Flooding with brackish water disrupts Liriodendron tulipifera sap flux in an ecosystem-scale experiment

Abstract

Coastal forests are retreating due to sea‑level rise and extreme precipitation events, yet the physiological mechanisms that precede woody plant decline remain poorly constrained. We used the ecosystem‑scale TEMPEST (Terrestrial Ecosystem Manipulation to Probe the Effects of Storm Treatments) experiment; annual, plot‑level freshwater and saltwater flooding of a mid‑Atlantic deciduous forest, to quantify short‑ and long‑term responses of sap flux in three tree species: Acer rubrum, Fagus grandifolia, and Liriodendron tulipifera. We analyzed five years of sap flux density measurements (2021–2025) using a Before–After Control–Impact (BACI) framework and linear mixed‑effects models to assess short and long-term changes in tree hydraulic function. Across the study period, species differed markedly in their responses to repeated flooding. A. rubrum showed no significant change in sap flux density under any treatment, while F. grandifolia exhibited plot‑specific increases or decreases consistent with relief from drought in the freshwater treatment plot. L. tulipifera sap flux density increased over time in all plots, but increases were dampened in the freshwater plot and strongly suppressed under saltwater flooding. Soil electrical conductivity rose progressively in the saltwater plot and emerged as a key predictor of both immediate and cumulative reductions in L. tulipifera sap flux density. As L. tulipifera constitutes a large share of canopy basal area in this coastal forest, continued saltwater intrusion is poised to accelerate mortality and drive ecosystem state-change. Our findings illustrate how ecosystem‑scale manipulations can uncover physiological thresholds relevant to coastal forest vulnerability under future sea‑level rise.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00442-026-05976-5.

Introduction

Sea-level rise and intensifying precipitation regimes alter the structure and function of coastal ecosystems (Boon 2012; Laurent et al. 2022). These environmental changes directly impact the structure and function of coastal forests (Hopple et al. 2023), resulting in rising rates of woody plant mortality over the course of transition from upland forest to wetland ecosystems (Kirwan and Gedan 2019; McDowell et al. 2022). Understanding and predicting such state changes remains challenging in part because little is known about the sensitivity of trees to infrequent storm events or the conditions that precede tree mortality and ecosystem state change. In North America’s Chesapeake Bay, a global hotspot for sea-level rise, recent work indicates that ghost forests are more widespread than previously realized (Yeung et al. 2025). Expectations that rates of sea-level rise are accelerating highlight the need to accurately predict and detect early changes in tree health.

Intense rainfall events lead to flooding and saturated soil, resulting in root mortality from hypoxia and reduced water uptake (McDowell et al. 2022). In addition, if floodwater is saline, salt accumulation can limit tree hydraulic conductance, photosynthesis, and growth (Kozlowski and Pallardy 1997; Longstreth and Nobel 1979; Perri et al. 2019). In instances of extreme weather events, interactions between environmental drivers (e.g., precipitation, king-tides) and site characteristics (e.g., topography) impact tree physiology, leading to hydraulic failure, carbon starvation, and eventually mortality (Wang et al. 2019; Zhang et al. 2021a, b; Zhang et al. 2021a, b). Thus, a thorough understanding of the impacts of such events is necessary to assess how and to what extent trees are affected by freshwater and saltwater flooding.

One crucial process in trees is sap flux (often, confusingly, referred to as sap flow), the upwards movement of water through a tree’s water-conducting xylem tissue driven by the water potential gradient between the soil and atmosphere (Bovard et al. 2005). Sap flow refers to the rate of sap movement, whereas flux incorporates the volume of sap moving at that rate. Another closely related concept is sap flux density, which is the sap flux scaled to the sapwood area of a given tree (Steppe et al. 2010). However termed, sap flux and associated transpiration are required for carbon acquisition through photosynthesis and is therefore a useful indicator of tree vitality (Wang et al. 2019). Hypoxia can reduce transpiration through root mortality, and elevated soil salinity reduces soil water potential, weakening the water potential gradient and thus the rate of sap flux. Reduced sap flux velocity is an early sign of stress in trees formerly growing in well-drained soils, in the process of conversion to a wetland ecosystem state (Wang et al. 2019; Teobaldelli et al. 2004).

