Reduced transpiration response to precipitation pulses precedes mortality in a piñon–juniper woodland subject to prolonged drought



  • Global climate change is predicted to alter the intensity and duration of droughts, but the effects of changing precipitation patterns on vegetation mortality are difficult to predict. Our objective was to determine whether prolonged drought or above-average precipitation altered the capacity to respond to the individual precipitation pulses that drive productivity and survival.
  • We analyzed 5 yr of data from a rainfall manipulation experiment in piñon–juniper (Pinus edulisJuniperus monosperma) woodland using mixed effects models of transpiration response to event size, antecedent soil moisture, and post-event vapor pressure deficit. Replicated treatments included irrigation, drought, ambient control and infrastructure control.
  • Mortality was highest under drought, and the reduced post-pulse transpiration in the droughted trees that died was attributable to treatment effects beyond drier antecedent conditions and reduced event size. In particular, trees that died were nearly unresponsive to antecedent shallow soil moisture, suggesting reduced shallow absorbing root area. Irrigated trees showed an enhanced response to precipitation pulses.
  • Prolonged drought initiates a downward spiral whereby trees are increasingly unable to utilize pulsed soil moisture. Thus, the additive effects of future, more frequent droughts may increase drought-related mortality.


Anthropogenic climate change has the potential to cause rapid drought-related forest mortality, altering land cover, hydrologic and fire regimes, and ecosystem services (Allison et al., 2009; Adams et al., 2010; Allen et al., 2010; Royer et al., 2011) with feedbacks to regional and global climate (Bonan, 2008; Jackson et al., 2008; van der Werf et al., 2009; Mildrexler et al., 2011; Pan et al., 2011). The physiological mechanisms of drought-related forest mortality are currently the subject of debate and ongoing research (Adams et al., 2009; Leuzinger et al., 2009; Sala, 2009; McDowell & Sevanto, 2010; Sala et al., 2010, 2012; McDowell, 2011), delaying their explicit representation in global climate models and dynamic global vegetation models (Fisher et al., 2010; McDowell et al., 2011). Developing the needed mechanistic understanding requires knowledge of the underlying plant responses and their physiological manifestations during the progression to mortality. Long-term data describing tree physiology during drought (Breshears et al., 2009) are rare, but rainfall manipulation experiments with appropriate controls provide a valuable perspective on the physiological cause of drought responses. In this study, we used plant responses to individual precipitation events in a semi-arid woodland as an integrated measure of plant function across treatments in a long-term rainfall manipulation experiment.

In the Southwestern USA (SWUSA), recent analyses suggest that, by 2050, the frequency and extent of drought-induced forest mortality will increase above that of the worst mega-droughts in the last millennium (Williams et al., 2012). Regional aridity has already been suggested to have increased (Balling & Goodrich, 2010; Sheffield et al., 2012), and is predicted to increase further in the 21st Century (Notaro et al., 2012; Ruff et al., 2012; Seager et al., 2012; Dai, 2013), especially as a result of drier winters (Seager & Vecchi, 2010). Low precipitation appears to be as important as rising temperature (and its effects on vapor pressure deficit) in driving vegetation impacts in the SWUSA (Notaro et al., 2012; Williams et al., 2012). Moreover, Williams et al. (2012) suggest that future vegetation mortality in the SWUSA is likely to be sensitive to possible changes in the amount and/or timing of precipitation associated with the North American monsoon (Notaro & Gutzler, 2012; Taylor et al., 2012; Cook & Seager, 2013).

Much of the SWUSA is semi-arid, with plant transpiration driven by relatively short periods of soil moisture availability following precipitation (Noy-Meir, 1973; Loik et al., 2004). Pulse response by plants is determined not only by abiotic factors such as pulse size, infiltration depth, and evaporative conditions following the pulse, but also by species-specific biotic factors such as rooting depth, hydraulic conductivity of root and stem xylem, and leaf area : root or leaf area : sapwood area ratio. Defining species according to such ‘hydraulic functional types’ (Mitchell et al., 2008) can help explain differential responsiveness to precipitation events, both among species within a site (Burgess, 2006) and within species across a temporal gradient of precipitation distribution (Williams & Ehleringer, 2000). It is unclear, though, how a species’ pulse response changes as result of changing precipitation patterns within-site, or to what extent the determinants of pulse response (e.g. cavitation resistance) also influence the survival or mortality of a species under a changing precipitation regime.

Drought intensity and duration determine plant function not only by limiting gas exchange during rainless periods but also via structural and physiological changes that alter the ability of plants to utilize soil water following precipitation (Blackman et al., 2009; Resco et al., 2009; Brodribb et al., 2010). Some physiological changes such as leaf drop or hydraulic isolation via mortality of fine roots may promote immediate survival but decrease a tree's ability to respond to soil moisture when it becomes available. Similarly, xylem cavitation, a consequence of low plant water potential, may provide water for transpiration via capacitance (Meinzer et al., 2009). During prolonged (versus diel) drought, however, cavitation has negative effects, because both woody tissue growth and xylem refilling (Bucci et al., 2003; Salleo et al., 2009; Zwieniecki & Holbrook, 2009) are apparently carbon-costly processes, and the consequence of not replacing lost conducting tissue is a lower potential for gas exchange as a result of reduced xylem hydraulic conductance. Thus, drought can initiate positive feedback whereby plants suffer increasing drought stress but are also increasingly unable to take advantage of precipitation pulses.

