Tree dieback and subsequent changes in water quality accelerated the climate-related warming of a central European forest lake

Water temperature affects both in-lake physical–chemical properties such as the concentration of dissolved oxygen, extent and duration of water stratification, evaporation, length of the ice-free period, and water–sediment interactions and the living conditions for many water organisms (Whitehead et al. 2009; O'Reilly et al. 2015; Dokulil et al. 2021; Ahonen et al. 2023). During ongoing climate change, temperature is increasing faster in higher latitudes and in mountain regions than in environments at lower latitudes and elevations (Schneider & Hook 2010; Moser et al. 2019). The warming lakes are generally freezing later than they did in the past, while their thaw dates show a trend towards earlier ice breakup in the spring (Thompson et al. 2005; Preston et al. 2016; Moser et al. 2019). For example, Benson et al. (2012) showed that the ice cover of lakes in the Northern Hemisphere has decreased at an average rate of 0.8–1 day decade⁻¹ since the mid-1800s due to later freeze dates and earlier thaw dates, but this trend has sharply accelerated during the last few decades (Sharma et al. 2021; Noori et al. 2022a). The lengthened ice-free season and warmer water affect the biodiversity of waterbodies (Parker et al. 2008; Bates et al. 2013) and may shift the community size structure of aquatic organisms to smaller sizes (e.g., Winder et al. 2009; Rasconi et al. 2015; Svitok et al. 2021) and even contribute to algal blooms, including harmful, toxin-producing cyanobacteria (Elliott 2012; Trtanj et al. 2016).

Another important factor affecting the physical–chemical and biological processes in waters is the effect of climate change on the terrestrial ecosystem of their catchments. These effects include (1) changes in the timing and extent of snowmelt, seasonal patterns of water flow, and sediment loads (Whitehead et al. 2009; Preston et al. 2016); (2) elevated terrestrial export of dissolved organic carbon (DOC) (Rasconi et al. 2015; de Wit et al. 2021); and (3) accelerated insect outbreaks and tree dieback in temperate and boreal regions, resulting in elevated nutrient (nitrate and phosphorus (P)) and DOC leaching (e.g., Mikkelson et al. 2013; Kopáček et al. 2017, 2019). The elevated concentrations of DOC and nutrients supporting elevated algal production also decrease water transparency and the depth of water adsorbing the incident solar radiation (Wetzel 2001).

The combined effects of accelerating climate change and decreasing acidic deposition during the last three decades have formed novel and complex freshwater conditions, with chemical reversal and biological recovery of acidified waters occurring along different trajectories than during the acidification period (Keller et al. 2019; Svitok et al. 2021). Climate change has induced compositional shifts in the freshwater biota of recovering lakes towards non-analogous communities (Marchetto et al. 2004; Sivarajahe et al. 2017), potentially causing unanticipated synergistic effects, ultimately leading to recovery trajectories that are very difficult to predict (Yan et al. 2003; Keller et al. 2019; Svitok et al. 2021).

Chemical reversal of the Bohemian Forest lakes (Czech Republic) from atmospheric acidification after steep declines in central European depositions of sulphur and nitrogen compounds (Kopáček et al. 2016, 2017) enabled the biological recovery of their zooplankton (Vrba et al. 2016), macrophytes (Čtvrtlíková et al. 2016,, 2023), macrozoobenthos, and fish (Petruželová et al. 2023). This biological recovery has recently accelerated (Čtvrtlíková et al. 2023; Petruželová et al. 2023). One possible reason for this promising trend is an increasing nutrient availability in the lakes originating from the synergistic effects of climate change and the reversal of soil and water composition from acidification. In-lake P availability has been elevated by increased terrestrial co-export with DOC (Kopáček et al. 2011), whose leaching has been increasing due to decreasing acidic deposition and climate change (Meyer-Jacob et al. 2020; de Wit et al. 2021), as well as following bark beetle–induced tree dieback in some Bohemian Forest catchments (Vrba et al. 2014; Kopáček et al. 2019). Another possible reason for accelerating biological recovery could be increasing water temperatures and the length of ice-free periods due to significantly increasing air temperatures in the catchments of the Bohemian Forest lakes (Kopáček et al. 2020). However, these potential climate-related variables have not yet been evaluated due to the absence of long-term temperature data. Another unanswered question is whether changes in water quality following tree dieback in catchments affect lake water warming and thermocline depth.