There is limited data on the effects of flooding in coastal forests due to the unpredictability of storm events in observational studies and the lack of large-scale manipulative studies, making causal inference difficult (Ward et al. 2020; Yang et al. 2022). To better understand the impacts of sea level rise and increased storm frequency, the Terrestrial Ecological Manipulation to Probe the Effects of Storm Treatments (TEMPEST) experiment was designed to simulate extreme, disruptive, hydrologic events in a deciduous coastal forest on the Chesapeake Bay in Maryland, U.S.A. (Hopple et al. 2023). In this ecosystem-scale experiment, large (2000 ) plots are annually inundated with fresh or saltwater. An extensive sensor network monitors ecosystem responses (Pennington et al. 2024), including sap flow in three tree species: Acer rubrum L. (ACRU, red maple), Fagus grandifolia Ehrh. (FAGR, American beech), and Liriodendron tulipifera L. (LITU, tulip poplar).

Here, we analyze sap flux density data from TEMPEST through a robust Before-After, Control-Impact (BACI) statistical framework, with the goal of testing how species vary in their response to fresh- and saltwater flooding. We hypothesize that LITU sap flux density will exhibit a strong negative response to saltwater inundation, that ACRU will exhibit no response, and that FAGR will exhibit an intermediate response. These hypotheses stem from these three species’ physiological characteristics: LITU is a primarily upland species and thus assumed to be intolerance of flooding (McKnight et al. 1980). Conversely, ACRU is a generalist species with a known tolerance for saturated conditions (Wang et al. 2020; Ludwig et al. 2021), while FAGR is intermediate in terms of its susceptibility to flooding (Tubbs and Houston 1990; Busby et al. 2009). Furthermore, LITU occupies a higher position in the canopy, allowing for greater exposure to photosynthetically active radiation (PAR) and an increased demand for water uptake during the growing season (Leblanc et al., 2020). For freshwater saturation, we hypothesize that there will be little to no cumulative effect on sap flux density for any of the species at this early stage of treatment intensity because growth is likely water limited, in addition to the absence of the additional negative impacts of saltwater inundation.

Methods

TEMPEST experimental design and focal species

The TEMPEST experiment consists of three 50 m × 40 m unreplicated experimental plots—freshwater, brackish (i.e., estuarine, saltwater), and control, with physical separation of between 5 and 20 m—located in a mid‑successional temperate deciduous forest on the western shore of the Chesapeake Bay, Maryland, U.S.A. (38.876°N, 76.553°W). The region is classified as Köppen Cfa, a humid temperate climate characterized by warm summers, mild winters, and year‑round precipitation. Mean annual temperature is 13.6 °C, mean annual precipitation is roughly 1100 mm, and rainfall is distributed throughout the year with peaks during late spring and summer convective storms. Seasonal soil moisture is therefore highly variable, and episodic summer dry periods occur regularly in the mid‑Atlantic.

Treatment plots lie at varying distances from the Chesapeake Bay, with the control plot situated closest at 40 m from the estuary. The surrounding forest is approximately 65 years old and typical of coastal forests in the region (Box 2015), with the three focal species well represented. ACRU is a mid‑canopy generalist (Abrams 1998; Fei and Steiner 2007), tolerant of variable moisture conditions, including moderate salinity and short‑term anoxia (Ludwig et al. 2021). FAGR, by contrast, is an upland, shade‑tolerant species (Burns, 1990) that is sensitive to both drought and prolonged flooding (Stephanson and Coe, 2017; Bose et al. 2017). LITU occupies high‑quality mesic soils and exhibits rapid growth (Leblanc et al., 2020), but has relatively high water demands and is considered sensitive to drought (Canham and Thomas 2010; Adams et al. 2017). These contrasting traits provide a multi‑species perspective on how freshwater and saltwater flooding influence tree water use within this coastal forest. In addition, the freshwater treatment serves as a secondary control for the saltwater treatment to isolate the effects of salt from soil saturation.