To determine how the precipitation regime affects the ability of plants to utilize water from individual rainfall events, we compared transpiration following precipitation in piñon pine (Pinus edulis) and oneseed juniper (Juniperus monosperma) across differing precipitation treatments of an ecosystem-scale rainfall manipulation experiment from 2007 to 2011. The drought treatment experienced 45% year-round precipitation removal, and led to significant piñon mortality (Plaut et al., 2012; Gaylord et al., 2013), while the irrigation plots received 57–112 mm in supplemental growing-season irrigation (total precipitation was 117–142% of ambient) starting in 2008. The multiple treatments in our study and their long duration allowed us to test whether pulse responses in the drought treatment were simply proportional responses to smaller events (as a consequence of the 45% precipitation reduction) or whether they were smaller than the control response would be to an event of similar size. Our goal was to determine how the transpiration pulse response of these species changes under prolonged precipitation manipulation. This analysis does not differentiate between the physiological mechanisms underlying pulse response differences, but rather focuses on the integrated, whole-plant response of the trees to long-term precipitation changes.

Materials and Methods


We sought to differentiate the instantaneous effects of pulse characteristics (event size and antecedent soil moisture) from the accumulated effects of 4 yr of drought or irrigation. To this end, we used mixed effects models to predict the transpiration responses of piñon (Pinus edulis Engelm.) and juniper (Juniperus monosperma (Engelm.) Sarg.) trees to ambient, experimentally enhanced, and experimentally reduced precipitation events. Our null hypothesis was that there was a species-specific response function that was unchanged across treatments (i.e. no treatment effect); the observed differences in absolute transpiration (Pangle et al., 2012; Plaut et al., 2012) would be explained by differences in event size and soil moisture under rainfall manipulation. For example, piñon in all the treatments would have the same response to a rainfall event of a given size. Alternatively, a model with treatment effects might be required because, within-species, trees in different treatments respond differently to similar-sized events. For instance, piñon in the drought treatment might have a reduced response to a 10-mm event because of emergent effects of prolonged drought on tree physiology such as hydraulic conductance or allometry.

Study site

Our study site was in an open-canopy woodland at an elevation of 1911 m on the eastern slope of the Los Pinos mountains in central New Mexico, USA (34°23′11″N, 106°31′46″W; Pangle et al., 2012; Plaut et al., 2012). The mean annual precipitation is 358 mm (Sevilleta LTER Cerro Montoso station #42, elevation 1976 m, 2.2 km from site, 21 yr record; and the mean annual temperature is 12.7°C.

Experimental design

The experiment consists of three replicated blocks of four treatments: (1) irrigation; (2) drought; (3) cover control; and (4) ambient control (see Pangle et al., 2012 for details). The irrigation plots were designed to receive ambient rainfall plus six c. 19-mm equivalent events, with reverse-osmosis water applied via overhead sprinklers, but full irrigation did not occur in 2008–2009 because of supply problems (Table 1). Drought plots were equipped with clear plastic troughs mounted on waist-high rails to remove c. 45% of ambient precipitation year-round. Cover control plots received the same plastic coverage, but troughs faced downward, allowing precipitation to reach the plot surface (Pangle et al., 2012). The ambient control treatment received no infrastructure. These treatments were replicated in: (1) a flat block; (2) a north-aspect block; and (3) a southeast-aspect block, selected to represent landscape heterogeneity. Within each plot, five piñon and five juniper trees that were centrally located were selected as target trees for physiological measurements. Initially, block 3 was selected for intensive physiological measurements, assuming that drought effects would be most intense on the slopes with the shallowest soils and most direct solar radiation exposure (Pangle et al., 2012). Full instrumentation of blocks 1 and 2 occurred in 2009 (see the following three sections), while block 3 was fully instrumented for physiological measurements in 2008 (Supporting Information Fig. S1).

Table 1. Irrigation dates and amounts, and annual total supplemental irrigation applied to each plot in the irrigation treatment
YearDateAmount (mm)Annual total (mm)
20073 October (system test)22
200824 June1957
15 July19
26 August19
200924 April12.569.5
19 May19
30 June19
28 October19
20105 May19112
2 June19
29 June19
3 August19
5 October17
201119 April14107
17 May19
21 June19
19 July17
23 August19
4 October19

Observed mortality and branch dieback

In August 2008, 1 yr after drought and cover control plots had been completed, droughted piñon in blocks 2 and 3 were attacked by bark beetle (Ips confusus (LeConte)) and infected with Ophiostoma fungi, and subsequently died (54% in 2008 and the remaining mature trees by July 2009), in some cases having also been attacked by twig beetle (Pityophthorus opaculus (LeConte)). Extensive stem and branch dieback occurred in drought-plot juniper in blocks 2 and 3 over the course of the experiment (Gaylord et al., 2013). By the end of 2011, 25% of drought-plot juniper site-wide had < 15% green canopy remaining, including the 12.5% that were completely dead. On nine non-drought plots, five piñon died during the experiment. When possible, dead target trees were replaced with trees of sufficient size and distance from the plot edge. After July 2009, however, no living mature piñon remained within blocks 2 and 3 of the drought plots.

Data collection and timeline

Sensor installation increased over the life of the experiment (see Fig. S1 for a timeline of measurements). Before the August 2007 start date, soil volumetric water content (VWC) was measured at −5 cm depth under each species and in open spaces on all 12 plots (model EC-20; Decagon Devices, Pullman, WA, USA) using CR-1000 dataloggers (Campbell Scientific, Logan, UT, USA). During 2007, thermocouple psychrometers (model PST-55; Campbell Scientific) were installed in profiles under every original target tree in block 3 for a total of = 10 profiles per plot (see Plaut et al., 2012 for installation details). Psychrometers were installed at −15, −20 cm, and as deep as possible (40–100 cm). Following the rapid piñon mortality already described, additional soil VWC sensors (model EC-5; Decagon Devices) were installed in similar profiles in blocks 1 and 2 in 2009. EC-5 profiles were installed under three target trees of each species and at three intercanopy locations, for a total of = 9 profiles per plot in blocks 1 and 2.

