Recurrent hot droughts cause persistent legacy effects in a temperate Scots Pine forest

Abstract

• Recent hot‐dry events have caused significant impacts and legacy effects in temperate ecosystems. Here, we investigate legacy effects of the 2018 hot drought on a Pinus sylvestris L. forest in southwestern Germany and the effects of post‐2018 recurrent hot‐droughts on ecosystem carbon fluxes.

• We combined ecophysiological, remote sensing (Enhanced Vegetation Index, EVI) and micrometeorological (Net Ecosystem Carbon Exchange, NEE) measurements to assess past and present ecosystem functioning.

• We found strong and persistent legacy effects and high tree mortality of P. sylvestris, with deciduous understorey trees slowly replacing P. sylvestris. After 2018, EVI clearly followed the pattern of a deciduous‐dominated forest, indicating changes in canopy structure, type and seasonality in NEE. Significant legacy effects in NEE were found and the ecosystem shifted from a carbon sink (NEE = −391 ± 204 g C m⁻² year⁻¹, 2003–2006) to carbon neutral (NEE = +13 ± 28 g C m⁻² year⁻¹) in 2021, a cold and wet year. All other years post‐2018 were hotter and drier than the long‐term average (1991–2020), and the ecosystem was turning into a carbon source, with highest values in 2022 (NEE = +329 ± 19 g C m⁻² year⁻¹). These compound events of atmospheric and edaphic drought led to strong ecosystem carbon release post‐2018.

• Our data show that the ecosystem most likely experienced strong drought legacy effects, such as 2018, at species, community and ecosystem scales. These negative effects were further exacerbated by recurrent atmospheric and edaphic droughts, shifting the ecosystem to a net carbon source after 2018.

A series of hot drought events caused persistent legacy effects in a Scots Pine forest, with severe negative impacts on ecosystem carbon fluxes.

INTRODUCTION

In recent years, temperate ecosystems have experienced significant droughts (e.g. Schuldt & Ruehr ²⁰²²), which can evoke distinct drought responses, for example, reduced carbon (e.g. Van der Woude et al. ²⁰²³; Scapucci et al. ²⁰²⁴) and water (e.g. Haberstroh et al. ²⁰²¹; Knüver et al. ²⁰²²; Gebhardt et al. ²⁰²³) fluxes. Plant stress is further enhanced when these periods are accompanied by (untypical) high air temperature and high vapour pressure deficit (VPD) (Grossiord et al. ²⁰²⁰). While the immediate response of ecosystems to such abiotic conditions can have substantial impacts on seasonal and annual ecosystem water and carbon budgets, drought legacy effects can have long‐lasting (years to decades to permanent) negative effects on ecosystems (e.g. Anderegg et al. ²⁰¹⁵; Kannenberg et al. ²⁰²⁰). Drought legacy effects are defined as changes or responses of ecosystem properties or processes in the post‐drought phase. These alterations can be positive or negative (Müller & Bahn ²⁰²²; Vilhonen et al. ²⁰²²), however negative legacy effects are more common in forest ecosystems (Anderegg et al. ²⁰¹⁵; Vilhonen et al. ²⁰²²). Such drought legacy effects in the temperate forest ecosystems of Central Europe are increasing (e.g. Bastos et al. ^2020a; Yu et al. ²⁰²²; Pohl et al. ²⁰²³) and thus, in the focus of current scientific research. However, there is still debate about how to assess drought legacy effects in a common framework (Kannenberg et al. ²⁰²⁰), especially because they occur at different spatial and temporal scales and interact with other abiotic and biotic stressors (Kannenberg et al. ²⁰¹⁹). Müller & Bahn (2022) suggest assessment of drought legacies (duration and size) in the post‐recovery or post‐drought period, that is, after ecosystem recovery, against a pre‐defined baseline without stress to allow for better comparability between studies.

According to Müller & Bahn (2022), drought legacy effects can occur on three different scales: the species, ecosystem, and community scales, which are all inherently connected and influence each other. Legacy effects at the species scale include, among others, reduced growth rates (e.g. Kannenberg et al. ²⁰²⁰) or reduced water fluxes (e.g. Gebhardt et al. ²⁰²³; Ruehr & Nadal‐Sala ²⁰²⁶) of single trees following drought. At the community scale, shifts in species composition or dominant tree species can alter the whole ecosystem structure (Müller & Bahn ²⁰²²). Often, changes in vegetation composition follow drought related tree mortality events and have been observed across multiple biomes globally (Batllori et al., ²⁰²⁰). For example, in a Patagonian forest, an extreme drought event led to a shift in dominance from Nothofagus dombeyi towards the drought‐tolerant species Austrocedrus chilensis (Suarez & Kitzberger ²⁰⁰⁸). An experimental drought treatment in a Mediterranean holm oak ecosystem indicated a vegetation shift from Quercus ilex to Phillyrea latifolia under progressing drought (Liu et al. ²⁰¹⁸). In the Swiss Rhone valley, P. sylvestris forests are developing into Quercus pubescens dominated ecosystems as a consequence of drought (Rigling et al. ²⁰¹³). Often, these changes in species composition are permanent (e.g. Hillebrand & Kunze ²⁰²⁰; Müller & Bahn ²⁰²²).

Legacy effects at the community scale are often directly interconnected to the ecosystem scale, which includes, among others, ecosystem carbon and water cycling or plant–plant interactions (Müller & Bahn ²⁰²²). For example, after the 2018 hot drought, in 2019 net ecosystem productivity was reduced by 93 g C m⁻² (−25%) in a temperate mixed deciduous forest compared to pre‐drought years (Pohl et al. ²⁰²³). Effects at the ecosystem scale are highly connected to legacy effects at the species and community scales with possible feedbacks (Müller & Bahn ²⁰²²).

The accurate assessment of drought legacy effects at different scales is complicated by the co‐occurrence of other abiotic and biotic stressors, which can affect the different scales in multiple manners. Moreover, interactions of different stressors are complex and difficult to predict, but often lead to a stronger impact than one stressor alone (Zscheischler et al. ²⁰¹⁸). For example, the occurrence of high VPD during periods of edaphic drought has been linked to reduced photosynthesis and stomatal conductance, but also to an increased risk of plant mortality (Grossiord et al. ²⁰²⁰). Such compound events (Zscheischler et al. ²⁰¹⁸) have and will increase, especially in the combination of hot‐dry events (Bevacqua et al. ²⁰²²). In Central Europe, 2018 (Schuldt et al. ²⁰²⁰; Schuldt & Ruehr ²⁰²²) and 2022 (e.g. Copernicus Climate Change Service ²⁰²²; Schmied et al. ²⁰²³; Scapucci et al. ²⁰²⁴) can be classified as hot‐dry compound events with devastating effects on ecosystems (e.g. Schuldt et al. ²⁰²⁰; Bastos et al. ^2020b). Compared to the hot drought in 2003, which used to be the benchmark for hot droughts in Europe, 2018 was even more intense and affected an area approx. 1.5‐times the size of 2003 (Buras et al. ²⁰²⁰; Bastos et al. ^2020a). In addition, 2018 was preceded by very dry conditions, starting already in 2014 (Moravec et al. ²⁰²¹), and followed by a series of dry and hot years in Central Europe, such as 2019, 2020 and 2022 (Knutzen et al. ²⁰²⁵). Thus, impacts on ecosystems in Central Europe of the 2018 hot drought were more severe than in 2003 (e.g. Buras et al. ²⁰²⁰). For example, Thompson et al. (2020) found that the temperate forests of Europe reduced their carbon uptake by 90 ± 60 Tg C year⁻¹ in 2018 compared to 2009–2018. Likewise, Van der Woude et al. (2023) quantified the reduction of carbon uptake by the biosphere to be 56–62 Tg C in the drought affected area of Central and Southeastern Europe in the summer months (June–August) of 2022.

