Does optimality partitioning theory fail for belowground traits? Insights from geophysical imaging of a drought‐release experiment in a Scots Pine forest

Summary

• We investigate the impact of a 20‐yr irrigation on root water uptake (RWU) and drought stress release in a naturally dry Scots pine forest.

• We use a combination of electrical resistivity tomography to image RWU, drone flights to image the crown stress and sensors to monitor soil water content.

• Our findings suggest that increased water availability enhances root growth and resource use efficiency, potentially increasing trees' resistance to future drought conditions by enabling water uptake from deeper soil layers.

• This research highlights the significant role of ecological memory and legacy effects in determining tree responses to environmental changes.

Introduction

The rate at which climate change affects forests is faster than the evolutionary adaptation of trees to the changing environment, which raises the question of their phenotypic acclimation potential (Gessler et al., ²⁰²⁰). Higher temperatures followed by increased water vapour pressure deficit (VPD; Treydte et al., ²⁰²⁴) lead to increased evaporative demand and transpiration (Grossiord et al., ²⁰²⁰). Along with negative precipitation anomalies during the growing season (Ciais et al., ²⁰⁰⁵) – that is when the observed precipitation is lower than the average value – tree water availability decreases, resulting in increased water stress (Hunziker et al., ²⁰²²; Meusburger et al., ²⁰²²). Drought stress can lead to reduced tree growth rates (Gazol et al., ²⁰¹⁷; Trotsiuk et al., ²⁰²¹), increased vulnerability to pests and diseases (Senf & Seidl, 2021), and crown dieback and tree mortality (Allen et al., ²⁰¹⁵; Frei et al., ²⁰²²; Gazol & Julio Camarero, ²⁰²²). However, forest responses to climate change are inherently challenging to characterise because they depend on above and belowground traits. Prediction of trait plasticity in response to changing environmental conditions and interactions between both compartments, including below‐ and aboveground trait coordination, remains difficult (Anderegg, 2023).

Remote‐ and proximal‐ (e.g. drone‐based) sensing enabled leaps in characterising aboveground traits of forest stands, often via vegetation indices, which help to identify functional and structural vegetation characteristics and assess vegetation dynamics, bridging the gap between ground‐based and global scales (Vicente‐Serrano et al., ²⁰¹³). Drones equipped multispectral cameras can be used to compute vegetation indices (Fawcett et al., ²⁰²⁰) such as the Photochemical Reflectance Index (PRI; Gamon et al., ¹⁹⁹⁷) which targets changes in the xanthophyll pigment cycle, a photoprotective mechanism (Demmig et al., ¹⁹⁸⁸). Decreases in PRI show that plants increasingly protect themselves from too much light by reducing their photosynthetic efficiency when there is a lack of water (Gamon et al., ¹⁹⁹⁷; Garbulsky et al., ²⁰¹¹). In principle, PRI can enable the detection of early, nonvisible signs of water stress preceding the occurrence of visible symptoms such as discoloration and defoliation of crowns.

Mapping belowground traits of trees, for example root distribution or processes such as water uptake at depths over extended spatial scales, is inherently challenging. The traditional approach in soil science is to perform isolated measurements such as soil moisture and water potential (Whalley et al., ²⁰¹³) or soil water isotopic composition (Gessler et al., ²⁰²⁰), or to dig soil transects to assess rooting patterns (Brunner et al., ²⁰⁰⁹; Billings et al., ²⁰¹⁸). Such methods provide detailed belowground information but are sparse, limited in penetration depth and often fail to characterise the full complexity in the presence of rocky or highly heterogeneous soils.

Observing deep roots and processes at depth is time‐consuming, expensive and often requires a combination of several methods (Maeght et al., ²⁰¹³). On top of these challenges, information on the presence and amount/volume of roots at a given depth does not necessarily indicate their activity in water uptake (Volkmann et al., ²⁰¹⁶). Therefore, the role of deep roots in buffering drought periods through water uptake is still poorly understood relative to shallow roots (Pierret et al., ²⁰¹⁶). The choice remains between direct methods, such as soil excavation, trenching, coring, ingrowth cores and rhizotron measurements, vs indirect methods, which aim to infer deep‐root functions by quantifying water and nutrient budgets or through geophysical imaging (Balwant et al., ²⁰²²; Loiseau et al., ²⁰²³).

