration pathways for the 21st century. In order to achieve an equal representation of ensemble members, the RCM ensemble was designed considering completeness of the RCM × GCM matrix (Table 1). The RCM simulations were provided in 0.11° rotated-pole grid (∼12.5 km; EUR-11 domain) in daily temporal resolution. For the historical 1976–2005 period, the RCMs were evaluated against the E-OBS gridded data set (version 24.0e with the 0.1° grid; Cornes et al., 2018) that was transformed to the same rotated-pole grid using the nearest neighbor method. The domain encompasses the whole of continental Europe, spanning approximately 35–72°N and 12°W–43°E.
| Institute | RCM | GCM | Abbreviation |
|---|---|---|---|
| Climate Limited-area Modeling Community (CLMcom) | COSMO-CCLM4-8-17 Sørland et al. (2021) | CNRM-CM5 (r1i1p1) | CLM-CCLM-CNRM |
| ICHEC-EC-EARTH (r12i1p1) | CLM-CCLM-ICHEC | ||
| MOHC-HadGEM2-ES (r1i1p1) | CLM-CCLM-MOHC | ||
| The Royal Netherlands Meteorological Institute (KNMI) | RACMO22E van Meijgaard et al. (2008) | CNRM-CM5 (r1i1p1) | KNMI-RACMO-CNRM |
| ICHEC-EC-EARTH (r12i1p1) | KNMI-RACMO-ICHEC | ||
| MOHC-HadGEM2-ES (r1i1p1) | KNMI-RACMO-MOHC | ||
| Swedish Meteorological and Hydrological Institute (SMHI) | RCA4 Kjellström et al. (2016) | CNRM-CM5 (r1i1p1) | SMHI-RCA-CNRM |
| ICHEC-EC-EARTH (r12i1p1) | SMHI-RCA-ICHEC | ||
| MOHC-HadGEM2-ES (r1i1p1) | SMHI-RCA-MOHC |
2.2 Definition of Dry–Hot Season
Advantages of this method include its limited requirements for input data and good results compared to the more data-demanding Penman–Monteith model (Lang et al., 2017). Moreover, unlike the extended Thornthwaite formula, Oudin formula is applicable for daily mean temperatures as low as −5°C (Oudin et al., 2005). Because just T and Re are used for the PET calculation, the RCM simulations can be evaluated against the E-OBS data set. In the next step, 365 daily differences between P and PET were calculated, thus obtaining CWB series for each year and grid point. These series were transformed into cumulative time series with a daily time step, in order to delimit dry–hot seasons. Their onset and termination for a given year and grid point were identified based on the largest difference between local maximum and subsequent minima of the cumulative CWB series (Figure 1).
Illustration of dry–hot seasons' definition based on daily P—PET difference (example for selected E-OBS grid points, year 2000). (a) Scandinavian tundra (∼69°N, 20°E), (b) Western Europe (∼50°N, 6°E), and (c) Mediterranean (∼38°N, 14°E). Red (blue) line indicates onset (termination) of the dry–hot season. Lengths of the given dry–hot season are also shown. Note different scales of y-axes.
For the 1976–2005 historical period, this procedure yielded 30 values of dry–hot seasons' onset and termination in each grid point. In the southern edge of the domain, dry–hot seasons occasionally extended across more than 1 year (Figure S1 in Supporting Information S1; if almost no precipitation occurred in that period). In such cases, however, the methodology terminates the dry–hot season on 31 December and assigns its length to 365 days in order to obtain the identical number of dry–hot seasons (30) across the domain. Note that only 587 out of total 100,244 grid points in the domain (i.e., 0.59%) were affected in this way. In the next step, the median, maximum, and minimum lengths; median onset and termination dates; and standard deviation of the length were calculated, thereby obtaining spatial patterns of these characteristics over Europe.