Yet, also 2019, 2020 and 2023 showed above average air temperature and low precipitation in many regions in Central Europe (Copernicus Climate Change Service ^{2019, 2020, 2023}), including extreme events, such as heatwaves and strong soil moisture deficits. Thus, the occurrence of recurrent drought and heat, prior to the complete recovery from the previous extreme event has been a frequent phenomenon in temperate ecosystems of Central Europe in the last decades, which further challenges the correct assessment of drought legacy effects (Müller & Bahn ²⁰²²; Werner et al. ²⁰²⁴). Only 2021 was wet and cool enough in Central Europe (Climate Change Service ²⁰²¹) to allow the possibility for proper ecosystem recovery. Thus, an increasing number of ecosystems are at the edge of permanent and persistent damage (e.g. Lloret & Batllori ²⁰²¹; George et al. ²⁰²²), some even reaching their tipping point, where a new ecosystem state is establishing due to increased mortality rates of former dominant key species (e.g. Batllori et al. ²⁰²⁰; Haberstroh et al. ²⁰²²).

A species widely distributed over Europe, which is severely suffering from (recurrent) drought events is Scots Pine (Pinus sylvestris L.), even in areas of its climate optimum (Bose et al. ²⁰²⁴). Recurrent extreme drought events of recent years, in combination with biotic agents, such as mistletoe infection or insect infestation, have led to increased dieback of P. sylvestris (Lemaire et al. ²⁰²²) in several areas of Europe, spanning from Spain to northern Germany (Bose et al. ²⁰²⁴). Although long‐considered a drought‐ and frost‐tolerant species with low site requirements (Krakau et al., ²⁰¹³), the suitability of P. sylvestris for (productive) forestry under future climate change is increasingly questioned (Brichta et al. ²⁰²³; Diers et al. ²⁰²⁴; Knocke et al. ²⁰²⁴).

A P. sylvestris site with a high level of drought impact can be found in SW Germany, in the Upper Rhine valley at the Hartheim Forest Research Site, which has provided continuous climate and tree‐level observations since 1978. The site is an associated ecosystem monitoring site of the Integrated Carbon Observation System (ICOS, Site code DE‐Har, Heiskanen et al. ²⁰²²). Even before 2018, there were signs of drought stress in this ecosystem. However, during the 2018 hot drought, midday needle water potential (Ψ) of P. sylvestris trees dropped to −7.5 MPa, triggering xylem cavitation and embolism formation. Consequently, roughly half of P. sylvestris trees at DE‐Har had died up to September 2020, and the ecosystem shifted to a carbon source in 2019 (Haberstroh et al. ²⁰²²). Thus, this ecosystem is ideally suited to investigate drought legacy effects, but also the impact of recurrent drought and heat events of recent years.

The main objectives of this study are (1) to identify persistent legacy effects in this drought‐prone P. sylvestris forest at different scales, as well as potential feedbacks between scales, and (2) to investigate the impact of recent hot drought events on critical ecosystem processes, such as ecosystem carbon fluxes. We combine measurements at the species (radial sap flow and water potential 5 years after 2018), community (vegetation dominance as depicted by the enhanced vegetation index) and ecosystem level (net ecosystem carbon exchange, NEE) to elucidate the impact of the 2018 hot drought and subsequent compound events until 2023.

We hypothesize that (a) the ecosystem, but especially P. sylvestris, has severely suffered from drought legacy effects of the 2018 hot drought at all scales, and that this may induce a shift in species dominance in favour of deciduous species. Further, (b) we postulate that co‐occurring atmospheric and edaphic droughts (i.e. compound droughts) since 2018 in combination with persistent drought legacy effects may shift the ecosystem to an annual net carbon source.

MATERIALS AND METHODS

Experimental site and design

The study was carried out at the permanent Hartheim Forest Research Site DE‐Har (7.59814° E, 47.93391° N, 201 m a.s.l.). DE‐Har is located in the Upper Rhine Valley, SW Germany. The field site was established in 1974 and is located in a slow‐growing Pinus sylvestris L. forest planted in 1963 (Holst et al. ²⁰⁰⁸). The mean annual air temperature (T _air ) is 10.6°C and the mean annual precipitation is 646 mm (1991–2020). In the growing season (April–September), mean T _air is 16.1°C, with a mean precipitation of 391 mm (1991–2020) (Christen et al. ^2024a). Access to the canopy is currently provided by two towers (18 and 30 m; Fig. 1 and Fig. S1), the taller tower provides continuous measurements of above‐canopy radiation fluxes, sensible heat, water and CO₂ fluxes by eddy covariance as part of ICOS.

Fig. 1.

Aerial view of the Pinus sylvestris ecosystem with the large canopy access tower (30 m) in May (A) and September (B) 2023. Dead P. sylvestris trees, gaps in the forest canopy and a dense layer of understorey trees are visible throughout the forest.

The P. sylvestris canopy reached a height of 14.3 m with a density of about 800 trees ha⁻¹ and a leaf area index (LAI) of 1.9 m² m⁻² in 2003 (Schindler et al. ²⁰⁰⁶). No silvicultural management treatment, such as thinning, has occurred since then. Between 2003 and 2014 the P. sylvestris canopy grew to a height of 17–18 m and levelled to 19 m height from 2014 to 2023. On site, a dense understorey of broadleaved species has developed, primarily since 2015 with an average height of 10 m in 2023, leading to an LAI under the dense understorey of up to 5.05 m² m⁻² in June 2023. The LAI of the whole forest in 2021 (2.81 m² m⁻²), 2022 (2.66 m² m⁻²) and 2024 (2.60 m² m⁻²) (Plapp et al. ²⁰²⁵) increased compared to 2003. Yet, the forest is still composed of two layers: larger P. sylvestris trees and several different deciduous tree species in the second layer (from here on ‘understorey’) (Fig. 1). The main species in the understorey are Tilia spp. (Tilia cordata Mill. and Tilia platyphyllos Scop.) (32% of understorey trees), Carpinus betulus L. (24%) and Fagus sylvatica L. (15%) (Plapp et al. ^2024b), which are all species capable of high photosynthesis rates under full sunlight conditions (e.g. Legner et al. ²⁰¹³). As a consequence of the 2018 hot drought, >60% of P. sylvestris trees had died until 2023, which is an increase of ~13% compared to tree mortality in September 2020 (Haberstroh et al. ²⁰²²). In 2023, the stand density of P. sylvestris declined to 375 trees ha⁻¹ of which 116 trees ha⁻¹ were considered standing dead trees (Plapp et al. ^2024b). Dead trees were not felled until they had fallen or endangered infrastructure. Until winter 2023/2024, dead wood remained on the ground (Fig. 1).