Electrical resistivity tomography (ERT) is a nondestructive geophysical technique regularly used to study the subsurface in forest ecosystems (Attia al Hagrey, 2007; Koch et al., ²⁰⁰⁹; Rodríguez‐Robles et al., ²⁰¹⁷; Pawlik & Kasprzak, ²⁰¹⁸). The methodology is to inject a current between a pair of electrodes into the soil while measuring changes in the potential difference between another pair of electrodes (Binley & Slater, 2020). With electrodes along a transect, the measurement combinations can be used to create a 2D map, or tomogram, of electrical resistivity as a function of distance and depth. An advantage of ERT is that time‐lapse measurements, that is the repetition of measurements at different times along an experiment to assess belowground processes (Parsekian et al., ²⁰¹⁵; Singha et al., ²⁰¹⁵), is straightforward if the electrodes are in place and well coupled. Carrière et al. (2020) showed that ERT can effectively image water uptake at depth and in particular in drought conditions, Zenone et al. (2008) have used it to study biomass in pine forests and several others studied spatiotemporal soil moisture changes in forests (Mares et al., ²⁰¹⁶; Dick et al., ²⁰¹⁸; Fäth et al., ²⁰²²; Rieder & Kneisel, ²⁰²³). Of particular interest in the critical zone are root‐zone processes (Furman et al., ²⁰¹⁵), where ERT in principle helps track water movement and redistribution in the soil, and where time‐lapse surveys can be used to assess dynamic root water uptake (RWU) and availability at depth (Jayawickreme et al., ²⁰⁰⁸; Boaga et al., ²⁰¹³). One major challenge in interpreting time‐lapse ERT signatures is that changes in resistivity are affected by changes in soil moisture, temperature, salinity, soil compaction and other processes (Llera et al., ¹⁹⁹⁰).

Our long‐term experimental research site is located in the naturally dry Scots pine forest of Pfynwald (Valais, Switzerland) and home to a long‐term (2003‐current) drought/irrigation experiment. The application of irrigation water, which doubled the amount of natural precipitation of c. 600 mm yr⁻¹, led to a release from drought and had distinct ecological consequences for the Scots pine forest ecosystem (D'Odorico et al., ²⁰²¹; Rissanen et al., ²⁰²¹; Bose et al., ²⁰²²). In 2013, the irrigation was stopped in one‐third of each irrigation treatment (thereby called the ‘irrigation‐stop’ treatment), which led to a decrease in soil water content similar back to that of the control conditions. The long‐term legacy of irrigation (and thus environmental conditions in general) in the Pfynwald site was measurable in photosynthesis, PRI, growth and morphology (Zweifel et al., ²⁰²⁰; Gao et al., ²⁰²¹; Schönbeck et al., ²⁰²²) and suggests that irrigation‐stop trees have still not had their functional traits converged to those of control trees (Vitali et al., ²⁰²⁴). One explanation might be the acclimation to the previously increased soil water availability on the belowground traits, as indicated by the differing responses of root‐respiration and carbon allocation to short‐term rain events between irrigated and control trees (Joseph et al., ²⁰²⁰; Gao et al., ²⁰²¹). For instance, due to irrigation, the rhizosphere expanded by 2.8 times laterally, and the root biomass increased in the upper 70 cm of the soil (Brunner et al., ²⁰¹⁹; Gao et al., ²⁰²¹); however, no information of root biomass at soil layers below 1 m has so far been available.

The observed changes in root distribution in Pfynwald conflict with the optimality partitioning theory, which suggests that irrigated trees with ample access to water neglect the investment into the root network and over‐proportionally invest in the aboveground traits. Conversely, control trees grown under water scarcity would increase their investment in the root network at the cost of aboveground traits (Hatton et al., ¹⁹⁹⁷). Consequently, when irrigation was stopped for the formerly irrigated trees (irrigation‐stop trees), optimality partitioning theory suggests that they would suffer more than control trees because their root network would not be able to support their large crowns. Unexpectedly, the irrigation stop led to a cascade of downregulations: Biophysical processes such as stomatal conductance and transpiration exhibited a rapid response within days, while needle and shoot lengths (Dobbertin et al., ²⁰¹⁰), crown transparency, and radial stem growth took up to 4 yr to return to control levels (Zweifel et al., ²⁰²⁰). Irrigation‐stop trees did not show higher, but instead lower drought stress than the control trees (D'Odorico et al., ²⁰²¹) and even higher radial stem growth despite ambient water input of c. 600 mm yr⁻¹ by natural precipitation. An enlarged root biomass (Brunner et al., ²⁰¹⁹; Gao et al., ²⁰²¹), which also extends to deeper soil layers, where direct measurements are not possible, could have improved water uptake capacity and explain the latter observations.

In this study, we investigate RWU acclimation under different water availability regimes using ERT measurements before the irrigation season started and in the middle of the 2022 summer drought. Under dry summer conditions, we also assessed crown stress via PRI derived from drone‐based multispectral imaging above the ERT transects. In contrast to the optimal partitioning theory, our hypothesis (H0) is that the irrigation‐stop trees have invested in deeper roots grown during 10 yr of irrigation conditions, despite receiving additional water from the soil surface. Now, with larger crowns than control trees they demand more water from their roots, thus drying out the soil more rapidly and into deeper layers. Irrigated trees receive water at the soil surface and are therefore expected to require less water from deeper root uptake than the control or irrigation‐stop trees. As a consequence, irrigation‐stop trees show intermediate drought stress levels at their crowns (as would be indicated by the PRI) due to their ability to partially compensate for the water demand by extensive RWU.