2.3 Dry–Hot Seasons in RCMs and Their Projected Changes
The same characteristics described in the previous section were determined in each RCM simulation for the 1976–2005 historical period and their ensemble means were visualized. For climate change scenarios, the median lengths of dry–hot seasons and median dates of their start and end were calculated for 3 time slices—2006–2035 (simulated present climate), 2036–2065 (near future), and 2066–2095 (end of the 21st century)—and two concentration scenarios (RCP 4.5 and 8.5). Extensions of dry–hot seasons in comparison to historical climate were then visualized by subtracting the corresponding values simulated for 1976–2005.
The robustness of projected changes was indicated by the number of RCM simulations agreeing on a given sign of change (analogously to IPCC reports; Lee et al., 2021). Empty circles represent regions where at least 7 (out of 9) RCM simulations yielded the extension of dry–hot seasons, while filled gray circles indicate that the extension was projected in all (9) RCM simulations. In order to distinguish between effects of RCMs and GCMs on prolongation of dry–hot seasons, the RCM simulations were assigned into 6 sub-ensembles designed to contain either (a) a single RCM driven by 3 different GCMs (e.g., composed of KNMI-RACMO-CNRM, KNMI-RACMO-ICHEC, and KNMI-RACMO-MOHC), or (b) a single GCM driving 3 different RCMs (e.g., CLM-CCLM-ICHEC, KNMI-RACMO-ICHEC, and SMHI-RCA-ICHEC).
Extensions of dry–hot seasons are driven by changes in temperature and precipitation patterns. In order to estimate the sole effect of the temperature change at the end of the 21st century under RCP 8.5 (in order to obtain the strongest climate change signal), CWB was calculated using an artificial series of PET from the 2066–2095 and P from the 1976–2005 time slice. For example, the CWB value for year 2080 was obtained as P1990–PET2080. An analogous procedure was performed to estimate the sole effect of the precipitation change but CWB was assessed through an artificial series of P from 2066 to 2095 and PET from 1976 to 2005. Although underlying physical links between daily P and PET were not preserved, the calculated lengths of artificial dry–hot seasons were not substantially affected due to their much longer timescales.
3 Results
3.1 Dry–Hot Seasons in Observed (E-OBS) Data
Observed characteristics of dry–hot seasons were first analyzed in the historical 1976–2005 period using the E-OBS data (Figure 2). The median length of dry–hot seasons had distinct meridional and elevation gradients and ranged from less than a month to nearly a whole year (Figure 2a). In the Mediterranean, dry–hot seasons were characterized by a median onset in late winter/early spring (February and March) and termination during autumn (October and November). In the southernmost parts of the domain (Northern Africa, Middle East, parts of the Iberian Peninsula), median length of dry–hot seasons neared the full year. Around 50°N (excluding high mountain ranges), the median length of dry–hot seasons approximately corresponded to the warm half of the year (from April to September). By contrast, in Northern Europe, the Alps and Carpathian mountains, dry–hot seasons were relatively short and tended to span from late spring to August only (Figures 2d and 2e).
Observed (E-OBS) characteristics of dry–hot seasons over Europe in the 1976–2005 historical period. (a) Median length, (b) minimum length, (c) maximum length, (d) median month of onset, (e) median month of termination, and (f) standard deviation of the dry–hot season length.
Over a majority of Europe, the minimum length of dry–hot seasons was shorter than 3 months. Exceptions were the Mediterranean and lowlands in the south-eastern part of the domain, where the minimum length was between 5 and 7 months (Figure 2b). This duration is analogous to the maximum observed length of dry–hot season over Northern Europe. The lowest values of the maximum length (1–3 months) were found over the Scandinavian mountains and the Alps. In the Mediterranean, by contrast, the dry–hot seasons may occasionally persist throughout the whole year (Figure 2c and Figure S1 in Supporting Information S1). Only slightly lower maximum length was observed in Central Europe (9–11 months), which is related (together with their relatively short minimum length) to the largest variability of the dry–hot seasons length across Europe (standard deviation greater than 50 days, Figure 2f).