Due to its location in the floodplain of the Rhine valley, the sediment consists of a carbonate‐rich (pH = 7.8), two‐layer Haplic Regosol (Goffin et al. ²⁰¹⁴). As a result of historical river regulation of the nearby Rhine between 1817 and 1959, the groundwater level is around 7 m (Brandes et al. ²⁰⁰⁷) and not within reach of tree roots.

Meteorological measurements

In‐canopy air temperature (T _air ) at DE‐Har was measured 2 m above ground level at the base of the tall tower with an actively ventilated psychrometer using a PT1000 based on the design of Frankenberger (1951). Precipitation was recorded on top of the small tower at a height of 18 m above the ground using a heated ombrometer Model 5.4031 (Adolf Thies, Göttingen, Germany). Although the P. sylvestris canopy is at 17–19 m, no trees with such a canopy height grew in close vicinity to the precipitation measurement on the tower, which was unaffected by trees (Fig. S1). Vapour pressure deficit (VPD) was measured at 27 m on the tall tower above the canopy using a CS215 temperature/humidity probe in a passively ventilated screen (Campbell Scientific, Logan, UT, USA). Shortwave irradiance was measured using a thermopile pyranometer (CM21; Kipp & Zonen, Delft, Netherlands) at 30 m on the tall tower. Volumetric water content reported in this experiment was taken as the average of six sensors inserted in two spatially separated profiles in 0.05, 0.10 und 0.20 m depth using CS655 soil sensors (Campbell Scientific) according to ICOS instructions (Op de Beeck et al. ²⁰¹⁷).

Ecophysiological measurements 2023

To compare the current ecophysiological response of P. sylvestris with deciduous understorey trees in 2023, C. betulus, as one of the dominant understorey tree species (24% of all broadleaf understorey trees; Plapp et al. (2024b)), was chosen due to its proximity to the canopy access tower and accessibility of the tree crown for water potential measurements, similar to the studied P. sylvestris trees. Both species had been monitored already during the 2018 hot drought for water potentials (Haberstroh et al. ²⁰²²). Water potential of P. sylvestris (n = 5) and C. betulus (n = 6) was measured pre‐dawn (Ψ _PD ) (before sunrise, 04:00–06:00 CEST) on terminal branchlets, representing leaf water potentials, using a Scholander‐type pressure chamber (1505D; PMS Instrument, USA) (Scholander et al. ¹⁹⁶⁵). Branchlets were cut with long‐reach pruning shears from the ground or from canopy access towers. Measurements were taken on six different days between April and September 2023.

Sap flow sensors (SFM1; ICT International, Armidale, Australia) were installed on P. sylvestris (n = 7) and C. betulus (n = 6) in early May 2023. Sensors were placed in the trunk 1.20 m above ground and consisted of three 35 mm stainless‐steel needles. Measurements were taken at two sapwood depths: at 12.5 (outer xylem) and 27.5 mm (inner xylem) below the cambium. All trees had a sapwood depth > 27.5 mm. Heat pulses (30 J) were generated by a heating element in the middle needle every 20 min; data were stored on internal loggers. Calculation of sap flux density (J _s ) followed the Heat Ratio Method (Burgess et al. ²⁰⁰¹). Total J _s was determined by averaging J _s in the two different measurement depths for both species. To account for nocturnal sap flow, J _s was corrected for offsets from the baseline on nights with unlikely environmental conditions for nocturnal sap flow (photosynthetically active photon flux density <5 μmol m⁻² s⁻¹, VPD <0.05 kPa, T _air  >0°C, RH > 95%) between May and November 2023. Due to sensor failure, J _s of three P. sylvestris trees was removed from the dataset starting from 24/08/23, 06/09/23 and 21/09/23. For comparison, max. J _s of P. sylvestris measured at DE‐Har were derived from Brandes et al. (2007) for the years 2003 and 2004.

Selected P. sylvestris trees had a height of 16.6 ± 0.4 m and a diameter at breast height (DBH) of 24.8 ± 1.1 cm. C. betulus trees were smaller (10.7 ± 0.3 m) and had a lower DBH (16.8 ± 0.4 cm). Note that not all P. sylvestris trees measured for sap flow were accessible for water potential measurements.

Enhanced vegetation index (EVI) and phenocams

Vegetation indices, such as the enhanced vegetation index (EVI) are an ideal tool to track ongoing changes in vegetation abundance and seasonality (Huete et al. ²⁰⁰²). Values for EVI typically vary between 0 and 1, where higher EVI values denote greener vegetation than lower EVI values (Swain et al. ²⁰¹⁷). EVI is responsive to variations in canopy structure, such as LAI, canopy type or architecture (Gao et al. ²⁰⁰⁰; Huete et al. ²⁰⁰²), which naturally vary between deciduous‐ and conifer‐dominated forests. For other vegetation indices, such as the normalized difference vegetation index (NDVI), it has been shown that deciduous‐dominated forests have higher maximum values than coniferous forests in summer (e.g. Soudani et al. ²⁰¹²) and show clearer differences in reflectance from summer to winter than evergreen species due to a leafless period (Wang et al. ²⁰²³). Data for EVI was downloaded and analysed from the Moderate‐resolution Imaging Spectroradiometer (MODIS; MOD13Q1 V6.1; Didan ²⁰²¹) via the Google Earth Engine at a spatial resolution of 250 m. For the exact calculation of EVI, please see Methods S1, equations S1 and S2 in the Supporting Information—S1. Data for DE‐Har were centred on the tall tower location and cover an area of 62,500 m² (250 × 250 m). Thus, the area covered by MODIS is larger than the actual field site. However, as the actual research plot is embedded within a larger P. sylvestris forest, it can be assumed that EVI data with a resolution of 250 m are sufficient to detect changes at the field site. Likewise, the EVI resolution fits the eddy covariance footprint well: during daytime (shortwave radiation >10 W m⁻²) 70% of the fluxes originate from an area, which covers ~54,000 m² and 80% from an area which covers ~125,000 m².

One data point was available for every 16‐day period and averaged per month from 2000 to 2023. If all input pixels were cloudy, a computation based on a historical database occurred (Didan 2021). Nevertheless, we filtered data, which were lower than the 5% confidence interval in winter to account for unrealistically low values caused by extreme winter conditions. Differences between evergreen and deciduous vegetation are especially pronounced in winter and summer. Thus, we calculated average values for EVI between July and August (summer), as well as for December to February (winter). In winter, EVI can be influenced by snow cover, however differences between vegetation types should still be visible (cf. Wang et al. ²⁰²³). In addition, we calculated seasonal courses of EVI for pre‐ (2000–2017) and post‐2018 years (2019–2023) to depict monthly changes in EVI after 2018.