Materials and Methods

Study site

This study was conducted at the long‐term experimental research site in Pfynwald, outlined in Fig. 1, which is a >130‐yr‐old Scots pine (Pinus Sylvestris L.) forest located in the Rhone Valley in Switzerland (46°18′N, 7°36′E, 615 m above sea level). With a mean annual temperature of 10.6°C and a mean annual sum of precipitation of 576 mm (1995–2014, from nearby MeteoSwiss station in Sion), this forest is at the very dry edge of the Scots pine distribution (Zweifel et al., ²⁰²⁰). The pine forest has an average tree height of c. 12 m, a stand density of 680 stems ha⁻¹ and a basal area of 24 m² ha⁻¹ (Hunziker et al., ²⁰²²).

Fig. 1.

Experiment setup: (a) The eight plots of the Pfynwald long‐term experimental research site along with the separation in control, irrigation and irrigation‐stop treatments. For this study, we focused on Plots 5 and 6, shown in (b) along with the soil sensors/profiles and resistivity transects. In (b) the location of Tef Pinus Sylvestris L. (Scots Pines) is shown.

The geology of the Pfynwald site is composed primarily of debris‐flow deposits from the nearby Ill‐graben system (Schuerch et al., ²⁰¹⁶), which extends at least 20 m deep. In Fig. 2(a), we show an outcrop located roughly 1.5 km northeast of the study site, at the outskirts of the deposits, which shows the approximate depth of the debris flow. To analyse the topsoil on the site, we dug three 2 m deep soil profiles during the experiment. One such profile reveals a topsoil layer of humus, more precisely classified as Regosol, with a low water‐holding capacity of 135 mm and soil water content varying between 11% and 47% (Fig. 2b). This layer reaches a depth between 15 and 45 cm and contains a high density of roots. The lower stratum has a high content of boulders, with an inter‐boulder filling composed roughly of 51% sand, 42% silt, 6% clay and 1% gravel.

Fig. 2.

Photographs taken from the Pfynwald forest: (a) Outcrop near the long‐term irrigation experiment at Pfynwald, showing the scale of above‐ (Pinus sylvestris L.) and belowground biomass compared with debris‐flow deposits. (b) A soil profile dug in the irrigation treatment reveals the debris‐flow deposits buried under < 0.5 m of humus layer and large boulders in the soil. (c) A view of soil sensor SM‐30, located near the electrical resistivity transect running from North to South as seen in Fig. 1(b).

Since 2003, a long‐term irrigation experiment has been continuously conducted with initially four plots being irrigated and four control plots receiving natural precipitation only, as indicated in Fig. 1(a). The irrigated plots received double the annual precipitation of c. 1200 mm yr⁻¹ from sprinklers at 1 m aboveground, delivered at night‐time during the frost‐free period (from April to October). By the end of 2013, irrigation was stopped in the upper third of each irrigated treatment. The irrigation of the remaining treatments, referred to as ‘irrigation’ continues to date. Several permanent sensors exist in Pfynwald to measure soil moisture/temperature, air temperature and humidity, radial growth (band‐ and point dendrometers) and other experiment‐specific parameters.

Electrical resistivity tomography

Along each resistivity transect (North–South, NS, and East–West, EW) shown in Fig. 1(b), we installed 50 electrodes at 1 m spacing. We used a Leica TS07 Total Station (Leica Geosystems) to survey each electrode position and obtain the relative height between electrodes to sub‐cm precision. To position the transects on the map, we used tree locations, which have been measured to cm precision during a previous surveying campaign (D'Odorico et al., ²⁰²¹). We measured electrical resistivity with a GeoTom apparatus (GEOLOG, Stamberg, Germany), and the timeline of our campaigns is shown in Supporting Information Table S1. More explanation on the design of the experiments, as well as the forward and inverse geophysical modelling, is found in Methods S1. In the results section, we present the inversion results as resistivity models with a transparency applied to model cells that have a low sensitivity to the data. Both sensitivity and resolution are taken into account when discussing our results.

Long‐term measurements of soil water content

Since 2014, 18 soil profiles have been established with volumetric soil water probes (10‐HS; Decagon Devices, Pullman, WA, USA) and soil water potential (MPS‐2; Decagon Devices) probes at 10 and 80 cm depths. The soil moisture sensors were calibrated using a soil‐specific calibration and the uncertainty in measurements is expected between ±0.02 m³ m⁻³ (±2% VWC). For the MPS‐2 sensors, the manufacturer suggests an uncertainty of ±25% of a reading (+2 kPa) from −9 to −100 kPa; however, no indication is given for values below −100 kPa. Two sensors were already present in the three chosen subplots for each treatment, resulting in six sensors per treatment available for analysis. These sensors continuously recorded measurements at 15‐min intervals. In postprocessing, we calibrated the data to account for the temperature sensitivity of the MPS‐2 sensors, ensuring accurate and normalised data (Walthert & Schleppi, 2018). Fig. S1 displays the temperature and temperature‐corrected volumetric water content recorded at 80 cm depth for the six MPS‐2 sensors and 10 HS sensors.

