Microclimactic Effects of the Loss of a Foundation Species from New England Forests

. Foundation species have a major impact on biotic and abiotic processes and create a 8 stable environment for many other species. Eastern hemlock ( Tsuga canadensis ), a foundation 9 tree species native to North America, is currently declining due to infestation by an invasive 10 insect, the hemlock woolly adelgid ( Adelges tsugae ). Loss of hemlock canopies can greatly alter 11 the dark, cool, and damp microclimate of hemlock forests. We studied five years of 12 microclimatic changes following logging or girdling (to simulate physical effects of adelgid) of 13 hemlocks in a multi-hectare-scale experiment in a New England forest. Both logging and 14 girdling of hemlocks caused large changes in light availability, air and soil temperature, and soil 15 moisture. Even though the impact of logging was more rapid than the effect of gradual hemlock 16 mortality after girdling, the microclimatic changes in these two canopy treatments converged 17 over time. The microclimate in hardwood control plots, which represent the predicted forest 18 composition 50 years after hemlock loss, was intermediate between the two canopy treatments 19 and the hemlock control plots. Our fine-scale results were generally consistent with average 20 microclimatic effects observed in comparative studies but revealed additional changes in 21 variance and seasonal rhythms, and the importance of stochastic events such as ice storms. The 22 variance in air temperature, but not in soil temperature, greatly increased after loss of hemlock. 23


Introduction
Biotic and abiotic processes in forest ecosystems often are strongly influenced by the structural or functional characteristics of certain tree species (Ellison et al. 2005a).Such foundation species (sensu Dayton 1972) create locally stable conditions for other species and control essential ecosystem processes.An important foundation species in eastern North America is the late successional conifer, eastern hemlock (Tsuga canadensis (L.) Carr.) (Orwig et al. 2002, Ellison et al. 2005a).When hemlock dominates a stand, it creates a unique forest environment with a cool and dark microclimate, and acidic, slowly decomposing litter atop nutrient-poor soils (Rogers 1978, Jenkins et al. 1999, Ellison et al. 2005a).Throughout the central and southern part of its range (from Massachusetts south to Georgia), eastern hemlock is currently declining as stands are infested by two non-native insects: the hemlock woolly adelgid (Adelges tsugae Annand) and the elongate hemlock scale (Fiorinia externa Ferris) (Preisser et al. 2008).In addition to morbidity caused by both the adelgid and the scale, and mortality caused by the adelgid, hemlock has also been cut and salvaged in anticipation of further expansion of the adelgid (Orwig et al. 2002(Orwig et al. , 2012)).Because foundation tree species often define or control the local microclimate, major changes in abiotic conditions are expected as eastern hemlock declines and is replaced progressively by early successional hardwoods, including birches (Betula spp.), red maples (Acer rubrum L.), and oaks (Quercus spp.) (Orwig et al. 2002).These changes in abiotic conditions may be associated with changes in abundance of a number of animal species, including arthropods (Rohr et al. 2009, Sackett et al. 2011), birds (Tingley et al. 2002), and aquatic invertebrates (Snyder et al. 2002), as well as with changes in species interactions, successional pathways, and the rates of cycling of energy and nutrients (Orwig et al. 2008, Gandhi andHerms 2010).
Despite the importance of variance in abiotic conditions for the population, community, and ecosystem dynamics (e.g., Benedetti-Cecchi et al. 2006), most contemporary studies of forest ecology simply report overall means, for example in temperature or light availability (e.g., Lebrija-Trejos et al. 2011, Moore et al. 2011).Fine-scale temporal dynamics in abiotic conditions are rarely quantified, but time-series analysis may reveal particular or episodic events that have long-term, cascading effects on the system.Temporal changes in abiotic conditions also can provide important insights into the pace of succession or the recovery of systems following disturbance.
Here, we describe fine-scale temporal changes in microclimate in a multi-hectare-scale manipulation designed to study the response of forested ecosystems to the loss of eastern hemlock (Ellison et al. 2010).We studied in detail five years of changes in air and soil temperature, light availability, and soil moisture in forest stands in which hemlock was killed in place (to simulate death by adelgid) or removed (as in a salvage logging operation).We chose these variables because they often are tightly associated with changes in species composition and ecosystem dynamics.For example, changes in air and soil temperature, light environment, and soil moisture are associated with changes in increased ant species richness and abundance (Ellison et al. 2005b) but decreased abundance of eastern red-backed salamanders (Mathewson 2009) as hemlock stands decline and are replaced by hardwoods.Soil nitrogen mineralization and soil respiration rates (carbon flux) in many forest types, including hemlock forests, are directly related to soil temperature and soil moisture (e.g., Savage and Davidson 2001, Templer and McCann 2010, Brzostek and Finzi 2011).