Each TEMPEST treatment uniformly applies 300 m³ of freshwater or saltwater over 10 h in a single day, equivalent to a 15 cm rainfall event (a ~ 10‑year storm for the region). Plots were established in 2021, and treatments began in June 2022 (Hopple et al. 2023). Although estuary water from the Rhode River is mesohaline (5–18 ppt; Jordan et al. 1991), we use the term “saltwater” for simplicity when referring to the brackish treatment. Saltwater pulses in this region occur during storm‑driven estuarine surges and extreme high tides, whereas freshwater flooding results from intense rainfall events and runoff from the upland watershed. Both types of events occur intermittently—several times per decade—and are projected to increase in frequency under sea‑level rise and changing precipitation regimes.

Soil and ancillary sensor data

Temperature, soil moisture, and electrical conductivity have been continuously monitored in the TEMPEST plots since establishment. A network of TEROS 12 (Meter) sensors records temperature (°C), volumetric water content (cm³ cm^− 3, VWC) and electrical conductivity (µS cm^− 1, EC) at 15 min intervals across three depths (5, 15, and 30 cm) (Ward et al. 2025). Each plot has 41 sensors in 36 locations, ensuring thorough spatial coverage. Soils are Typic Hapludults, consisting of well-drained, fine sandy loams in the top 30 cm, with increasing clay content with depth. Below the rooting zone there is a silty clay layer, transitioning to silty sand below 1.6 m. Data from 2021 indicate no differences in pre-treatment VWC, though plots did differ in their baseline EC (Hopple et al. 2023).

Photosynthetically active radiation (µmol m^− 2 s^− 1, every 15 min, PAR) and precipitation as 24-h rainfall totals (mm) are drawn from a meteorological station located in the Global Change Research Wetland, directly adjacent to the TEMPEST site. While VWC and EC are available for the full experimental period, PAR and rainfall data are not available for the year 2021.

Temporal scope

For this analysis, the 15 days prior to and following flood events were designated as pre-flood and post-flood periods, respectively (Table 1). To serve as a control for sap flux temporal dynamics before any treatment was applied, four weeks of data were chosen from the middle of June 2021, representing the average mid-point of treatment dates. The same window in June of 2025 was selected to represent the long-term response of sap flux dynamics.

Table 1.

TEMPEST flood event dates and the “pre-flood” and “post-flood” periods used in this analysis. Each consecutive year of TEMPEST added an additional consecutive day with a 10-hour flood (e.g., TEMPEST in 2022 had one flood event over 24 h; 2023 had two flood events over two consecutive days, etc.). Note that there were no flooding events in 2021 and 2025

Year	TEMPEST Event Date(s)	Pre-Flood Dates	Post-Flood Dates
2021	N/A - no treatments applied*	May 29 - June 12	June 13 - June 27
2022	June 22	June 7 - June 21	June 23 - July 4
2023	June 6, June 7	May 22 - June 5	June 8 - June 22
2024	June 11, June 12, June 13	May 27 - June 10	Jun 14 - June 28
2025	N/A - no treatments applied	May 29 - June 12	June 13 - June 27

*300 m³ freshwater were applied once each to the freshwater and saltwater plots to test the experimental design on 25 Aug and 9 Sep 2021, respectively (Hopple et al. 2023)

Sap flux measurements

Sap flux velocity in focal species (ACRU, FAGR, LITU; n = 6 trees of each species in each treatment plot) was measured using a thermal heat probe dissipation system (Granier 1985), containing two 3.5 cm-long probes with a 2 cm effective sensing length (PlantSensors, Darwin, Australia). Both probes were inserted perpendicular to the tree, one approximately 1.3 m above the ground on the north-facing side of the tree, and the second probe approximately 10 cm below the first probe (Hopple et al. 2023). The upper probe, downstream of the lower probe in the rising xylem sap, was heated at a constant power of 0.2 W while the unheated lower probe acted as a reference for temperature difference measurements due to heat dissipation (Lu et al. 2004), a proxy that scales linearly with sap flow. The probe systems were covered with reflective insulation to ensure thermal isolation of the system. To test for possible non-uniform sap flow velocity across sapwood, 50% of the sap flow trees were outfitted with additional 5 cm-long probes with a 2 cm effective sensing length. There was determined to be no significant difference between sap flow measurements obtained from shallow-sensing and deep-sensing probes.