Drone flights and Photochemical Reflectance Index

We acquired narrow‐band multispectral images with a Micasense Dual camera (Micasense, Seattle, WA USA), mounted on a DJI Matrice 300 quadcopter (DJI Technology Co. Ltd, Shenzhen, China). The camera simultaneously operates 10 separate imaging sensors in 10 narrow bands within the visible and near‐infrared spectrum. We selected the bands to target specific absorption features of photosynthetic pigments and performed the drone flights under clear, sunny conditions close to local solar noon to minimise shadows and variations in acquisition geometry. We describe all image processing and postprocessing of the drone data in Methods S2. We compute the PRI at the scale of single, 6 cm pixels and the crown of individual trees.

Results

Soil conditions

The soil water potential varied substantially during the vegetation period between the treatments (Fig. 3). Before the irrigation started in mid‐May 2022, irrigated trees with large canopies and transpiration rates dried out the soil at 80 cm faster than the control treatment equivalents, as is indicated by the lower soil water potential. This resulted in a distinct difference between the soil water potential of the irrigated treatment and the irrigation‐stop and control treatments at 80 cm soil depth. There were no major differences observed between treatments in the top 10 cm at the onset of the irrigation. The onset of irrigation (May 11) immediately increased the soil water potential, and higher water availability persisted for irrigated trees throughout the rest of the vegetation period. The differences in soil water potential between the irrigation‐stop and control treatments were less pronounced; however, an overall decrease was observed from May to July.

Fig. 3.

Soil water potential in kPa (as the mean of two sensors) was measured for the three different treatments (red = control, blue = irrigated and green = irrigation stop) at 10 cm and 80 cm belowground. The rainfall from a nearby meteorological station (Sion) is plotted in the top row. The uncertainty in these measurements is indicated as shaded areas surrounding the lines. Grey dashed lines indicate the two field campaigns in mid‐May and late‐July. Irrigation started in the same week as the first experiment (see Supporting Information Table S1).

Imaging the belowground with electrical resistivity tomography

The electrical resistivity tomograms for the different treatments are shown in Fig. 4. Based on the sensitivity computation, these tomograms are informative to depths of roughly 10 m. However, the resolution matrices indicate a shallower depth of investigation, with the resolution dropping below 10⁻⁴ (unitless, e.g. see Day‐Lewis et al., ²⁰⁰⁵) at a depth of roughly 8 m (see Methods S1 for more information). For each tomogram, we set the starting electrode to a z and distance value of 0 m. Negative values of z refer to increasing depth, but as the profiles have a slight elevation, we avoid calling the vertical axis depth. Instead, we compute the depth for each transect with respect to the model's ground surface.

Fig. 4.

Comparison of electrical resistivity tomography results for North–South (NS; a, b) and East–West (EW; c, d) transects measured in May (left column) and July (right column) respectively. The intersection between the two transects is indicated with a dashed vertical line. The extent of each treatment is indicated on the top of each panel. Tomograms are made transparent in areas of low sensitivity.

In May (Fig. 4a,c), we observe a high‐resistivity band in the uppermost c. 2 m of soil, which is present throughout the experimental treatments but strongest below the irrigated trees. Along the NS transect (Fig. 4a), we observe a weakly resistive band at depth with the highest resistivity in the middle of the transect at c. 30 m. A low‐resistivity band is visible below the irrigated trees (Fig. 4c) and extends from 0 m to 30 m along the transect and between 6 m to 10 m in depth.

In July, the resistivity along the NS transect (Fig. 4b) increased compared with May. At the irrigation‐stop treatment, we observe higher resistivities and at larger depths compared with the control treatments. A similar pattern is observed along the EW transect (Fig. 4d) where a resistive anomaly below the irrigation‐stop trees is strongly visible compared with the May campaign, and is also seen to cross over to the irrigation treatment. On the contrary, below the irrigated trees (Fig. 4d) we observe very low resistivities extending to the maximum‐resolved depth of the tomogram. We present a validation of our inversion results in Methods S1.

Relative changes between the two electrical resistivity campaigns

In Fig. 5, we show the relative (percentile) changes between the May and June campaigns, that is relative change = 100 × (R2 − R1)/R1, where R2 refers to the inverted resistivities from July and R1 from May. Changes below the control treatment in the NS transect reveal processes occurring in the first 5–6 m belowground, while the changes below the irrigation‐stop treatment extend to c. 10 m. This observation is consistent along the EW transect when noting the intersection of the two transects. Relative changes along the EW transect suggest processes occurring mainly below the irrigation‐stop treatment, where resistivities increase by almost 200%. Below the irrigated trees, the relative changes show a roughly 50% negative shift, meaning resistivity is reduced at depth during the dry summer months.

