Scots pine responses to drought investigated with eddy covariance and sap flow methods

Abstract

Scots pine, as one of the dominant European tree species in the temperate zone, is experiencing intensified water deficits, especially in north-western and central Poland, where it suffers from frequent droughts and generally low precipitation. This work investigates drought impact on forest functioning, by analysing ecosystem transpiration under normal as well as dry conditions. Therefore, eddy covariance (EC) and sap flow measurements (using the thermal heat balance, THB, method) were combined to estimate transpiration (T) in two different-aged Scots pine (Pinus sylvestris) stands in north-western Poland: Mezyk (ME; 26 years old) and Tuczno (TU; 67 years old). Transpiration (T) estimates regarding EC measurements were derived from gross primary productivity (GPP) fluxes and vapour pressure deficit (VPD) dependence, considering their common relationship with stomatal activity. In 2019, the year following severe drought in Poland and Europe in general, total annual transpiration estimated based on sap flow measurements (T_SF) was significantly lower than EC-derived transpiration (T_EC) at both sites. The total ratio of T_SF/T_EC for the growing season (March–August) was 0.64 and 0.41 at ME and TU, respectively. We thus speculate that the understory, which was more abundant in TU than in ME, and which could only be observed by the EC system, may be responsible for the observed discrepancies. Bigger differences between T_SF and T_EC occurred under dry and wet conditions, while both were fairly similar under moderate conditions. The analysis of the relationships between T_SF and soil water content (SWC) at depth of 10 cm revealed that there is a thresholds (SWC ~ 3.5%) at which T_SF starts to decrease sharply, presumably due to stomatal closure. However, the decrease in GPP fluxes at the same time was less pronounced, indicating the impact of additional non-stomatal factor on water conductivity. We generally conclude that care should be taken if the conclusion of the occurrence of drought stress of some plants is derived from a bulk evapotranspiration flux, as it is commonly done with EC measurements averaging over the whole ecosystem. Our results also support the notion that non-stomatal water losses are an important element during extreme dry conditions, and that these may appear not only when stomata are already closed.

Introduction

With intensified climate change caused by the growing human impact on the environment, as well as the increasing severity of droughts in Europe, it is important to address and understand the major role that forest ecosystems play in the global carbon and water cycling. Ecosystem disturbances originating from human activities (e.g. harvesting, land cover changes) and natural causes (e.g. drought, wildfires, windstorms) significantly affect hydrological processes at any ecosystem by impacting vegetation dynamics (Wang et al. 2018). It is commonly known that stomata play a dominant role in regulating the amount of water transpired (T) by vegetation (Jarvis and Mcnaughton 1986; Pieruschka et al. 2010). On a global scale, the transpiration fraction of evapotranspiration (ET) is estimated to be 56–74% (median = 65%, mean = 64%) (Good et al. 2015). It has been previously reported that T/ET varies among different types of vegetation which is though the highest in forests (Moran et al. 2009; Hu et al. 2018). It was also found that the average T/ET in temperate coniferous forests is 55 ± 15% s.d. (standard deviation), while for boreal forests it is ca. 65 ± 18% s.d. (Schlesinger and Jasechko 2014).

Understanding the relationship between T, ET and precipitation (P) allows proper planning and implementation of water management in terms of ecosystem efficiency and water protection (Aiken and Klocke 2012). However, estimating the transpiration flux at the ecosystem level is difficult due to inter-species differences as well as variability between trees of the same species. Therefore, a mechanistic understanding is needed to estimate the transpiration flux of individual trees. Also, specific environmental conditions can affect the strategy of plants to maintain the highest photosynthesis rate at the expense of the lowest water loss. It is particularly important to identify when/in which conditions such strategy begins to affect plants’ health, as this can eventually lead to tree mortality. It is projected that drought-related tree mortality will most probably increase in the future on the global scale, yet the knowledge required to accurately predict forest functioning under drought stress and the occurrence of tree mortality itself is still limited (Plaut et al. 2012; Trugman et al. 2021). Therefore, any research aiming at understanding water scarcity impact on forests functioning helps to improve the estimates and future predictions related to drought resistance.

The general impact of drought can be described as a strong influence on water and carbon fluxes, and consequently the productivity of all terrestrial ecosystems (Granier et al. 2007). For these reasons, thorough analysis of water cycling, its individual components and its interactions with the carbon cycle in ecosystems is crucial for understanding the factors controlling these processes. One of the most crucial links in water cycle of land ecosystems is the soil, because it is the first element of soil–tree–atmosphere continuum. Water deficit in the soil can thus affect several stages along that continuum substantially (Bréda et al. 2006). Besides, drought stress alters both soil–root and leaf–atmosphere exchange, which further threatens the integrity of the liquid phase continuum from soil to leaves. It has been shown on the example of urban trees that an increase in drought stress and susceptibility to drought can lead to the development of embolism, suggesting a stress-induced deterioration in hydraulic conductivity (Savi et al. 2015). Drought stress reduces the exchange of both water and CO₂ fluxes and consequently limits tree growth. It comes from the fact that water availability is one of the most important drivers for both processes (Levesque et al. 2017). It is assumed that under drought-related stress conditions, the rate of water and carbon exchange with the atmosphere should decrease. The suggested underlying mechanism of that reduction is the closing reaction of stomata which has evolved to protect against excessive dehydration and physiological damage to plants (Oren et al. 1999).

Partitioning ET into T and E is challenging, but can be done using various methods including: eddy covariance (EC) observations, chamber measurements, isotopes analysis, lysimeters and sap flow systems (Kool et al. 2014; Sulman et al. 2016; Anderson et al. 2017; Rafi et al. 2019). The ET partitioning utilizing EC point measurements, which are nowadays available from numerous sites, can be done indirectly by the use of empirical relationships or estimated from above- and below-canopy systems working simultaneously (Paul-Limoges et al. 2020). Nevertheless, at many flux tower sites the measurement setup is located only above the tree canopy.

In this work, in addition to direct sap flow measurements at several trees, the transpiration of the entire forest ecosystems was obtained by partitioning ET fluxes (T_EC) derived from one above-canopy EC system, using the relationship between gross primary productivity (GPP) and vapour pressure deficit (VPD) (after Berkelhammer et al. 2016). It is well known that the higher the VPD, the higher the transpiration at the leaf level (Yong et al. 1997; Broughton et al. 2021). Simultaneously, stomata regulates transpiration via adjusting closing/opening range to avoid a decrease in water potential (Oren et al. 1999). The interplay of these are described by stomatal optimization theories that maximize carbon gain for a given amount of water loss in the rooting system per unit leaf area. The assumption of optimizing stomatal control makes T proportional to GPPxVPD^0.5 (Katul et al. 2009). In addition, the method assumes that the T can be approximated by ET during ideal conditions for transpiration (Stoy et al. 2019). The T = ET line is determined from half-hour eddy covariance data by regressing the minimum ET values defined as the 5th percentile of ET values in each GPPxVPD^0.5 bin. A linear regression of these minimum ET values defines the T = ET line, and any point below this line has T/ET equal to 1 by definition. Due to stochastic nature of half-hour values of eddy covariance, some data points are below the T = ET line (Berkelhammer et al. 2016). Moreover, many water use efficiency (WUE)-derived methods for ET partitioning are based on these same assumptions (Berkelhammer et al. 2016; Zhou et al. 2016; Boese et al. 2017; Stoy et al. 2019).

