The study was conducted in the Agua Salud Project site within the Panama Canal Watershed (9◦13′ N, 79◦47′W, 330 m amsl). We worked in the 30-ha T. grandis plantations that were established in 2008. Trees were planted at a spacing of 3 m between trees and 2.6 m between rows (total of 1,111 trees per hectare), such that trees were planted in a triangular configuration to reduce potential erosion on steeper slopes. Prior to plantation establishment, the land was cleared of forest in the 1970s with the predominant land use being cattle grazing (Weber and Hall, 2009). Since tree establishment in 2008, four yearly understory cleanings occurred from May through August to prevent additional competition with the planted trees. The cleanings reduced to three and then two times a year as the tree canopies began to close. Teak trees were also pruned to maintain 4.9 m (16 ft) of a branch free bole. Teak is characterized as dry season deciduous, but we found it does not lose all of its leaves at once at our site (Fernández-Moya et al., 2014). The topography of the area is characterized by both flat areas and areas with short and steep slopes (Hassler et al., 2011; Mayoral et al., 2018) and the soils are silt clay to clay with pH values ranging from 4.4 to 5.8 (Batterman et al., 2018).

The sampling design was meant to estimate water use in T. grandis stands. During June of 2014, we established five teak plots, measuring 25 m by 25 m in area with <10 degrees of inclination to avoid logistical problems with the sapflow cables. At all sites, trees were interacting aboveground, but were not overtopping one another. Between June and August of 2014 and 2015, all trees within each plot were measured for height and diameter at breast height (DBH, 1.3 m above ground). A total of 351 trees were inventoried. Between July and August of each year, the understory of the 30 ha teak plantation was cleaned, which included the teak plots where inventories were established.

In June 2014, we established four subplots within a subset of the monocultures to measure sap flow in selected trees (hereafter referred to as sap flow trees). We selected eight trees per plot for sap flow measurements based on the following criteria: (1) The tree crowns had to be interacting (i.e., the stand has reached canopy closure); and (2) The trees had to be within a 10 m radius of each other to be within reach of the sap flow cables. A total of 32 trees were measured, 17 of which were used for analysis. Trees were excluded from analysis if they were missing >20% of data points. Large data gaps were caused by ant attacks, termite infestations, or lightning. We refer to the two plots that were not thinned as Control 1 and Control 2.

Two of the teak plantation plots received a thinning treatment in June 2015, and the other two were left unthinned as controls (Table 1). The thinned plots are referred to as Treatment 1 and Treatment 2. The thinning treatment consisted of removing 50% of the individual trees, which is around the percentage of trees thinned in Panama teak plantations (Griess and Knoke, 2011). Trees were felled with a chainsaw or machete in a direction to avoid damage to study trees. Sap flow trees in the thinned sites had three competing trees removed so that each sap flow tree had three neighbors after the thinning. In the unthinned stands, each sap flow tree had 6 neighbors during the study period.

Central Panama experiences annual dry seasons that start in mid-December and end in mid-May, with the remainder of the year (May through early December) being considered the wet season. During the period of the study, the El Niño drought occurred such that the normal wet season was dry, essentially connecting the dry season of 2015 with the dry season of 2016, creating what we refer to as a prolonged drought. We selected the Standard Precipitation-Evapotranspiration Index to assess drought conditions based on the recommendation of Slette et al. (2019) for methods to define drought. We calculated the index using the SPEI package in R (Begueria and Vicente-Serrano, 2013). The SPEI calculates potential evapotranspiration (PET) using the Penman-Monteith equation. From the package, PET is calculated using the monthly minimum temperature, maximum temperature, average wind speed, sun hours, the site latitude, and site altitude in meters above sea-level. Negative values on the SPEI index (<−1) indicated drought conditions. We calculated the index based on 1 month periods and 6 month periods. We calculated the SPEI using data between 2002 and 2019. Data from 2002 to 2015 were collected from a nearby meteorological (MET) station on Barro Colorado Island. Data from 2015 onward were collected from the Agua Salud MET station (more details below). We also calculated monthly precipitation totals and subtracted precipitation from the PET for a 10 year period from 2009 to 2019 in order to determine the water deficit or surplus.