Fig. 5.

Relative changes in resistivity between May and July 2022 for the North–South (a) and East–West transects (b). The intersection between the two transects is indicated with a dashed vertical line. The extent of each treatment is indicated on the top of each panel. Tomograms are made transparent in areas of low sensitivity.

Drone imaging during the July campaign

We analyse the drone images by looking at the pixel‐based PRI index directly (Fig. 6a), and by selecting polygons for the extent of the crowns of each tree (Fig. 6b). The control trees are present in both the control treatment and on the outskirts of the EW transect (buffer zones between treatments) and reveal the lowest PRI values, c. 0.1, indicating the highest stress levels. Trees in the irrigation‐stop treatment revealed medium‐stress levels, with average PRI values between 0.13 and 0.17, and irrigated trees had the highest average PRI, up to 0.19, revealing the least‐stressed conditions (see also Methods S2).

Fig. 6.

Photochemical Reflectance Index (PRI) analysis from drone multispectral imaging during the July campaign. (a) Pixel‐level PRI values after removing ground, understory and shaded pixels. (b) Average PRI values for trees along the transects, outlined by polygons. Lower PRI values suggest higher stress conditions for trees and vice versa. In both figures, we indicate the extent of the treatments, the location of soil moisture sensors (triangles) and the electrical resistivity tomography (ERT) transects, using different colours and line styles for visibility. Fig. 1 explains the experimental setup in detail.

Discussion

Active root water uptake depth

We expect resistivity increases between May and July due to decreased water content from RWU (Fig. S2). At the soil surface, increased resistivity may also be the result of soil evaporation; however, soil evaporation is typically restricted to soil depth < 0.8 m (Lehmann et al., ²⁰⁰⁸). Therefore, any increase in resistivity observed below this depth can be interpreted as a decrease in soil moisture driven by RWU. That said, resistivity decreases can occur due to processes such as drainage during rain events and irrigation, capillary rise, or lateral flow. However, throughout the growing period, across all treatments and soil depths (Fig. S3) the measured soil water potential values remain below field capacity (< −30 kPa), suggesting that significant water drainage is unlikely, except possibly during high‐intensity rainfall events. We thus assume that any observed increases in resistivity are a consequence of RWU, at least at depths below 1 m.

Still, resistivity changes are only indirectly related to RWU, which allows statements about absolute moisture changes only after calibration with soil moisture sensors or through petrophysical transforms. We avoid both approaches because the tomographic models result from inverse modelling and not direct measurements (Day‐Lewis et al., ²⁰⁰⁵). Tracing water isotopes is an alternative direct method to assess RWU but even then the quantification to absolute amounts is challenging (Penna et al., ²⁰¹⁸).

Measurements of soil water potential (Fig. 3) suggest that the irrigated trees dried out the top 10 cm of soil more efficiently, whereas the control and irrigation‐stop trees were more efficient in water uptake at 80 cm. The spike in the irrigation‐stop 10‐cm sensor plot roughly a week after irrigation starts is likely linked to a rain event observed at the same period that was not captured by all sensors (Fig. S3). The low and negative water potentials measured in these plots suggest that lateral flow of water between treatments is unlikely. In addition, the lower elevation of the irrigation treatment makes it impossible for water to flow laterally on the surface between the various treatments.

The above observations are supported by the resistivity tomograms (Fig. 4) where at shallow depths and before irrigation (a and c) the top few metres are equally dry (resistive) between treatments. On the contrary, below the irrigation treatment a distinct low‐resistivity layer is visible, which is likely the wetting front (contrast between wet and dry soil) from the last winter phase. In July, after > 2 months of irrigation, we also observed increases in resistivity at depth on the irrigation‐stop treatment. However, a low‐resistive band between c. 3 and 8 m depths, also visible in the irrigation‐stop treatment (Fig. 4b) might be linked to large boulders redirect water flow and root growth (visible in Fig. 2a,b). The resistive band shown in Fig. 4(d) ‘crosses‐over’ from the irrigation stop to the irrigation treatment. Possible explanations for this ‘crossing over’ may be as follows: (1) there is a large population of irrigation‐stop trees along the transect to irrigation as is shown in Fig. 1(b) from a top view; (2) the interface between irrigation and irrigation stop is in reality not as sharp as shown in Fig. 1(b), since there is no buffer between these treatments; or (3) the roots of the irrigation‐stop trees are extending into the irrigation treatment.

