Infrared heater system for warming tropical forest understory plants and soils

Abstract The response of tropical forests to global warming is one of the largest uncertainties in predicting the future carbon balance of Earth. To determine the likely effects of elevated temperatures on tropical forest understory plants and soils, as well as other ecosystems, an infrared (IR) heater system was developed to provide in situ warming for the Tropical Responses to Altered Climate Experiment (TRACE) in the Luquillo Experimental Forest in Puerto Rico. Three replicate heated 4‐m‐diameter plots were warmed to maintain a 4°C increase in understory vegetation compared to three unheated control plots, as sensed by IR thermometers. The equipment was larger than any used previously and was subjected to challenges different from those of many temperate ecosystem warming systems, including frequent power surges and outages, high humidity, heavy rains, hurricanes, saturated clayey soils, and steep slopes. The system was able to maintain the target 4.0°C increase in hourly average vegetation temperatures to within ± 0.1°C. The vegetation was heterogeneous and on a 21° slope, which decreased uniformity of the warming treatment on the plots; yet, the green leaves were fairly uniformly warmed, and there was little difference among 0–10 cm depth soil temperatures at the plot centers, edges, and midway between. Soil temperatures at the 40–50 cm depth increased about 3°C compared to the controls after a month of warming. As expected, the soil in the heated plots dried faster than that of the control plots, but the average soil moisture remained adequate for the plants. The TRACE heating system produced an adequately uniform warming precisely controlled down to at least 50‐cm soil depth, thereby creating a treatment that allows for assessing mechanistic responses of tropical plants and soil to warming, with applicability to other ecosystems. No physical obstacles to scaling the approach to taller vegetation (i.e., trees) and larger plots were observed.


| INTRODUCTION
Ecosystems worldwide are responding to Earth's changing climate, with important implications for their structure and function, as well as for determining central feedbacks to future climate (IPCC, 2014).
These ongoing changes and associated consequences underscore the value of understanding and forecasting the effects of climatic change, yet our knowledge of how systems will respond to further increases in temperature remains notably poor (e.g., Shao, Zeng, Sakaguchi, Monson, & Zeng, 2013). This is particularly true for tropical forested ecosystems, which sustain exceptionally high biodiversity, cycle vast amounts of carbon, and represent a key uncertainty in considering terrestrial ecosystem feedbacks to future climate (Cavaleri, Reed, Smith, & Wood, 2015;Pan et al., 2011;Wood, Cavaleri, & Reed, 2012).
However, while the need for an improved understanding of warming effects on tropical forests is clear, enhancing our understanding of temperature effects on these and other ecosystems remains a significant research challenge (Cavaleri et al., 2015).
Common methods for investigating temperature controls over forest composition and processes include laboratory and greenhouse experiments, soil incubations, and studies along elevation gradients (e.g., Chen, Yu, González, Zou, & Gao, 2017). While the use of field experimental warming has been less common in forested ecosystems, it holds much promise for elucidating the consequences of increasing temperatures. The majority of forest warming experiments that currently exist have primarily been implemented in temperate forests using underground cables (Melillo et al., 2002;Pries, Castanha, Porras, & Torn, 2017) but there are other viable methods. Using infrared (IR) heaters is an approach to field warming that enables warming of the vegetation and soils under free-air, open-field conditions, with much less disturbance than buried cables (e.g., Jarvi & Burton, 2013). IR arrays that maintain a constant temperature rise of the heated plots above that of reference plots using proportional integrative derivative (PID) control have been reported for several ecosystems including grass (Nijs et al., 1996;Van Peer, Nijs, Reheul, & De Cauwer, 2004), tundra (Marchand, Mertens, Kockelbergh, Beyens, & Nijs, 2005;Nijs et al., 2000), grazing land (LeCain et al., 2015;Luo et al., 2010), wheat fields Wall, Kimball, White, & Ottman, 2011), paddy rice (Gaihre et al., 2014;Rehmani et al., 2011), soybean (Ruiz-Vera et al., 2013;Siebers et al., 2015), and maize (Ruiz-Vera, Siebers, Drag, Ort, & Bernacchi, 2015), as well as tree saplings (McDaniel et al., 2014;Rich et al., 2015). However, all of these studies had 3-m-diameter plots or smaller, and most warmed by only 1.5°C during daytime. Although short-term heat-wave simulations have achieved higher degrees of warming (e.g., De Boeck, Dreesen, Janssens, & Nijs, 2011;Marchand et al., 2005;Siebers et al., 2015;Van Peer et al., 2004), the highest reported season-long warming of a 3-m-diameter plot is 3.5°C (Ruiz-Vera et al., 2015). Our experimental system expanded plot size to 4 m diameter and the degrees of warming to 4°C. The projected warming in Puerto Rico is greater than the global averages (Henareh Khalyani et al., 2016).
Further, we adapted an IR heater system for a wet tropical forest with high humidity, hurricanes, power surges and outages, clayey soils, and a steep slope. The solutions described herein address these challenges to warming tropical forest understory plants and soils and provide valuable information for designing future warming experiments for taller vegetation in a number of underrepresented ecosystem types.