Heat dissipation sapflow probes were installed in block 3 by April 2007. Each of the five target trees per species had two sensors installed, ≥ 1 m (pathlength) from the ground, on stems ≥ 9 cm in diameter. The sapflow sensors were built at the University of New Mexico, and modified to correct for ambient axial thermal gradients associated with the open canopy structure at the site (Goulden & Field, 1994). Heated 10-mm needles were supplied with a constant 0.067 W, and sap flux density calculated according to the standard Granier equation (Granier, 1987) using a daily maximum temperature difference (ΔTmax). This amount of power consumption per probe is within the range used in other studies (Goulden & Field, 1994), and any effect on sapflow estimates would be consistent site-wide. The empirically derived coefficients in the Granier equation may introduce error in sap flux density calculation, although errors resulting from non-species-specific calibration are generally greater for ring-porous than diffuse-porous or tracheid-bearing species (Bush et al., 2010). Limited testing showed that radial decline in sapflow (Cohen et al., 2008) was minimal in piñon and juniper at this site (R. Pangle, pers. comm.), and thus the 0–10-mm sapflow rates were deemed accurate for quantification of sapflow response patterns. Furthermore, the scale of the project precluded installing probes at multiple depths site-wide. Piñon sapwood depth was consistently deeper than the 10 mm probe length, so the risk of heartwood installation was minimal. Juniper sapwood depth was circumferentially asymmetrical, and therefore great care was taken to install sensors in radial portions of the stem with the highest percentage of sapwood, avoiding dead zones. A very small subset of juniper probes with muted sap flux density, Js, and shallow sapwood depth were subsequently identified and removed from this analysis, thus avoiding the need to correct for partial probe insertion into heartwood (Clearwater et al., 1999). Sapflow sensors were measured with CR-1000 dataloggers equipped with AM 16/32 multiplexers (Campbell Scientific).

Target tree replacement and sensor replacement

Following the 2008 piñon mortality, one remaining piñon tree within the block 3 drought plot had sapflow sensors installed in November 2008; block 2 drought-plot piñon were also instrumented in 2008. Because of declining tree condition in block 2 and 3 droughted juniper, additional target trees were identified and instrumented in late 2009 (five in the block 2 drought plot and four in the block 3 drought plot). The remaining target trees in blocks 1 and 2 were instrumented with two sapflow sensors each during 2009, although, as already stated, all block 2 and 3 drought-plot piñon were dead by the end of 2009. Given the long-term nature of the experiment (5+ yr), it was unrealistic to assume that sapflow probes would remain robust and provide reliable data indefinitely. Therefore, we replaced sensors after 3 yr to ensure that our sapflow sensor network would continue to provide reliable measurements during the latter years of the experiment.

Micrometeorological data

A micrometeorological station at the research site measured air temperature and relative humidity (HMP45C; Campbell Scientific) and precipitation (TE525 tipping bucket rain gauge with Series 525 snowfall adapter; Texas Electronics, Dallas, TX, USA). Sensors measurements were recorded using a Campbell Scientific CR10X datalogger, and continuous data were summed (rain gauge) or averaged over 15-min intervals. To model responses to precipitation events of different sizes, precipitation on the drought plots was assumed to be 55% of the measured ambient precipitation after August 2007 (Pangle et al., 2012). On the irrigation plots, the application of a known volume of water over a known area allowed calculation of ‘event size’. The standard application of c. 31.23 m3 of water over an area of 1600 m2 produced c. 19 mm after accounting for 3.3% water loss in the irrigation system (see Table 1 for actual amounts).

Description of data used in the models

Our mixed effects models analyzed the response of daily sapflow to individual precipitation events in the context of environmental conditions before and after each event. We used as our response variable the mean daily sum of sapflow over the first 5 d following a precipitation event. Events were defined as days with precipitation sums > 1 mm. Thus, in the case of two precipitation events on consecutive days, only the second event was considered for the purposes of the mixed effects model, although pre-event soil moisture would reflect the first event.

Data input to the model included environmental conditions before and after each precipitation event (natural or experimental irrigation; Table 1). We represented pre-event conditions with the soil VWC at −5 cm on the day before the event and predawn plant water potential measured up to 7 d before a pulse, with no intervening precipitation. Tree-specific data were used when possible. For trees without VWC sensors, the cover type mean VWC (piñon or juniper) for the plot was used. Post-event meteorological conditions were recorded as the average midday photosynthetically active radiation (PAR) and vapor pressure deficit (VPD) within 5 d. PAR is rarely limiting at this site, and therefore it was used as a filter, and events with post-event PAR < 600 μmol m−2 s−1 were excluded from the analysis. Temperature is highly correlated with VPD, and therefore inappropriate to include in the models, but we restricted the analysis to the period 1 March to 31 October in order to exclude freezing conditions, which down-regulate tree function and also complicate the interpretation of sapflow measurements. VPD, VWC, soil water potential (Ψs), and event size were mean-centered to aid interpretation. Sufficient data were available for block 3 during the period 2007–2011 and for blocks 1 and 2 during the period 2009–2011.

The majority of root area for both these species is certainly below the −5 cm of the shallow soil VWC sensors already described. To determine pre-pulse soil moisture at greater depths, we used the deepest soil VWC sensors in the block 1 and 2 profiles and the deepest soil psychrometers in the block 3 profiles. As we did not have paired profiles in all three blocks, and the soil type and depth vary considerably, we could not generate moisture release curves. Therefore, in the modeling framework described in the ‘Mixed effects models’ section below, we treated the block 1 and 2 2010–2011 data (with VWC sensors) and the block 3 2008–2011 data (with soil psychrometers) as separate data sets.