In Fig. 7, we show the arithmetic average of the relative changes in resistivity (Fig. 5) as a function of depth, for each treatment. The control trees exhibit increases in resistivity (e.g. drying) in the upper soil until c. 6 m from May to July, and decreased resistivity below that depth. The resistivities below the irrigated trees exhibit an irregular pattern with predominantly decreasing resistivity (e.g. wetting) in the upper 2 m between May and July and increasing changes (drying) below. The irrigation‐stop trees behave similarly to the control trees in the upper 5 m of the soil, increasing in resistivity by 55% from May to June, implying that this upper layer dries out considerably during the summer. This agrees with the observations from soil moisture sensors (Fig. 3, lower panel) when comparing irrigation‐stop and control treatments. These results suggest that in the first few metres, the response of irrigated trees vs irrigation‐stop and control trees varies strongly. The irrigation supplies water at shallow depth which results in a reduction of resistivity of up to 47% from May to July, at depths above c. 2 m.

Fig. 7.

Averaged relative changes (between May and July) in resistivity vs soil depth for the three treatments. The values are averaged for each treatment, taking the arithmetic mean of all inversion cells at a given depth. Positive values indicate an increase in resistivity (e.g. drying) from May to July whereas negative values indicate a percentage decrease (e.g. wetting). The varying shading indicates how well resolved each measurement is, with darker shades denoting higher resolution.

While we expect that the observations are less trustworthy as depth increases, even for depths of 6–8 m we observe a decrease in resistivity suggesting increasing water availability. This is expected for the irrigated trees since we see the wetting front of the winter moving down. This is also visible (but less pronounced) for the control trees indicating that the roots of these trees cannot access the previous year's wetting front. Below the irrigation‐stop trees, where we find a large (20–40%) percentage increase in resistivity as a function of depth. This result suggests that during the dry summer months' irrigation‐stop trees seek water from deeper soil layers than the control trees.

Relation between belowground resistivity and aboveground drought stress

Assuming a similar root distribution between the treatments, we would expect a negative relationship between PRI and resistivity changes. An intuitive explanation is that low drought stress (high PRI) correlates with high‐water availability (reduced or minimal resistivity changes). In Fig. 8, we compare the relation between changes in belowground resistivity and the drought stress proxy (PRI) for the different treatments. We achieve this by averaging the pixel‐based PRI response along the resistivity transects within a radius of 2 m, which is the average crown size. For the resistivity, we average the tomograms for each treatment between a depth of 1 m (assuming the start of RWU, e.g. Fan et al., ²⁰¹⁷) and the depth of maximum sensitivity (Figs 4, 5), to focus on the deeper root processes.

Fig. 8.

Kernel‐density‐estimation and marginal probabilities of the relationship between Photochemical Reflectance Index (PRI) and resistivity for each treatment. Lower PRI values indicate higher drought stress levels and positive changes in resistivity suggest drying out or increased root water uptake (RWU). The treatments show distinct and nonlinear patterns, suggesting different responses to stress and RWU depths.

Irrigated trees showing low drought stress (high PRI) shift towards negative resistivity values (higher water availability due to irrigation) in July. Conversely, the control trees suffer from the highest drought stress (lowest PRI) and show only minor changes (peak c. 25%) in resistivity between May and July. This suggests that RWU was marginal in this period. The behaviour of the irrigation‐stop trees was the most interesting as they showed median PRI values between control and irrigated trees, although we would expect them to have higher PRI values because of their large crown. The latter observation is supported by the largest increase in resistivity at depths for all three treatments, indicating a pronounced RWU of these trees.

The observed rooting depth for the control trees of c. 5 m is in the range of reported values between 5 and 8 m, as reviewed by Stone & Kalisz (1991) and a maximum rooting depth of up to c. 8 m for temperate coniferous forests suggested by Fan et al. (2017). The RWU depth of the irrigation‐stop trees exceeds these reported maximum rooting depths in the literature. Unlike natural sites, the irrigation may have allowed investment in the belowground biomass production (Brunner et al., ²⁰¹⁹; Joseph et al., ²⁰²⁰; Gao et al., ²⁰²¹). For instance, Joseph et al. (2020) found that increased relative soil moisture at a tipping point of 15% increased whole‐tree carbon allocation to the belowground organs, which did not exist for trees under naturally dry control conditions. Gao et al. (2021) showed that the increased belowground carbon allocation increased the root projection area (i.e. the horizontal spatial extent of the rooting system of a single tree) upon long‐term irrigation. Here, we provide indications that the vertical extension of the rooting system reaches these previously unexplored depths.

Link between the crown stress indicators and RWU

Due to their larger crowns, irrigation‐stop trees require more water than control trees but can compensate for parts of their demand by exploiting deeper soil water sources, resulting in an intermediate water stress level indicated by intermediate PRI values (Fig. 9). Other studies found intermediate stress levels in the irrigation‐stop trees by PRI (D'Odorico et al., ²⁰²¹) and tree water deficit (Zweifel et al., ²⁰²⁰). Access to deeper water is likely to explain the buffering of drought stress (Burns et al., ²⁰²³) and even improved growth during an initial phase after the stop of irrigation (Zweifel et al., ²⁰²⁰). However, from these existing studies it is quantitatively difficult to rule out the effect of downregulation of tree physiological (e.g. sap flow) and morphological (e.g. needle length) responses.