| Site
The experimental site is located in northeastern Puerto Rico near the USDA Forest Service Sabana Field Research Station in the Luquillo Experimental Forest (LEF; 18°18′N, 65°50′W) (Figures 1 and 2). The site is classified as a subtropical wet forest according to the Holdridge Life Zone System (Holdridge, 1967) and is a secondary forest, regenerated naturally from pasture since the early 1950s. Mean annual temperature is 24°C, with month-to-month variation of just 4°C, and mean annual rainfall is 3,500 mm (García-Martino, Warner, Scatena, & Civco, 1996). While the driest period generally occurs from January to March, there is no significant "dry season," as every month receives F I G U R E 1 Geographic site map showing location of Puerto Rico with respect to continental United States and insert showing location of USDA Forest Service Sabana Field Research Station within Puerto Rico greater than 200 mm and precipitation is highly variable (Heartsill-Scalley, Scatena, Estrada, McDowell, & Lugo, 2007). Located at 100 m elevation, the site is characterized by relatively steep slopes ranging from 15 to 26°, with an average slope of 21° (Table 1). Soils are classified as Ultisols and are deep, clay-rich and acidic, with high aluminum and iron content (Scatena, 1989). Root biomass is concentrated (>60%) in the top 10 cm of mineral soil (Silver & Vogt, 1993).