Two phenocams (Star Dot NetCam SC 5MP D/N; Star Dot Technologies., Buena Park, CA, USA) have been continuously operated since 2018 at the tall tower, taking recordings of the canopy (29.6 m) and understorey (8.4 m) facing towards North at 30‐min intervals. Pictures can be accessed from the phenocam network (see data availability statement). From the cameras, green chromatic coordinate values (G_CC) are calculated for selected regions of interests according to Richardson et al. (2018), which have been shown to closely track leaf phenology (Sonnentag et al., ²⁰¹²). We chose the 90th percentile across a 3‐day moving window to minimize the effects of atmospheric variations on G_CC values (Richardson et al. ²⁰¹⁸; Seyednasrollah et al. ²⁰¹⁹). Finally, we extracted the dates of the maximum yearly value of G_CC (90th percentile) and the date where 10% and 50% of this maximum value was reached (Richardson et al. ²⁰¹⁸; Seyednasrollah et al. ²⁰¹⁹), to depict the approximate timing of leaf flush of the understorey, that is, the time point when the deciduous understorey became photosynthetically active (Dougherty et al. ¹⁹⁷⁹; Greco & Baldocchi ¹⁹⁹⁶).

Ecosystem carbon fluxes

Ecosystem CO₂ fluxes between the forest and the atmosphere are available from eddy covariance from 2003 to 2006 and from 2019 to 2023. Details on the instrumentation and data processing of the eddy covariance data from 2005 are described in Holst et al. (2008). The measurement set‐up and data processing from 2019 to 2023 are identical to Haberstroh et al. (2022). The set‐up consists of a CSAT‐3 ultrasonic anemometer‐thermometer (Campbell Scientific) co‐located with a closed‐path infrared gas analyser (Li‐7200; Licor, Lincoln, NB, USA) on the tall tower at a height of 27 m (Fig. S1). Fluxes were processed in 30‐min blocks using Eddy‐Pro. Data with quality control flags 0 and 1 and a flow rate >12 l min⁻¹ are further used in this study. Data availability of measured CO₂ fluxes was 59% (2019), 64% (2020), 73% (2021), 71% (2022) and 77% (2023). CO₂ fluxes under low turbulence were removed using an iterative friction velocity threshold (Papale et al. ²⁰⁰⁶). Missing CO₂ fluxes were then gap‐filled following Reichstein et al. (2005) using REddyProc (v. 1.3.3; Wutzler et al. ²⁰¹⁸). With REddyProc, fluxes were partitioned into gross primary productivity (GPP; Fig. S2A) and ecosystem respiration (R_eco; Fig. S2B) based on nighttime fluxes (Reichstein et al. ²⁰⁰⁵) and data from air temperature at 2 m and above‐canopy short‐wave irradiance.

It should be noted that Holst et al. (2008) used an open‐path sensor in their study compared to the closed‐path sensor used in the data from 2019 to 2023. Open‐path sensors can be subject to carbon flux underestimation in winter in cold climates, due to artificial heating of the instruments surface (Burba et al. ²⁰⁰⁸). While in Holst et al. (2008) there was no correction applied for instrument‐related sensible heat flux (Burba et al. ²⁰⁰⁸), we are confident that potential errors in NEE are small. Carbon flux winter underestimation is a major problem when T _air drop <−10°C (Cassidy et al. ²⁰¹⁶), which was rarely the case at DE‐Har. In 2005, only on 6 days min. T _air reached values <−10°C and only 6 days reached average T _air  < −5°C (Christen et al. ^2024b). Similarly, differences between open‐path and closed path sensors are only a major issue at T _air  < −10°C and negligible at T _air between 0 and 20°C (Cassidy et al. ²⁰¹⁶, supplement).

Assessment of legacy effects on net ecosystem carbon exchange

The analysis of legacy effects requires a predefined baseline without marked stress conditions against which data can be assessed (Müller & Bahn ²⁰²²). Unfortunately, no data for ecosystem carbon fluxes were available from 2007 to 2018 (Table 2), due to lack of resources, staff and instrumentation. However, remote sensing products, such as the NDVI and EVI, which are often used as proxy for gross primary productivity (GPP) (e.g. Sims et al. ²⁰⁰⁶), provide evidence that the P. sylvestris forest was in a rather stable condition before 2018 (Haberstroh et al. ²⁰²²). Thus, we decided to compare NEE of post 2018 years against NEE of 2005, which depicted optimal growth conditions (Holst et al. ²⁰⁰⁸) with growing season (April–September) T _air (15.8°C, −0.3 K relative to 1991 climatic reference period) and precipitation (381 mm, −10 mm) resembling growing season averages (1991–2020), thus stressing its suitability as baseline to assess legacy effects. Other available years, such as 2003/2004 or 2006 were characterized by distinct summer drought periods (Schindler et al. ²⁰⁰⁶; Holst et al. ²⁰⁰⁸). As Holst et al. (2008) did not partition NEE in GPP and R_eco, we used NEE for this comparison. We calculated monthly deviations in NEE from 2005 as in equation 1:

$${\mathrm{\Delta }}_{\mathrm{NEE}2005}={\mathrm{NEE}}_{\mathrm{X}}-{\mathrm{NEE}}_{2005}$$ (1)

where NEE_X is monthly NEE of the years 2019, 2020, 2021, 2022 and 2023, and NEE₂₀₀₅ is the corresponding monthly NEE measured in the year 2005.

Table 2.

Net ecosystem carbon exchange (NEE), gross primary productivity (GPP) and ecosystem respiration (R_eco) for the Pinus sylvestris forest at DE‐Har for 2003–2006 and 2019–2023 with difference of NEE to NEE of the year 2005 (ΔNEE₂₀₀₅).

Year	NEE (g C m−2 year−1)	ΔNEE2005 (g C m−2 year−1)	GPP (g C m−2 year−1)	Reco (g C m−2 year−1)	references
2003/04	−195	–	–	–	Schindler et al. (2006)
2005	−603	–	–	–	Holst et al. (2008)
2006	−376	–	–	–	Holst et al. (2008)
2019	229 ± 18	+832	1,403 ± 141	1,632 ± 128	This study
2020	71 ± 15	+674	1,418 ± 24	1,489 ± 14	This study
2021	13 ± 28	+616	1,561 ± 107	1,573 ± 130	This study
2022	329 ± 19	+932	1,287 ± 75	1,616 ± 89	This study
2023	63 ± 1	+666	1,909 ± 199	1,973 ± 199	This study

Analogous to equation 1, we calculated the difference in NEE of the year 2021 to NEE of all years post‐2018 (Δ_NEE2021). 2021 was the only year after 2018 with above average precipitation (693 mm, +47 mm) and below average T _air (9.9°C, −0.7 K). Thus, it is the only year post‐2018, which qualifies as baseline, according to Müller & Bahn (2022).

Statistical analysis

To compare the ecophysiological behaviour of both species in 2023, linear mixed effect models (lmer) (lme4, Bates et al. ²⁰¹⁵) were calculated for the response variables Ψ _PD and J _s . For Ψ _PD , species and day of year, and for J _s species were treated as fixed effect; TreeID was set as random effect in both models. For J _s , a second model was conducted to compare J _s at the two different xylem depths per species. Significant differences in monthly EVI before and after 2018 were detected using a linear mixed effect model (lmer) with the response variable ‘EVI’ predicted by the variable ‘month’ and ‘period’ (pre‐ and post‐2018). Year was set as random effect. Significant differences in all of these models were assessed with pairwise comparisons (Tukey adjustment) with the function emmeans (Lenth 2024) in R (R Core Team ²⁰²²). Ψ _PD was square‐root transformed to meet model assumptions. For all J _s models, model assumptions were not met, but model results were confirmed with a non‐parametric Wilcox test.