Diameter at breast height (DBH) was measured for all instrumented trees at 1.37 m above the ground at the time of probe installation and annually thereafter. Across treatment plots, the focal species spanned a representative range of mature tree sizes for this mid‑successional coastal forest. In the control plot (C) in 2021, ACRU averaged 30.5 ± 7.4 cm DBH, FAGR averaged 33.6 ± 7.7 cm, and LITU averaged 46.8 ± 7.2 cm. Freshwater‑plot (FW) trees were similar for ACRU (30.6 ± 7.3 cm) and FAGR (34.1 ± 5.1 cm), but included larger LITU individuals (65.4 ± 9.6 cm). In the saltwater plot (SW), DBH values were slightly lower overall, with ACRU at 23.2 ± 6.4 cm, FAGR at 28.2 ± 6.0 cm, and LITU at 48.6 ± 8.2 cm. DBH measurements supported subsequent conversion of sap flux velocity to sap flux density.

Sap flux data used for this analysis was recorded at 15-minute intervals from January 1, 2021 to December 31, 2025. During the growing season, daily cycles of sap flux are visually confirmed from sensor outputs, ensuring that sensors remain within active sapwood. Raw data was outputted as a difference in voltage between the two probes, which was then converted to a difference in temperature (℃) following Granier (1985):

1

In which = the difference in temperature between the heated and reference probe, = the maximum difference in temperature (i.e., no sap flux) between midnight and 4 am for each day, and = sap flux cm³ cm⁻² s⁻¹. The constants and =  1.231 (Eq. 1) are derived from the original Granier probe calibration and depend on the magnitude of constant power (0.2 W) supplied to the probe. We followed the unit convention for sap flux as outlined in Steppe et al. (2010). Sap flux calculations were then allometrically scaled to sapwood depth using species-specific constants derived from Brantley et al. (2016), resulting in final units of sap flux density.

Computational methods

Sap flux density data were filtered to midday (11 am − 2 pm EST), when sap flux rate was found to be at its peak, and averaged daily across this period for individual trees in each plot during our study window (Table 1; Fig. 1). Some natural mortality occurred, resulting in two ACRU being removed from the C plot dataset, and later replaced. Outliers were removed before averaging by calculating the inner quartile range for each unique species, plot, year combination and excluding the above and below 1.5 times those bounds.

Fig. 1.

Overview of midday sap flux density (cm³ cm⁻² s⁻¹) for five growing seasons: one pre-treatment year (2021), three treatment years of progressively intensifying inundations (2022 to 2024), and one post-treatment year (2025), ranging from early March to late October. Points are daily means for individual trees and lines are smoothed plot averages. Note that species vary in their y-axis scale

We first analyzed the long-term impact of TEMPEST treatments by isolating data from 2021 (pre-treatment) and 2025 (post-treatment year with no flooding event). We treated year, plot, and species as fixed effects and included their three-way interaction as well as a random effect of tree ID to explain variation in June midday sap flux density data. From this mixed-effects model (lme4 v1.1-35; Bates et al. 2003) we extracted the estimated marginal means (emmeans v1.10.4; Length & Piaskowski, 2017) and contrasted the long-term ‘Before’ and ‘After’ (Fig. 2), fluxes for each species within a given plot. Diagnostic plots and the ANOVA table for this initial test are available in the supplement (Table S1, Figure S1 A-F).

Fig. 2.

Long-term Before-After, Control-Impact (BACI) of sap flux density (cm³ cm⁻² s⁻¹) for all TEMPEST species (ACRU, FAGR, LITU) and treatment plots. Values are estimated marginal means and 95% confidence intervals. Differing letters between treatments in 2021 and 2025 indicate statistically significant change in the estimated marginal mean sap flux density across plots within that species and year combination

This initial analysis indicated that long-term treatment impacts are strongest for LITU in response to SW (relative to C) and thus we focused our second analysis on treatment-year effects on LITU responses to SW. We analyzed the response of LITU sap flux density using an iterative approach adapted from Pardini et al. (2018).

Data were fitted using a linear mixed model (Eq. 2) in which the response variable, F_avg, is sap flux density averaged by individual tree for each mid-day 15-minute interval in pre- and post-treatment periods (i.e., 15 days before and after flood events, see Table 1). BA (Before-After) and CI (Control-Impact) are fixed effects, indicating the pre- and post-flood period as well as the control and treatment plot, respectively. The BA*CI term tests for significance of disturbance effects. Random effects of individual trees (1|ID) and each year’s specific characteristics (1|Year), as well as the performance of individuals within years (1| ID: Year), were included in the model. During our iterative testing we found that EC served as a close proxy for saltwater treatments and greatly increased the explanatory power of the BACI model. All models were compared for their fit to the data using an Akaike Information Criterion (AIC) comparison.