3.2 Evaluation of Dry–Hot Seasons in RCMs for the Historical Period (1976–2005)
Dry–hot seasons were evaluated in the ensemble of 9 RCM simulations (Section 2.1) for the same 1976–2005 historical period. The ensemble mean reproduced spatial patterns of the median length relatively well, but too-short dry–hot seasons were simulated especially in the Mediterranean and Western Europe (their median length was underestimated by 1–2 months; Figures 3a and 3d). These biases were related primarily to a tendency of the RCMs to simulate late onset of dry–hot periods (Figures 3b and 3e). By contrast, their median termination was reproduced realistically across Europe, with a slight negative (positive) bias in the western (eastern) parts of the domain (Figures 3c and 3f).
Simulated (model mean) characteristics of dry–hot seasons over Europe in the 1976–2005 historical period. (a) Median length, (b) median month of onset, and (c) median month of termination. (d–f) The same characteristics as (a–c) but showing differences between model mean and observed data.
While a relatively good reproduction of the minimum length of dry–hot seasons was found (Figures 4a and 4d), the RCMs tended to underestimate their maximum length (Figures 4b and 4e). This underestimation was larger compared to that related to the median length and was found across almost all of Europe (except Eastern Europe, Figures 4b and 4e). The too-short maximum length of dry–hot seasons contributed to underestimated standard deviations of their length (Figures 4c and 4f). Over the Alps, Carpathians, and Caucasus mountain ranges, opposite biases in characteristics of dry–hot seasons compared to the rest of Europe were found. In these regions, the RCMs simulated longer dry–hot seasons in the historical period compared to observed data. This relates both to their earlier onset and later termination (Figure 3), and it was demonstrated also in longer maximum length.
Simulated (model mean) characteristics of dry–hot seasons over Europe in the 1976–2005 historical period. (a) Minimum length, (b) maximum length, and (c) standard deviation of the dry–hot season length. (d–f) The same characteristics as (a–c) but showing differences between model mean and observed data.
In general, no sub-ensemble (Section 2.1) outperforms others in terms of reproducing observed characteristics of dry–hot seasons. The negative biases in their median length were more pronounced in the RCA RCM and CNRM GCM sub-ensembles, while CCLM RCM simulated distinctly longer dry–hot seasons in the east (Figure S2 in Supporting Information S1).
3.3 Climate Change Scenarios of Dry–Hot Seasons
Changes in the median length of dry–hot seasons were studied for three time periods (2006–2035–present climate, 2036–2065–near future, and 2066–2095–end of the 21st century) and two greenhouse gas concentration scenarios (RCP 4.5 and 8.5) while taking the simulated historical period (1976–2005) as reference data.
For the present climate (2006–2035), spatial patterns of changes in the dry–hot season length were analogous in both RCP 4.5 and 8.5 (Figures 5a and 5d). The most pronounced extensions within Europe were projected over the Mediterranean but these were relatively small (up to 10 days in most areas). No prolongations of dry–hot seasons with respect to the historical period were simulated over vast parts of Eastern Europe and Scandinavia.
Projected extension of the dry–hot season length in 2006–2035, 2036–2065, and 2066–2095 periods against the simulated 1976–2005 period under RCP 4.5 (top) and 8.5 (bottom) greenhouse gas concentration scenarios. White color indicates no extension of dry–hot seasons. Empty circles represent regions where at least 7 (out of 9) regional climate model (RCM) simulations yielded extension of dry–hot seasons, while filled circles mean that an extension was projected in all (9) RCM simulations.
In the near future (2036–2065), RCM simulations yielded regionally specific extensions of dry–hot seasons. Under the RCP 4.5 scenario (Figure 5b), the Iberian Peninsula and adjacent parts of Northern Africa were projected to experience considerably longer dry–hot seasons (about 1 month longer compared to 1976–2005). The RCM simulations also agreed on relatively large prolongation in Western Europe (10–20 days). Only minor changes in the length of dry–hot seasons were found over Eastern Europe and Scandinavia. Although overall magnitude of the extensions was similar under both RCP 4.5 and 8.5 scenarios, their spatial patterns differed. Under RCP 8.5 (Figure 5e), larger prolongations of dry–hot seasons were projected in the Mediterranean and the Balkan Peninsula while extensions over Western Europe were smaller (compared to RCP 4.5).