Monthly Δ_NEE2005 was correlated with average monthly T _air and maximum monthly VPD (VPD_max) as proxies for atmospheric drought, and with average monthly VWC and monthly precipitation sum as proxies for edaphic drought. The relationship of Δ_NEE2005 and environmental parameters was assessed with linear regressions (lm) in R (R Core Team ²⁰²²). For these regressions, only data points where monthly mean T _air exceeded 15°C were used to isolate months with a high likelihood of atmospheric and/or edaphic drought conditions. For 2019, no reliable VWC data were available on a monthly basis due to instrument communication failures. Likewise, 2 months in 2021 (June and July) were removed from the analysis due to similar technical problems with VWC. Assumptions for all models were checked with the Shapiro–Wilk test (normal distribution of residuals) and the function simulateResiduals in the package DHARMa (Hartig 2022).

The relationship of Δ_NEE2021 and environmental parameters (assessed with piecewise two segment linear regressions with the function segmented for T _air , VPD_max and VWC (segmented, Fasola et al. ²⁰¹⁸) and linear regression for precipitation) can be found in the Figs. S4 and S5.

RESULTS

Meteorological conditions between 2018 and 2023

2018 was the driest year at DE‐Har, with only 417 mm of precipitation (−230 mm, −36%) compared to the long‐term annual precipitation of 646 mm in the 1991–2020 reference period (Fig. 2A). All following years, except 2021, were drier and warmer than the 1991–2020 reference period, with conditions in 2020 (463 mm, −183 mm, −28%) being almost as dry as 2018. 2022 was characterized by both, above‐average in‐canopy T _air (12.0°C, +1.4 K) and below average precipitation (553 mm, −93 mm, −14%). Only 2021 was colder (9.9°C, −0.7 K; 693 mm, +47 mm, +7%) than the long‐term average (1991–2020). In contrast, 2023 was the hottest year (+12.3°C, +1.7 K) on record for the investigated ecosystem, with only minor deviations from the long‐term precipitation mean (−40 mm, −6%, 1991–2020). In particular, 2023 was characterized by a hot summer with two distinct periods of high T _air in July (T _max : 36.0°C) and August 2023 (T _max : 36.2°C) (Fig. 2B). Topsoil moisture (first 0.20 m) declined in May and early June, but did only recover to pre‐drought levels in mid‐November despite frequent precipitation during summer (Fig. 2C).

Fig. 2.

Yearly in‐canopy air temperature (T _air ) and above‐canopy precipitation anomaly to the long‐term climate reference period (1991–2020) of the years 2003, 2005 and 2018–2023 (A). All other years since 1978 are illustrated by points in the background. Mean air temperature (T _air ) with minimum and maximum T _air , cumulative precipitation (B), volumetric soil water content (VWC) in the top 0.20 m and precipitation (C) for the year 2023.

Ecophysiological tree responses in 2023

Pre‐dawn water potentials (Ψ _PD ) were similar between both investigated species in May and July 2023, but significantly differed in June (p < 0.001) with higher values for C. betulus (Fig. 3A). From early August onwards, only a slight drop in Ψ _PD from −0.26 ± 0.08 MPa to −0.52 ± 0.03 MPa occurred for P. sylvestris, while Ψ _PD of C. betulus differed significantly (p < 0.001), decreasing strongly from −0.53 ± 0.08 MPa to −2.26 ± 0.17 MPa. In September, Ψ _PD of.

Fig. 3.

Pre‐dawn water potential (Ψ_PD ) ± 1 SE for Pinus sylvestris (n = 5) and Carpinus betulus (n = 6) between May and September 2023 (A). Total daily sum of sap flux density (J _s ) ± 1 SE for P. sylvestris (n = 7) and C. betulus (n = 6) (B) with max. J _s for P. sylvestris in 2003 and 2004 (Brandes et al. ²⁰⁰⁷) and daily sum of sap flux density (J _s ) ± 1 SE in the inner and outer xylem for P. sylvestris (n = 7) (C) and C. betulus (n = 6) (D) between May and November 2023. Significance differences between species in panel (A) is given at P < 0.001***. Please note the different scale of panel (C) to panels (B) and (D).

C. betulus recovered to values close to 0 MPa, while Ψ _PD of P. sylvestris remained at values of −0.49 ± 0.05 MPa (p < 0.001).

Sap flux density (J _s ) of both species demonstrated vast and persistent differences between species water fluxes in 2023 (p < 0.001). J _s of P. sylvestris rarely exceeded 50 cm³ cm⁻² day⁻¹ in spring (max.: 55.6 ± 13.7 cm³ cm⁻² day⁻¹), which is lower than J _s measured in 2003 and 2004 (Fig. 3B, Brandes et al. ²⁰⁰⁷). J _s of P. sylvestris nearly ceased in the first dry‐down phase in June (Fig. 3B) and remained at a low level until the end of the year. Over the whole measurement period, J _s in the inner xylem was significantly reduced by 49% on average compared to J _s in the outer xylem (p < 0.001) (Fig. 3C). J _s of C. betulus peaked in spring (228.1 ± 26.7 cm³ cm⁻² day⁻¹), and was four times higher than J _s of P. sylvestris throughout the summer. After the first dry‐down in June, J _s recovered to values exceeding 150 cm³ cm⁻² day⁻¹ until the end of August, when J _s , as well as Ψ _PD of C. betulus dropped strongly, but recovered quickly in September (Fig. 3B). For C. betulus, J _s of the inner and outer xylem were also significantly different (p < 0.05). Until September, outer J _s exceeded inner J _s , while this pattern was reversed afterwards (Fig. 3D).

Enhanced vegetation index (EVI) and green chromatic coordinate (G_CC ) values

Long‐term records of EVI show that from 2000 to 2017, the summer EVI of the ecosystem at DE‐Har was relatively stable with values ranging between 0.405 ± 0.015 and 0.442 ± 0.016 (Fig. 4A). Only in 2003, an extreme dry year, a drop in EVI (0.390 ± 0.015) was evident. Similarly, during the 2018 drought, values decreased to 0.404 ± 0.024 (Fig. 4A). Interestingly, after 2018, an increase in summer EVI to 0.490 ± 0.031 (2023) was observed, with the exception of the extreme drought in 2022 (0.388 ± 0.026). Contrarily, in winter, EVI strongly dropped from 0.267 ± 0.016 after 2018 to values of 0.219 ± 0.004 in 2023 (Fig. 4A). Seasonal courses of EVI before and after 2018 confirm these seasonal dynamics (Fig. 4B). In January, March, November and December, EVI significantly decreased after 2018 (p < 0.05), while a significant increase could be observed from May to July (p < 0.05).

Fig. 4.

Enhanced vegetation index (EVI) of the P. sylvestris ecosystem at a spatial resolution of 250 m centred on the main tower location at DE‐Har from 2000 to 2023 for summer (July and August) and winter (December–February) ± 1 SE (A) and seasonal course of EVI for the periods 2000–2017 and 2019–2023 (B). Significant differences between months given at *P < 0.05, **P < 0.01.

From green chromatic coordinate (G_CC) values, it is evident that leaves of deciduous understorey trees started to develop in late March, reaching 50% of the max. G_CC value in mid‐April (Table 1). In early May, G_CC values reached their annual max. values, indicating full leaf development. Variations within years were small, however leaf development in 2019, after the 2018 hot drought, and in 2021, a particularly cold and wet year, were slower than in 2020, 2022 and 2023 (Table 1).