Predawn plant water potential (Ψpd) was measured approximately monthly at the site, and on irrigation and control trees before and after each irrigation (Plaut et al., 2012; Limousin et al., 2013). A subset of those measurement dates also preceded precipitation events, thereby allowing us to examine the effects of pre-pulse plant water status on pulse response.

Mixed effects models

To test the above hypotheses, we built mixed effects models for each species (piñon and juniper) to predict mean daily sum Js over the first 5 d following each pulse (named Js in the model; see Table 2 for variable names and definitions), based on event size and soil and atmospheric conditions. Shallow (−5 cm) soil VWC (named VWCs) and either deep soil VWC (−40 to −100 cm; named VWCd) or deep soil water potential (−20 to −100 cm; named Ψs) on the day before the event, predawn plant water potential before the event (Ψpd), event size, and mean midday VPD in the 5 d following the event (named VPD) were included as fixed effects in the model (Table 2), with the rainfall manipulation treatment as an interaction term. Year, aspect (block), and individual tree ID were included as random effects (Table 2). As a preliminary step, we built general linear models using the same fixed effects but excluding the random effects. The general linear models were judged significant, but the residuals were highly non-normal and heteroskedastic. The mixed effects model approach was implemented to avoid this departure from statistical assumptions. This approach allows testing for treatment effects with respect to certain variables of interest while controlling for heterogeneity across groups (i.e. plots) and individuals (i.e. trees). By including interaction terms in the model, the fitted coefficients can be interpreted as slopes and intercepts. That is, conditional on the effects of other variables, the response to each variable has its own intercept and slope. For further details, see Methods S1, and also Pinheiro & Bates (2000) and Gelman & Hill (2007).

Table 2. Response and predictor variables for the mixed effects models
Responses J s Total daily sapflow averaged over the 5 d post-event (kg m−2 d−1)Continuous
PredictorsVPDPost-event mean midday VPD (kPa)FixedContinuous
VWCsPre-event soil VWC at −5 cm (m3 m−3)FixedContinuous
VWCdPre-event soil VWC at −40 to −100 cm (m3 m−3)FixedContinuous
ΨsPre-event soil water potential at −40 to −100 cm (MPa)FixedContinuous
Event sizeAmount of rainfall (mm)FixedContinuous
TreatmentRainfall manipulation treatmentFixedNominal



Cover control

Ambient control

YearYear of measurementsRandomNominal2007–2011
AspectAspect of the measurement blockRandomNominal

Block 1 = flat

Block 2 = north-facing

Block 3 = southeast-facing

Tree IDIndividual tree IDRandomNominal1–129

The different combinations of variables in each model were as follows: the mortality+ model included VPD, event size, VWCs, and Ψs; because soil psychrometers were only present in block 3, in this model all of the droughted piñon and some of the droughted juniper died. The mortality− model included VPD, event size, VWCs, and VWCd. The VWCd sensors were only installed in blocks 1 and 2, and as sapflow sensors were not installed in those blocks until 2009, all of the trees included in that data set survived during the time period modeled. Finally, the predawn model included VPD, VWCs, event size, and Ψpd, thereby including data from all three blocks and all 5 yr, albeit with a greatly reduced sample size because of the noncontinuous nature of plant water potential measurements. Adding Ψpd to the mortality+ and mortality− models reduced the sample size to unacceptable levels, so unfortunately we were unable to consider deep soil moisture simultaneously with plant water status.

Data set preprocessing was performed in Matlab (R2011a; The Mathworks, Natick, MA, USA) and modeling was performed in R (R Core Team, 2012) using the package lme4 (Bates et al., 2012).


The ambient and experimental precipitation variation during the 5 yr of this study provided a large variety of event conditions. Total annual precipitation ranged from a 2011 low of 152 mm in the drought treatment to a 2010 high of over 400 mm in the irrigation treatment (Fig. 1). Ambient precipitation ranged from 250 mm in 2011 to 340 mm in 2007. The drought treatment, initiated in August 2007, was probably more intense than what the trees had experienced in the past century. In Socorro, NM, 50 km southwest of the field site, 1942–1956 precipitation was 76.2% of, or 0.86 standard deviations below, the 1914–1941 mean (Western Regional Climate Center, Soccorro station, accessed 7 November 2012; In Mountainair, NM, 31 km east-northeast of the field site, 1953–1956 precipitation was 54.8% of, or 1.58 standard deviations below, the 1914–1952 mean (Western Regional Climate Center, Mountainair station, accessed 7 November 2012; Applying the experimental precipitation reduction (55% of ambient) to the 2008–2011 annual precipitation at the nearby Sevilleta LTER Cerro Montoso micrometeorological station, the effective rainfall in the drought treatment was 44.9% of, or 2.3 standard deviations below, the 1991–2007 mean, which represents the available record. The irrigation treatment, which was only fully implemented in 2010 and 2011, brought precipitation up to the long-term average, while ambient precipitation was below average for most of the study period (Table 1, Fig. 1). The mean event size was 2.1, 3.8, and 4.5 mm in the drought, control, and irrigation treatments, respectively.

Figure 1.

Precipitation over the course of the experiment showing the cumulative precipitation for each treatment in each year of the study. Drought (red line) and cover control (black line, equal to ambient) treatments were initiated in August 2007; irrigation (blue line) was implemented in 2008. The horizontal gray line indicates the 21-yr mean annual precipitation at the Sevilleta LTER Cerro Montoso meteorological station near the site; open circles indicate total annual precipitation at that station. The inset shows the distribution of ambient event size since treatment inception in August 2007.