| Equipment
The basic design of the IR heater system was that of the T-FACE (temperature free-air controlled enhancement) described by Kimball et al. (2008), which produced uniform warming across hexagonal plots with six IR heaters (Figure 3), and which was able to maintain a constant temperature rise of the heated plots above that of control or reference plots for growing-season-long periods of time. In our case, the slope (~21° ; Table 1), saturated soils, and high clay content provided additional challenges. Galvanized 38 mm steel pipes (nominal 1½ inches) were installed to 0.7 m depth in 0.3 m × 0.8 m concrete footings for each of the six hexagon points. The footings and posts with attached cross-bars provided stable structures and prevented slippage down slope. The heaters were bolted to cross-bars (38 mm galvanized steel pipes), which spanned the space between posts on the periphery, and which enabled the heaters to face the center of the plots. The cross-bars were attached to the vertical posts with Kee Klamps (Kee Safety Inc., Buffalo, NY), allowing heaters to be mounted at the same height above the vegetation regardless of slope. The upslope and downslope heaters were deployed with long axes horizontal, while the long axes of the side heaters were parallel to the slope.
The short axes of all heaters were tilted 45° with respect to their vertical support posts. The variability of the height of the vegetation provided another challenge, so heaters were mounted at 0.4 times plot diameter above three-fourth the height of the tallest plants to provide as uniform warming as possible without exposing the tallest vegetation to too much heat. A small junction box was installed for each heater and connected to a larger junction box for each heated plot via a waterproof, flexible "super service" cable. This cable along with the use of Kee Klamps enabled the heaters to be raised and lowered for repairs and to adjust for changing vegetation heights.
As specified by a civil engineer, the concrete footings, posts, and cross-bars with the heaters should be able to withstand hurricaneforce winds with the heaters tilted vertical at a maximum operating height of 3.6 m with no allowance for a windbreak from the trees.
Nevertheless, when a hurricane is forecast, the plan is to shut off the system and lower the heaters to just above vegetation height. Based on our observations from the two storms, the infrastructure damage from fallen trees was more substantial when the heaters were not lowered. Nevertheless, all damage is repairable, and we expect the experiment will be operational again by the time power is restored to the area. Thus, there is promise for the design even in areas where disturbances such as hurricanes are common.
Based on theory (Kimball, 2005) and experience (Kimball et al., 2008) and the weather data available at the design stage, larger heaters were required to warm 4-m-diameter plots than had been used previously   (Kimball et al., 2008). The amount of reflected radiation R f (W/m 2 ) was calculated from: where R H is the heater capacity (W/m 2 ), f 8-14 is the fraction in the 8-14 μm waveband (assumed 0.13), ρ is the vegetation surface reflectance (assumed 0.02), and η is the heater array efficiency computed from: where u (m/s) is wind speed (Kimball & Conley, 2009;Kimball et al., 2008). To account for stochastic events (e.g., falling palm fronds) and (1) High humidity resulted in components rusting, growing mold, and creating an environment with high potential for arc flash and electrical shorts. We used high heat Rustoleum paint to seal any cuts made to the steel posts to deter rust. The control panel casing was made of painted steel. We added additional sealing to the doors of each of the control panels to reduce humidity entering the casing. We also implemented a regular 3-month scheduled shut-down, examination, and maintenance of the internal components of the control panels with a certified electrician and a 6-month inspection of electrical connections, cabling, and grounding. Finally, we added code to the control program in the datalogger to prevent the PID signal from dropping below 0.05 V, which keeps the heaters and other equipment at a minimum of 0.5% of capacity to maintain enough residual warmth for the components inside the control panels to stay dry as long as the 480 V power is on.
A weather mast was installed near the middle of the experimental area to collect understory microclimatic data ( Figure 2). Instruments Step

3-phase infrared heaters
Wires in weatherproof conduits that provide flexibility to adjust heights of the heaters ± 1.5 m from a nominal height of 3.1 m.

V surge protector 480 V surge protector
The deeper probes were installed in a single 5-cm-diameter hole at a 45° angle into the soil wall.
Power lines to the field site were conveyed in above-ground electrical conduits with concrete supports at about 1.5-m intervals (Figures 2 and 3). Power lines were above-ground to minimize disturbance to the root and soil system of the study site, to facilitate regular maintenance of the junction boxes, and to facilitate removal when the project is terminated. At each of the power line junction boxes associated with the three control panels, we included 120 V power outlets to accommodate scientific equipment and small tools ( Figure 2). Twelve-volt-DC sensors and dataloggers were connected to a marine battery, which in turn was connected to a 120 V AC trickle charger and a backup generator. Therefore, data collection continued during power outages and system maintenance, even when there was no power supply to the heaters. The battery also provided additional lightning protection to the logger and sensors.

| Actual and theoretical power use calculations
For six 8,100 W heaters deployed over a 4-m-diameter plot, if the system were 100% efficient, the available heating capacity would be 3,867 W/m 2 . By multiplying 3,867 times the PID signal (divided by 10) squared and then dividing by the measured temperature rise, the actual average hourly power use per degree of temperature rise for each heated plot was obtained.
The theoretical average hourly power use per degree of temperature rise for each heated plot was calculated following Kimball (2005). adapted to growing in full sunlight. Leaf area index was estimated to be about 0.5, and the soil was covered with dead leaves that were dry except when covered with dew or rain. Therefore, canopy resistance was estimated to also be about 1,000 s/m. However, the IR heating requirement proved quite insensitive to the estimate of canopy resistance due to the very humid conditions with low evapotranspiration.
In contrast, there was more uncertainty in the aerodynamic resistance introduced by the high stall speed (0.5 m/s) of the anemometer and the calm understory conditions. Under stall conditions, wind speed was assumed to be 0.25 m/s. Next, the radiative efficiency of the heaters was calculated following Kimball (2005), and again the poor wind measurements under the very calm conditions introduced uncertainty. Finally, the theoretical IR heater requirement was divided by the radiative efficiency and the geometric efficiency for a hexagonal IR heater array (0.372 from Kimball et al., 2012) to obtain the theoretical average hourly power use per degree of temperature rise.