Experimental design
The Harvard Forest Hemlock Removal Experiment (HF-HeRE; Ellison et al. 2010) has two types of canopy manipulation applied to 90 × 90m plots; a girdling treatment that mimics slow death by HWA infestation and a logging treatment analogous to a pre-emptive salvage logging operation aimed at extracting income from a forest stand before the adelgid would have arrived.HF-HeRE is located in two experimental blocks at the Simes Tract at Harvard Forest, a Long Term Ecological Research Site (LTER) in Petersham, MA, USA (42.47° to 42.48° N, 72.21° to 72.22° W, 215-300 m a.s.l): a "Valley" block located on a gently sloping, low terrain, and a "Ridge" block located on a forested ridge (Fig. 1).Each block consists of four plots: the two aforementioned treatments (girdling, logging) and two control plots: one dominated by hemlock, the other by young hardwoods (birches, maples, and oaks), which is the type of forest that is expected to develop after hemlock dies (Orwig et al. 2002).In the girdled plots, the bark and cambium of each hemlock (including small saplings) was girdled using a chain saw or knife.
In the logged plots, all hemlocks > 20 cm and all commercially valuable large white pine (Pinus strobus L.) and red oak (Quercus rubra L.) were cut and removed from the site with a skidder.
Plots were located in 2003, pre-treatment observations began in summer 2004, and experimental treatments were applied between February and May 2005.In 2010, the adelgid colonized the site, changing the focus of the experiment from a comparison between treatment and controls to a comparison between the physical effect of girdling and the additive or interactive effects of the adelgid itself on forest dynamics.Ellison et al. (2010) provide a full description of the experimental design and analytical approach.Here we focus on the data collected prior to the arrival of the adelgid (2005 -2009) to compare the initial response of the canopy manipulation plots to the hemlock and hardwood controls.

Light availability
Hemispherical canopy photographs have been taken twice yearly since 2005: in April before leaf bud-break by the deciduous trees ("leaf-off" condition), and in September, before the deciduous trees have dropped their leaves ("leaf-on" condition).Twenty-five photographs are taken in each plot on a 15-m grid.The photographs are taken with a Nikon 8mm fish-eye lens, from a position 1-2 m above ground.From 2005 through April 2008, the lens was mounted on a Nikon F-3 film-camera body; from September 2008 onward the same lens has been mounted on a Nikon D-3 digital camera body.The canopy photographs were analyzed using the HemiView software, version 2.1 (Delta-T Devices, Cambridge, UK).For each photograph, we calculated the global site factor (GSF), which ranges from 0 to 1 and is the estimated proportion of (direct + diffuse) solar radiation reaching the camera (Rich 1989); here, GSF measures light available to the forest understory at 1-2 m above ground.Initial analysis found no evidence of spatial autocorrelation at the scale of the 15 m grid, showed that each of the 25 photograph locations within each plot showed similar patterns over time, and that differences among them were constant (A.M. Ellison and M. Levine, unpublished data).Thus, to compare the two canopy manipulations and the hardwood controls with the hemlock controls, we used the mean GSF value from all 25 photographs in a plot at each measurement time.Raw data, along with the digitized (from film) or digitally original canopy photographs are available from the Harvard Forest LTER data archive, dataset HF-107 (Ellison 2005b).
The GSF data were analyzed with a mixed model ANCOVA, using the lme function in the nlme package in R version 2.13.2 (Pinheiro et al. 2011, R Development Core Team 2011): GSF was the response variable and treatment entered the model as a fixed factor, with four levels corresponding to the four plot types.We used block as a random factor and time as a continuous covariate; the leaf-on and leaf-off data were analyzed separately and consisted of yearly measurements from 2005 to 2009.We included both the additive and the interaction term of time and treatment in the model.The main effects of the logged, girdled, and hardwood treatments were evaluated relative to the hemlock controls.