Shallow soil VWC was lower in the drought treatment relative to the controls and higher in the irrigation treatment following the irrigations (Fig. 2a). A comparison of shallow and deep pre-pulse VWC in the two blocks with similar sensors (blocks 1 and 2) showed that drought treatment VWC was lower at both depths relative to the other treatments (Fig. 3; see Table 2 for variable definitions). Sapflow was lower in the drought treatment and higher in the irrigation treatment following irrigations, relative to controls (Fig. 2b,c). The irrigation treatment appears to have had a stronger effect on Js in 2010 and 2011 (Fig. 2b,c), when more additions were applied compared with 2007–2009 (Table 1). Mortality over the period 2007–2011 was highest in the drought treatment and higher in piñon than in juniper (Fig. 4).

Figure 2.

Time series of the treatment effect on soil volumetric water content (VWC) at −5 cm (a), and maximum midday sapflow (Js) for piñon (b) and juniper (c). Lines represent the mean of (treatment – ambient control) for each block for the irrigation (blue), cover control (gray) and drought (red) treatments. The vertical dotted line indicates initiation of the drought treatment. See Table 1 for irrigation amounts.

Figure 3.

Relationship between soil volumetric water content (VWC) measured at −5 cm (shallow; VWCs) and −40 to −100 cm (deep; VWCd) from sensors under piñon and juniper trees used in the mortality− model. Data shown are only from flat and north-facing blocks where shallow and deep sensors are of the same type. Symbols represent irrigation (blue), cover control (gray), ambient control (black) and drought (red). Error bars indicate ± 1 SE from the mean.

Figure 4.

Proportion of piñon (solid lines) and juniper (dotted lines) target trees surviving after treatments were implemented in August 2007, in the irrigation (blue), cover control (gray), ambient control (black) and drought (red) treatments. Juniper mortality is defined as 0% green canopy.

The intercept terms given in the model output (Tables 3,4, S1) describe overall treatment differences in post-pulse Js, while the slope terms indicate the strength of the response to each environmental parameter (VPD, event size, VWCs, Ψs, VWCd, or Ψpd). The significance of each term indicates whether it is different from zero (in the case of the ambient control treatment) or different from the ambient control (in the case of the other treatments).

Table 3. Summary of the mortality+ linear mixed effects model with mean daily sum sap flux density, Js, as the response variable
 Parameter estimateSEt-value
  1. VPD, vapor pressure deficit; VWCs, volumetric water content (shallow); Ψs, soil water potential.

  2. Bold values indicate significance at the < 0.05 level.

Intercept 451.9797 65.33 6.918
VPD 60.3632 34.8449 1.732
Event size 11.0568 2.2846 4.84
VWCs 4089.5435 425.4538 9.612
Ψs 22.048 10.9339 2.016
Treatment irrigation−54.633559.8945−0.912
Treatment drought300.2966 108.1564 2.777
Treatment cover control−43.61661.4605−0.71
VPD × irrigation15.402438.09390.404
VPD × drought−114.2906108.3744−1.055
VPD × cover control49.011847.69941.028
Event size × irrigation−3.4762.9223−1.189
Event size × drought−3.997713.5131−0.296
Event size × cover control6.0725.58531.087
VWCs × irrigation2705.0167 516.8926 5.233
VWCs × drought3799.1067 903.6793 4.204
VWCs × cover control2907.1832 635.3474 4.576
Ψs × irrigation0.330514.86840.022
Ψs × drought−37.526330.0502−1.249
Ψs × cover control21.043918.211.156
Intercept 375.9524 62.6675 5.999
VPD 90.9779 23.6358 3.849
Event size 5.45 1.6681 3.267
VWCs 1909.3003 279.4807 6.832
Ψs 36.3937 7.8052 4.663
Treatment irrigation34.399552.03180.661
Treatment drought133.2049 48.573 2.742
Treatment cover control28.207858.37210.483
VPD × irrigation−1.561426.6164−0.059
VPD × drought−27.544230.182−0.913
VPD × cover control56.801334.48221.647
Event size × irrigation1.1252.16720.519
Event size × drought−0.61823.7417−0.165
Event size × cover control−2.71354.1372−0.656
VWCs × irrigation−473.6862375.946−1.26
VWCs × drought1018.7232 390.9633 2.606
VWCs × cover control−14.9134483.4467−0.031
Ψs × irrigation0.871310.49360.083
Ψs × drought40.4049 10.1249 3.991
Ψs × cover control3.730613.740.272
Table 4. Summary of the mortality− linear mixed effects model with mean daily sum sap flux density, Js, as the response variable
 Parameter estimateSEt-value
  1. VPD, vapor pressure deficit; VWCs, volumetric water content (shallow); VWCd, volumetric water content (deep).

  2. Bold values indicate significance at the < 0.05 level.

Intercept 271.662 40.119 6.771
Event size 8.457 2.716 3.114
VWCs 2208.928 293.074 7.537
VWCd 2397.987 468.413 5.119
Treatment irrigation 211.232 59.25 3.565
Treatment drought154.82687.8681.762
Treatment cover control 188.165 57.06 3.298
VPD × irrigation 61.144 24.336 2.512
VPD × drought−51.55332.782−1.573
VPD × cover control−23.64825.304−0.935
Event size × irrigation 6.976 3.339 2.09
Event size × drought−1.7568.495−0.207
Event size × cover control3.6263.8440.943
VWCs × irrigation−169.484495.032−0.342
VWCs × drought−373.2291117.612−0.334
VWCs × cover control 1640.863 473.135 3.468
VWCd × irrigation989.238765.5181.292
VWCd × drought1632.9382175.0020.751
VWCd × cover control−1121.07711.829−1.575
Intercept 411.439 65.146 6.316
VPD 52.707 19.491 2.704
Event size3.8132.7971.363
VWCs 2754.063 321.298 8.572
VWCd 5400.491 842.771 6.408
Treatment irrigation117.61285.0461.383
Treatment drought−108.19984.594−1.279
Treatment cover control89.89486.5511.039
VPD × irrigation28.45222.9491.24
VPD × drought−47.14427.022−1.745
VPD × cover control−27.31824.807−1.101
Event size × irrigation 7.86 3.214 2.445
Event size × drought−8.1215.961−1.362
Event size × cover control−1.4593.627−0.402
VWCs × irrigation148.252473.3610.313
VWCs × drought−764.485517.905−1.476
VWCs × cover control1703.081 449.12 3.792
VWCd × irrigation2027.96 1006.875 2.014
VWCd × drought629.6351661.3690.379
VWCd × cover control791.431204.7720.657