Two MET stations located within the Agua Salud Project study area collected local climate data for the 2014–2015 study period. From June 2014 through January 2015, MET data were collected from the Property 2 site while data after February 2015 were collected from the Celestino Tower site. Climate data from the Celestino Tower included air temperature (°C) and relative humidity (RH, %) using an HMP60 (Vaisala, Vantaa, Finland), and precipitation (mm) using a 260–250-A tipping bucket (NovaLynx, CA, USA), Vapor pressure deficit (VPD, kPA) was calculated from the air temperature and RH data following (Allen et al., 1998). Small gaps in the dataset exist due to either sensor malfunction or due to the transition period between taking down the Property 2 tower and constructing the Celestino Tower.

The 2014 dry season began December 21, 2013 and ended May 6, 2014 and the 2015 dry season began December 14, 2014 and ended May 16, 2015 (Paton, 2016). Mean annual rainfall for 2014 was 2,203 mm and for 2015 it was 1,810 mm (Paton, 2016). Generally, about 80% of the average annual precipitation falls between May and mid-December. Mean daily maximum and minimum temperatures are 32 and 23°C, respectively (http://striweb.si.edu/esp/physical_monitoring/descrip_bci.htm).

Soil volumetric water content (VWC) was measured using DeltaT PR2 sensors (DeltaT, Cambridge, United Kingdom) at six soil depths (100, 200, 300, 400, 600, and 1,000 mm) starting in December 2014. At each site, three trees with sap flow sensors were randomly selected. An access tube was placed 0.5 m in a random cardinal direction from the bole of the tree. Soil moisture measurements were collected for each tube every 1–4 days. A mean VWC was calculated for the first three depths (100, 200, and 300), where changes in VWC throughout the year were apparent. An ANOVA with a post-hoc Tukey test was performed to analyze mean VWC difference between plots.

We analyzed data from June 15, 2014 through December 15, 2015. Tree-level water use estimates are based on data obtained from individual sap flow trees. Stand-level measurements are scaled using plot inventories and the relationship between DBH and water use (Hernandez-Santana et al., 2015). Sap flow was measured using the heat ratio method (HRM; Burgess et al., 2001). On each tree, one sensor was installed 1.30 m above the base of the tree facing north. Each sensor contained three probes (a heater probe and two temperature probes, installed equidistantly upstream and downstream from the heater probe, 0.6 cm). Each temperature probe contained three thermocouples located at 0.5, 1.7, and 3.0 cm from the bark of the tree. A heat pulse was automatically sent to the sensors every 15 min. The speed of the heat (Vh) was calculated according to Burgess et al. (2001) every 15 min. Heat pulse velocities were corrected (Vc) for errors (probe misalignment and wounding) following Burgess et al. (2001). Estimates of each tree's daily sap flux density (J_s) were obtained from Vc (Green et al., 2003):

where ρ_d is the density of sapwood, ρ_s is the density of water, MC is the volumetric water content of the sapwood, C_dw is the thermal conductivity of dry wood, and C_s is the thermal conductivity of water.

Cores were taken from trees where sensors were not actively measuring sap flow to determine conducting sapwood area. No cores across the range of sampled tree diameters (7.1 to 13.3 cm at breast height) revealed heartwood, so the entire diameter cross section of each study tree was treated as conducting sapwood for the transpiration calculations. We found a significant but weak relationship (p < 0.001, R² = 0.04) between DBH and whole tree water use. We calculated the daily water use of each sap flow tree (Equation 2):

Where Q is the daily water use of an individual tree and DBH is the diameter at breast height of the tree. Each Q was summed for the site and then divided by the ground area of the site to measure transpiration in mm day⁻¹.

The middle sap flux density (thermocouple 1.7 cm from the bark) was divided by the outer sap flux density (thermocouple 0.5 cm from the bark) to determine where sap flux density was greatest throughout the study period. If the sap flux ratio (J_s ratio) is >1, the sap flux density is greater in the outer portion of the tree than the middle. Conversely, if the ratio is <1, the sap flux density is greater in the middle portion of the tree than the outer. We calculated the ratio for every 15 min reading and then averaged the ratio by week for thinned and unthinned sites across the study length (1.5 years).