At the end of the 21st century (2066–2095) under RCP 4.5, only slight changes were found with respect to the previous period (cf. Figures 5b and 5c). Additional extensions of dry–hot seasons were projected over the British Isles, the Balkan Peninsula, and Anatolia (in total 10–20 days compared to 1976–2005), while no major changes were found in the other regions compared to 2036–2065. Under RCP 8.5, by contrast, substantial prolongation of dry–hot seasons was found at the end of the 21st century (Figure 5f). Over the Mediterranean, all RCM simulations agreed on dry–hot seasons extension by approximately 1 month (compared to 1976–2005). In certain regions (the Iberian Peninsula, adjacent Northern Africa, and part of Anatolia), the prolongation is even larger and exceeds 40 days.
RCM simulations also agreed on considerable extensions of dry–hot seasons in Western Europe (about 1 month longer in 2066–2095 compared to 1976–2005, Figure 5f). Relatively large extensions (approximately 20 days) were found also in Central and North-eastern Europe, but agreement among the RCMs was weaker in these regions. In general, the median length of dry–hot seasons was projected to increase across the whole of Europe but the increment was relatively small over the Scandinavian mountains, which retain short dry–hot seasons even in projections for the end of the 21st century. These results suggest a strengthening of the meridional gradient of the length of dry–hot seasons.
Increased median length of dry–hot seasons was related to both their earlier start and later end. Under the high concentration RCP 8.5 scenario, regional variations were found (Figure 6). The prolongations of dry–hot seasons over the Mediterranean were related to their earlier onset rather than later termination. This feature is present already in 2006–2035 (the RCM agreement in Anatolia, Figure 6a) and distinct in 2066–2095 (cf. Figures 6c and 6f). In the latter period, dry–hot seasons were mostly projected to start earlier (by about 20–30 days), while the RCMs simulated smaller extensions at the end of the dry–hot seasons (10–20 days in the Western Mediterranean, fewer in the east). The RCMs yielded similar patterns for Scandinavia, where the extensions were also linked rather to earlier onset of dry–hot seasons than to their later termination.
Same as Figure 5 but for extension of the dry–hot season length split between earlier onset (a–c) and later termination (d–f) under the RCP 8.5 greenhouse gas concentration scenario.
The prolongations of dry–hot seasons over Western and Central Europe, by contrast, were primarily driven by their later termination. While only minor shifts of dry–hot seasons' start toward the beginning of the year were found in these regions, the RCM simulations yielded much larger prolongations toward the end of the year. This feature was especially apparent in the 2066–2095 period, when the RCMs tended to agree on positive shifts of the dry–hot season end, while their onset remained almost unchanged (Figure 6c).
3.4 Differences Between Climate Models and the Role of Changing Precipitation Patterns
The extensions of dry–hot seasons are related to simulated changes in temperature and precipitation patterns. In this section, the RCM ensemble was divided into 6 sub-ensembles, each of them containing 3 simulations based on RCM × GCM combinations (Section 2.1). For each sub-ensemble, changes in the dry–hot season length for the end of the 21st century (2066–2095) under the high-concentration RCP 8.5 scenario compared to 1976–2005 were analyzed. In addition, the individual contributions of increased temperature and altered precipitation (Section 2.2) to projected changes in dry–hot season lengths were identified.
The 3 RCM sub-ensembles (each containing a single RCM driven by 3 GCMs) agreed on magnitude and spatial patterns of the dry–hot seasons extensions over the Mediterranean (25–50 days; Figures 7a–7c). In this region, the prolongation was linked to both increased temperature and decreased precipitation. The projected reduction of precipitation was more important with respect to extended dry–hot seasons than increased temperature in parts of the Iberian Peninsula. The decreased precipitation (around the time when dry–hot seasons occur) was linked to prolonged dry–hot seasons also over Western Europe and the Alps, but increased temperature generally had a larger effect there (except for the Alps in the CCLM sub-ensemble; Figures 7d–7i).