In this paper, we estimated ET and T fluxes in two managed Scots pine stands of different age classes (young and mature) located in north-western Poland by combining EC and sap flow measurements during the year 2019, with the main focus on the influence of drought conditions. We thus follow the rationale of previous studies (Zhou et al. 2016; Nelson et al. 2018) that focus on T/ET mainly for across-site comparisons and land surface model development, suggesting that T/ET ratio is likely more useful than absolute T and ET values. Since sap flow measurements were only available in 2019, comparisons of both water fluxes (T_SF and T_EC) were only available over such a short term. Reliable estimates of T, especially during dry conditions, are very important for recognition and diagnosis of forest ecosystem status. In general, it is difficult to identify the ideal reference method because each has its advantages and disadvantages. Our dataset allowed for an insight into ecosystem as well as tree-scale transpiration utilizing robust upscaling method (based on 20 sensors per site). Moreover, the investigated areas were recently exposed to prolonged and repetitive droughts. In general, the entire region is one of the poorest in Europe regarding freshwater supplies with a large fraction of a Scots pine forests being managed. Thus, the investigation of how Scots pine trees are adjusting to these harsh conditions, especially in relation to tree hydraulics coupled with carbon exchange, is important at the national scale.

The main aim of this study is to determine differences between the stand transpiration (T) calculated using the EC-derived water vapour fluxes and upscaled sap flow data. Specifically, we determine the suitability of the two methods to detect changes in T in drought conditions. The purpose of using two different-aged ecosystems with a similar species composition and soil conditions was to explore whether the established relationships are age- and stand-structure dependent. The hypothesis is that T estimated from the upscaled sap flow measurements (T_SF) is technically and methodologically more sensitive to changes caused by drought compared to EC-derived transpiration (T_EC) obtained from directly measured ecosystem-scale evapotranspiration (ET) and is thus better suited to detect drought-induced effects on trees. On the other hand, the relationship between low soil water content reflecting drought conditions and an increase in vapour pressure deficit (e.g. Grossiord et al. 2020) should result in T_EC to be sensitive to drought similarly to T_SF obtained at tree scale. Low values of soil water content (SWC) were used as an indicator of drought on a daily scale for each site individually. It was also investigated how GPP changes with soil water conditions, since it is included in T_EC calculation and coupled with T_SF via stomata activity. Finally, we discussed the possible role of understory (shrubs and other hardwood species) transpiration in its total stand value, which is only included in T_EC, not in T_SF. Thus, the discrepancies between T_SF and T_EC were used as a rough estimate of understory transpiration.

Material and methods

Site description

This research was conducted at two EC study sites. The first site, Mezyk (ME), is located in north-western Poland (52°50'N 16°15'E). This 26-year-old forest was established after a fire in 1992 and is currently dominated by Scots pine (Pinus sylvestris) and approximately 1% admixture of silver birch (Betula pendula). The understory (i.e. other hardwood species and shrubs) is very poorly developed, mainly due to the young age of the ecosystem and the fact that it was established in a postfire area. It has been estimated that within the 1-hectare area, other forest ecosystem elements such as dead wood (both above- and belowground), herbaceous plants, bryophytes and woody plants up to 0.5 m high together constitute approximately 6.3% of the total dry biomass. The estimated tree density within the 500 m footprint of the EC system is about 4260 trees per hectare. The average tree height is approximately 10 m, and the average diameter at the breast height (DBH) is 9.5 cm. The organic layer was about 2 cm; along the top soil layer up to 2 m where profile measurements were performed, the dominating granulometric soil type was loose sand. The second site, Tuczno (TU), is located 50 km north of ME (53°11’N, 16°05’E) in a mature, 67-year-old Scots pine forest. The main tree species composition at TU is very similar to that at ME–Scots pine accounts for 99% of the tree species composition, with an admixture of birch (Betula pendula Roth). The understory is dominated by beech (Fagus sylvatica L.) and hornbeam (Carpinus betulus L.) (Ziemblińska et al. 2016a). Other abundant species of the forest floor include sea buckthorn (Pleurozium schreberi), fern (Dryopteris carthusiana), red raspberry (Rubus idaeus) and two grass species (Deschampsia flexuosa and Calamagrostis epigejos). The approximate tree density is 600 trees per hectare, and the average height and DBH are approximately 28.5 m and 31.2 cm, respectively. The soil profile was similar to ME, where loose sand is the main soil type up to 2 m, with the organic layer reaching 4 cm, composed mainly of freshly fallen pine needles and branches. According to the classification in the World Reference Base for Soil Resources (2006), the soil at both study areas is classified as a Brunic Arenosol (Dystric; (WRB-IUSS 2007). Similar soil properties at both sites allowed for a reliable comparison considering soil water conditions. At both sites the soil was highly permeable, as evidenced by the grain size composition and fluvioglacial origins (related to the North Polish Glaciations). The soil fluvioglacial origin and a thick layer of permeable sands caused the groundwater to be deep (> 10 m). Basic characteristics of the investigated sites are summarized in Table 1.

Table 1 Stand characteristics for Tuczno (TU) and Mezyk (ME) Scots pine forest sites (as for the year 2019)

Eddy covariance (EC) and sap flow measurements

Eddy covariance and meteorological measurements

Measurements of the CO₂ and H₂O exchange between forest and the atmosphere have been carried out at ME by an EC system installed at the top of a 22-m-high flux tower since May 2018. The EC system includes a three-dimensional (3D) sonic anemometer (model WindMaster Pro, Gill Instruments, Hampshire, UK) and an open-path infrared gas analyser IRGA (model 7500DS, LI-COR Inc., Nebraska, USA). Continuous meteorological measurements were simultaneously performed, including radiation balance components, air temperature (Ta), air humidity, precipitation (P) as well as soil water content (SWC) and temperature. Both soil temperature and volumetric water content were measured at a depth of 10 cm in the mineral soil using Digital TDT sensors (Acclima, USA). All measurements were averaged to 30 min. The data collected from the uppermost layer were used as a proxy for sub-daily and daily soil water status as well as drought stress intercomparison between the two stands, as the sites have the same soil type to at least 2 m depth. It does not represent the all water available for trees at deeper layers. The same measurements were performed at TU at a 38-m-high steel scaffold flux tower (2 m × 2 m) since 2008. The EC system comprised a 3D non-orthogonal sonic anemometer (model CSAT3, Campbell Scientific Inc. (CSI), UK) and an IRGA (model LI-7500, LI-COR Inc., Nebraska, USA). The instruments were positioned 39 m above the ground until January 2019 and then moved to 44 m due to tree growth. A detailed description of the TU station’s meteorological equipment can be found in the study of (Ziemblińska et al. 2016b). At the TU site, temperature probes (model T-107, CSI, USA) were installed at 2, 5, 10, 30 and 50 cm below the soil surface. Volumetric water content was measured directly beside the temperature sensors at depths of 10, 30 and 50 cm using CS-616 sensors (CSI, USA). The measurements of SWC in two separate profiles show that the values obtained at 10 cm depth have the greatest diurnal variability in response to changing meteorological conditions and drought occurrence (Fig s2. supplementary materials). Moreover, this supports the argument that although water deficit exists in the shallow soil layer (especial during April and June), more water is still available in the deeper soil layers. Also, because measurements were performed only down to 130 cm depth, most probably pine trees could still benefit from water below that threshold. The magnitude of the SWC measured between 10 and 130 cm in profile 2 was 5%, and thus, the variability of the SWC between different soil levels was not large. Since there were technical problems with the rain gauge at both sites, daily precipitation records were replaced by those collected at the Mialy precipitation station (ca. 2 km from the ME site) and Człopa precipitation station (ca. 10 km from TU) of the National Institute of Meteorology and Water Management (IMGW-PIB 2021). We used EC and meteorological measurements at the ME site from 2019 only, while the measurements from TU were used from the data range between 2012 and 2019.