Air and soil temperature
Air and soil temperature are measured every minute using thermocouples located in the center of each plot.Air temperature sensors are located 1 m above the soil surface; soil temperature sensors are buried in the mineral soil, ~10 cm below the surface.Every hour the mean, maximum and minimum temperatures are stored in Campbell 21-X data loggers (Campbell Scientific, Logan, Utah, USA).We began collecting temperature data for the hemlock control, logged, and girdled plots in 2005.Temperature data collection in the hardwood plots started in July 2008.Sporadic gaps (< 5% of the data) in the data have occurred due to logger failure, lightning, damage by wild animals, and other stochastic events.Thus, all data were filtered for questionable data that seemed to be caused by data logger errors (Appendix A).Average monthly air and soil temperatures were calculated from the mean hourly temperatures (raw data in Harvard Forest data archive, dataset HF-108 : Ellison 2005a).Monthly data were used for three reasons.First, biotic changes and ecosystem dynamics in these plots are routinely measured seasonally or annually, and only occasionally monthly (Ellison et al. 2010).Thus, monthly data provide finer resolution than do biotic or ecosystem data, have enough signal relative to daily or weekly noise to be interpretable, and can also be scaled up (aggregated) when assessing impacts of a changing microclimate on overall (annual or decadal) ecosystem dynamics.Second, timeseries analysis requires evenly spaced, regular data.There are consistently 12 months in a year, but there are not a regular number of weeks in a year, and leap years add additional temporal complexity.Thus, time-series analyses work most effectively with daily (or even finer-scale), monthly, or annualized data.We chose to use monthly data as it allowed us to deal with missing (daily) data in consistent ways (see Appendix A).
We analyzed the data using time-series analysis, focusing on the monthly temperature differences between the hemlock control plots and the logged, girdled, or hardwood control plots.The monthly temperature differences were monthly averages of the hourly temperature differences between the hemlock control and each of the other three treatments.We also calculated the differences among treatments in the variance of all hourly measurements in each month.We decomposed the monthly time series into seasonal, trend, and residual components using STL, a decomposition procedure based on local regression ("loess": Cleveland 1979, Cleveland et al. 1990).First the seasonal component was estimated using loess smoothing; then a linear trend was fit to the data minus the seasonal component.We used the seasonal Mann-Kendall test to test for monotonic trends in the time series (Hipel and McLeod 1994).This test was specifically developed for monthly environmental data exhibiting seasonality and missing values.The test is implemented in the Kendall package in R.
We used phenological data that have been collected at Harvard Forest since 1990 (raw data in Harvard Forest data archive, dataset HF003: O'Keefe 2000) to identify the start and end of each year's growing season.On permanently tagged individuals at Harvard Forest, bud-break and leaf development were observed at 3 -7 day intervals in spring, and leaf coloration and leaffall were observed at weekly intervals in fall.For this study, the period between 50% bud-break and 50% leaf-fall was used as an indicator of the growing season.We used data from four species -Acer rubrum, Betula alleghaniensis Britton, Quercus rubra and Q. alba L. -to calculate the mean bud-break and leaf-fall dates for the years 2005-2009.

Soil moisture
The percentage of moisture in the soil was measured in the hemlock control, logged, and girdled plots using a hand-held probe that was inserted vertically into the soil profile.

Light availability
In the logged plots, between 60% and 70% of the stand basal area was cut and removed, whereas in the girdled plots, the hemlocks died over the course of 30 months (Ellison et al. 2010).Regrowth in the understory of the logged and girdled plots became apparent around 2006, but did not reach the height of the canopy-photograph cameras until 2009 (Fig. 2).During the growing seasons (i.e., leaf-on conditions), there was a significant time × treatment interaction in light availability (F 3,40 = 11.59,P < 0.001; Fig. 3).We observed a gradual increase in light availability at the forest floor as the hemlocks died (t = 2.97, d.f.= 31, P = 0.006; Fig. 3).In contrast, we observed in the logged plots an abrupt increase in light availability at the forest floor followed by a decrease in light availability as regrowth commenced (t = -2.93,d.f.= 31, P = 0.006; Fig. 3).There was no significant change in GSF in the hardwood control plots over time (t = -0.12,d.f.= 31, P = 0.91; Fig. 3).Overall (across time), the hardwood and hemlock control plots had nearly identical GSF (t = 0.12, d.f.= 31, P = 0.91; Fig. 3), but GSF in both the logged and the girdled plots were significantly higher than those in the hemlock control plots (logged: t = 2.94, d.f.= 31, P = 0.0061; girdled: t = -2.96,d.f.= 31, P = 0.006; Fig. 3).
During leaf-off conditions (autumn, winter, and early spring), there was also a significant time × treatment interaction in light availability (F 3,40 = 6.41,P = 0.002; Fig. 3).However, only the girdled treatment showed a significant change in light availability through time, as it gradually increased (t = 2.29, d.f.= 31, P = 0.03; Fig. 3).The girdled treatment also was the only treatment that differed significantly from the hemlock controls (t = 2.29, d.f.= 31, P = 0.029; the girdled plots were brighter).