(a–c) Differences in the median dry–hot seasons length between 2066–2095 and 1976–2005 in 3 regional climate model (RCM) sub-ensembles (each containing a single RCM driven by 3 GCMs). (d–f) Contribution of temperature change to the altered dry–hot season length. (g–i) Contribution of precipitation change to the altered dry–hot season length. Filled circles in (a–c) indicate regions where all (3) simulations in the given sub-ensemble yielded extension of the dry–hot seasons.
In the rest of Europe, precipitation is projected to have a rather suppressing effect on extension of dry–hot seasons. In the CCLM sub-ensemble, this effect is rather small compared to remaining sub-ensembles (as much as −15 days in most regions; Figure 7g) but the CCLM sub-ensemble yields also relatively small dry–hot season prolongations due to increased temperature, especially over Eastern Europe. In certain parts of this region, the extension of dry–hot season due to higher temperatures is projected to be smaller compared to the suppressing effect of precipitation, thus resulting in no prolongation (or even slight shortening) of its length. A similar pattern is projected also by the RACMO sub-ensemble but in relation to large extensions of dry–hot seasons due to increased temperature (25 days and more) compensated by their substantial shortening due to higher precipitation amounts (of similar magnitude; Figures 7e and 7h).
In the RCA sub-ensemble, changes in median length of the dry–hot season north of 48°N were projected differently. In Central and Eastern Europe, the suppressing effect of precipitation was lower compared to that in the RACMO sub-ensemble (from 5 to 20 days; Figure 7i). Jointly with relatively large extension due to increased temperature, those regions experience substantially larger prolongation in the RCA sub-ensemble compared to the CCLM and RACMO sub-ensembles. As the RCA sub-ensemble yielded considerable shortening of dry–hot season due to higher precipitation over Scandinavia not compensated by temperature increases in many areas, the largest north–south gradient of changes in the dry–hot season length was projected in this sub-ensemble (Figure 7c).
Figure 8 shows results for the 3 GCM sub-ensembles (each containing 3 RCMs driven by a single GCM). Because the spatial patterns of the dry–hot seasons extensions due to increased temperature were simulated similarly in all sub-ensembles (Figures 8d–8f), the projected changes in their length are predominantly modulated by altered precipitation patterns. The CNRM sub-ensemble yielded considerable shortening of dry–hot seasons due to precipitation effects that were not compensated by increased temperature over large parts of Europe, resulting in no (or only minor) extensions of dry–hot seasons in these regions (Figure 8a).
Same as Figure 7 but for 3 GCM sub-ensembles (each containing 3 regional climate models driven by a single GCM).
The suppressing effects of precipitation, by contrast, were limited mainly to Northern Europe in the MOHC sub-ensemble (Figure 8i), which was linked to the largest prolongation of dry–hot seasons in Eastern Europe and the Iberian Peninsula (compared to the other sub-ensembles). The ICHEC sub-ensemble was characterized by substantial extensions of dry–hot seasons over Western Europe (30–50 days, a similar magnitude as in the Mediterranean; Figure 8b). In general, larger differences in the length of dry–hot seasons were found among GCM than RCM sub-ensembles (cf. CNRM and MOHC sub-ensembles).
4 Discussion
Regional hydroclimates determine many aspects of ecosystems and society and their shifts toward drier conditions are usually associated with negative consequences (Rakovec et al., 2022). Padrón et al. (2020), using observation-based reconstructions for 1902–2014, showed a reduction of water availability during dry seasons in Europe and linked these changes to increased evapotranspiration. This is in accordance with other studies reporting increased frequency of shorter warm seasons' droughts throughout Europe due to rising temperatures and related higher PET (Manning et al., 2019; Markonis et al., 2021). Concerns that this trend will continue in a future climate are prompting further investigation of this topic.