Meteorological conditions and climate classification

In reference to the widely used Köppen–Geiger climate classification, the climate at the research area was characterized as Cfb (warm temperate, fully humid with warm summer (Kottek et al. 2006). Monthly precipitation totals measured at the Człopa meteorological station, averaged for the period 2012–2019 station (Fig. s1, Supplementary materials), show characteristic annual course—with low values especially in spring (average ~  < 40 mm), and the highest precipitation in summer (> 80 mm in July) and most of the winter months (December, January > 60 mm). Average annual precipitation reached almost 700 mm, with a maximum in 2017 when the annual P sum was 1075 mm (Table 2). In 2018, the annual P sum reached the lowest value of all investigated years. As for the thermal conditions, the highest average monthly temperature was measured in July and August (> 18 °C), while the lowest in January (monthly average < − 2 °C) (Fig. s1, Supplementary materials). In 2018 and 2019, the annual average air temperature reached the highest values among all eight analysed years. During the investigated period, significant droughts occurred not only in Poland but also in other European countries, particularly in 2015, 2018 and 2019 (Ionita et al. 2017; Boergens et al. 2020). Consequently, soil water content in 2018 and 2019 was the lowest compared to the previous years, except for the very dry year of 2015.

Table 2 Annual means/totals of basic meteorological elements and soil water content for TU site during the period 2012–2019. P—annual precipitation sum, ET—annual evapotranspiration sum, SWC—mean soil water content, T_a—mean air temperature, T_soil—mean soil temperature

As for ME site in 2019, when measurements at both sites were performed simultaneously, some meteorological conditions, such as T_a and global solar radiation (R_g), were fairly similar to those observed at TU on an annual scale, since they are both located in the same climatic region. However, there were noticeable differences regarding P and ET annual totals which were lower in ME than for TU, reaching 449 and 563 mm, respectively. Average annual daily mean value of SWC was 7.6% at ME and 6.7% at TU, while the average annual T_soil at the 10 cm depth was 9.7 and 11.3 °C, respectively.

In order to detect and compare drought conditions in the investigated areas, the Standardised Precipitation and Evapotranspiration Index (SPEI) over a 3-month period was calculated (Vicente-Serrano et al. 2010). Long-term data used to calculate historical background values were retrieved from the Global Drought Monitor database (Beguería et al. 2020). It is assumed that values of SPEI below 0 indicate transition to drier conditions, values below − 1.5 indicate severe drought, and those below − 2 reflect extreme drought. The index together with SWC at the uppermost soil depth is used to characterize the drought conditions (Fig. s6. supplementary materials). Daily precipitation and SWC variations recorded in 2019 as well as SPEI values are compared for both sites in the Results section.

EC data processing

The forest within the flux footprint of the towers can be considered homogeneous from the perspective of aerodynamic properties. At the both sites the 80% flux footprint extends c. 500 m away from the tower (Kormann and Meixner 2001). Although the measurement height is higher at TU, the footprint is the same as in the ME site because the TU forest is also taller (aerodynamic displacement height is higher). Due to the asymmetric type of anemometer used at TU, the data recorded from wind directions that were disturbed by the instrument construction were omitted, resulting in the exclusion of 27% of data over the four years. The raw data obtained from the EC system were used to calculate covariances of the vertical wind speed (w) component and the quantity of interest: Ta, water vapour (H₂O) and carbon dioxide (CO₂), to finally derive turbulent sensible heat (H), water vapour (which is ET) and CO₂ (F_C) fluxes. The rate of change in CO₂ and ET (ΔS_CO2, ΔS_ET) in the air column below the EC system (i.e. the storage flux) was calculated during the first step of the data processing in EddyPro software version 7.0.1 (LI‑COR; Inc. 2019). Storage fluxes are commonly obtained as a product of the estimation of air density, height of EC measurement, and the rate of CO₂ or water vapour concentration change at EC measurement height (using the difference between the average concentration at the beginning and end of the corresponding 30 min single-point EC measurement period). However, it has been shown that this storage flux calculation yields similar results to those obtained with the profile approach used in similarly tall vegetation (Morgenstern et al. 2004). Both ΔS_CO2 and ΔS_ET values were accounted for in the final fluxes calculations.

During data processing, the following corrections were applied: the double rotation calculation, tilt correction, WPL correction, covariance maximization and analytic correction of high-pass and low-pass filtering effects (Webb et al. 1980; Moncrieff et al. 1997, 2004; Burba et al. 2012). Final datasets of carbon and water fluxes include quality control and gap-filling procedures. Additionally, all fluxes recorded during precipitation events at both sites were excluded. To filter out fluxes measured under insufficient turbulence conditions, friction velocity (u*) threshold (u ∗ th) was used, as obtained from the breakpoint detection method (Barr et al. 2013; Wutzler et al. 2018), performed with the ReddyPro package (Wutzler et al. 2018) in R software (R Core Team 2020). The gap filling of carbon and water fluxes as well as NEE fluxes partitioning into GPP and ecosystem respiration (R_eco) was also performed using the ReddyPro package. The partitioning was done following the Reichstein night-time approach (Körner 1995; Falge et al. 2001; Gilmanov et al. 2003; Reichstein et al. 2005; Lasslop et al. 2010). Ecosystem respiration was estimated according to Lloyd and Taylor’s (1994) regression model, to fit the ecosystem respiration (R_eco) as a function of soil (Ts) or air temperature (Ta) (Lloyd and Taylor 1994). In this work, the soil temperature was used in the partitioning procedure.

The gross primary productivity GPP is estimated by:

$$ {\text{GPP }} = R_{{{\text{eco}}}} \; - {\text{NEE}} $$

(1)

where R_eco is the total ecosystem respiration and NEE is the net ecosystem exchange.

Sap flow measurements

At both ME and TU site, three test plots with an area of 1080 m² were chosen randomly within the 20–70% flux footprint domain, mainly to the west of the tower (within approximately 100 m from the tower, Fig. 1), since this was the prevailing wind direction at both sites. At each site, 25 “model trees” were selected (Fig. 1), all equipped with sap flow sensors. Sap flow has been measured at both sites since October 2018. The EMS 81 sap flow system used here consists of sensors and SDI12 modules produced by EMS Brno (Czech Republic). Each sap flow sensor (SF 81) consists of stainless-steel electrodes inserted into the slots, which assure that sap flow values are nearly independent on the radial profile of sap flow density (Kučera 2018). The greatest advantage of this approach is the direct heating of a relatively large volume of xylem, which ensures high accuracy of the sap flow estimation. High electrical resistance of some trees’ tissue (not related to their age) caused poor quality of measurements in six samples at both sites (mainly due to internal stem damages (drying, cracks cavitations, etc.); these were excluded. Also, the measurements for the thinnest trees in ME had low quality and were thus excluded from the further analysis—the instruments used here have been designed for trees with a circumference greater than 40 cm. One sap flow sensor per tree was installed (north side) at DBH level (Fig. 2).