The aim of this study was to evaluate data on air and water temperatures and ice-on/off dates at two intensively studied Bohemian Forest lakes, of which one was also affected by a sharp increase in DOC and P concentrations after tree dieback in its catchment (Kopáček et al. 2019), along with available trends in climate characteristics (air temperature, cloudiness, wind, and incident solar radiation at a nearby meteorological station). We hypothesise that (1) increasing air temperature resulted in prolonging ice-free and summer water stratification periods, and (2) changes in water quality contributed to an extension of the period with epilimnetic temperatures ≥20 °C and a decrease in thermocline depth in the lake with the damaged forest in its catchment.

The catchments of Plešné (PL) and Čertovo (CT) lakes are situated in the Bohemian Forest at 48.774°N, 13.860°E and 49.167°N, 13.192°E at elevations of 1,087–1,378 and 1,027–1,343 m, are northeast and southeast oriented, and have areas of ∼60 and ∼79 ha, respectively (Fig. SM-1 in the Supplementary material). Both lakes are of glacial origin and are dimictic. Their surface areas, volumes, and maximum depths are 7.2 and 10.7 ha, 0.55 and 1.86 million m³, and 17.7 and 35.4 m for the PL and CT lakes, respectively (Šobr & Janský 2016). According to the climate classification of Peel et al. (2007), the study area belongs to the cold climate category, with monthly mean air temperature exceeding 10 °C only in three summer months. Current annual mean air temperatures and precipitation are ∼5 °C and ∼1.3 m, respectively, at the elevation of ∼1,100 m (Kopáček et al. 2020).

Both catchments are unmanaged and belong to the Šumava National Park, declared in 1991. The catchment soils are shallow, sandy (∼75%), low in clay (∼2%), and dominated by cambisols, haplic podzols, and organic rich rankers (A/C soils) (Kopáček et al. 2023). Both catchments were >90% covered by an unmanaged mature Norway spruce (Picea abies) forest, with a minor contribution of rowan (Sorbus aucuparia), birch (Betula pubescens and B. pendula), and European beech (Fagus sylvatica), with an average spruce age of 160 and 140 years in the PL and CT catchments, respectively, in 2000. A bark beetle (Ips typographus) outbreak occurred in the summers of 2004 and 2006 in the PL catchment as a result of accelerating climate change (increasing air temperature, drought periods, and winds) and killed >75% of its original mature spruce trees over the next 2–3 years (Kopáček et al. 2024). Natural tree regeneration and the biomass of understory herbs started to increase within 1–3 years after tree dieback, when the availability of sunlight on the forest floor increased due to the thinning of canopies. Effects of these changes on hydrology and microclimate characteristics in the PL forest area have been described elsewhere (Kopáček et al. 2020, 2024). The CT catchment was partly affected by windthrows in 2007 and 2008 and a subsequent bark beetle outbreak from 2007 to 2011. Altogether, the total area of damaged forest (with >50% dead trees) increased from ∼4 to ∼18% between 2000 and 2011 (Kopáček et al. 2016). However, the overall forest damage in the CT catchment was relatively small compared to the heavily affected PL catchment until 2020, when a massive bark beetle outbreak also occurred in the CT catchment. Most mature spruces were infested but were still standing, and canopies were still unbroken at the end of this study in spring 2023. Consequently, changes in water composition (especially concentrations of DOC and P) associated with changes in forest vegetation were more pronounced in the PL compared to the CT lake (Kopáček et al. 2016, 2019).

Lake outflow water and air temperatures (2 m above ground) were measured using an MS4016 automatic weather station (Fiedler AMS, Czech Republic) equipped with RVT13 and TEP1/H sensors close to lake outlets at 15-min intervals. The lake's outflows originated from the surface water layer of both lakes. The weather stations were situated in places unaffected by tree dieback. In the PL catchment, the station was in an open area without trees and has been operating since January 1999 and 2001 for water and air, respectively. At the CT lake, the station was in a shallow valley along the outlet stream in a forested area. This station has been operating for both air and water temperature since January 1998. Mature spruce stands in the surroundings of the CT station were continuously infested in 2021–2022 and partially cut down in autumn 2022. The effect of the change in vegetation cover on the 1998–2022 trend was thus short and assumed to be negligible.