The overall treatment differences indicated by the intercept term varied depending on the model. The mortality+ models included VPD, event size, VWCs, and deep Ψs as fixed effects, representing the responses of block 3 trees. The piñon mortality+ model indicated that the drought treatment had a lower Js pulse response, while the ambient control, cover control, and irrigation treatments were all similar (Table 3, Fig. 5a–d). All of the droughted piñon in this model died, so we might expect to see a reduced pulse response in this model. Further, the irrigated piñon were minimally irrigated for the first 2 yr included in this data set, which may explain why the irrigation treatment is not different from the ambient control. The mortality− model used VWCd rather than Ψs, representing the responses of block 1 and 2 trees. The mortality− model indicated that the irrigation and cover control treatments had greater Js pulse response than the ambient control and drought treatments (Table 4, Fig. 6a–d). The data set for this model includes only droughted piñon that survived (block 1), as well as irrigated trees that had been irrigated for a growing season before the years included, which may explain the treatment effects observed. We do not have a satisfactory explanation for the greater pulse response observed in the cover control treatment.

Figure 5.

mortality+’ mixed effects model predictions of mean total daily transpiration (Js) over 5 d following a pulse in piñon (a–d) and juniper (e–h) in response to vapor pressure deficit (VPD; a, e), event size (b, f), volumetric water content (VWCs) at −5 cm (c, g), and soil water potential at −40 to −100 cm (d, h) for the irrigation (blue), cover control (gray), ambient control (black) and drought (red) treatments. Data are from block 3, 2008–2011. Closed symbols on the y-axis represent intercepts that are significantly different from zero (for ambient control) or significantly different from the ambient control (for other treatments). Solid lines represent slopes that are significantly different from zero (ambient control) or significantly different from ambient control (other treatments). Dashed lines have slopes that either are not different from zero (ambient control) or are not different from the ambient control (other treatments). Shaded bars indicate the range of observed values for each treatment if different from the maximum range.

Figure 6.

mortality−’ mixed effects model predictions of mean total daily transpiration (Js) over 5 d following a pulse in piñon (a–d) and juniper (e–h) in response to vapor pressure deficit (VPD; a, e), event size (b, f), volumetric water content (VWCs) at −5 cm (c, g), and VWCd at −40 to −100 cm (d, h) for the irrigation (blue), cover control (gray), ambient control (black) and drought (red) treatments. Data are from blocks 1 and 2, 2010–2011. Filled symbols on the y-axis represent intercepts that are significantly different from zero (for ambient control) or significantly different from the ambient control (for other treatments). Solid lines represent slopes that are significantly different from zero (ambient control) or significantly different from ambient control (other treatments). Dashed lines have slopes that either are not different from zero (ambient control) or are not different from the ambient control (other treatments). Shaded bars indicate the range of observed values for each treatment if different from the maximum range.

VPD was not a significant driver of Js for any of the treatments in the mortality+ model (Fig. 5a). Event size had a positive effect on Js, which was not different among plots (Fig. 5b). VWCs also had a positive effect on Js, which differed among plots: the drought treatment had a greatly reduced response to increasing VWCs compared with the ambient control, while the irrigation and cover control responses were intermediate (Fig. 5c). Decreasing Ψs had a negative effect on Js, and the responses did not differ significantly among treatments (Fig. 5d).

In the mortality− model, increasing VPD did not affect Js except in the irrigation treatment (Fig. 6a). The positive effect of increasing event size was similar among the ambient control, cover control, and drought treatments, and greater in the irrigation treatment (Fig. 6b). This is in contrast to the mortality+ model in which all four treatments had a similar response (Fig. 5b). VWCs had a similar positive effect in the ambient control, irrigation, and drought treatments, and a greater effect in the cover control treatment (Fig. 6c). Increasing VWCd also had a positive effect on Js, which was similar among treatments (Fig. 6d).

Deep soil representation also affected overall treatment effects for juniper. Similar to piñon, the mortality+ model indicated that the drought treatment had lower Js, while the rest of the treatments were comparable (Fig. 5e–h). The mortality− model did not find any treatment effect on overall Js (Fig. 6e–h). The predawn model indicated that the controls were similar, while the irrigation treatment had higher and the drought treatment had lower Js (Table S1, Fig. S2e–h).

In the mortality+ model, all of the included environmental variables had a significant effect on juniper Js (Fig. 5). The only significant treatment interactions were that the response to increasing VPD was enhanced in the cover control treatment, and the responses to VWCs and Ψs were dampened in the drought treatment (Fig. 5e,g,h).

In the mortality− model, all of the included environmental variables had a significant effect on Js except event size (Fig. 6e–h). The cover control treatment had a dampened response to increasing VPD and VWCs, while the water addition treatment had a greater response to event size, relative to ambient control (Fig. 6e,f). Droughted juniper did not differ from the ambient control at all in this model.

In a final analysis, we used predawn plant water potential in lieu of Ψs or VWCd. The predawn model indicated that, for both species, the irrigation treatment had overall higher Js compared with the ambient control (Table S1, Fig. S2). Including Ψpd improved model fit (relative to a model without Ψpd), but pre-pulse Ψpd did not drive variation in Js except in cover control piñon and irrigated juniper (Fig. S2d,h).