2

The explanatory power of the model with and without the BACI interaction term was quantified using a likelihood ratio test and a parametric bootstrap test.

In addition to our primary hypotheses, we tested several assumptions of our analytical approach (Table S2, Fig. S3 A - H) and abiotic drivers (Fig. S2, Fig. S4). Diagnostic plots for visual assessment of assumptions underlying linear mixed effect models for our final model of LITU sap flux density (Eq. 2) can be found in the supplementary material (Fig. S3). To determine if differences in sap flux density before and after treatments could be attributed to changes in PAR as a driving force of photosynthetic activity and water demand, we excluded days during which the maximum PAR fell below 500 µmol m^− 2 s^− 1. We also considered that the two-week study window may be influencing our results and tested an increased temporal scope, selecting sap flux density from all daylight hours (7:00 am to 7:00 pm EST) and the full growing-season (May 15th to October 1st, Appendix I). Modified linear models excluding the BACI interaction term and random effects were also tested to determine the impacts of VWC (Fig. S2), PAR (Fig. S4), and EC (Fig. 3b) on sap flux density.

Fig. 3.

(a) Four-year trend in soil electrical conductivity (µS cm^− 1) in TEMPEST treatment plots (b) Long-term before-after control-impact (BACI) for TEMPEST soil electrical conductivity (µS cm^− 1). Plotted values are estimated marginal means and 95% confidence intervals

All analyses were conducted using R version 4.5.0 (R Core Team 2025), with data manipulation conducted using dplyr version 2.4.0 (Wickham et al. 2014) and visualizations using ggplot2 version 3.5.32 (Wickham et al. 2007). Estimated marginal means were calculated following Lenth and Piaskowski (2017). All data are publicly available on ESS-DIVE (See Open Research section and Pennington et al. 2024). Complete analytical code for main results and supplementary material is available at https://github.com/COMPASS-DOE/Srinivasan-Sapflow.

Results

Long-term effects

Four years after the initiation of TEMPEST treatments, species responses to fresh and saltwater saturation differed. ACRU was the least responsive to treatments, with no statistically significant changes in sap flux density from the pre-treatment year (2021) to post-treatment year (2025) in any plots. In contrast, there is strong evidence that FAGR sap flux density declined in C (p < 0.001) and increased in FW (p < 0.001). For LITU, all plots increased significantly over time, with SW plot LITU lagging behind FW and C (Fig. 2).

To understand which species’ treatment effects to target for further analysis, we compared post-treatment data from 2025 across plots for FAGR and LITU. FAGR sap flux density was lower in C than in FW or SW (p < 0.0001 and p = 0.003, respectively), but the two treatment plots did not differ from each other (p = 0.07). LITU sap flux density was similar in C and FW, and both of these plots had significantly higher sap flux density than SW (C vs. SW, p = 0.002; FW vs. SW, p = 0.035).

We found the low sap flux density values from FAGR in C surprising, and conducted a pairwise comparison test on FAGR sap flux density values for all years in C. This revealed sap flux density for FAGR in C to be consistently lower in 2025 than all other data years, which can be observed in Fig. 1. To confirm that similar year-to-year variability did not drive sap flux density differences among plots for LITU, we conducted the same pairwise comparison test across all years for LITU sap flux density in C. In contrast to FAGR, LITU sap flux density in C shows consistent, consecutive increase in sap flux density from one year to the next.

Treatment impacts on EC

C and SW differed in their baseline (i.e., year 2021) EC (Fig. 3a and b). Pre-treatment C soil averaged 120 ± 9 µS cm^− 1, and SW soil averaged 50 ± 4 µS cm^− 1, likely due to C’s higher proximity to the estuary. As salt from treatments accumulated in SW soil, EC increased in the plot and declined back towards ambient levels more slowly each year (Fig. 3a); by 2025, soil electrical conductivity was approximately equal between C and SW (Fig. 3b).