Relative growth rate (RGR) was calculated for all sap flow trees (n = 32) using aboveground tree biomass (AGB) and (Equation 3):

For the year before thinning, RGR was calculated using the 2015 (AGB₂) and 2014 (t₁, AGB₁), while 2016 (AGB₃) and 2015 (AGB₂). Diameter at breast height measurements for each year were used to calculate AGB based on equations from Kraenzel et al. (2003).

Leaf area index (LAI) of the teak canopies were estimated using hemispherical photography photos. Photos were taken at two different time points—May 2015 (pre-treatment) and October 2015 (post-treatment). The teak canopies were fully foliated in May 2015 and October 2015, but during October 2015, the photos represent the end of the drought period. Photos were taken with a Canon Eos Xs 1000D (Canon, Ota, Tokio, Japan) with an attached Sigma 4.5 mm fisheye lens with an equisolid projection (Sigma, Ronkonkoma, New York, USA). A total of six photos were taken per plot at three different time points. Tubes were placed at each point so the same location could be used over the study period. At each tube, three photos were taken with a self-timer and automatic exposure compensation and bracketing at −2.0, −1.0, and 0. The camera was placed on a tripod so that the lens was one meter above the ground and level. If there were leaves at a distance <0.6 meters from the lens, they were pulled from the view of the lens. Photos were taken only if there was no direct sunlight or rain (typically occurring during the late morning. We took photos during the wet season prior to the drought/thinning and after the drought/thinning. Hemispherical photo analyses were performed with the Hemiphot package (ter Steeg, 2018).

All statistical analyses were performed in R (R Core Team, 2017) and figures were produced with ggplot2 (Wickham, 2016). The lme4 package (Bates et al., 2015) was used to perform mixed effects models to determine differences in (1) pooled J_s by temperature and VPD and (2) stand transpiration by treatment. All mixed effects models included an additional interaction of pre-treatment or thinning. The dabestr package (Ho et al., 2019) was used to create estimation graphics for both RGR and LAI analyses. We selected the Ho et al. (2019) method of estimating effect sizes and their uncertainty to take advantage of plotting full sampling curves and effect sizes to show the variability in the data.

Based on the SPEI calculation, the time period of the study that experienced a drought in 2015/2016 coincided with an El Niño Southern Oscillation Event (ENSO). The SPEI was <−1 (moderately dry) at the beginning of the ENSO event, approaching −2 (extremely dry) toward the end of the ENSO event (Figure S1). Prior to and after the ENSO event, the annual precipitation was greater than the annual PET. There was always a PRCP-PET deficit during the dry seasons, but the deficit was longer (8 months) during the ENSO event than the other years (4–5 months). Additionally, the higher PRCP-PET during the ENSO event was lower than the previous years (Figure S2).

After thinning, 50% of the trees were removed in two of the plots. In the other two control plots, no trees were thinned. Around the same time of the thinning (June 2015), Panama experienced an El Niño Southern Oscillation Event (ENSO). At the sites, there was a 54% decrease in precipitation in comparison to the previous year (Figure 1E) and nearly a 32% reduction from the average (since 1927) in cumulative rainfall, based on a nearby site of Barro Colorado Island (Paton, 2016). The ENSO event also corresponded with increases in net radiation (40%) and VPD (20%). Volumetric water content (VWC) was not significantly different between treatments at the beginning of the study (dry season 2015) but started to become significantly different after the thinning. Notably, the thinned sites had 10% higher VWC than the unthinned sites from July until October 2015 (p < 0.001). After October, the VWC of the unthinned and thinned sites were no longer significantly different.