Table 1.

Dates for green chromatic coordinate (G_CC) values (90th percentile), when 10% and 50% of the maximum G_CC was reached (transition date) and when the maximum G_CC was reached per year. These dates indicate the timing of the leaf flush and development of understorey trees at DE‐Har (phenocam at 8.4 m – hartheim2).

Year	2019	2020	2021	2022	2023
Transition date (10%)	28/03	26/03	30/03	29/03	25/03
Transition date (50%)	15/04	9/04	21/04	16/04	14/04
Maximum GCC	20/05	1/05	11/05	5/05	5/05

Ecosystem carbon fluxes

Before the 2018 drought, the ecosystem was a considerable carbon sink with average values for NEE of −391 ± 204 g C m⁻² year⁻¹, including dry years, such as 2003 or 2006 (Schindler et al. ²⁰⁰⁶; Holst et al. ²⁰⁰⁸) (Table 2). In 2005, the predefined baseline, NEE reached values of −603 g C m⁻² year⁻¹. Carbon uptake was especially high during the vegetation period, however, also in winter the ecosystem was close to carbon neutral (Fig. 5A). In the past, during single dry and hot years, such as 2003 (Fig. 2A), and 2006, which was characterized by high summer air temperature, the ecosystem was still a significant carbon sink (Table 2).

Fig. 5.

Monthly net ecosystem carbon exchange (NEE) of the P. sylvestris ecosystem for the years 2005 and 2019–2023 (A) and monthly difference in NEE of 2005 (Δ_NEE2005) to the years 2019–2023 (B).

By comparing NEE of 2005 and 2021 (+13 ± 28 g C m⁻² year⁻¹), both years without marked stress conditions during the growing season (Figs. 2A, 5A), it becomes evident that the ecosystem has lost a considerable proportion of its carbon sink capacity (Δ_NEE = +616 g C m⁻² year⁻¹; Table 2). Compared to 2005, NEE decreased in all months post 2018 (positive Δ_NEE2005) (Fig. 5B). Between November and February, this difference was quite stable at (~ + 50 g C m⁻² month⁻¹), but increased strongly thereafter, peaking in August (except 2021) with the largest difference observed in August 2022 (Δ_NEE2005 = +181 g C m⁻² month⁻¹) (Fig. 5B). This large difference in August 2022 was mostly driven by a large drop in GPP (Fig. S2). The largest yearly difference to 2005 was also reached in the hot and dry year 2022, with values of +932 g C m⁻² year⁻¹ (Δ_NEE2005) (Table 2). Notably, GPP has continuously increased since 2019, with the exception of 2022 (Table 2, Fig. S2), reaching a maximum of 1,909 ± 199 g C m⁻² year⁻¹ in 2023. These patterns fit the overall increase in EVI (Fig. 4), which is depicted by the highly significant correlation between GPP and EVI (p < 0.001, R ² = 0.91) (Fig. S3). However, R_eco similarly increased with EVI (p < 0.001, R ² = 0.70) (Fig. S3), partially offsetting the increase in GPP (Fig. S2B, Table 2). In 2023, R_eco reached a maximum of 1,973 ± 199 g C m⁻² year⁻¹.

These patterns in GPP and R_eco led to the ecosystem being a (strong) carbon source, for example, in 2022 (+329 ± 19 g C m⁻² year⁻¹), but also in the comparably wet, but hot, 2023 (+63 ± 1 g C m⁻² year⁻¹), with 2020 (+71 ± 15 g C m⁻² year⁻¹) and 2019 being in between (+229 ± 18 g C m⁻² year⁻¹) (Fig. 5A and Table 2).

As the ecosystem was exposed to recurrent drought and heatwaves after 2018, all post‐drought years (except 2021) were drier and hotter (2019, 2020, 2022, 2023) than the long‐term average (Fig. 2A). Fig. 6 illustrates the impact of these conditions on Δ_NEE2005. Especially under atmospheric (and edaphic) drought conditions, NEE significantly decreased in comparison to 2005, that is, with rising T _air and VPD_max (p < 0.001), but also with declining VWC (although not significant) (Fig. 6A, B, D). However, this decrease in NEE was only evoked after a certain threshold was exceeded (T _air  > 15°C). Under colder and wetter conditions, Δ_NEE2005 did not show a clear pattern with T _air , VPD_max and VWC and fluctuated at values of Δ_NEE2005 ~ +50 g C m⁻² month⁻¹ with some minor exceptions. In addition, NEE declined non‐significantly with increasing precipitation sum (Fig. 6C). However, some of these large precipitation events co‐occurred as convective precipitation events with high T _air in the summer months, such as in August 2019, 2020 or 2022. All of the above‐mentioned patterns with environmental parameters, but especially VWC, were confirmed and strengthened when 2021 instead of 2005 was used as baseline. Here, NEE strongly decreased in comparison to 2021 when the following thresholds were crossed: T _air  > 16.1 ± 1.0°C, VPD_max >2.4 ± 0.4 kPa, VWC < 10.1 ± 1.0% (Fig. S4, segmented linear regressions).

Fig. 6.

Relationship of monthly difference in net ecosystem exchange (NEE) of 2005 (Δ_NEE2005) to the years 2019–2023 and monthly mean air temperature (T _air ) (A), monthly maximum vapour pressure deficit (VPD_max) (B), monthly precipitation sum (C) and monthly mean soil volumetric water content (VWC) (D). The same analysis for Δ_NEE2021 can be found in Fig. S4 in the Supporting Information—S1. For linear regressions in panels A, B, D, only data from months with a mean T _air  > 15°C were used. Significance at P < 0.001***. Large symbols in panels A, B and D were used for the linear regressions.

In conclusion, the decrease in monthly NEE compared to 2005 (and 2021) was stronger with both atmospheric and edaphic droughts, which complicates assessment of the individual impact of both stressors on Δ_NEE2005. Yet, Fig. 7 and Fig. S5 depict that NEE especially decreased in comparison to 2005 (or 2021) when both stressors co‐occurred in a compound event, that is, high T _air and low VWC. These compound events preferentially materialized in the summer months and led to a strong carbon release from the P. sylvestris forest. In Fig. 7, 2 months at moderately high T _air (14.2°C and 14.9°C), but high VWC (19.9% and 22.7%) stand out. These months (September and October 2022) followed the extreme drought and heat month of August 2022, where the difference in NEE from 2005 was highest in the study period (Δ_NEE2005 = +181 g C m⁻² month⁻¹) (Figs. 5B and 7). The winter months (blue points) did not significantly contribute to deviations in NEE from 2005 (Figs. 5 and 7). Similar patterns were found when Δ_NEE2021 instead of Δ_NEE2005 was used (Fig. S5).

Fig. 7.

Monthly difference in net ecosystem exchange (NEE) of 2005 (Δ_NEE2005) to the years 2020–2023, monthly mean air temperature (T _air ) and monthly mean volumetric soil water content (VWC). Larger circles indicate more positive Δ_NEE2005. Months are colour‐coded from 1 (January) to 12 (December). Different months are illustrated with a colour scale. The same analysis for Δ_NEE2021 can be found in Fig. S5 in the Supporting Information—S1.