Progressive analysis

Short-term responses of LITU to TEMPEST saltwater treatments were examined using a progressive analytical approach. The initial model comparisons indicated that inclusion of the BACI interaction term (Eq. 2) marginally increased the explanatory power for LITU sap flux in the SW and C, both before and after treatments (likelihood ratio test: p = 0.075; parametric bootstrap: p = 0.059). When EC was also included in the model, the explanatory power of the BACI interaction term increases further (likelihood ratio test: p < 0.001, parametric bootstrap: p = 0.010). We plotted sap flux density data from two weeks before and after flood events for each treatment year (Fig. 4) and found relatively stronger long-term responses, suggesting that treatment impacts on sap flux rates persist well beyond the flooding event, and that there is a considerable lag before the full effects on sap flux density are observed.

Fig. 4.

Short-term responses of sap flux density (cm³ cm⁻² s⁻¹) to saltwater treatments for Liriodendron tulipifera in TEMPEST. Panels from left to right correspond to strong saltwater flooding treatments (see Table 1)

The results above did not depend on the specific choices made with respect to data window lengths or abiotic drivers. For example, including sap flux density data from the full growing season and all daylight hours did not differ meaningfully from that presented above (Appendix I, supplement). PAR before and after flooding events differed significantly in 2024, but was the same in all other years (Fig. S4). No long-term impacts on VWC were observed from any of the treatments (Fig. S2).

Discussion

We used a manipulative ecosystem-scale experiment to probe the long and short-term impacts of forest flooding treatments on three focal tree species – Acer rubrum (red maple), Fagus grandifolia (American beech), and Liriodendron tulipifera (tulip poplar) – that differ in key ecophysiological characteristics and occupy distinct ecological niches. Sap flux density responses to the treatments largely support our hypotheses that red maple would exhibit the most resilience to flooding treatments, tulip poplar would exhibit the most sensitivity, and beech would fall between these two extremes.

Changing soil conditions and sap flux in the TEMPEST plots

Tree growth is fundamentally but not exclusively determined by soil conditions (Passioura 2002), and the TEMPEST treatments’ first effects occur in the soil column (Regier et al. 2025). In this study, soil EC in the saltwater plot increased over the four-year treatment period (Fig. 3B). Interestingly, at the end of this period, saltwater plot EC was comparable to that of the control plot which, situated closer to the estuary, presumably experiences higher windblown salt deposition. Significant short-term EC excursions occurred during the treatment floods, as expected (Fig. 3A). In addition, over multiple years of progressively intensifying saltwater treatments, soil EC levels took longer and longer to return to baseline levels, suggesting both accumulating salt in the soil of the saltwater plot and decreasing resilience of the system (Mathes et al. 2021). Such increasing salt loads would be expected to produce ecophysiological changes limiting water uptake and carbon assimilation (Zhao et al. 2020), reducing growth (Anderson et al. 2024), and eventually increasing tree mortality rates (McDowell et al. 2022). Specific leaf area, a measure of leaf morphology, has been found to vary with saltwater exposure in this coastal area and specifically the TEMPEST plots (Bond-Lamberty et al. 2023), consistent with the expectation that tree growth responds to external stressors such as salt exposure.

Throughout our study period, neither red maple nor American beech exhibited consistent, statistically significant responses to the TEMPEST saltwater treatment. Red maple sap flux density was unchanged in both the control and treatment plots, supporting the null hypothesis of no change (Muff et al. 2022). This is unsurprising from a physiological and ecological perspective, as A. rubrum is considered a generalist with high morphological variability (Ludwig et al. 2021; Saeki et al. 2011). High levels of salt exposure can inhibit its physiology and induce early senescence (Anderson et al. 2024), but it is generally considered to be salt tolerant at low levels (Wang et al. 2020), consistent with its lack of response to experimental treatment here (Fig. 2).

We had hypothesized that any beech response would be intermediate between that of red maple and tulip poplar; in our analysis, F. grandifolia sap flux density exhibited no long-term change in the saltwater plot, but significantly decreased in the control plot and increased (p < 0.001) in the freshwater plot. This latter effect suggests an alleviation of drought stress for this sub-canopy species that generally has shallow roots and reduced growth under drought conditions (Wang et al. 2015; Bennett 1985; Elizabeth and Arain 2024); notably, 2024 and 2025 were both drier-than-normal years (Fig. S5, Keune et al. 2025). In 2024, three consecutive TEMPEST events would have supplemented the freshwater plot with 45 cm of additional ‘rainfall’ (Table 1).