J_s of teak generally increased with increasing VPD and temperatures, independent of treatment (Figure 2). At 28°C, J_s began to decline as VPD increased, with the exception of 34°C. The slopes of the lines ranged from 1.20 for 20°C to −0.01 for 28°C, and did not follow a consistent pattern (i.e., decreasing slopes with increasing temperatures). All models relating J_s with VPD and temperature were significant, but the R² ranged from 0.24 to <0.01. Unlike the relationship of VPD and temperature, radiation had a pattern of decreasing slopes with increasing VPD. J_s increased with increasing VPD and net radiation, until radiation reached 450 W m⁻² (Figure 3), and notably, the rate of change decreased at each increase in net radiation of 50 W m⁻².

Prior to thinning, both treatments had trees with similar J_s (Figure 1A), except for a period immediately preceding the thinning, possibly due to a small rainfall event and differences in leaf cover among sites. The thinned sites began to have greater J_s that corresponded with increases in VWC and the thinning, with the exception of a brief period in August when J_s did not significantly differ. An increase in the ratio of outer to middle J_s occurred in the thinned sites after the thinning (with no change in the control sites; Figures S3).

Relative growth rates (RGR) for all trees declined during the drought (Figure 4). The effect size was calculated to compare the thinning and drought treatment to the control (unthinned sites). The RGR effect size was negative for thinning + drought. The distribution of changes for RGR was greater for sites that were thinned than for the controls (Figure 4). Leaf area index (LAI) before the drought/thinning and after the drought/thinning did not vary (Figure 5). The effect size of the drought year was not significantly different from the control, regardless of whether the sites were thinned. Stand transpiration (E) was significantly different among sites pre-treatment, with one site having the greatest E with 7.60 mm day⁻¹. All sites had lower E during the drought/post-thinning period. The two thinned sites had lower E (2.94 and 1.77 mm day⁻¹) than one of the control sites which had the highest E during the second half of the study (3.50 mm day⁻¹) (Table 1).

Stand transpiration was on average lower in the thinned sites than the unthinned sites because of the reduction of trees. Additionally, thinning increased tree sap flux density for a short period and not enough to compensate for the lower stand density experienced in the thinned sites (Figure 1, Table 1). This result suggests that stands can be thinned to regulate stand transpiration. During the dry season or a drought event, when there is significantly less rainfall than the wet season and a normal year, respectively, reducing the evaporative demand of plantations by reducing the density can potentially benefit both individual tree growth and downstream water resources in the short-term, although long-term changes (i.e., 1 year +) in stand transpiration are still not fully understood. Stand transpiration was notably variable (Table 1), which may be partially attributable to the differences in stand LAI (Figure 4), but also could be caused by differences in soil nutrients or even slope, which have an effect on the growth of some native species in the adjacent area (Mayoral et al., 2018). All of these factors, when scaled to the stand-level, can result in significant differences in transpiration.

Although we expected teak to reduce LAI during the drought (like it does during the dry season), the teak trees did not fully lose leaves. In fact, LAI trended toward an increase during the drought regardless of whether the stands were thinned or not (Figure 5). Higher growth rates and altered leaf flushing to take advantage of the higher than normal radiation experienced during an El Niño event is common among some species in secondary forests (Detto et al., 2018; Schnabel et al., 2019). Our work shows, however, that even within the same species, drought can differentially affect tree-level and stand-level growth rates, J_s, and transpiration.

One limitation of our conclusions at the stand-level is that the number of plots is a constraint and limits the generalizability of our results. However, this is a common limitation inherent to sapflow studies, given the high costs, labor, and resources involved in fully instrumenting and maintaining the plots. Although plot sample size was small, we were able to collect sapflow data on 32 trees, which we argue provides sufficient statistical power to draw robust conclusions for our specific study sites about sap flux density and stand transpiration.

Although thinning may decrease competition for soil moisture, the remaining trees after the thinning may experience higher levels of radiation on the sides of their crowns. Simultaneously, greater exposure of soil to radiation may increase evaporation from the upper soil layer faster than in unthinned plots due to reduced shading by trees. Based on thinning principles, we expected that relative growth rate of released trees to be greater than the year before the release, but we did not find this to be true. Both the controls and thinned stands had lower RGR during the ENSO event (Figure 4) suggesting that the drought had a negative impact on aboveground growth that overrode any positive growth impacts associated with thinning. However, despite decreases in RGR, LAI a year after thinning and the drought returned to pre-thinning levels (Figure 5). Since RGR is based on increases in DBH, this metric does not represent any below ground growth that could have been occurring in response to drought (Markesteijn and Poorter, 2009).