In addition to weather stations, water and air temperatures were measured at hourly intervals, using UTBI-001 and U23-001A sensors (Onset, USA). For water and air, the respective sensors were situated in about the centres of the lakes (above the maximum depth) at depths of ∼20 cm, and in solar radiation shields fixed on the shaded, northern sides of wooden poles at 2 m above ground at nearby terrestrial research plots (Fig. SM-1 in the Supplementary material). The accuracy of these sensors was ±0.2 °C. Prior to use, the sensors were kept for several days at 4 °C and then at a laboratory temperature of ∼20 °C, and only sensors that met the accuracy range of ±0.2 °C at both temperatures were used.

Data on the lake water composition (epilimnetic concentrations of DOC, P, chlorophyll a (Ch-a), and absorbance at 254 nm (A₂₅₄)), Secchi disk depth (SDD), and water temperature profiles come from Vrba et al. (2016) and Kopáček et al. (2016, 2019, Unpublished data). SDD and water temperature profiles (0.5–1 m intervals) were measured irregularly (<15 times per year) at the deepest points of the lakes. These data were used to calculate the thermocline depths (see Section 2.3). Concentrations of DOC, P, Ch-a, and A₂₅₄ were determined at 1 to 3-week intervals.

Daily average data on air temperature, cloudiness, snow cover, incident solar radiation, and wind direction in the Bohemian Forest come from the Churáňov meteorological station (CMS; Czech Hydrometeorological Institute) situated at 49.0682°N, 13.6148°E (i.e., ∼35 km northwest and southeast of the PL and CT catchments, respectively), and at an elevation of 1,118 m.

Theoretical incoming solar radiation (under cloudless conditions) on the lake surfaces and to the CMS was modelled as described by Kopáček et al. (2024). In short, solar radiation was calculated in System for Automated Geoscientific Analyses (SAGA) version 8.1 (Conrad et al. 2015) at daily intervals for a calendar year, using digital elevation models (spatial resolution 1 m for PL and 2 m for CT and CMS). The model parameters (e.g., sky radiation parameters) were determined as the best fit between the SAGA values and direct ground measurements of incident radiation at the CMS for cloudless days throughout the year.

In this study, we use (1) data on water composition, temperature profiles, and SDD for the summer periods (July–September, similar to Schneider & Hook 2010 and Noori et al. 2023) from 1998 to 2022; (2) daily and monthly average air and water temperatures measured by the PL and CT weather stations at lake outlets; and (3) daily, monthly, annual, and winter (December–February) mean values of climate characteristics measured at the CMS from 1961 to 2022. The daily average air and water temperatures measured by the sensors in the PL and CT catchment–lake systems were almost identical to those at the weather stations at lake outlets (R² > 0.97 and p < 0.001 for all relationships) and were only used instead of missing data at lake outlets in isolated periods of weather station failures (<8% of days). The trends in outflow water temperatures were considered to represent surface water temperatures.

The ice-on and ice-off data for the PL and CT lakes were based on daily to about weekly observations by Šumava National Park rangers and institutional sampling teams, and on the measured water and air temperatures. Based on in situ experience, ice-on usually occurred when the water temperature at outflows sharply decreased below 2 °C and daily mean air temperatures were <0 °C for at least 2 days. Similarly, ice-off usually occurred when the water temperature at outflows sharply increased above 3 °C and daily mean air temperatures were >0 °C for about a week. These water and air temperature criteria were used to estimate changes in ice cover for the periods between two direct observations when the precise date of the change in ice cover was missing.

The length of the summer thermal stratification of lakes for each year was computed as the number of days with daily mean outflow water temperature ≥4 °C. In addition, we computed the annual number of days with daily mean outflow water temperature ≥20 °C. We selected this temperature threshold because it is often assumed to be critical for numerous lake water biota (Thomas et al. 2017; Dokulil et al. 2021).

The analyses of trends in air and water temperature, cloudiness, solar radiation, wind directions, SDD, and summer (July–September) concentrations of DOC, A₂₅₄, P, and Ch-a were performed in the R environment for statistical computing (R Core Team 2023). The non-parametric Mann–Kendall test (Pohlert 2023) was used to detect monotonic trends. The thermocline depths were calculated using the R package rLakeAnalyzer from temperature profiles (Winslow et al. 2019; R Core Team 2023).