In light of this, the consistent sap flux density of A. rubrum over time indicates imperviousness to year-to-year variation in annual precipitation, in addition to salt exposure, while F. grandifolia is seen to vary its sap flux density with the relative favorability of rainfall, and appears more tolerant of salinity than drought (Fig. 2). Perhaps the most surprising is the long-term increase in June sap flux density of L. tulipfera in all treatments, a species that is generally considered drought sensitive (Fahey et al. 2013). Drought sensitivity is strongly site and individual specific, with age class, genotype, and canopy position all playing a role (McGregor et al. 2021). A. rubrum is known to have especially high genetic variability in its leaf hydraulic traits (Sjöman et al. 2015). Without additional data on relatedness, genotype, root depth, or growth phenology for individual sap flow trees, our ability to assign current responses to the freshwater treatment is limited.

Over time, we found progressively decreasing sap flux density in L. tulipfera in the saltwater plot relative to the control (Figs. 1 and 4). One possibility is that sap flux density gains in the long-term are indicative of normal tree growth over the four-year period (Tyree and Ewers 1991; Steppe et al. 2010), whereas smaller increases in freshwater and saltwater reflect suppressed growth due to accumulating physiological stress. However, higher sap flux density can also result from greater canopy exposure to incident radiation, or strengthened hydraulic architecture (such as smaller vessel diameter, Meinzer et al. 2005), possibly indicating an acquisitive rooting strategy (Tran et al. 2026). Environmental variability may also have played a role in L. tulipfera year-over year increase in sap flux density (McGregor et al. 2021). Regardless of the driver, the smaller magnitude increase in sap flux density in the freshwater and especially saltwater plots indicate that these trees are not keeping pace with the hydraulic upscaling observed in the control plot, consistent with chronic stress from saturation and salinity. This pattern occurred despite naturally high soil EC in the Control plot, supporting the interpretation that TEMPEST treatments, rather than underlying site conditions, are responsible for the hydraulic suppression observed in treated trees. Post-treatment responses, captured using 15‑day pre‑ and post‑flood windows in 2022–2024 indicate that the BACI design has the strongest explanatory power when EC is included in the statistical model (Fig. 4), highlighting the central role of salinity in shaping sap flux density responses, even on shorter timescales. These findings are reinforced by field observations of premature leaf senescence in saltwater‑plot tulip poplars beginning in late 2024, consistent with previous reports of foliar damage and early senescence in Liriodendron (LeBlanc et al. 2020) and a related magnolia species under high salinity stress (Askerova 2022).

How will these specific-specific stresses, changes, and mortality interact at the ecosystem scale? Based on the effects of three TEMPEST saltwater treatments on tulip poplar sap flux density, as well as the observed leaf senescence in tulip poplar, we infer that continuing treatments will increase hydraulic stress and result in widespread tulip poplar mortality (McDowell et al. 2022; Perri et al. 2019). Similarly, if the steady sap flux density of F. grandifolia in response to saltwater treatments is indicative of stalled growth, these individuals may be less capable of dealing with other stressors, such as beech leaf-disease (Ewing et al. 2019). Conversely, the freshwater treatment may reduce baseline soil salinity, and/or provide some degree of drought alleviation during summer, which could lead to higher carbon assimilation and photosynthetic activity for trees. The juxtaposition of relatively unchanged sap flux density in response to saltwater treatment and the visible foliar damage in tulip poplar trees additionally highlights the need for further data regarding factors related to hydraulic failure, such as root productivity and leaf gas exchange, to untangle these hypotheses. Given that L. tulipifera is one of the primary canopy dominants of this mid-Atlantic coastal forest, accounting for 20–30% of basal area (A. Stearns, unpublished data), such senescence will produce a degree of ecological disruption and reorganization (Curtis and Gough 2018; Gough et al. 2022), providing opportunities for late-successional flora as well as fauna (Holm et al. 2013). Such moderate disturbances can be surprisingly difficult to model and predict (Haber et al. 2025; Ding et al. 2023), although detection of individual-tree mortality has become increasingly possible (Yeung et al. 2025; Kirwan and Gedan 2019). Finally, the frequency and duration of saltwater treatments, and timing with respect to the growing season, are crucial in driving tree stress and mortality (Anderson et al. 2024), given the overall low salinity level of the Rhode River estuary (Jordan et al. 1991).