Our analysis of the sapflow response to precipitation pulses across treatments highlights how precipitation patterns affect tree response to transient soil moisture following precipitation pulses. Specifically, intense drought reduces the pulse response capacity in trees that die. A key strength of the long-term data set used here is that it provides the sample size necessary to use tools such as mixed effects models to differentiate between the effects of environmental drivers and other effects of the treatments. In the mortality+ model, reduced pulse response in piñon that died was an effect of treatment independent of the environmental drivers in the model, while in the mortality− model those drivers did explain the difference between ambient control and surviving droughted piñon. Likewise, the juniper model containing the most dead or dying trees (mortality+) indicated a significant effect of the drought treatment beyond the environmental drivers of sapflow included in the model (Figs 5, 6e–h). Pulse response in the ambient treatment (both species) was enhanced by increased VPD, larger precipitation pulses, and wetter pre-pulse conditions (higher VWCs, Ψs, and VWCd; Figs 5, 6). Interestingly, Ψpd did not affect pulse response except in cover control piñon and irrigated juniper (Fig. S2d,h). We hypothesize that incomplete recovery of whole-plant hydraulic conductance, Kh, following seasonal drought may decouple plant water status (i.e. Ψpd) from Js pulse response.

Loss of hydraulic conductance may constrain pulse response in droughted piñon

Previous work has emphasized the importance of stomatal closure in minimizing cavitation in piñon during drought (Linton et al., 1998; West et al., 2008). As a result of its relatively narrow hydraulic safety margin, however, piñon still experiences reduced hydraulic conductance during seasonal drought, with subsequent recovery after rain events (West et al., 2007b). Piñon at our site experienced low xylem water potentials in 2007, before drought treatment installation (as low as −3.5 MPa; Plaut et al., 2012). The ambient trees included in the mortality+ data set recovered Kh while the drought treatment trees in that set had lower modeled Kh going into 2008 (Plaut et al., 2012), and started dying in August of that year. Reduced leaf area (Limousin et al., 2009) could also drive the lower Js (Plaut et al., 2012) and pulse response described here, but needle fall did not begin until late August 2008 and reduced new growth would not have had such a dramatic effect. Thus, we hypothesize that the effect of drought treatment which is not explained by the environmental variables included in the mortality+ model (Fig. 5) is reduced Kh. Maximum stomatal conductance (gs) was not significantly different in drought and ambient control piñon within block 1 during 2010–2011 (Limousin et al., 2013), but root and shoot specific conductivity, Ks, values were lower in droughted piñon (P. Hudson, unpublished data). Apparently, though, the Ks difference was not great enough to cause a detectable drought treatment effect in the mortality− model (Fig. 6a–d).

Loss of both leaf area and hydraulic conductance may reduce pulse response in droughted juniper

Some of the same physiological changes can be invoked to explain the mortality+ model drought treatment effect of lowered pulse response in juniper (Fig. 5e–h), but juniper differs from piñon in key attributes. Droughted juniper appear to die via a gradual loss of whole branches (Gaylord et al., 2013), in contrast to piñon, which maintains all of its foliated branches until death. Canopy loss would probably change the leaf area : sapwood area ratio and decrease the sapwood-specific metric of transpiration used here. Additionally, modeled Kh was lower in the block 3 drought treatment in 2007–2008 (Plaut et al., 2012), and in 2010–2011 block 1 droughted juniper had lower maximum gs compared with the ambient control (Limousin et al., 2013). In view of the latter result (from block 1), the lack of a significant effect of the drought treatment in the mortality− model (Fig. 6e–h) was surprising and suggests that the absolute difference in droughted juniper pulse response is attributable to drier soils and reduced event size.

Irrigation treatment responses

Two of the three piñon models (mortality− and predawn) also show an overall enhanced pulse response by irrigated piñon, relative to the ambient control. We do not have measured or modeled estimates of Kh over the time period described here, but plant water potentials in the irrigation treatment generally did not reach the minima experienced by the other treatments (Pangle et al., 2012; Limousin et al., 2013), and Kh may well have been greater in the irrigation treatment (R. Pangle, unpublished data). Maximum gs was higher in 2010–2011 for block 1 irrigated piñon (compared with the ambient control and drought treatments; Limousin et al., 2013), and those piñon may have increased their leaf area : sapwood area ratio since treatment inception (A. Boutz, unpublished data). Any of these phenomena (increased Kh, gs, or leaf area : sapwood area ratio) would increase the sapwood-specific transpiration measurements modeled here.

Irrigated juniper also had higher overall Js (Fig. 2c), but the models incorporating deeper soil moisture did not indicate a systemic increase in pulse response beyond the event size effect in the mortality− model (Fig. 6f). The predawn model did indicate an irrigation treatment effect (Fig. S2e–h); that model may be more sensitive to irrigation effects because of the preferential sampling of Ψpd before supplemental irrigations. Similar to piñon, an increase in Kh or gs, or in the leaf area to sapwood area ratio may explain the increased pulse response.

Plant water status did not drive variation in pulse response for either species

It is interesting that Ψpd had no significant effect on pulse response for either species despite the importance of the two deep soil moisture metrics (Ψs in mortality+ and VWCd in mortality−; Figs 5, 6, S2d,h). The isohydric habit of piñon, which closes its stomata at a leaf water potential, Ψl, between −2.3 and −3.5 MPa (West et al., 2007b; Plaut et al., 2012), may decouple Ψpd from Ψs under dry conditions (Plaut et al., 2012; Fig. 4). Variation in Ψs or VWCd clearly affects the magnitude of the pulse response, but if Ψpd does not reflect variation in soil moisture below some threshold, it would not have the same effect in the model. For both species, plant water status (e.g. Ψpd) may not correlate directly with hydraulic function (e.g. Kh) because of hysteresis in the relationship between hydraulic conductance and xylem water potential. Piñon roots can refill cavitated xylem following precipitation events (West et al., 2007b), but refilling could be hampered by antecedent drought length or intensity. Further, cavitated xylem in other parts of the tree may only be refilled or replaced on a seasonal time-scale. This effect may be even more pronounced for juniper, which may not even refill roots following precipitation (West et al., 2007b).