Linear regression was used to evaluate the significance of relationships between (1) monthly mean air temperatures measured at the CMS, PL, and CT stations; (2) monthly mean surface water temperatures in the PL and CT lakes (outflow water); and (3) time and the dates of ice-on/off, length of the ice-cover period, the annual number of days with water temperatures ≥4 and ≥20 °C, and annual and winter (December–February) mean air temperatures and daily records on snow cover at CMS from 1961 to 2022.

From 1998 to 2022, air and water temperatures significantly (p < 0.01) increased at all stations, with Sen's slopes of seasonal Mann–Kendall tests of 0.42–0.56 °C decade⁻¹ for air and 0.32–0.47 °C decade⁻¹ for water (Table 1). The mean air temperature observed at the CT outlet was lower than at the PL outlet and at the CMS (Table 1) by 6 and 2% for monthly means, respectively (Fig. SM-3A, C in the Supplementary material), despite its lower elevation. This disproportion probably resulted from differences in the surroundings of individual stations. The monthly mean air temperatures were 5% higher at the PL outlet than at the CMS (Fig. SM-3B in the Supplementary material), due probably to the lower elevation of the PL station. Despite these differences, the monthly mean air temperatures were closely correlated at all three stations (R² > 0.98; p < 0.001; Fig. SM-3 in the Supplementary material). In contrast to the air, the monthly mean water temperatures were similar at both lake outlets (Fig. SM-3D in the Supplementary material), with a long-term average of ∼8 °C (Table 1).

The increases in air and water temperatures resulted in part from the increasing incident solar radiation on both annual (Table 1) and monthly (Figure 2) bases. The increase in cloudiness in May was accompanied by significant decreases in air temperature at all three stations (Table SM-1 in the Supplementary material). However, slopes of changes in air temperature were positive in all other months, with significant trends in June–July and October–December at all stations and in February at the lake outlets (Figure 2). Water temperature significantly increased at both lake outlets in most months, except for May at the CT outlet and February and December at the PL outlet (Figure 2).

The rates of water temperature increases observed in the Bohemian Forest lakes were within the ranges recently reported for numerous lakes within the Northern Hemisphere. For example, Dokulil et al. (2021) observed that annual maximum lake surface temperatures in 10 European lakes increased on average by 0.58 °C decade⁻¹ from 1966 to 2015. Noori et al. (2002b, 2023) observed an average increase in summer lake surface water temperature of 0.25 °C decade⁻¹ in Lake Inari from 1961 to 2020 and 0.41 °C decade⁻¹ in Lake Konnevesi from 1984 to 2021. Similar warming rates (with averages of 0.45 and 0.52 °C decade⁻¹ for all studied sites and Northern Hemisphere, respectively) were reported for July–September temperatures in 167 large inland water bodies by Schneider & Hook (2010) during the 1985–2009 period. Even higher long-term warming (up to 0.77 °C decade⁻¹) has been reported for some streams and rivers in Europe and the USA since the 1980s (Kasushal et al. 2010; Gizińska & Sojka 2023). On a global scale, using more than 25 years of satellite temperature data and ground measurements of 235 lakes on six continents, O'Reilly et al. (2015) found that the lakes warmed at an average rate of 0.34 °C decade⁻¹ between 1985 and 2009.

The effects of current climate change on lake ice phenology (i.e., the timing of ice formation and loss) generally include delayed ice-on, earlier ice-off, and shorter ice duration (e.g., Preston et al. 2016; Sharma et al. 2021; Li et al. 2022). Sharma et al. (2021) assembled records on ice phenology for 60 Northern Hemisphere lakes with time-series extending over 100 years. They found that, on average, ice-on was 1.1 days decade⁻¹ later, ice-off was 0.7 days decade⁻¹ earlier, and ice duration was 1.7 days decade⁻¹ shorter during the 1917–2016 period. However, these long-term trends were ∼6 times less steep than the averages for the 1992–2016 period, when they increased to 7.2, 4.5, and 10.6 days decade⁻¹ for ice-on, ice-off, and ice duration, respectively (Sharma et al. 2021). The respective rates observed for the Bohemian Forest lakes from 1998 to 2022 were thus similar to those reported by Sharma et al. (2021).