| Uniformity of warming
Theoretically, a hexagonal array of IR heaters should produce uniform warming across a circular plot (Figure 7a). However, the understory vegetation in this experiment was very nonuniform, with varying height and ground coverage (Figures 3 and 8a). Consequently, the warming also appears very nonuniform (Figure 8b). Closer comparison between Figure 8a and b, however, revealed that the hot spots corresponded to areas where dry leaves on the soil surface were visible.
The areas of green vegetation, at which the IR thermometers were pointed, were being warmed close to the target +4°C. The warming of live foliage was therefore relatively uniform, and thus, the deployment heights and angles of the six heaters were able to provide consistent warming to the understory vegetation.
The slope of the site also affected the IR radiation distribution.
Hexagonal heater arrays theoretically and actually can produce uniform warming on sites with zero slope (Figure 7a; Kimball et al., 2008Kimball et al., , 2012, and if the posts were installed at an angle from vertical equal to the slope, a distribution like 7a should be possible on sloped plots. However, it is not easy to bore holes in the soil, pour concrete, and set posts at nonvertical angles, and then, the resultant structure would have a constant torque, so our posts were installed vertically. If the heaters are tilted at 45° with respect to a sloped soil surface and the heater posts are vertical, theoretically the area of uniform radiation would be displaced upslope (Figure 7c). In our case, we tilted the heaters at 45° with respect to their vertical posts (Figure 7b), rather than the soil surface, which theoretically distributed the warming more evenly over the plots, although some radiation still falls outside the plot area on the high side while the downslope edge of the plot receives less (Figure 7d). Nevertheless, the theoretical proportion of total radiation falling within the plot was only reduced to 36% compared 39% for level plots.
As expected, the IR heater array warmed the shallow soil ( Figures 9b, 10b,e and 11). At the 0-10 cm depth in the center of the heated plots, soil temperatures were ~1°C warmer than the understory vegetation temperatures at night, with less of a differential at midday (Figure 9b). Under control conditions, there was very little spatial variability of the 0-10 cm soil temperatures across the plots from center to edge (Figure 10b). The plot center position in the heated plots, however, was slightly hotter (<1°C) than either edge or midway (between center and edge) positions ( Figure 10b). Moreover, the IR system consistently maintained the 4°C target temperature rise in surface soils during the whole season, except when there were power outages, as evidenced by the downward spikes in the curves for the heated plots (Figure 10b).
The temperature of the surface soils (0-10 cm) of the heated plots rose to approximately 3.6°C warmer than the control plots within 6 days of the start of warming. In contrast, the soil temperatures at both 20-30 and 40-50 cm depths were very similar to those of the control plots at the start of the experiment (Figures 10e and 11), but the temperature difference steadily increased for about a month, leveling off at approximately +3°C warming at 20-30 and 40-50 cm depths, respectively, for the remaining observation period (Figures 10e   and 11). Averaged over the whole six months, the averaged warming was 2.9 and 2.6°C for the two depth increments, respectively (Table 2). other chambers, the warming from IR heaters generally extend outside the defined plot area (Figures 7 and 8). For a hexagonal array, 63% of the IR radiation theoretically falls outside the plot area , warming the soil under the plot edges and serving as a buffer around the periphery. Overall, the consistent warming of surface soils to 3.6°C and the loss of just 1°C warming with depth demonstrates a strong heat holding capacity of these clayey soils, despite the high rainfall environment. Taken together, these findings suggest this IR warming design can be utilized to examine the effects of warming on soil processes as deep as 50 cm, if not more.