Fig. 1

Location of the three selected test plots (1080 m2; yellow rectangles) within the footprint of the EC system (calculated for unstable conditions by the approach described by (Kormann and Meixner 2001)). Diameter and number of trees were directly measured at ME (left) and TU (right) sites. Green circles indicate a group of 5 sap flow sensors; red dots represent the location of the EC flux towers. Colour scale represents the areal contribution of the flux (in %) from the footprint area. (Color figure online)

Fig. 2

Tree diameter at breast height (DBH) distributions at ME and TU sites measured in 2019. ME—grey bars; TU—black bars. (Color figure online)

Due to difficulties of sap flow and eddy covariance observations during winter (Foken et al. 2012; Kittler et al. 2017; Frank and Massman 2020), transpiration data were only analysed during the growing season (from March to the end of August; in the case of ME, to the end of September, because there was no downtime in the operation of the equipment). The gaps in transpiration were filled with the MDS (marginal distribution sampling) method using ReddyPro.

There are several techniques for sap flow measurements, including the tissue heat balance (THB) method (Čermak et al. 1973; Čermák and Deml 1974; Čermák et al. 1976, 2004; Kučera 1977; Kucera et al. 1977). The “balance family” of sap flow measurements (including mainly the THB method) are the only methods that directly measure sap flow rate (Flo et al. 2019). In the THB method, thermal balance is calculated for a given heated space. According to its assumptions, the input energy is divided between conductive heat losses and warming of the flowing water (Čermák et al. 2004). The amount of water passing through the measuring point in the stem is calculated based on the electric power input consumption and the increase in temperature of water flowing through the area that is heated (Kučera and Urban 2012)$.$ Whole tree sap flow (Q_tree−kg time⁻¹) was derived by multiplying Q by the tree circumference, after the bark and phloem layer was excluded (Szatniewska et al. 2022). The phloem layer was measured during the equipment installation.

As mentioned earlier, thermal methods for measuring sap flow are based on heat loss detected by the sensor. Therefore, when tree sap flow data are processed, the zero transfer value (“baseline”) must be determined, which varies depending on the sensor and the time of year (Oishi et al. 2016). When performing this procedure, the possibility of nocturnal sap flow appearance should also be taken into consideration, since night sap flux is thought to be driven by atmospheric evaporative demand (Forster 2014). In our work, raw sap flow data were post-processed by applying the automated baseline subtraction based on the Exponential Feedback Weighting method (Kučera et al. 2020) developed for mini32 software (EMS 2020a). This method considers the nocturnal sap flow and removes outliers due to natural temperature gradients. It has been found that the best results were computed by the 5-day weighting average (Kučera et al. 2020), and thus, the same averaging interval was applied here.

Upscaling model tree sap flow values to the stand level

We applied the diameter class technique for sap flow results upscaling, which is based on the assumption that sap flow rates depends only on the DBH (Čermák and Kučera 1987, 1990) (Fig. 3b).

Fig. 3

Relationship between daily averages of sap flow rates (Q [kg/h]) and diameter at the breast height (DBH) used for upscaling the tree-level transpiration to stand-level transpiration (TSF)

For the purpose of upscaling sap flow from the tree to stand level, a histogram of the DBH distribution was created for the sampled areas (Fig. 1), then proportionally upscaled to 1 ha (Fig. 2). The reliability of this procedure is supported by the homogeneity of both forests’ structure.

An exponential function was fitted to the relationship between sap flow rates and DBH (Cristiano et al. 2015). Differences in physiology of 27-year-old and 67-year-old trees raised doubts about using one common relationship for both sites, since trees of the similar size but different age can have different features related to water transport. Thus, upscaling was done with parameters specified separately for each ecosystem. Specifically, the Q_tree ~ DBH original exponential relationships were fitted for a 7-day window from 05.05.2019 to 15.05.2019, as the average daily sap flow quickly reached high values in May, with the highest observed in June. It is suggested that regression parameters for such relationships should be calculated over a few days when soil water content is roughly constant and VPD is relatively high—conditions that characterize high evaporative demand.

Average daily sap flow (kg h⁻¹tree⁻¹) for individual DBH classes was first calculated based on regression parameters obtained from the exponential relationship for sample trees between these two factors (Fig. 3). Then, the values of the daily average tree sap flow [kg/h] for individual DBH classes were multiplied by the number of trees in a given class and total average daily sap flow of all DBH classes were aggregated for the unit stand area (1 ha). Total daily sap flow per ha was finally divided by the sum of the average daily sap flow of the sample trees in the same period—this ensured that despite different maximum sap flow rates during different periods, the obtained sap flow rescaling factor from the sample trees to the stand transpiration (SC) was very similar regardless the calculation period.

The final stand transpiration expressed in kg m⁻² (or mm m⁻²) was obtained with the following calculation:

$$\mathrm{TSF} =\frac{\sum_{i=1}^{n}({Q}_{{tree)}_{i}} x Fs }{A}$$

(3)

where A (m²) is stand area, Q_tree is sap flow rate (kg/h), and Fs is the scaling factor. If the investigated stand is not homogeneous, different species must be analysed separately, similar to the situation when different canopy layers occur.

ET Partitioning

The method used for EC-derived ET partitioning was based on the assumption that ET is linearly related to GPP × VPD^0.5 only when T is the dominant term in ET. To estimate the T/ET value for each 30 min period, the product of GPP × VPD^0.5 was plotted against ET (for 2019 at the ME site; for 2012–2019 at the TU site; Fig. s3—supplementary materials), and the minimum value of ET was then selected as the 5th percentile for each equal-sized GPP × VPD^0.5 bin (after Berkelhammer et al. 2016). Fifty bins of equal numbers of points were implemented for both sites. The number of points in each bin differed between sites due to the larger dataset in TU. The linear regression determined for these bins gives the ET value for which T = ET. Similarly, for the points below the regression line, it was also assumed that T = ET (Räsänen et al. 2022). ET partitioning was done for EC datasets after excluding night-time data, as well as observations during precipitation events and when relative humidity was greater than 80%.

To determine the water loss per carbon gain for the forest stand, we calculated the daily water use efficiency (WUE) as GPP/T_SF (unit: g C/(kg H₂O m² per day). This way a better understanding of water use efficiency of Scoots pine trees, which are the main scope here, can be achieved than with the use of ET flux that includes also other components like evaporation from the soil and interception which were not directly measured. Since both forests are composed in 99% of one tree species for which transpiration was upscaled from sap flow measurements, thus it was assumed that T_SF should represent stand-level tree transpiration fairly well. We also assumed here that GPP represents only dominant pine trees since the contribution of other minor tree species—mainly birches, understory and soil vegetation, is negligible. Additionally, we calculated the canopy stomatal conductance from the upscaled foliage transpiration T_SF (in kg/m² s) by diving each estimate by corresponding VPD value converted to units kg/m³.

Results

Annual relationships between the components of the water balance at Tuczno 2012–2019

The data collected during 2012–2019 period at the Tuczno site were used to investigate what are the interannual differences between the annual totals and proportions of water balance components (Table 3). These components and their interrelation were then calculated for the 2019 growing season when data from both measurement methods were available for both sites at a higher time resolution. Annual T_EC total was the highest in 2012 (421 mm) and 2013 (407), when annual P sum > 700 m. In the wettest year of 2017, T_EC constituted only 33% of the annual P sum, while in the driest 2018 and 2019, it was noticeably higher (61% and 60%, respectively). Even though the lowest annual cumulative ET was recorded in 2018, the calculated T_EC was the lowest in 2015. Still, the ratio of T_EC/ET was the lowest for dry 2019 year which was only 0.65.