Thinning had an initial effect on tree J_s, but this effect subsided when precipitation increased. Initially, there was an increase in tree J_s after the thinning compared to the trees in the control sites, despite the presence of a drought. During a brief rainy period during the drought in September where precipitation was >PET (see Figure S1), J_s of the trees in thinned and control sites were not significantly different. However, when PET shifted and was higher than precipitation, the differences between J_s were more pronounced, with tree J_s in thinned sites being higher than J_s of trees in control sites. Soon after, when precipitation returned to more normal conditions and precipitation was higher than PET, J_s of both treatments were no longer significantly different. The results from PET and precipitation also follow closely with VWC, in which VWC was initially higher in thinned sites than controls, but closely tracked differences in PET and precipitation. Based on these results, the increase in tree J_s due to thinning was significant, but the effect was temporary (<6 months) and did not necessarily override the effect of the drought. This suggests that thinning might be a useful tool to use when trees are experiencing water shortages (e.g., low VWC), to increase access to VWC while a drought persists. However, since the effect appears short for teak, thinning may be more useful for longer droughts or for species that are slower to recover LAI. Although not directly tested here, thinning can affect the boundary layer and canopy roughness, which can result in trees losing more water. For example, trees growing next to forest edges in the Amazon and spruce trees in Denmark have higher water use due to changes in the boundary layer (Hernandez-Santana et al., 2011; Ringgaard et al., 2012; Kunert et al., 2015). A change in the boundary layer, combined with a reduction in competition for soil water for remaining trees after thinning, can in the short-term increase water use. However, for open stands (like the ones in this study), droughts can dry out the top soil layer faster, limiting the accessibility of water for the remaining trees, which will fade the effect of thinning over the long-term.

Also notable is that there was a transition to higher J_s in the outer portion of the sapwood area after the thinning (Figure S3). Generally, J_s is higher toward the bark of the tree, where newer wood is being added (Alvarado-Barrientos et al., 2013). The change to higher J_s in the outer thermocouple post thinning, despite the drought, can be explained by the LAI of the thinned stands. The fact that the thinned stands were able to return to pre-thinned LAI within a year, suggests an increase in leaf production, which would necessitate greater water use by the tree, primarily in the outer portion of the tree which is preferentially connected to the newest leaves. Similarly, the control plots did not have an increase or decrease in LAI between the pre-treatment and treatment + drought years, which is expected.

In addition to tree species or site-specific characteristics (Hassler et al., 2018), atmospheric conditions and soil water availability are the primary controls of hourly, daily and yearly changes in sap flux density (J_s) (Bovard et al., 2005). While some meteorological variables that affect J_s are highly correlated with one another, others can interact and have opposite effects on J_s. For example, while VPD, temperature, and radiation are tightly linked, they can have opposite effects on J_s. For example, J_s increases in the morning with increasing VPD and then often declines in the afternoon because radiation forces stomatal closure despite similar VPD levels (Zeppel et al., 2004).

Although J_s generally increases with VPD and temperature, we found that teak hits a temperature and radiation threshold at high VPD (Figures 2, 3). Irrespective of treatment, J_s increased with increasing VPD until 28°C, at which point, J_s began to decline. In contrast, locally adapted species in adjacent sites had temperature thresholds up to 30°C (Sinacore et al., in revision). Given that climate change induced temperature increases are expected at our sites and may influence the ability of teak to regulate water, these results are particularly notable, especially when compared to native species alternatives. Radiation also had a negative effect on J_s at higher VPD, with the threshold 450 W m⁻², which was near the maximum radiation during the course of our study. Despite trees in tropical forests generally having positive responses to increasing VPD and radiation (i.e., through higher growth rates), predicted increases in temperatures may cause an opposite effect in the future, a concept that requires further investigation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.