Air and soil temperature
The average monthly differences in air temperature between the canopy treatments and the hemlock controls ranged from approximately -0.4 to +2.6 °C in the logged plots and from -0.1 to +2.3 °C in the girdled plots (Fig. 4).The soil temperature differences were slightly greater, ranging from -1.1 to +3.1 °C in the logged plots and from -1.5 to +2.4 °C in the girdled plots (Fig. 4).The magnitude of the temperature differences changed over time; in general, the temperature deviations in the logged plots decreased through time, whereas the temperature deviations in the girdled plots increased through time.An exception is the soil temperature on the Ridge, where the difference between the logged plot and the hemlock control also increased over time.By 2009, the air and soil temperature deviations of the logged and girdled plots had converged in both blocks (Fig. 4).The slopes of six out the eight trends were significantly different from zero; although consistent in direction with the other treatments, in the Valley block, neither the air temperature deviation in the girdled plot nor the soil temperature deviation in the logged plot changed significantly over time (P > 0.05, seasonal Mann-Kenndall test).
There were clear seasonal patterns in the temperature deviations of the canopy treatments from the hemlock controls (Fig. 5).Both the logged and the girdled plots were warmer than the hemlock controls in summer and colder in winter.The amplitude of the seasonal pattern was greater in the logged plots than in the girdled plots, and greater on the Ridge than in the Valley.
In general, the temperature deviation from the hemlock controls started increasing before bud break, reached a peak within the growing season, and then decreased until it reached its minimum in winter.An additional striking observation was a clear saw-tooth pattern in the air temperature deviations in the Valley, consisting of a small peak before bud break followed by a larger peak after bud break (Fig. 5).This pattern was weaker on the Ridge and was almost absent in the seasonal patterns of soil temperatures.We could not seasonally decompose the hardwood control data because of the short time frame of data collection in those plots (1.5 years), but a graphical inspection of the seasonal plots suggests that the peak deviations of the hardwood plots relative to the hemlock controls occurred before the peak deviations of the girdled and logged plots relative to the hemlock controls (Fig. 5).
The monthly variance in air temperature was much greater in the logged and girdled plots than in the hemlock control plots (Fig. 6).The difference between the logged plots and the hemlock controls remained constant over time, while the difference between the girdled plots and the hemlock controls increased over time (P <0.001, seasonal Mann-Kendall test), converging towards the logged plots.The difference in air temperature variance was greatest in early spring and summer.In contrast, the differences in monthly variance in soil temperatures between the treatments and the hemlock controls were much smaller (Fig. 6).In the Valley, the deviations of the logged and the girdled plots from the hemlock controls converged over time (negative trend in the logged plots, P = 0.006; positive trend in the girdled plots, P < 0.001).On the Ridge, the differences between the treatments and the hemlock controls remained constant over time.

Soil moisture
There was no treatment × year interaction in the analysis of soil moisture data (F 2,69 = 0.39, P = 0.68), but an additive model revealed significant effects of both treatment (F 2,69 = 6.92,P < 0.001) and year (F 1,69 = 25.19,P < 0.001).Both the logged and the girdled plots were significantly moister than the hemlock controls (logged: t = 2.85, P = 0.006; girdled: t = 3.49, P < 0.001), and in all plots the percent soil moisture decreased over time (Fig. 7).In the Valley, soil moisture was highest in the girdled plot, while on the Ridge, soil moisture was highest in the logged plot.