Table 3 Annual EC-derived: transpiration total (T_EC) and the ratios of evapotranspiration to precipitation (ET/P), T_EC/P and T_EC/ET at the TU site during 2012–2019

Drought conditions in 2019 indicated by SPEI, daily sums of precipitation and mean daily SWC

We compared T derived with the two different methods throughout the growing season of 2019, when a full set of measurements (sap flow and EC) were available at both study sites. The growing season T_EC total at ME was 304 mm, and its share in precipitation was 60%, similar to the value calculated at TU. In order to compare water conditions at the two investigated sites during the growing season of 2019, daily precipitation and soil moisture patterns are presented in Fig. 4. Slight differences related to precipitation distribution reflect the local conditions characterizing the TU site, where more frequent thunderstorms during the summer season occurred—these result in strong, local, convective precipitation in summer. In terms of meteorological drought, two periods (April and June) were identified at both sites, due to the prolonged rainless/dry periods and the increase in or persistently high temperatures. These findings were confirmed by the results derived from drought monitoring and analysis based on SPEI index (Fig. 4). For the grids covering the area of the two investigated sites, April, June and August 2019 were identified as severe (SPEI < − 1.5 and > − 2) or extremely dry (SPEI <  = − 2) depending on the site (Fig. 4). Precipitation total in March was higher than in April, when drought has occurred. We assumed that trees benefited from this water supply which helped maintaining their functioning (including transpiration) in the spring.

Fig. 4

Daily precipitation totals (P) and mean daily soil water content (SWC—depth of 10 cm) at ME a and TU b during the growing season of 2019, on the background of monthly values of the SPEI index. SPEI below -2 (marked with a black horizontal line) indicates extreme drought conditions. (Color figure online)

Variability of water fluxes at young (ME) and mature (TU) scots pine forest

To test the hypothesis that T estimated from the upscaled sap flow measurements (T_SF) is better suited to detect drought-induced effects than EC-derived T, the results of both methods were compared under different precipitation conditions, including extreme drought. During the growing season of 2019, the cumulative sum of T_SF was higher at TU (137 mm) than at ME (189 mm, Fig. 5), and total growing season T_EC amounted to 222 mm and 283 mm for ME and TU, respectively. Apart from the differences between transpiration fluxes derived with the two methods, ET sums from EC measurements also differed substantially between sites (287 mm at ME and 389 mm at TU). The calculations of the share of the T_EC sum from March to August (assumed growing season period) in the respective ET totals resulted in 0.77 for ME and 0.73 for TU. However, the T_SF/ET ratio was much lower and accounted for only 49% of ET flux at both the young and mature Scots pine stands (Fig. 4). Overall, the T_SF/T_EC was similar between sites: 0.60 and 0.65 for ME and TU, respectively. Moreover, despite the fact that precipitation totals for this period were almost equal for both sites, the total T_SF/P relationship was different; it reached 0.74 for ME and 0.95 for TU. Interestingly, the T_EC sum in both cases exceeded the precipitation sum during the growing season in 2019: at ME, the T_EC/P reached 1.20, while at TU it was ca. 1.43. For both sites during this period, cumulative ET strongly exceeded the value of total precipitation (Fig. 5).

Fig. 5

Cumulative components of water balance (WBC) during the growing season of 2019 at ME a and TU b. Transpiration from sap flow (TSF), transpiration from ET partitioning (TEC), evapotranspiration (ET), evaporation from partitioning (EEC = ET-TEC) and precipitation sum (P). TSF/ET, TEC/ET and TSF/TEC ratios represent total values for the growing season. Changes of daily mean SWC values are presented on the right y axis

In order to further explore the relationships for the entire growing season, monthly totals of transpiration (T_EC and T_SF) for both sites are shown together with precipitation during the growing season of 2019 (Fig. 6). Despite the differences in total sums of T_SF and T_EC for the growing season, there are similar patterns regarding high and low monthly sums at ME and TU, for both T fluxes. In particular, the estimated T_EC was higher than the T_SF in all months except April at TU. In May, the highest precipitation total was recorded at both sites: 58 mm at ME and 69 mm at TU. The lowest precipitation sums—ca. 10 mm or less—at both sites were recorded in April and June. In July, T_SF reached one of the lowest values at ME and TU. Also, at ME site, the sap flow sum reached its lowest value in August, in contrast to the TU site. Highest monthly T_SF was observed in June and May for ME and TU, respectively. T_EC was the highest for both sites in June.

Fig. 6

Monthly sums of EC-derived evapotranspiration (ET), transpiration from sap flow (TSF), transpiration from ET partitioning (TEC) and precipitation (P), as well as the ratios of TEC/P, TSF/P, TSF/ET and TSF/TEC for the growing season of 2019 at ME a and c, and TU b and d, respectively

It was also shown that in specific months, precipitation sums exceeded both T_EC and T_SF similarly at both sites. T_EC exceeded precipitation in almost all months, excluding March and May when P sums were higher than transpiration. T_SF was higher than monthly P only in April and June, when drought was detected at both sites by SPEI index. At ME, the share of T_SF in ET was the highest in May, when it reached 0.70, with very similar values in the preceding and following month. At this site, T_SF/ET was the lowest in the summer months of July and August (below 0.20; Fig. 6c). At the mature pine stand (TU), the highest T_SF/ET was detected in April (0.72) and May (0.66); the lowest (ca 0.12) was observed in March (Fig. 6c, d). The highest T_SF/T_EC ratio was calculated in April and May at TU (Fig. 6d), while at ME the highest was in May (Fig. 6c)—it reached 0.99, which means that T_SF and T_EC sums were nearly identical in this month. The lowest values of monthly T_SF/T_EC at ME were found in July and August, while at TU it occurred in July when T_SF/T_EC was less than 0.25.

Comparison of daily transpiration fluxes estimated by sap flow (TSF) and eddy covariance (TEC) methods

Since soil water content measurements were used to indicate drought occurrence on a daily time scale, transpiration fluxes were related to changing SWC conditions. The estimates of daily sums of T_EC were compared to T_SF for the 2019 growing season for each site separately. When summer data with SWC < 3.5% were excluded, which was the case especially in June and August for ME and July for TU, an additional correlation between the two fluxes was derived, to exclude the influence of days with drought related to deficits of water in the soil. This was represented by an increase in R² to 0.73 and 0.80 at ME and TU, respectively (Fig. 7a, b). Data with precipitation events have been excluded in this analysis. The regression line slopes of the relationships between daily values of these two transpiration fluxes over the growing season have been different for each site and approach (all data together vs low SWC days excluded) and ranged from 0.70 to 0.79 (Fig. 7a, b). T_SF values in July (when the T_SF/ET was also low) indicate a reduction in tree transpiration during this time. T_EC was not consistent with the T_SF values after prolonged drought (Fig. 7a, b), which was observed mainly by the relationship between T_SF ~ T_EC in June at TU, and in August for ME, when SWC reached the lowest values. It should be noted that transpiration after upscaling to the stand level from the sap flow measurements only corresponds to the transpiration from pine trees. The remaining flux from other plants in TU—young beech and hornbeam trees—was also in the range of the EC system and thus only visible in T_EC measurements, even though in principle both stands consist of 99% Scots pine trees (Table 1). In ME, both T_SF and T_EC were high under high SWC in May and June, and there was a good agreement between those T estimates. In TU, higher daily T_EC values were also observed in May, while the lowest was in April and March, similar to ME. However, lower SWC values—especially in August—were accompanied with higher T_SF and T_EC at TU than ME. The daily T_EC and T_SF courses during the growing season of 2019 are presented in the supplementary materials (Fig. s4, supplementary materials).