| Effects on soil moisture
As expected, the warming treatment resulted in a stronger drydown of the soil of the heated plots than in the control plots during periods of low rain, especially in the shallow soil layer (Figures 9e   and 10a,d). However, the high rainfall environment led to rapid rewetting with overall no difference in average soil moisture values between heated and control plots. Although soil moisture was much more temporally variable than temperature (Figure 10b), there appeared to be little or no spatial variation in soil water content from the centers of the plots to the edges (Figure 10a). Water contents at the deeper depths were greater and fluctuated less dramatically than those at 0-10 cm (Figure 10d). From about Day 20 through Day 100, there was ample rain (Figure 10c), and consequently, the effects of heating on soil water content were not as pronounced (Figure 10a,d). However, there were four distinct drying periods during which the heated plots dried faster and more than the controls, peaking around 20, 125, 145, and 175 days from the start of the experiment (Figure 10d). Nevertheless, the average soil moisture remained adequate for the plants. While there were stronger drying events in the heated plots, over the course of 6 months of the warming treatment, the differences in mean soil moisture between the control and heated plots were not different in the surface soils (0.35 m 3 /m 3 ; Table 2). The observed drying effect was overall stronger in the deeper, more buffered soils, with the heated plots approximately 0.036/m 3 /m 3 drier than the control plots at both 20-30 cm and 40-50 cm depths.
It is not surprising that the warming treatment caused more soil drying. The IR warming system warms foliage and soil surface, but does not directly affect the absolute humidity of the air. Some observations and global circulation models suggest that relative humidity, not absolute, will remain more or less constant with global warming (Dessler & Sherwood, 2009). On the other hand, more recent calculations by Sherwood and Fu (2014) suggest that, in contrast to the whole earth with land and ocean surfaces, global warming over the continental land surfaces is likely to lead to overall drying, that is, a reduction in relative humidity and closer to warming at constant absolute humidity. Thus, the T-FACE system likely produced conditions similar to those expected with future global climate change.

| Reliability and Power Issues
The primary obstacle to having a highly reliable warming system was the unreliability of the main power supply to the station, which could be a common issue at numerous sites around the world.
During the first 6 months of operation from 28 September 2016 F I G U R E 1 1 Profiles of the differences in soil temperature between heated and control plots at the 0-10, 20-30, and 40-50 cm depths that were observed at midnight 2 days before and 180 days after the start of the 4°C IR heating treatment. The sensors were at the centers of the 4-m-diameter plots. Also shown are the corresponding differences in vegetation canopy temperatures, which have been plotted at a height of 5 cm for convenience T A B L E 2 Soil temperature (T) and volumetric soil moisture contents (M) for 0-10, 20-30, and 40-50 depth increments in the control and heated plots that were averaged over a year before the start of the warming treatment and for 6 months after the start of the treatment were working in the plots, amounting to another 1% of the total off time in the first 6 months. Fortunately, after power outages, the system rapidly re-equilibrated once power was restored. For example, there was a power outage for about 6 hr on the afternoon of DOY 37 (Figure 9). During these hours, the temperatures of the heated plots dropped close to those of the control plots ( Figure 9b).
After power was restored, the 4°C target was achieved within an hour.
During the first 6 months, four of the 18 heaters failed and were replaced. Condensate may have formed during a power outage, resulting in a short circuit when power was restored. Three of the heaters were in the same plot, so alternatively perhaps there was a lightning strike near this plot.

| Performance
The forest understory microclimate was dark, humid, and calm 120 V surge protector where the power goes to the EZ Zone controller.
Backup battery and generator power allowed data collection to occur during power outages.
High humidity Liquid tight welding.
Rustoleum paint to seal any cuts made to the steel posts.
Control panel casing made of painted steel.
Added additional sealing to the doors of each control panel.
Implemented a scheduled shut-down every 3-months to examine the internal components of the control panels and provide maintenance as needed.
Added code to the control program so the PID signal could not drop below 0.05 V, allowing the components inside the control panels to maintain enough residual warmth to stay dry.
Steep slope with weak soils when wet Tilt of heaters 45° with respect to vertical posts (rather than soil surface).
Vertical posts set in concrete.