Recently, Zhang et al. (2022) studied outputs of CMIP6 GCMs and found out a globally increased magnitude of compound dry–hot events (defined as heat waves occurring within a dry month) in a future climate. While the heat waves were based on daily temperature data, the droughts were calculated using the 3-month temporal scale SPEI-3 index (Vicente-Serrano et al., 2010), and therefore not excluding the possibility of rain during heat waves due to the rather coarse temporal resolution of SPEI. In our study, we proposed an alternative approach based on dry–hot seasons, that is, several months long periods prone to occurrence of compound dry–hot events. Lengths of dry–hot seasons are tracked through CWB on a daily temporal scale. In contrast to the standardized indices (e.g., SCDHI; Li et al., 2021), CWB allows estimating soil moisture deficits/surpluses in absolute numbers. Therefore, dry–hot seasons can be characterized as time periods of gradual soil desiccation due to compound effects of precipitation and temperature, while soil moisture is replenished during the rest of the year, naturally mimicking the annual cycle in temperate hydroclimates.
Future prolongations of dry–hot seasons are thus linked to changes in precipitation and PET. A number of methods for estimation of PET were introduced but none of them outperforms others globally, as their performance is often linked to certain climatic region (Valipour et al., 2017). Some studies have reported that more complex radiation-based models perform better under climate change conditions (Sheffield et al., 2012) but they demand variables (e.g., relative humidity, solar radiation, and wind speed) often laden with errors in RCMs (Lhotka & Kyselý, 2022b; Vautard et al., 2021), thus increasing uncertainties originating from climate models' selection. In addition, the limited requirements for input data in the Oudin's method allow evaluating the climate models against the established gridded observation-based products (such as the E-OBS data set; Cornes et al., 2018).
In general, uncertainties related to future climate projections originate from different greenhouse gas concentration pathways, climate models, and internal variability of models' climate (initial conditions uncertainty; Knutti & Sedláček, 2013). In the present study, we addressed the first two sources of uncertainties by analyzing low- and high-greenhouse gas concentration scenarios (RCP 4.5 and 8.5) using a complete RCM × GCM matrix (Section 2.1). The internal variability of models' climate was not assessed, because only 2 out of 9 ensemble members are available with a subset of different initial conditions, but Déqué et al. (2012) showed that this source of uncertainty tends to be relatively low.
Under both low- and high-concentration pathways, the largest prolongations of dry–hot seasons are projected over the Mediterranean. In this region, the observed rate of temperature increase exceeds global trends (Cramer et al., 2018), and the recent record-breaking 2021 heat wave (Lhotka & Kyselý, 2022a) was an example of an unprecedented event with major societal and environmental consequences. Cook et al. (2016) reported drying in the Mediterranean, which is projected to continue in a future climate (Carvalho et al., 2022; Spinoni et al., 2018). We showed that the extension of Mediterranean dry–hot seasons is mainly due to their earlier onset rather than later termination and that the prolongations are largest in the western Mediterranean. Assessment of underlying physical mechanism is beyond the scope of this study, but Kim et al. (2019) reported increasing air subsidence over the Western Mediterranean, favoring dry anticyclonic conditions in relation to possible future intensification of the Asian monsoon. The origin of the marked difference between spring/autumn extensions of dry–hot seasons is, however, not clear as the pattern varies across Europe. In Central Europe, for example, the extension of dry–hot seasons is primarily related to their later termination, which is in contrast to the projected changes in the Mediterranean.
The fact that the climate models in the ensemble are not independent of each other (implying that the ensemble mean may be biased; Abramowitz & Bishop, 2015; Knutti et al., 2013) may affect the projections. Therefore, we presented the climate change scenarios also as divided into six sub-ensembles based on RCM and driving GCM. While all sub-ensembles agree on major prolongations of dry–hot seasons over the Mediterranean, differences between them are considerably larger in other European regions. These variations between the sub-ensembles are linked to future changes in temperature, precipitation, and the mutual relationships between these two. Lhotka and Kyselý (2022b) showed that CORDEX RCMs have difficulties in reproducing links between summertime precipitation and temperature on the monthly scale and that these biases are RCM-dependent. In addition, differences between sub-ensembles are probably also linked to simulations of future changes in atmospheric circulation patterns (Zappa & Shepherd, 2017). We found prolongations of dry–hot seasons even in Scandinavia, a region that is often projected to become rather moister under climate change (Chen et al., 2021). In the majority of the sub-ensembles, increased precipitation did not compensate higher temperatures and related evapotranspiration demand.