Fig. 7

The relationship between daily sums of transpiration derived from ET partitioning (T_EC) and transpiration from upscaled sap flow measurements (T_SF) at ME a and TU b during March–August 2019. Solid red lines represent linear regressions for all available water fluxes. Black solid lines represent simple linear regressions with the exclusion of data when SWC was < 3.5% in June and July for TU site and June–August for ME site. The colour of each point indicates corresponding mean daily SWC; numbers indicate individual months; dashed lines indicate the 1:1 regression line. (Color figure online)

The comparison of the responses of daily T_SF and T_EC fluxes to the daily mean SWC variations showed that both calculated T fluxes have similar dynamics at the same site (Fig. 8). The 9th decile of T sums daily averages was chosen as the threshold, above which values of T_SF and T_EC were considered high. The figures show that under similar SWC conditions, most of the values higher than the 9th decile for both T fluxes occurred similarly at each site. T_SF and T_EC reached the highest values, and T_SF/T_EC was closer to 1 for a SWC of 5–9% (ME) and 4–6% (TU) (Fig. 8c). Water use efficiency increases steeply below SWC values of 4%, since carbon uptake (GPP) was maintained at relatively high level (Fig. s5, supplementary materials), while the differences between T_SF and T_EC were distinct, especially at the ME site. The enhanced water use efficiency at low SWC (Fig. 8d) reflects the strategy of trees to minimize T while keeping GPP rate at a high level. When SWC was above 9%, lower transpiration and GPP values were observed. This is usually associated with evaporative demand reduction at the end or early beginning of the growing season together with lower VPD and lower radiation conditions. With low VPD values both GPP and T_SF decreased (Fig. 8e, f). At high values, a reduction in transpiration was observed, which was not as pronounced for GPP at ME site. At the TU site, corresponding reduction of both fluxes was observed at high VPD. Additionally, results from the ME site show the gradual increase in the conductance as a function of SWC (< 5.5%)—Fig s6. Supplementary materials.

Fig. 8

Relationship between SWC and TSF (orange) and TEC (purple) at a ME and b TU. c boxplot presenting the relationship of TSF/TEC at different SWC levels (bin width of 1.8% of SWC), at ME (red) and TU (blue). d Relationship between WUETSF and SWC at ME (red) and TU (blue) for SWC < 5.5%. e, f Relationship between daily sums of GPP, TSF and VPD is presented for ME and TU; light-shaded points represent all data of daily T and GPP totals; highlighted points and lines represent binned data; dashed horizontal lines a, b indicate the 9th decile (90%) of daily mean SWC at ME (orange) and TU (purple); vertical lines represent the intervals with daily values above the designated threshold for both sites. (Color figure online)

Discussion

Differences between water fluxes estimated for both sites

The investigated sites are located in a region that is exposed to water shortages and has experienced several rainfall anomalies in recent years (IMGiW 2020). For the studied 2019 growing season, total T_SF was higher at the mature stand (TU) than at the 26-year-old site (ME), which can mainly be explained by differences in the characteristics of the stand—larger trees and basal area at the TU site. Also, total T_EC at ME was lower than at TU, mainly due to the less understory at the ME site, as well as lower precipitation totals. Tree height, basal area and number of trees were the most distinguished differences between sites. The basal area was 1.5 times greater in Tuczno, where the sum of T_SF in the growing season was also greater by a factor of 1.4. As shown by Zimmermann et al. (2000), density and sapwood area in western Siberian pine were strongly correlated, which results from the exponential decrease in stem number with age (self-thinning) and the linear increase in sapwood area per tree (Zimmermann et al. 2000). Thus, transpiration was higher for stands with the higher sapwood area compared to the average transpiration of all other stands, by as much as a factor of 2. Growing season T_SF total at ME constituted 74% of precipitation, while at TU it was as much as 95%. In our study, the growing season T_SF constituted 48% of its annual value both at the ME and TU site. Despite the differences in absolute sums of ET and T_SF between sites, the tree transpiration has contributed similarly to ET at both sites. For T_EC, its share in ET was much higher than corresponding T_SF/ET and equal too: 77% and 73% at ME and TU, respectively. Since the decrease of T_EC (derived from the EC-derived ET fluxes partitioning method) during drought conditions was not as pronounced as for the stand-level T_SF, it seems that the EC method is not as sensitive to extreme conditions as the upscaled sap flow measurements at individual trees. Annually, T_EC/ET at the TU site varied throughout the study period (2012–2019), reaching the highest values in 2018—the year with the lowest annual precipitation sum. This is likely to be driven by high evaporative demand during days with sufficient water availability. The variability in transpiration rates origins from uncertain supply from soil layers deeper than 50 cm as well as transpiration from ground vegetation and tree species not considered in the investigation. Similarly, in a study that partitioned evapotranspiration derived from the EC method into transpiration and evaporation (Li et al. 2019), it was found that the T/ET and mean annual precipitation relationship was also weak and non-significant on an annual timescale (R² = 0.09, p value = 0.11). Another study based on data from sites that are part of the Fluxnet network, which only consider evergreen coniferous forests, showed that T/ET varies throughout the year, with a minimum in the winter and a maximum (reaching almost 0.8) in the summer (Nelson et al. 2020). In our study, it was found that the highest T_EC/ET occurred in spring: in May at both sites and additionally in April 2019 in the mature forest only. Estimated E at our sites (as a simple difference between measured ET and T_EC) was close to semi-arid conditions (Fig. 5) and has contributed to ET by 23% and 27% at ME and TU, respectively. It has been shown that soil E can reach up to 26% of ET under semi-arid conditions with a very water saving vegetation (Qubaja et al. 2020), while in alpine forest with ample water supply it was up to 40% (Wieser et al. 2018). Thus, despite the uncertainties related to the remaining water balance elements, not attributed to pine transpiration, our results indicate a close relation to the investigations under semi-arid conditions. The assessment of the share of transpiration in total ET is important for improving models of stomata behaviour, as well as understanding the mechanisms that control the amount of T in various environmental conditions, especially in ecosystems exposed to water deficits. We also found that even when the T_EC/P annually fluctuated between 0.50 and 0.60 over the years, T_EC/P calculated for the growing season only increased significantly in relation to its annual value. In the same year, considering the total for the growing season (March to August), values of T_EC/P exceeded 1, reaching 1.20 and 1.43 at ME and TU, respectively. For this period, the ET sum at both sites exceeded the value of precipitation reaching 287 mm at ME and 389 mm at TU, which was not the case for the annual ET and P totals calculated for 2012–2019 at TU. ET/P ranges annually from ca. 0.76 to 0.92 depending on the precipitation conditions, with the exception of 2017 (which was very wet), when it was only 0.51 (Table 2). The differences in the ET/P ratio for the growing season and the whole year indicate that the use and replenishment of water in the sandy soil is affected by the intra-annual precipitation pattern. ET exceeds P over the growing season whereas precipitation predominates over ET in autumn and winter. Although the summer-time ET is thus already high, it is predicted to increase further due to the increasing temperature and associated hydrological changes resulted from altering climate conditions (Okoniewska and Szuminska 2020). The high ET/P values for growing season indicate the dependence of transpiration on soil water storage. In the presented study, it was particularly noticeable in June 2019, when the SPEI index indicated extreme drought along with high transpiration, which contributed to reduced SWC over the same period. Transpiration which occurred in June under low SWC values indicates an increasing use of stem water storage. Subsequently, the low T_SF values in July were measured by the sap flow method at both sites independently, indicating a similar ecosystem reaction to changing environmental conditions—mainly low values of precipitation and SWC in the preceding month. Previous research has shown that for Scots Pine, relative stored water use ranged from less than 1% up to 44% of the daily transpiration (Verbeeck et al. 2007). After exceeding a certain site- and species-specific soil moisture threshold, plants attempt to optimize the carbon uptake against transpiration by stomatal regulation (Cowan 1978; Hari et al. 1986; Katul et al. 2000; Farquhar et al. 2002; Tang et al. 2006; Beer et al. 2009). If the water consumption of the plant exceeds soil water recharge, this can cause restrictions on plant water uptake, reduction of stomatal conductance (g_o), and lead to feedback on leaf-related processes and evaporative losses (Beer et al. 2009).