Hurricanes
Civil engineer-specified design of concrete footings for the posts to withstand code design winds for the region with the heaters tilted vertical at maximum operating height (3.6 m) with no allowance for windbreak effects of the trees [withstood Hurricane Irma and Maria winds (but not falling trees)].
Movable cross-bars to lower during hurricane.
Easy power off switch at the main circuit breaker.

Heterogeneous understory
Heater height set at 0.4 times plot diameter above three-fourth of the height of the tallest plants. The height was applied with respect to the soil surface at the bottom and top of the plots accounting for slope.

Vegetation above heaters
The welded lip of each IR lamp was placed around the hot front side, which was pointed downwards to minimize leaves and other debris being caught by the lip.
increase of the understory vegetation temperatures (Figures 9b,c and 12). Although there were minute-to-minute variations of the temperature rise with gusts of wind and passing clouds, the hourly average temperature rise was consistently within 0.1°C of the 4°C target (omitting hours with power outages) during the first 6 months of operation (Figure 12), which is excellent performance. The heaters were below 35% of capacity 95% of the time, so for this tropical forest understory site, we could have tripled plot areas or used equivalently smaller heaters. Surprisingly, the system maintained relatively constant heating under rainy conditions. For example, 11 mm of rain was received on DOY 38, enough to significantly increase soil moisture at the 10 cm depth (Figure 9e), yet the temperature rise of the heated plots dropped only one degree below the 4°C target (Figure 9b,c).
The actual measured power consumption per degree of temperature rise was generally only 100 W m −2 °C −1 (Figure 9d), which is somewhat low, but consistent with the calm, humid conditions. The theoretical curve is generally lower than the measured values but near the lower error bar values, which again gives credence to the theoretical equations that can be used to aid in the design of future experiments. Because tropical species are adapted to very narrow thermal niches, their ability to tolerate and acclimate to increased temperatures may be much more limited than in higher-latitude ecosystems (Janzen, 1967). If true, relatively small increases in temperature could have very large consequences for the substantial amounts of carbon cycled by tropical forests. Until now, logistical challenges such as high rainfall and humidity and inconsistent power have meant that the field warming experiments, which have been so influential in our understanding of ecosystem responses to warming in other biomes, have been absent in tropical forests. Here, we describe solutions developed in response to these unique challenges, providing a significant step forward in our ability to experimentally warm a wider range of ecosystem types.

| CONCLUSIONS AND IMPLICATIONS
As a whole, the IR Heater system maintained the target 4°C rise in temperature of the heated plots above that of the controls when power was available (Figure 12), and uniformity of the treatment effect was satisfactory ( Figure 8). Furthermore, the system achieved significant soil warming of 2.6°C on average to a depth of 50 cm with minimal soil drying. To achieve this success, several obstacles were addressed (Table 3). Thus, based on our experience reported herein and that of many prior experiments listed in the Introduction, our system can be used to study tropical and nontropical forest understory plants and soils, as well as agricultural crops and rangeland, up to a vegetation height of at least 1.3 m (one-third the plot diameter) and a soil depth of approximately 50 cm (one-eighth plot diameter).
Tackling the larger issue of the effects of global warming on whole forest ecosystems and their climate feedbacks would ideally be achieved with larger plots with adult trees. Here, we successfully extended both plot size and degrees of warming with heaters 4-8 times larger than previously used, which were formerly untested for this application. No scaling issues were encountered, with relatively uniform warming of vegetation and soils of 4-m-diameter plots, so it should be feasible to combine hexagonal arrays into honeycombs as suggested by Kimball et al. (2012) and extrapolate to 30-m-tall trees and the 100-m scale plots needed for forest and orchard experiments.

ACKNOWLEDGMENTS
We gratefully acknowledge the technical assistance of Carlos Estrada,

CONFLICT OF INTEREST
None declared.

AUTHOR CONTRIBUTIONS
TW, MC, and SR involved in conceptualizing and planning experiment; BK, TW, and GG involved in design of warming system and power infrastructure; TW, BK, GG, and AA involved in construction and installation of heater system and power upgrades; TW and AA performed maintenance of experiment; SR, MC, TW, and AA performed installation of soil and plant sensors and making measurements; all involved in writing and editing the manuscript.