The formation of ice cover on lakes in the winter and its disappearance the following spring depend on climate factors such as air temperature, cloud cover, and wind. A particularly important factor is earlier snowmelt in catchments causing earlier seasonal runoff, which influences the timing of ice-off (Sadro et al. 2019). For example, Preston et al. (2016) observed that ice-off dates of high-elevation alpine lakes in Colorado, USA, shifted 7 days earlier over the 1981–2014 period, and snowfall was the most important variable affecting the annual variation in ice-off timing. In the Bohemian Forest, winter mean air temperatures increased from –4.4 to –1.9 °C at the CMS from 1961 to 2022, with some averages already above the freezing point (Figure 1). The increasing winter air temperature and mid-winter melting periods accompanied by rain events decreased the snowpack in extent and duration. Less snow on the ice (particularly in spring) decreased its insulation effect and contributed to more rapid melting. Less snow in catchments, especially along lake shores, shortened the period of their high albedo. Snow-less lake shores exposed to sunshine warmed more rapidly than in years with high snowpack, which accelerated ice melt along sun-exposed shores and further decreased the albedo of the overall lake surface.

The increases in air temperatures in February–March and in October–December (Figure 2(c) and 2(e)) were probably the major reason for earlier ice-off and later ice-on dates, respectively, at the study lakes. The increasing air temperature delayed ice-on of the CT lake by 6 days more than at the PL lake. This between-lake difference was probably associated with higher volume (and thus higher heat capacity) of the CT lake. However, the between-lake differences in air temperature and lake volumes cannot explain the almost three times faster ice melt of the PL compared to the CT lake. We assume that the greater acceleration of ice melting on the PL lake was probably related to the more frequent occurrence of warm downwind episodes over the mountain ridge from the south, which we often observed in later winter periods. In contrast, similar situations at the CT lake were rare because the lake lies further behind the mountain ridge. This orographic and foehn phenomenon (Sharples 2018; Fig. SM-4 in the Supplementary material) probably contributed to the higher air temperatures at the PL compared to the CT lake in March (often >0 °C) and April (Figure 2(c) and 2(e)). A possible influence of wind on earlier ice melt was also indicated by the increasing proportion of southerly winds during February–March and south-westerly winds during most months at the expense of decreasing westerly winds during 1998–2022 (Table SM-2 in the Supplementary material). Moreover, the orography of the PL catchment tends towards the more frequent formation of foehn winds in the direction of south-westerly winds compared to the CT catchment (Fig. SM-4 in the Supplementary material).

The decrease in the ice cover period was accompanied by an increase in the length of summer stratification (i.e., period with epilimnetic temperature ≥4 °C) by 30 days, i.e., with average rates of 12.4–12.7 days decade⁻¹ (Figure 3(b)). During the mid-winter ice-free episodes, lakes water columns were not thermally stratified and mixed at 1.8 to <4 °C. The excess lake water warming was influenced both by the advancement in the timing of ice melting, as observed elsewhere (Li et al. 2022), but also by delayed ice-on dates. The effect of earlier ice-off was manifested by the steepest increases in water temperatures that occurred in April, despite less pronounced increases in air temperatures (Figure 2). The increasing solar radiation in November–December and increasing air temperatures in October–December resulted in significantly (p < 0.001) increasing water temperatures over 4 °C at both lake outflows in December (Figure 2). The prolonging ice-free period, and thus the longer time of accumulating solar energy, also likely explains why (1) lake water temperature increased in more months than air temperature (Figure 2) and (2) the ice cover became less stable and more susceptible to melting in December–January during our study.

Increases in the number of days with potential critical water temperatures ≥20 °C have been documented for other European lakes (Dokulil et al. 2021). Such increases also occurred in both Bohemian Forest lakes, but the change was steeper in the PL lake (Figure 3(c)). The more intensive warming of the PL epilimnion influenced its mixing dynamics and expectedly (Moser et al. 2019) affected the intensity of its thermal stratification. The PL thermocline depth thus significantly (p < 0.05) decreased by ∼1 m from 1998 to 2022, while the change was not significant in the CT lake (Figure 3(d)). For examples of thermal stratification of the study lakes, see Fig. SM-5 in the Supplementary material.

Key drivers of surface water temperature increases include air temperature, solar radiation, cloudiness, humidity, ice cover, wind, and are also mediated by local morphological terrain properties (Toffolon et al. 2014; O'Reilly et al. 2015). These variables may offset simple relationships between lake and air temperatures (Karl et al. 2015; O'Reilly et al. 2015). Most climatic variables were similar for both study lakes, however, and could not explain why the PL epilimnion warmed more steeply than the CT epilimnion (Figure 3) despite only small differences in air temperature trends (Figure 2(c) and 2(e)) and annual cycles of energy inputs (Fig. SM-2 in the Supplementary material). So the question arises: What was the dominant between-lake difference, contributing to the more rapid warming of the PL epilimnion?