Linking reduced pulse response to mortality

Wetter antecedent conditions enhanced piñon pulse response (Figs 5, 6c,d), but droughted piñon that died did not respond at all to VWCs (mortality+ and predawn models; Figs 5, S2c). Similarly, the mortality+ model, which includes the highest proportion of droughted juniper that died, also indicated a reduced sensitivity to VWCs in those trees (Fig. 5g). This suggests that droughted trees that died did not have live absorbing roots in the shallow soil. Presumably, roots in very dry shallow soil must either be hydraulically disconnected from that soil, or be supplied with water redistributed from other parts of the root system in order to remain functional enough to respond to small pulses. If those shallow roots lose enough hydraulic conductivity, or simply die, then the resulting inability to utilize soil moisture from small precipitation events could have a critical effect on plant carbon balance. This is especially true for piñon, which was shown previously to rely on shallow roots more than co-occuring Juniperus species (West et al., 2007a). Small pulses that wet only the shallow soil are more frequent than large pulses that lead to deep soil recharge (Fig. 1 inset), and probably represent a significant proportion of annual opportunity for carbon gain. Shallow roots may therefore be a key link between hydraulic impairment and carbon limitation, which together lead eventually to mortality (McDowell et al., 2011).

The consequences of shifts between dry and wet periods may become increasingly important under climate change scenarios

An outstanding question is the effect of physiological adjustments to drier or wetter periods on survival of subsequent drought. In particular, the effects of adjustment to above-average precipitation on survival or mortality during subsequent drought remain unexamined. Drought intensity affects the ability of some species to recover function following rewetting via effects on photosynthesis and stomatal conductance (Yan et al., 2000), and hydraulic conductance of roots (Trifilò et al., 2004; Domec et al., 2010), stems (Gallé et al., 2007; Schuldt et al., 2011), and leaves (Lo Gullo et al., 2003; Blackman et al., 2009; Brodribb & Cochard, 2009). Some studies have documented lagged mortality from drought as a result of cavitation fatigue (Anderegg et al., 2013) or depletion of carbon stores (Galiano et al., 2012). Our results show that prolonged drought reduces the transpiration response to pulses of precipitation. This decreased pulse response limits the ability of trees to assimilate new carbon required to grow new tissue (absorbing roots, xylem, or leaf area) and replace carbon reserves. In semi-arid regions subject to pulsed precipitation, more intense or frequent drought may preclude complete recovery, leading to greater susceptibility to mortality during subsequent droughts.

This additive effect of drought on trees subject to pulsed precipitation suggests increased forest mortality under climate change projections. Precipitation projections span a range of uncertainty (Notaro et al., 2012; Ruff et al., 2012; Seager et al., 2012; vs Taylor et al., 2012), but the effects of temperature on VPD are better constrained and indicate that in the SWUSA trees in the future will essentially be spending more water on assimilating carbon. This increase in evaporative demand will enhance drought experienced by the trees, with predicted widespread mortality impacts (Williams et al., 2012). Our analysis supports that projection in that prolonged drought is associated with reduced pulse response and increased mortality in piñon and juniper. Any change in the hydrologic regime which results in greater frequency of dry periods would therefore contribute to this pulse response reduction.

Soil moisture dynamics are key to understanding both pulse responses and the effects of future hydrologic regimes

The Js response to soil moisture observed in both species (Figs 5, 6) highlights prior conditions as drivers of plant response following a rain event (Huxman et al., 2004; Emmerich & Verdugo, 2008; Zeppel et al., 2008; Resco et al., 2009). Reduced response in the drought treatment may illustrate the legacy effect of prolonged drought on deep soil moisture. Shallow soil may saturate similarly across treatments, but be accompanied by much drier deep soil in the drought treatment compared with the others. Indeed, the range of VWCs values for which the drought treatment overlaps with the other treatments represents drier VWCd in the drought treatment (Fig. 3).

Data from several semi-arid systems indicate that recharge of deep soil moisture is a rare but important event, which requires large storms (Reynolds et al., 2004; Yaseef et al., 2010). A key unanswered question, then, is whether a future, more intense hydrologic regime could actually reduce the soil drought experienced by semi-arid systems (Knapp et al., 2008). This appears to be the case for semi-arid grasslands (Heisler-White et al., 2009; Thomey et al., 2011), but in shrublands and forests, the greater inter-pulse drought could offset the effects of any larger precipitation events (Ross et al., 2012). We were not able to test multiple scenarios with the fixed-trough rainout shelter design, but the effect of precipitation intensity on shrubs and trees remains an important knowledge gap.


We gratefully acknowledge the efforts of Jim Elliot, Judson Hill, Nathan Gehres, Don Natvig, Renee Brown, Jennifer Johnson, Julie Glaser, Clif Meyer, Sam Markwell, Matt Spinelli, Greg Brittain, Jake Ring, and numerous undergraduate students in implementing this experiment and collecting much of the data. This project was supported by an award to N.G.M. and W.T.P. from the Department of Energy's Office of Science (BER) and to J.A.P. by the National Science Foundation's Graduate Research Fellowship Program. W.D.W. was supported by NIH Grant NCI T32 CA096520 at Rice University. The project was also supported by the resources and staff of the Sevilleta LTER (funded by NSF DEB 0620482), the Sevilleta Field Station at the University of New Mexico, and the US Fish and Wildlife Service, who provided access to the Sevilleta National Wildlife Refuge.