Our results also revealed that on a daily timescale, transpiration was the highest with non-extreme/moderate SWC values, for both T_SF and T_EC. Moderate values of daily average soil water content (6–8%) were found to be optimal for promoting higher T_SF and T_EC at ME. The highest differences between the two methods (T_SF and T_EC) occur under SWC values lower than 3.5 and 4% for ME and TU, respectively (Fig. 7). Similarly, after exceeding the above-mentioned moderate values, differences between T_SF and T_EC increased as the soil water content increased. Low soil moisture reduces T due to stomata closure and decrease in hydraulic conductivity (Duursma et al. 2008). High soil moisture is often associated with conditions of decreasing vapour pressure deficit and radiation during days with precipitation, which reduce evaporative demand, also causing reduction of T. Similar non-extreme values of soil moisture (~ 10%) have been found to be optimal for higher daily sap flow sums, for example, in high-elevation five-needle pines Nevertheless, it was suggested that the differences in species’ characteristics (like leaf area index, tree age and density, root characteristics) are possible factors underlying the different response of T to soil water content changes (Ji et al. 2016; Jiao et al. 2019). Moreover, there are some studies showing that finding a clear relationship between T and SWC is difficult for many forests (Kumagai et al. 2014; Ghimire et al. 2014; Jiao et al. 2019). One of the main reasons for this is the distribution of roots which are sensitive to heterogeneously distributed soil water both in time and space (Kume et al. 2007; Brito et al. 2015). Furthermore, it is difficult to separate the relative influence of soil water from other cross-correlating environmental factors (e.g. VPD).

The dependencies between available water for plants and high vapour pressure deficit (Grossiord et al. 2020) should cause a decline in T_EC, obtained at the ecosystem scale. It is thus sensitive to drought, similar to T_SF obtained by measurements performed at the tree scale. Some relationships regarding sap flow measurements, especially the decrease in tree transpiration (T_SF) at very low SWC values, were not as pronounced in results obtained for water fluxes derived from EC observations (Figs. 6, 7). Our findings concerning lower T_SF under drought stress conditions are consistent with the recognized role of stomata in protecting hydraulic function of the xylem by regulating transpiration rates in order to keep the water potential in this tissue below critical thresholds (Oren et al. 1999). If plants respond by reducing transpiration, ET is also reduced, and thus, soil moisture is conserved (Massmann et al. 2019). This is especially important during prolonged periods with no precipitation (as in April and June in this study). In principle, under such condition’s evaporation should be low, and the main part of ET should be transpiration which is likely to be reduced to prevent tree from excessive water loss. We also can assume that roots have access to deeper soil layers where water content was higher and which was not covered by measurements. Daily T_EC was generally higher than T_SF, especially when the SWC values were lower than 3.5% (Figs. 6, 7c). It should be emphasized that the soils had very similar properties at both sites (same soil type). The different dynamics of SWC may arise from a different interception caused by the specific structure and stand density at the sites, as well as the different thickness of the organic layer between sites.

Transpiration of understory trees and ground vegetation—which, for technical reasons, is neglected in the methodology of upscaled sap flow measurements, is one of the most possible explanations for the observed discrepancies between T_SF and T_EC. This applies especially to the case of the older, more structurally complex pine stand at TU, where beech and hornbeam saplings as well as some shrubs are present. Understory contributes to overall transpiration and can thus decrease the evaporative share of the dominant species’ transpiration. Research on the influence of understory removal on water, soil micro-climate, growth and physiology of dominant Scots pine trees showed that stomata closure occurred during dry periods in the control plot with retained understory, whereas the trees without understory were able to maintain a certain stomatal aperture—roughly 50% more open than the control trees (Giuggiola et al. 2018). However, T_EC was the closest to T_SF as long as there were appropriate conditions for high tree transpiration, i.e. stable water supply in soil (Fig. 7). Nevertheless, understory may not experience the same stress and can also have more anisohydric strategies than pines (e.g. birches). This may allow for relatively high understorey transpiration under dry conditions. Finally, the low T_SF contribution to measured ET may be influenced by non-stomatal loss of water, which becomes more important when stomata are closed but which may have a moderate significance in general. Thus, we conclude that drought stress of plants should not be derived solely with an integrated method such as EC. Our results also support the notion that non-stomatal water losses are important during extreme dry conditions.

Sources of uncertainties

After analysing the components of forest water balance with a focus on transpiration, some factors that influenced the obtained results were identified. We assumed that under stress conditions, direct upscaled sap flow measurements are technically and methodologically more sensitive to drought than the EC method. It has been shown that measuring sap flow in trees (sap fluxes) provides better estimates of water use at the stand level than those derived by other methods (Renninger and Schäfer 2012).

We chose to partition ET based on water use efficiency since flux partitioning does not require additional instrumentation aside from the commonly used eddy covariance, which has become the standard in many sites worldwide. Transpiration regulated by the stomata quickly responds to changes in VPD; therefore, T_EC was estimated at a resolution of every 30 min (Zhou et al. 2014, 2015). However, there are multiple sources of uncertainty in the estimation of the ET–GPP relationship itself (Sulman et al. 2016; Anderson et al. 2017). ET/VPD is a proxy for canopy conductance if the canopy is well coupled to the atmosphere, and if boundary layer resistance is low and leaf temperature is similar to air temperature (Beer et al. 2009). Another important aspect is the possibility of reliable estimation of the share of T in ET using the method based on water use efficiency ($ET \sim GPP*{VPD}^{0.5}$) in conditions of drought, as done here, since the accuracy of the linear relation during drought is unclear. The assumption whether T/ET approaches 1 during favourable conditions for transpiration has been reasonable assumption in ecosystem with higher LAI while rare in dry ecosystems (Stoy et al. 2019). Interestingly, previous research has shown that WUE values did not vary much under severe soil drought during summer in a boreal Scots pine forest in Finland (Gao et al. 2017). This would indicate that the transpiration of the boreal Scots pine forest was not disturbed by the drought event at the analysed site, although the stomatal conductance of plants decreased. T_EC based on ecosystem-scale ET measurements from the high tower includes information on not only main tree species (Scots pine) but also the understory, if present. As mentioned above, the thorough inventory of the studied Scots pine forest resulted in the conclusion that there is no significant understory in the younger forest (ME), while a much richer composition of small trees and shrubs are present in the older ecosystem (TU). However, we do not have quantitative information on the composition and biomass of other hardwood trees and shrubs at the TU site.