We assume that besides orographic profiles of terrain in lake surroundings (Fig. SM-4 in the Supplementary material), the effect of climate change on catchment forest and resulting changes in the water quality of the PL lake represented another important difference between the PL and CT lakes. The terrestrial export of DOC has been slowly increasing in both catchments since the late 1980s because of soil recovery from acidification, and later also due to a climate-related higher frequency of high precipitation and discharge events (e.g., de Wit et al. 2021). The increasing DOC leaching was accompanied by elevated terrestrial exports of P (Kopáček et al. 2011). Both the DOC and P leaching, however, accelerated in the PL catchment after the bark beetle–induced tree dieback (Kopáček et al. 2019). Because algal production of the Bohemian Forest lakes is P-limited, the elevated P availability resulted in increasing concentrations of Ch-a. The increasing DOC concentrations also caused water brownification, manifested by significantly (p < 0.001) increasing A₂₅₄ values (Table 2). The brownification was probably more related to increasing DOC concentrations than to changes in the quality of organic matter because the molar absorptivity of organic carbon (i.e., A₂₅₄:DOC) was similar and stable in both lakes (∼40 m² mol⁻¹ on average). Both the brownification and presence of more algae in the lake epilimnion resulted in decreasing water transparency in both lakes (Table 2). The lower transparency resulted in the solar energy being adsorbed in a shallower water layer and resulted in a higher (and increasing) number of days with water temperatures ≥20 °C (Figure 3(c)). The chemical changes were more pronounced in the PL lake, where transparency decreased from 1.6 m prior to tree dieback to ∼1 m in 2020–2022 (Table 2), resulting in steeper trends in water temperatures exceeding 20 °C and in decreasing (p < 0.05) summer thermocline depths during our study (Figure 3(d)). These results suggest that between-lake differences in the rates of climate-driven water warming may also be indirectly influenced by changes in forest status in their catchments.

We evaluated air and water temperatures, ice cover, and thermocline depth at two Bohemian Forest lakes from 1998 to 2022 along with trends in climatic data from a nearby meteorological station (1961–2022). Lake water temperatures significantly increased by 0.47 and 0.32 °C decade⁻¹ at lake outlets (Table 1), and ice cover periods decreased by 11.7 and 14.8 days decade⁻¹ (Figure 3(a)) in the CT and PL lakes, respectively, i.e., with rates similar to many other freshwaters in the Northern Hemisphere. These changes were associated with increasing air temperature at all measuring stations over the whole study period (Table 1), as well as during most individual months, except for May (Figure 2) when cloudiness increased (Table SM-1 in the Supplementary material). Decreasing snow cover in winter (by 3.8 cm decade⁻¹; Figure 1) also likely contributed to earlier lake ice melt. The steeper trend in ice-off observed at the PL lake was probably associated with its different geographical position and orography, which supported more frequent foehns during late winter. In contrast, the more delayed ice-on of the CT lake was probably associated with its higher volume (and heat capacity) compared to the PL lake.

The number of days with water temperatures ≥20 °C increased more steeply in the PL than CT lakes (Figure 3(c)), and the depth of the summer thermocline decreased in the PL lake (Figure 3(d)). This between-lake difference was supported by stronger brownification and algal production (and thus decreasing transparency) in the PL lake, associated with the accelerated terrestrial export of DOC and P after bark beetle–induced tree mortality in its catchment.

The bark beetle outbreak in the PL catchment was associated with higher air temperatures and summer droughts. Our results thus suggest that bark beetle–induced tree mortality and resulting changes in water quality accelerated climate-related changes in water temperature regimes in the PL lake. Consequently, the climatic effects on catchment vegetation should be considered as another possible factor contributing to the variability in lake water temperature responses to climate change.

We thank the Šumava National Park authorities for their administrative support, D. W. Hardekopf for proofreading, and four anonymous reviewers for their helpful comments. This study was supported by the Czech Science Foundation (project No. P503-22-05421S) and partially by the Czech Academy of Sciences (Strategy AV21, project VP20-Water for life).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.