DISCUSSION

The occurrence of recurrent hot‐dry events (compound droughts) and concurrent drought legacy effects severely threaten ecosystem processes, functioning and survival. In our P. sylvestris forest in SW Germany (DE‐Har), ecophysiological parameters, EVI and NEE, indicate that this particular ecosystem is suffering severely on all assessed scales. We attribute this to the legacy effects of the 2018 hot drought, combined with the dry and hot conditions that preceded and followed it. The forest has lost a considerable capacity for carbon uptake due to high P. sylvestris mortality, which induced a vegetation shift towards a more deciduous‐dominated forest. Although GPP did steadily increase after 2019, the ecosystem has mainly become a net source of carbon dioxide due to a subsequent increase of R_eco, exacerbated by recurrent atmospheric and edaphic droughts.

Progressive shift in species dominance as legacy effect of (recurrent) hot droughts with persistent impact on net ecosystem carbon exchange

Data from ecophysiological measurements in 2023 indicate that P. sylvestris trees were rather inactive with low tree water fluxes. While this is in line with its isohydric behaviour (Poyatos et al. ²⁰⁰⁸) and the dry growing season in 2023 (−81 mm of precipitation compared to 391 mm between 1991 and 2020), J _s of P. sylvestris at DE‐Har was higher in the extreme drought year 2003 (Brandes et al. ²⁰⁰⁷) (−129 mm of precipitation compared to 319 mm between 1991 and 2020 in the vegetation period). As needle water potentials of P. sylvestris dropped to −7.5 MPa in 2018 (Haberstroh et al. ²⁰²²) and Ψ ₈₈ values (−5.70 MPa; Martínez‐Vilalta et al. ²⁰⁰⁹) were crossed, it is likely that xylem conductivity decreased, for example, through embolism formation (Lüttschwager & Remus ²⁰⁰⁷; Haberstroh et al. ²⁰²²). The reduction of ~50% in J _s in the inner xylem might be species‐specific for P. sylvestris (Köstner et al. ¹⁹⁹⁶; Cermak & Nadezhdina ¹⁹⁹⁸; Nadezhdina et al. ²⁰⁰²). In contrast, the deciduous C. betulus, exhibited water‐spending behaviour with high J _s in the inner and outer xylem even during dry and hot periods in 2023. Signs of stress in C. betulus were only found in late August, but trees recovered quickly in September (cf. Haberstroh et al. ²⁰²⁵). This indicates the understorey vegetation is more vital and resilient than the former dominant P. sylvestris (Haberstroh et al. ²⁰²²). Most probably, understorey trees still profited from microclimatic buffering (Zellweger et al. ²⁰²⁰; Werner et al. ²⁰²¹), such as lower solar irradiance, maximum T _air and increased cooling effects of evapotranspiration (Zellweger et al. ²⁰¹⁹; Greiser et al. ²⁰²⁴). Concurrently, understorey trees potentially experienced a release from competition (Haberstroh & Werner ²⁰²²) of P. sylvestris due to ongoing mortality, and a higher resource availability, such as light, water or nutrients (Lloret et al. ²⁰¹²; Gessler et al. ²⁰¹⁷).

These patterns are well reflected in the dynamics of EVI after 2018, which suggest an up‐greening in summer, but reduction in greenness in winter, indicating a change in canopy type and structure (Gao et al. ²⁰⁰⁰; Huete et al. ²⁰⁰²) from a coniferous‐ to a deciduous‐dominated forest (cf. Soudani et al. ²⁰¹²). P. sylvestris decline and mortality most likely led to more opening of the upper canopy (Fig. 1), which exposed smaller and younger deciduous understorey vegetation with higher EVI values (cf. Fiore et al. ²⁰²⁰) and explains the opposing trend from increasing EVI values und decreasing carbon sink after 2018. Indeed, GPP has been steadily increasing with increasing EVI after the 2018 hot drought (Fig. S3).

From these observed patterns at the species and community scales, it is fair to assume that the contribution of P. sylvestris to NEE was rather diminished compared to pre‐2018 conditions, leading to a shift to the younger understorey vegetation. A strong change in seasonal NEE from pre‐ to post‐2018 can be seen in the winter and autumn months, where the ecosystem shifted from carbon sink/neutral to an average carbon source (Fig. 5A). Evergreen coniferous tree species are capable of fixing and using carbon on favourable, sufficiently warm days in these seasons due to their extended growing period (Falge et al. ²⁰⁰²; Wang et al. ²⁰⁰³; Bowling et al. ²⁰²⁴; Diers et al. ²⁰²⁴). The G_CC values from the understorey Phenocam indicate that leaf flush and development (and thus the capacity to fix carbon) of the deciduous understorey are only fully completed in early May (Table 1), which is also depicted by the large increase in GPP from April to May (Fig. S2). Thus, it is likely that the seasonal changes in NEE are mostly driven by a shift from coniferous to deciduous forest. Potentially, also warmer winter air temperatures have contributed to the increased carbon release in winter (Liu et al. ²⁰²⁰; Sanders‐DeMott et al. ²⁰²⁰) and should be considered in the future as driving force of winter ecosystem respiration (Van der Woude et al. ²⁰²³). Nevertheless, it should be noted that, eventually, differences in winter NEE between 2005 (Holst et al. ²⁰⁰⁸) and 2019–2023 might be smaller on certain days than depicted in Fig. 5 due to the missing correction for instrument‐related sensible heat flux (Burba et al. ²⁰⁰⁸) in Holst et al. (2008).

All of the above‐discussed patterns and processes led to a severely diminished net capacity of the ecosystem to sequester carbon. This is in line with reported legacy effects on NEE in other ecosystems (e.g. Yu et al. ²⁰²²; Pohl et al. ²⁰²³), however the persistent reduction in NEE at DE‐Har compared to pre‐2018 levels, ranging from +616 to +932 g C m⁻² year⁻¹ (Table 2), is particularly strong. Forests are an important terrestrial carbon sink (Pan et al. ^{2011, 2024}), but if more forests suffer from such strong reductions in carbon uptake, this invaluable ecosystem service will be severely threatened and potential carbon cycle feedbacks to climate change are likely to occur (diminished land sink). This is in line with the last forest inventory in Germany (Bundeswaldinventur), which found that German forests were an average carbon source between 2017 and 2022 (BMEL ²⁰²⁴), enhanced by disturbances, such as bark‐beetle impacts and high mortality rates.

After such disturbances with high tree mortality, NEE can be strongly impacted by decomposing deadwood (Russell et al. ²⁰¹⁵; Wijas et al. ²⁰²⁴) and increased mineralization processes in the soil (Goulden et al. ²⁰¹¹), which are caused by large gaps in the canopy and subsequent increased microbial activity and soil respiration (e.g. Dörr & Münnich ¹⁹⁸⁷; Curiel et al. ²⁰⁰⁷). Such processes would explain the observed increase in R_eco since 2019 (Table 1 and Fig. S2). Simultaneously, living biomass is reduced (e.g. by tree mortality) and requires time to reach pre‐disturbance levels (Goulden et al. ²⁰¹¹), which is confirmed by increasing GPP since 2019 (Table 1 and Fig. S2). From other studies, it is likely that the ecosystem, despite the shift in vegetation dominance, is developing again into a substantial carbon sink, once reaching a later successional state (Goulden et al. ²⁰¹¹; Gough et al. ²⁰²¹), which might even further enhance the carbon uptake capacity as the ecosystem is developing into a multi‐species forest with high structural complexity (Gough et al. ²⁰¹⁹). However, the development of ecosystem carbon fluxes is strongly dependent on the predominant environmental conditions, which have strongly impacted the ecosystem after 2018.