Fig. 9.

Conceptual model of the processes taking place in the different treatments and indication of the various sensing techniques, highlighting the link between crown stress, inferred from drone‐based Photochemical Reflectance Index (PRI), and root water uptake (RWU), inferred from electrical resistivity imaging.

Why did irrigated trees invest in their deep roots?

Our observation of deeper RWU in previously irrigated trees partially conflicts with the correlation between RWU depth and local soil water availability (Fan et al., ²⁰¹⁷). It also contradicts the optimal partitioning theory, which suggests that plants acclimate to different environments by allocating a significant proportion of biomass to the organs acquiring the most limiting resource (Gedroc et al., ¹⁹⁹⁶). Based on this principle, we would expect trees with sufficient water supply at the surface (here, irrigated (2003–2013) or irrigation stop (2013–current) trees) to invest in shoots at the expense of roots leading to shallower rooting systems. Conversely, trees in environments with limited water availability (here, the control trees) would enhance their root network investment to better cope with water scarcity and to exploit deeper soil layers. By contrast, however, we observe that formerly well‐watered irrigation‐stop trees established deeper rooting systems while the drought‐exposed control trees did not. A possible explanation for this might be provided by the observations of Joseph et al. (2020), who showed reduced belowground carbon allocation under strong soil moisture limitations in the control treatment. They assumed that under such conditions, the metabolic activity of the roots and the rhizosphere microbiome is strongly suppressed. A decrease in sink strength (Gessler & Zweifel, 2024) demands nonphotosynthetic tissues and tree‐related microbes for carbon compounds and leads to an accumulation of metabolites in the release phloem in the roots, reducing phloem allocation to belowground tissues with potential negative feedback on assimilation (Hagedorn et al., ²⁰¹⁶). However, under high‐water supply as in the irrigation treatment, a larger proportion of the new assimilates are transported belowground to roots and rhizosphere (Joseph et al., ²⁰²⁰). Thus roots can grow more intensively when they receive more water, not only horizontally as shown by Gao et al. (2021) but also vertically, as is supported by our indirect evidence of changing resistivity to depths below 6 m. These roots seem still to be present and active years after the irrigation stopped.

Our interpretation needs to be corroborated by results assessing tree carbon (C) relations. Here, we investigate two alternative hypotheses (H); First, greater water availability may have allowed trees to keep their stomata open longer, increasing overall photosynthetic carbon uptake and growth across all tree parts, including roots, without altering relative carbon allocation (H1). Second, differences in canopy size and whole‐tree photosynthesis rates may have led to differing absolute amounts of carbon invested in roots vs in aboveground growth without the relative allocation being affected (H2).

Table 1 summarises previous results on stomatal conductance, photosynthetic C assimilation and C allocation. Long‐term stomatal conductance measurements showed no significant differences between irrigated and control trees (Schönbeck et al., ²⁰²²), indicating that H1 may be rejected. Whole‐tree ¹³CO₂ exposure (Joseph et al., ²⁰²⁰) indicated comparable total CO₂ uptake in control and irrigated trees, allowing us to reject H2. Assessment of relative above‐ and belowground allocation of new assimilates 30 d after the pulse labelling showed that a significantly higher share of C taken up ended up in the belowground compartment, supporting our initial hypothesis (H0) that the irrigation‐stop trees have invested in belowground traits during the 10 yr of irrigation. In the same labelling experiment, Gao et al. (2021) showed that this higher proportional C allocation matched a larger horizontally projected area of the active rhizosphere, determined as the area around the stem where the labelled CO₂ was respired. The fact that irrigated trees allocated relatively (and also absolutely as canopy level assimilation is comparable) more carbon belowground allows us to assume that sufficient water supply is a prerequisite to build a deep‐reaching root system.

Table 1.

Stomatal conductance, photosynthetic C assimilation and allocation of the newly assimilated C between aboveground and belowground ecosystem compartments in the control and irrigated plots of the Pfynwald research site.

	Control	Irrigation	Significance	Reference
Needle‐level stomatal conductance	0.05 ± 0.03 mol m−2 s−1	0.07 ± 0.03 mol m−2 s−1	ns	Schönbeck et al. (2022)
Canopy level assimilation	59 ± 14 μmol tree−1 s−1	69 ± 13 μmol tree−1 s−1	ns	Joseph et al. (2020)
Relative aboveground allocation	82.8 ± 5.4%	63.6 ± 5.5%	P < 0.01	Joseph et al. (2020)
Relative belowground allocation	17.2 ± 2.2%	36.4 ± 10.8	P < 0.01	Joseph et al. (2020)
Projection area of the active rhizosphere	13.0 ± 1.1 m2	20.8 ± 2.0 m2	P < 0.01	Gao et al. (2021)