It should be noted that the RCMs' evaluation revealed their tendency to underestimate median and particularly maximum lengths of dry–hot seasons. The maximum length (in the 1976–2005 period) was simulated as too short over the majority of Europe compared to E-OBS, suggesting that the RCMs have difficulties to capture extreme seasons. This issue may propagate also into the climate change scenarios and may impose a limitation in their credibility with respect to high-impact events (i.e., major heat waves associated with drought). In the present analysis, this drawback was partially overcome by analyzing the future scenarios with respect to simulated historical climates (rather than applying bias-correction methods limited by non-stationarity of biases; Nahar et al., 2017). The presented climate change scenarios constitute the first attempt to quantify changing characteristics of dry–hot seasons and related uncertainties in the European hydroclimate of the 21st century and may provide useful information for impact modeling and decision making related to adaptation planning.
5 Conclusions
-
The median length of dry–hot seasons varies across Europe in relation to both latitude and altitude. Over the Mediterranean, dry–hot seasons tend to start in late winter/early spring and terminate in autumn, while in Northern Europe and high mountains, their median length is only about 1–3 months.
-
The climate models reproduced this spatial pattern well, but dry–hot seasons tended to be too short especially over the Mediterranean and Western Europe, mainly due to their late onset. The climate models underestimated interannual variability of their length, as they had difficulties to yield longer dry–hot seasons across Europe. The limited ability to capture extreme seasons deserves further investigation and may propagate also into climate change scenarios; this constitutes a possible limitation upon their credibility with respect to high-impact major heat waves associated with drought.
-
In the near future (2036–2065), the climate models agree on large extensions of dry–hot seasons over the Mediterranean (by as much as 1 month) and Western and Central Europe (10–20 days, under both low- and high-greenhouse gas concentration scenarios). By contrast, no clear changes in the median length were found over Eastern Europe and Scandinavia.
-
At the end of the 21st century (2066–2095), substantial prolongations were projected under the high greenhouse gas concentration scenario, with large increases also in Central and Northeastern Europe. In the Mediterranean, extensions of dry–hot seasons are mainly due to their earlier onset, while in most other regions, these are driven by their later termination.
-
While prolongations of dry–hot seasons are primarily due to increase in temperature, changes in precipitation have suppressing (amplifying) effects in the northern (southern) parts of the domain. This general spatial pattern was found in all RCMs and notwithstanding the driving GCM, which increases confidence in the projected changes.
Acknowledgments
The study was supported by the Czech Science Foundation, project 20-28560S, and benefited from COST Action CA17109 DAMOCLES and related INTER-COST project funded by the Ministry of Education, Youth and Sports of the Czech Republic (project no. LTC19044). We acknowledge the World Climate Research Programme's Working Group on Regional Climate, the Working Group on Coupled Modelling (former coordinating body of CORDEX) and the panel responsible for CMIP5 and the E-OBS data set (from the EU-FP6 project UERRA; http://www.uerra.eu). We also acknowledge the Copernicus Climate Change Service, and the data providers in the ECA&D project (https://www.ecad.eu).
Open Research
Data Availability Statement
The WCRP CORDEX RCMs (WCRP, 2022) used for creating climate change scenarios can be freely downloaded for research purposes from individual Earth System Grid Federation nodes (https://cordex.org/data-access/esgf/, retrieved 11 August 2022). The E-OBS gridded data set (ECA&D, 2022) used for the RCMs evaluation is freely available from C3S (https://doi.org/10.24381/cds.151d3ec6).