Compound events of heat and drought evoke strong ecosystem carbon release

Both atmospheric and edaphic droughts, as stand‐alone stressors, can have significant negative impacts on ecosystem processes, such as water flux or carbon assimilation. Both stressors lead to stomatal closure of plants, if certain thresholds are crossed (e.g. Johnson & Ferrell ¹⁹⁸³; Irvine et al. ¹⁹⁹⁸; Carminati & Javaux, ²⁰²⁰) to reduce transpirational water losses and reduce the risk of hydraulic failure (McDowell et al. ²⁰⁰⁸; Grossiord et al. ²⁰²⁰). Not only is photosynthesis reduced, but under persistent conditions of atmospheric and edaphic drought, plant respiration processes are enhanced (Teskey et al. ²⁰¹⁵; Scafaro et al. ²⁰²¹). Furthermore, the area of photosynthetic active leaf material is reduced due to premature leaf and needle shedding (Brun et al. ²⁰²⁰; Nadal‐Sala et al. ²⁰²¹), as occurred at the studied site in 2018 (Haberstroh et al. ²⁰²²) and most likely 2022 (Fig. 4A). The interaction of edaphic and atmospheric drought accelerates these negative impacts, as increased atmospheric water demand can initially increase transpirational water losses (McDowell et al. ²⁰⁰⁸; Will et al. ²⁰¹³), which in return causes faster depletion of available soil water resources, exacerbating the severity of the edaphic drought (Swann 2018; Zhou et al. ²⁰¹⁹; Grossiord et al. ²⁰²⁰). These hot‐dry events preferentially materialize and peak in late summer, where a negative impact on the C‐cycle has also been found in other ecosystems (e.g. Pohl et al. ²⁰²³). This is in line with reports from various ecosystems in Europe during 2018 and 2022, which suffered from severe reductions in NEE during hot‐dry events (Lindroth et al. ²⁰²⁰; Smith et al. ²⁰²⁰; Thompson et al. ²⁰²⁰; Van der Woude et al. ²⁰²³; Scapucci et al. ²⁰²⁴). These extreme conditions can even lead to net carbon losses during the growing season (Fig. 5A), which has rarely been observed before, but could become a regular phenomenon during future extreme climate events (Van der Woude et al. ²⁰²³).

As the DE‐Har ecosystem was severely affected by the hot drought of 2018 (and previous droughts; Haberstroh et al. ²⁰²²), it is difficult to disentangle long‐term legacy effects from current year effects (Müller & Bahn ²⁰²²), as it is challenging to determine whether the ecosystem has recovered from the previous extreme event before the occurrence of the next extreme event. For example, in other forest ecosystems there are reports of legacy effects of the 2018 hot drought on carbon fluxes in 2019, while NEE returned to pre‐2018 levels in 2020 (Pohl et al. ²⁰²³). At DE‐Har, 2019 was a year with high carbon losses (+229 ± 18 g C m⁻² year⁻¹), although meteorological conditions only deviated slightly from the long‐term average in the growing season. Thus, it is likely that the ecosystem still suffered from legacy effects of the year 2018 (Haberstroh et al. ²⁰²²). The delayed recovery of NEE in September and October 2022, despite refilled soil water resources (Fig. 7), similarly raises the question of legacy effects of the year 2022 in 2023. Future observations and longer time‐series of NEE will help to elucidate the legacy size and duration on ecosystem carbon fluxes at DE‐Har and other ecosystems. Furthermore, detailed studies of ecophysiological processes at the individual tree level, such as leaf gas exchange, sap flow and water potential dynamics of all major tree species growing at DE‐Har, will provide a better understanding of the mechanisms underlying the observed changes in ecosystem processes, such as the shift in species dominance and its impact on ecosystem carbon fluxes.

CONCLUSION

The impacts of the 2018 hot drought and the following years have had substantial negative effects at different scales on the P. sylvestris ecosystem at DE‐Har, which are most likely legacy effects of several recurrent hot drought events. This ecosystem is slowly shifting to a deciduous‐dominated forest, which is, however, still mostly a carbon source due to the frequent occurrence of compound droughts in recent years. If climatic extremes do occur at the same frequency as 2018–2023, this could markedly retard or even prevent ecosystem recovery, thus endangering more ecosystems in Central Europe that might turn into permanent carbon sources.

AUTHOR CONTRIBUTIONS

SH, AC and CW conceived the study. SH, AC, MS and FS collected and analyzed the data. SH, CW and AC interpreted the data. SH wrote the manuscript draft with inputs from AC, MS, FS and CW. All authors contributed critically to the drafts and gave final approval for publication.

Conflict of Interest

The authors declare no conflicts of interest.

Supporting information

Fig. S1. Illustration of the precipitation and eddy covariance measurement positions at the two towers at DE‐Har. Photo: University of Freiburg.

Fig. S2. Monthly gross primary productivity (GPP) (A) and ecosystem respiration (R_eco) (B) of the P. sylvestris ecosystem for the years 2019–2023.

Fig. S3. Relationship of monthly enhanced vegetation index (EVI) and monthly gross primary productivity (GPP) (A), net ecosystem carbon exchange (NEE) (B) and ecosystem respiration (R_eco) (C) of the P. sylvestris ecosystem for the years 2019–2023 assessed by linear regressions (lm).

Fig. S4. Relationship of monthly difference in net ecosystem exchange (NEE) of 2021 (Δ_NEE2021) to the years 2019, 2020, 2022, 2023 and monthly mean air temperature (T _air ) (A), monthly maximum vapour pressure deficit (VPD_max) (B), monthly precipitation sum (C) and monthly mean soil volumetric water content (VWC) (D). The relationship of NEE with T _air , VPD_max and VWC was assessed with piecewise two segment linear regressions with the function segmented (segmented, Fasola et al. ²⁰¹⁸).

Fig. S5. Monthly difference in net ecosystem exchange (NEE) of 2021 (Δ_NEE2021) to the years 2020, 2022, 2023, monthly mean air temperature (T _air ) and monthly mean volumetric soil water content. Larger circles indicate more positive Δ_NEE2021. Negative Δ_NEE2021 values are bordered in blue, positive Δ_NEE2021 bordered in red. Months are colour‐coded from 1 (January) to 12 (December).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The meteorological data and the flux data can be found on the ICOS data portal (https://www.icos‐cp.eu/data‐products/ecosystem‐release, site ID DE‐Har). Phenocam (e.g. Plapp et al. ^2024a) and long‐term meteorological data from DE‐Har (Christen et al. ^2024a) are available at https://zenodo.org/communities/de‐har/ in the community Hartheim forest research site. Time series of Phenocam pictures from both cameras can be accessed at https://phenocam.nau.edu/webcam/sites/hartheim1/ (29.6 m above ground, hartheim1) and https://phenocam.nau.edu/webcam/sites/hartheim2/ (8.4 m above ground, hartheim2).