Discussion
Microclimate influences rates of soil respiration (e.g., Savage andDavidson 2001, Brzostek andFinzi 2011), nutrient cycling (e.g., Templer and McCann 2010), and biomass production, among other ecosystem processes, as well as the distribution of many species (e.g., Ellison et al. 2005b, Mathewson 2009) in a wide variety of forests.Structural changes in the landscape can greatly alter microclimate (Chen et al. 1999), and loss of foundation species is a significant structural change.We observed large changes in both mean and variance of microclimatic variables after logging or girdling of eastern hemlock, a foundation tree species (Ellison et al. 2005a), to simulate the impacts of the hemlock woolly adelgid in the Harvard Forest Hemlock Removal Experiment (HF-HeRE: Ellison et al. 2010).
Hemlock mortality opened up the canopy and increased light availability in the girdled plots.These findings are consistent with both comparative (Orwig and Foster 1998, Jenkins et al. 1999, Orwig et al. 2008) and monitoring (Eschtruth et al. 2006) studies of infested and noninfested hemlock stands.We also observed a notable stochastic event that occurred during the data collection period: a severe ice storm that hit the northeastern United States in December 2008 (de la Crétaz et al. 2010).This singular event likely explains the increase in light availability observed in all treatments under leaf-off conditions in 2009 (Fig. 3).Overall, the girdled plots suffered more damage from the ice storm because the standing boles were already structurally unsound; data collection is underway to test this hypothesis (Ellison and Barker Plotkin 2009).
Air and soil temperature are very sensitive to changes in canopy cover; even small disturbed patches can have an increased temperature and a more variable microclimate (Mladenoff 1987, Chen et al. 1999).Adelgid infestation has been predicted to lead to an increase in soil temperature (Jenkins et al. 1999), andOrwig et al. (2008b) found a strong correlation between soil temperature and the thinning of hemlock crowns.Our findings are consistent with these patterns, but also highlight changes in variance and seasonal variability.For example, air and soil temperature differences between treatments and controls were greatest in summer, when the amount of solar radiation is highest.The amplitude of the seasonal pattern was greater in the more exposed logged plots and plots on the Ridge.The small peak in the saw-tooth pattern we observed in both the logged and the girdled plots (Fig. 5) occurred before the start of the growing season and we hypothesize that it is caused by the insulating hemlock canopy preventing snow from melting.Differences in air and soil temperatures between hardwood and hemlock controls also were greatest at this time.Although air and soil temperature showed very similar patterns in mean monthly temperatures, only air temperature showed a large increase in monthly variance compared to the hemlock controls (Fig. 6), most likely due to the greater thermal inertia of the soil.
Although soil moisture tends to be relatively high beneath hemlock canopies because of hemlock's relatively constant but low transpiration rate, we expected that soil moisture levels would initially increase as hemlock dies and forest transpiration declines.Subsequently, soil moisture should decrease beyond initial conditions as hemlocks are replaced by deciduous species that transpire approximately twice as much water as hemlock (Catovsky et al. 2002, Hadley et al. 2008).Our results supported this hypothesis.Soil in the logged and girdled plots was consistently moister than in the hemlock controls (Fig. 7).Five years after the canopy manipulations, decreased evaporation due to hemlock loss still has a greater impact on soil moisture at our site than does increased evapotranspiration by new hardwood trees; similarly, Jenkins et al. (1999) and Orwig et al. (2008b) found no difference in soil moisture between infested and non-infested stands.However, Kizlinski et al. (2002) found that much older, logged and adelgid-infested sites were dryer than recent harvest or undamaged sites.We note that our soil moisture data were collected only during the summer months, and do not present a complete picture of changes in this important variable.Continued monitoring of the HF-HeRE plots, extension of data collection into other parts of the year, and linkages with available precipitation data at weather stations within 10km (Boose 2001) all will allow us to better test our hypothesis that soil moisture levels should eventually decrease below initial levels under intact hemlock stands.
Overall, we observed a rapid change in microclimate in the logged plots and a slower change in the girdled plots, but over time the two treatments have converged in their responses.
Although we had fewer data from the hardwood control plots, the microclimate of the hardwood plots appears to be intermediate between the canopy treatments and the hemlock controls.
Continued monitoring of these experimental plots will reveal whether the logged and girdled plots will diverge again in the future, whether they will become more similar to the hardwood controls, and how these changes will impact biological diversity and ecosystem processes in these changing forests.As the spread of the adelgid continues (Fitzpatrick et al. 2011), it is necessary to make as complete an assessment as possible of the effect of the decline of eastern hemlock -our canonical foundation tree species -on forest ecosystems in eastern North America (Albani et al. 2010).

Figure Legends
Measurements were done 1 -3 times a month during the summer months (June, July, and August) of 2006 -2009.This method provides an estimation only of the relative differences in soil moisture among plots.Monthly averages of percent soil moisture were calculated for June, July and August of each year.The data were then analyzed in the same way as the light availability data using a mixed model ANCOVA, with the soil moisture percentage as the response variable and the year as a covariate.These data span only2006-2009, no  soil-moisture data were collected in 2005 (raw data in Harvard Forest data archive, dataset HF-130: Davidson and Savage 2009).

Fig. 2 .
Fig. 2. Canopy photos taken in the hemlock and hardwood control plots in 2009 (top) and

Fig. 3 .
Fig. 3. Mean global site factor (GSF) per plot in the Valley and Ridge blocks, during leaf-

Fig. 4 .
Fig. 4. Monthly means of hourly differences in air (top) and soil (bottom) temperature

Fig. 6 .
Fig. 6.Monthly variance in air (top) and soil (bottom) temperature of the two canopy

Fig. 7 .
Fig. 7. Percent soil moisture of the logged, girdled and hemlock control plots in the