Potential sources of error for eddy covariance measurements may also result from the fact that the open-path gas analyser (OP) was used. Often, the OP system accuracy is compared to enclosed-path (CP) gas analysers. The different design of both instruments and data processing is a potential source of bias for ongoing global flux synthesis activities (Haslwanter et al. 2009). For water flux measurements, open-path systems minimize spectral attenuation and therefore also require smaller correction factors than sensors with an inlet tube (Polonik et al. 2019). Gas analysers with heated tubes have been found to be less effective (losses of H₂O) under conditions of relative humidity over 60%. On the other hand, in a study by Haslwanter et al. (2009), the CP system had a tendency to underestimate H₂O fluxes in comparison to the OP system during high air temperature conditions, high wind speed, large global radiation, sun angles and low relative humidity (Haslwanter et al. 2009). However, open-path analysers do not work sufficiently under sub-optimal conditions such as during rain, fog and exposure to aerosols and dirt and therefore must be cleaned frequently to avoid signal weakening (Polonik et al. 2019). There is an increasing amount of evidence suggesting that observations of trace gases fluxes from open-path analysers can be influenced by errors related to the heating of open-path instruments (Novick et al. 2013).

Overall, the greatest analytical difficulty was mainly related to upscaling the results obtained from the sap flow sensors to estimate stand transpiration within the area covered by the EC footprint. Upscaling the point measurements to the stand or catchment level remains challenging and causes some uncertainties (Rabbel et al. 2018). The obtained results were also affected by differences between the stands and the limits of the upscaling method. Utilized exponential relationship between DBH and Q has described the relationship between the average sap flow and DBH within the range of measured samples relatively well, as it was presented in the literature (Cristiano et al. 2015). However, the occurrence of trees with DBH out of the measured range would increase the uncertainty and result in greater bias, especially for younger ecosystem, where trees with DBH out of range may constitute a great share of all trees. It has been suggested that the discrepancies in upscaled ET flux from the sap flux could also result from the systematic underestimation of sap flux by the actual sensors (Wilson et al. 2001; thermal dissipation probes, operated at the constant power principle (Granier 1987)). Therefore, more research is needed to explore the possibility of accurately upscaling tree- or leaf-level T estimates before confidently including the influence of species composition at the stand- or catchment-scale T (Ford et al. 2007). It is particularly demanding to accurately recognize the number of trees “seen” by the EC system (accurate footprint) that represent the characteristics of the overall research area. In this study, we measured the number of trees and their DBH only on a few sample plots; therefore, our measurements might not reliably represent the actual stand tree distribution.

Additionally, the wounding effect is a known issue that can cause sap flux measurements to become unreliable. According to Maranon-Jimenez et al. (2018), the wounding effect showed a progressive development up to 22 weeks after sensor installation in living stems(Marañón-Jiménez et al. 2018). The same research also revealed that faster and more efficient wound compartmentalization occurs in early growing season installation, when physiological mechanisms involved in defence and prevention of pathogen infections are very active. Higher temperatures can also enhance this effect. In our experiment, we installed sensors in late autumn 2018, to generate the most reliable measurements for the upcoming growing season. According to the literature, the wounding effect develops with time. As such, it is recommended to change the sensors position in the stem after every growing season. Point measurements like the heat dissipation method (HD) are more prone to the wounding effect than tissue heat balance (THB).

Other water balance components such as undergrowth transpiration, canopy interception and runoff were not directly measured, and hence, their importance is difficult to estimate. It is frequently assumed that interception is already accounted for in ET flux (Soubie et al. 2016; Gu et al. 2018). Since the T_EC was determined here based on the ET fluxes after excluding the precipitation events, it was assumed that the T_EC does not contain interception. Nevertheless, this issue is a potential source of uncertainty, since the lens of the EC open-path gas analyser can dry faster after rain compared to the canopy. Using data that have been observed after short periods of rain, when the canopy was still wet, can thus lead to biased interpretation if it is assumed that the canopy is already dry. Thus, methodology of calculating interception using ET measurements from EC, combined with simplified assumptions, can produce potential failure to close the water balance. Evaporation (E) flux was higher at TU than at ME, presumably because of the higher total precipitation and number of low intensity rain events (Paschalis et al. 2018). It has been shown that such conditions typically result in an increase in rainfall interception loss. Also, E from exposed soil increase after precipitation. Furthermore, wet canopy surface simultaneously enhances evaporation and hampers transpiration due to blockage of the leaf stomates (Gash 1979; Paschalis et al. 2018).

Conclusions

Our study of transpiration fluxes in two Scots pine stands of different age classes (young and mature) located in north-western Poland was focused mainly on comparing EC and sap flow measurements under drought conditions, and finding key drivers of these processes. Our results yielded the following conclusions:

• (1) For the growing season of 2019, total T_EC exceeded precipitation sums for both sites, while T_SF did not. Calculated as total for year T_EC did not exceeded precipitation (for TU site). This means that ecosystem is using winter precipitation stored in soil and stems to survive during prolonged droughts in summer and most likely plants have access to water retained in deep soil layers (up to 2 m or deeper), which was not directly measured.

• (2) T_SF and T_EC had similar seasonal variability, daily and monthly courses and dynamics in response to environmental conditions for both sites. This confirms that both observations reflect the same flux, although the scale and scope of measurement have been different (ecosystem vs. tree level). T_EC includes information on transpiration of other species and understory as well. Thus, we can conclude that for managed forests, where one species is clearly dominant, upscaled sap flow measurements can be sufficient to properly diagnose the forest life status (in case that all tree size classes are well represented). In contrary, upscaling of sap flow measurements for forests with a significant amount of other hardwood species and ground vegetation, that are not considered in sap flow measurements, is challenging and consequently the sap flow measurement solely is not suitable for forest drought stress recognition.

• (3) It was found that with adequate water availability (5–8% for MU and ca. 4–6% for TU), T_SF and T_EC were high, and both methods give similar total values for the entire growing season. For lower SWCs, GPP was still relatively high, and discrepancies between T_SF and T_EC increased for both young and mature forest. Transpiration which occurred in June under low SWC values indicates an increasing use of stem water storage. High WUE with low SWC values within days of drought indicates that both ecosystems have reduced their water loss per carbon gain. Also, under low SWC values in the summer (ca. 3.5%), T_SF strongly decreased at both sites. Thus, it was evident that functioning of Scots pines at the investigated sites has reflected the strategy of trees to increase their water use efficiency under dry conditions to a surprisingly high level.

• (4) The sensitivity of T to dry conditions was similar at both sites/stands, but larger reductions were observed in T_SF relative to T_EC and in ET in July and August, when SWC was extremely low. This also shows that for assessing drought-induced effects on trees at the stand scale, transpiration from upscaled sap flow measurements seems more useful than EC-derived transpiration.