Needle‐level stomatal conductance is the average value (±SE) from 13 measurement time points between 2014 and 2019, covering spring, midsummer and late summer (see Schönbeck et al., ²⁰²²). In 2017, whole‐tree ¹³CO₂ labelling experiments were performed for 3.5 h (see Joseph et al., ²⁰²⁰), and whole ¹³C uptake by the canopy of irrigated and control trees during that period was assessed (mean ± SD). After 30 d, the relative below‐ and aboveground allocation of the ¹³C taken up (mean ± SD) was calculated from the tree and microbial biomass pools (aboveground: needles, branches, stems; belowground: fine and coarse roots, soil microbial biomass) and from the cumulative respiratory fluxes. In the same labelling experiment, Gao et al. (2021) assessed the spatial dynamics of the ¹³C excess in soil‐respired CO₂ at different distances from the stem of the labelled trees. The active rhizosphere projection area after 30 d (mean ± SD) was defined to have a ¹³CO₂ signal strength of at least 25% of its maxima (that was always found within the first m from each stem). Differenes between treatments were assessed with linear mixed effect models.

Joseph et al. (2020) provided an explanation for the lower proportional allocation of C belowground under a lower water supply. They assumed that under such conditions, the metabolic activity of the roots and the rhizosphere microbiome is strongly suppressed. Low belowground sink strength leads to an accumulation of metabolites in the release phloem in the roots, thus reducing phloem allocation to belowground tissues. Increased soil water availability was shown to increase the metabolic activity of the rhizosphere immediately, resulting in increased allocation of assimilates belowground.

The way that plants allocate resources between their above‐ and belowground compartments seems spatially constant when looking at large‐scale datasets (Enquist & Niklas, 2002). However, at the local scale, it is more challenging to derive general patterns of biomass allocation (Raven, 1998; Poorter et al., ²⁰¹²; Caspersen et al., ²⁰⁰⁰; Puglielli et al., ²⁰²¹, most likely due to the fast temporal dynamics (from days to a few years) with changing environmental conditions (Joseph et al., ²⁰²⁰). This is only particularly true for belowground traits (Poorter et al., ²⁰¹²) since reduced water availability can lead to both reduced and increased allocation of carbon to the root system Ruehr et al., ²⁰⁰⁹; Hommel et al., ²⁰¹⁶). Long‐term legacy effects of previous conditions might play an important role for the rooting systems of trees (Ruehr et al., ²⁰⁰⁹). We might assume in our experiment that after drought release by irrigation, which started in 2003, the activation of the belowground metabolic activity boosted root growth, which led to increased transport of new assimilates to the belowground compartment as observed for beech trees previously (Hagedorn et al., ²⁰¹⁶). There is an indication that ecological memory affects plant growth and biomass allocation strategies in Scots pines even across generations (Bose et al., ²⁰²⁰). We might thus assume that the memory of Scots pine trees exposed to long‐term drought conditions before 2003 led to a strong biomass allocation to the roots that was only possible after irrigation was initiated (Gao et al., ²⁰²¹), resulting in a deep‐reaching rooting system. After the irrigation was stopped, the irrigation‐stop trees still benefited from this root system and were able to exploit deeper soil water resources in contrast to the control trees. This more efficient use of the belowground water resources led to a lower stress level and higher photosynthetic efficiency (D'Odorico et al., ²⁰²¹; Schönbeck et al., ²⁰²²) compared with the control trees, even though the natural precipitation input was the same.

Our results indicate that (1) long‐term legacy and memory effects might strongly determine the current resource use efficiency and thus the reaction to current environmental conditions, and (2) an increase in water availability for a decadal time period might not necessarily increase the susceptibility of trees due to a large crown but in contrast might increase the resistance to low precipitation over longer time periods as only sufficient soil water availability seems to allow producing a rooting system that can also exploit deeper water resources.

Competing interests

None declared.

Author contributions

AS, KS, GP, HM and KM designed the research. AS, RH, JG and KM performed the research. AS, KS, GP, PD, FMW and KM analysed the data. AS, RH, HG and KM collected the data. AS, RH, AG, KS, GP, PD, FMW and KM interpreted the data. MS served as site PI. AS, AG, KS, PD, MS and KM wrote the manuscript.

Supporting information

Fig. S1 Changes in resistivity with volumetric water content.

Fig. S2 Soil water potential measurements in 10 cm soil depth of each treatment and all plots of the Pfynwald experiment.

Fig. S3 Soil temperature and volumetric water content for measuring stations at 80 cm depth.

Methods S1 ERT data processing and modelling.

Methods S2 Drone data analysis and postprocessing.

Table S1 Dates of the various measurement campaigns.

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

All data supporting the findings of this study are available within the article and its Supporting Information files. The data used in this work can be accessed on Shakas et al. (2024) doi: 10.16904/envidat.550. A git repository is linked in the data repository and hosts the code needed to reproduce the results.