Limitations and next steps

It is important to note some limitations of this study and caveats about our conclusions. While TEMPEST is an unprecedentedly large-scale forest flooding manipulation, it is only halfway through its planned decade-long execution, and thus our results provide a relatively short-term snapshot of tree responses. In addition, the experiment is unreplicated at the 2000 m² plot level, meaning that we rely on the BACI approach to infer treatment effects (Hopple et al. 2023); the benefit of this design, however, is that the large plots provide insight into ecosystem-scale interactions and effects.

Sap flow was measured in only 6–8 trees per species in each plot; caution and care are needed when drawing inferences about the larger population of trees (~ 150–200 total trees per experimental plot). While steps were taken to mitigate errors in sap fluxcalculations via the Granier method, we note several potential sources of error. Wounding from probe insertion can underestimate sap flow measurements (Wiedemann et al. 2016)). Furthermore, although longer probes were deployed to compare with 3.5 cm sensors and test for nonuniform sap flow velocity (Clearwater et al. 1999; Shinohara et al. 2013), these probes were installed on only 50% of the study trees and may not have been representative of all individuals. Additionally, while sap flux is a strong proxy for plant water use and overall health (Juice et al. 2016), there are numerous other physiological responses that cannot be observed through sap flux (Chang et al. 2014), and sap flux changes may or may not be linked with future mortality (Choat et al. 2018; De Kauwe et al. 2022). While there is strong evidence that salinity exposure from saltwater flooding treatments is driving short-term sap flux density responses in tulip poplar, more data will be needed to understand longer-term responses and test our hypothesis that the current results are due to a combination of rapid tulip poplar growth and the negative physiological effects of salinity.

An understanding of these in situ mechanisms is crucial to generalize TEMPEST results for predicting coastal forest vulnerability, decline, and potential mortality (Yeung et al. 2025) and the broader potential impacts of these changes (O’Donnell et al. 2024). The combination of detailed site-specific measurements of ecophysiological processes and metrics such as sap flow is key. For model-data assimilation to take advantage of large ecological networks (Zhang et al., 2026), long-term experiments such as TEMPEST are crucial for building understanding at an ecosystem and species level (Luo et al. 2011). In the face of ongoing sea level rise and climate change, the resistance and resilience of deciduous coastal forests will strongly shape the timing and extent of ‘ghost forests’ and concomitant ecosystem reorganization (Ward et al. 2020) across the terrestrial-aquatic coastal interface.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

RVS and KAM conceptualized the statistical analysis; RVS wrote the code and initial manuscript draft with mentoring from KAM; SCP and BBL contributed to the acquisition and analysis of the data and critically revised the code; RVS, KAM, JPM, and NM interpreted the results; NM ensured that all sap flow calculations had been executed correctly and led TEMPEST tree sampling experimental design; JPM provided site access and senior leadership for the TEMPEST experiment; and AES managed installation, monitoring and upkeep of the sap flow network and documented issues that contributed to data QA/QC. All authors contributed to editing of the final manuscript.

Funding

This research is based on work supported by COMPASS-FME, a multi-institutional project supported by the U.S. Department of Energy, Office of Science, Biological and Environmental Research as part of the Environmental System Science Program. The Pacific Northwest National Laboratory is operated for DOE by Battelle Memorial Institute under contract DE-AC05-76RL01830. RVS’s time was supported in part by the U.S. Department of Energy, Office of Science, Office of Workforce Development for Teachers and Scientists (WDTS) under the Science Undergraduate Laboratory Internships (SULI) program.

Data availability

The data that support the findings of this study are available for download on ESS-DIVE. All code for processing this data is available on github. >Code: https://github.com/COMPASS-DOE/Srinivasan-Sapflow. Sap flow data: doi:https://doi.org/10.15485/2479200. Ancillary data: doi:https://doi.org/10.15485/2439400.

Declarations

Competing interests

The authors declare no competing interests.