A New Field Instrument for Leaf Volatiles Reveals an Unexpected Vertical Profile of Isoprenoid Emission Capacities in a Tropical Forest

Both plant physiology and atmospheric chemistry are substantially altered by the emission of volatile isoprenoids (VI), such as isoprene and monoterpenes, from plant leaves. Yet, since gaining scientific attention in the 1950’s, empirical research on leaf VI has been largely confined to laboratory experiments and atmospheric observations. Here, we introduce a new field instrument designed to bridge the scales from leaf to atmosphere, by enabling precision VI detection in real time from plants in their natural ecological setting. With a field campaign in the Brazilian Amazon, we reveal an unexpected distribution of leaf emission capacities (EC) across the vertical axis of the forest canopy, with EC peaking in the mid-canopy instead of the sun-exposed canopy surface, and moderately high emissions occurring in understory specialist species. Compared to the simple interpretation that VI protect leaves from heat stress at the hot canopy surface, our results encourage a more nuanced view of the adaptive role of VI in plants. We infer that forest emissions to the atmosphere depend on the dynamic microenvironments imposed by canopy structure, and not simply on canopy surface conditions. We provide a new emissions inventory from 52 tropical tree species, revealing moderate consistency in EC within taxonomic groups. We highlight priorities in leaf volatiles research that require field-portable detection systems. Our self-contained, portable instrument provides real-time detection and live measurement feedback with precision and detection limits better than 0.5 nmolVI m–2leaf s–1. We call the instrument ‘PORCO’ based on the gas detection method: photoionization of organic compounds. We provide a thorough validation of PORCO and demonstrate its capacity to detect ecologically driven variation in leaf emission rates and thus accelerate a nascent field of science: the ecology and ecophysiology of plant volatiles.


Introduction
Forest-atmosphere interactions are shaped not only by the exchange of carbon, water, and 55 energy, but also by the biological production of organic trace gases (Laothawornkitkul et  contribute to the formation of atmospheric aerosols (Heald et al., 2008). These 60 'secondary organic aerosols' formed through atmospheric chemistry with organic 61 gases-most conspicuous in the blue haze that forms over maritime forests in hot 62 conditions-affect the radiative forcing of the climate (Carslaw et al., 2013) and are the 63 source of up to half of all cloud condensation nuclei (Carslaw et al., 2010). 64 Understanding which plant species emit VI and how they are distributed across 65 landscapes is critical to determining regional and global emissions (P. Harley  control of the leaf environment using a sophisticated leaf cuvette, precision quantification 113 of emissions, and the ability to distinguish VI species. The disadvantages include limited 114 sample sizes due to the high cost of adsorption cartridges, the requirement for offsite 115 analysis, a lack of live measurement feedback, and the potential chemical degradation of 116 stored samples during transport. The second common method employs portable, 'online' 117 detection instruments such as handheld photoionization detectors (PIDs), which quantify 118 organic gas concentrations in real time, but lack the ability to distinguish gas species. A 119 portable online detector provides significant advantages in remote areas, affording 120 unlimited sample sizes and live measurement feedback (P. Harley  Here, we present a prototype instrument for precision online detection of leaf VI 126 emissions in the field. We field-validate the instrument with a measurement campaign in 127 the Brazilian Amazon, from which we produce an emissions inventory of 51 tropical 128 trees species, and estimate the vertical distribution of emission capacities within the 129 forest canopy. We call the instrument "PORCO", after the detection method-130 photoionization of organic compounds. Custom, light-controlled leaf cuvettes are 131 combined with optimized photoionization detection for high measurement precision and 132 repeatability. The components and methods are adaptable to a range of real-time gas-133 sampling objectives where required detection limits are around 5 ppb or greater (e.g., 134 greenhouse air, headspace analysis), and to systems requiring microenvironmental 135 control (e.g., moss growth chambers). The adaptations unique to PORCO enable greater 136 sample sizes, lower detection limits, real-time data, and better distinction between leaves 137 and species for more nuanced ecological hypothesis testing than was previously possible. 138 139

140
1 Overview of the 'PORCO' system 141 PORCO measures emissions from intact leaves attached to branches (Fig. 1) in a manner 142 similar to field measurements for leaf photosynthesis (Hunt, 2003). Leaves are enclosed 143 in custom acrylic cuvettes with custom light-emitting-diode (LED) panels providing 144 photosynthetically active radiation (PAR: 90% red, 660 nm; 10% blue, 460 nm) to the 145 leaf to drive photosynthesis and organic gas emissions. Air is purified and pumped to the 146 cuvette inlet at a controlled rate. Cuvette air is subsampled from an outlet while excess 147 flow is exhausted at a loose seal around the leaf petiole. Sample air is drawn through 148 tubing embedded in ice to remove water vapor, and into a commercial photoionization 149 detector (PID). The PID quantifies hydrocarbon gas concentrations, logs its readings at 1 150 Hz, and displays live numerical data as it is logged. Emission rates from the leaf are 151 calculated as non-steady-state fluxes. The entire system is battery powered and mounted 152 on an external-frame backpack for field portability. For a list of major system 153 components and sources, see Table S1.   processing.

202
Because measurement error can be positive or negative, monitoring and 203 correcting it requires a continuous non-zero data stream, ideally a raw voltage signal from 204 the instrument. However, commercial PIDs will only log internally processed data based 205 on onboard calibrations. The solution employed in PORCO is to conduct a 'false 206 calibration' of the PID that results in its generation of 'pseudo-raw data'. Pseudo-raw 207 data is the standard PID datalog in putative units of ppb, but calibrated to be a better 208 representation of the raw electrical signal. This returns control of data interpretation to 209 the user, so that corrections for environmental sensitivities and drift can be applied in 210 data post-processing. 211 The false calibration to generate pseudo-raw data is conducted by performing an 212 onboard 'zero' calibration after intentionally reducing the PID's baseline signal. This can 213 be achieved by one or a combination of the following: apply a zero-air airstream with 214 high humidity; reduce the temperature of the instrument; or install a UV bulb with a 215 lower strength (bulbs of any given model vary in output, and trials will quickly 216 demonstrate which of two bulbs is stronger). All three methods reduce sensor voltage and 217 thereby the signal that the onboard computer interprets as 'zero'. Upon return to the 218 configuration used during sampling, purified air will read far above zero, thus enabling 219 the perception of any decrease in baseline signal. Similarly, a false 'span' (non-zero 220 concentration) calibration can be employed to reduce the minimum signal interval from 1 221 ppb to less than 1 ppb. For example, conducting a span calibration set to 1000 ppb while 222 providing 500 ppb of calibration gas will induce the instrument to return signal at 0.5 ppb 223 intervals (essentially increasing the sensor 'gain'). At some point, however, signal noise 224 at the sensor exceeds the apparent resolution of the signal. 225 To demonstrate the consequences of not controlling for environmental 226 sensitivities and drift, we performed calibration experiments to quantify the effects of 227 humidity, temperature, and time on the PID baseline signal (calibration intercept) and 228 responsivity (calibration slope) (Figs. 2, 3). In the first experiment, the PID was stabilized 229 at a given temperature in a thermally controlled enclosure, and exposed to hydrocarbon-230 purified air ('zero-air') with a series of distinct relative humidity values. Dilutions of 231 isobutylene calibration gas in zero-air were applied, and the process was repeated at a 232 higher temperature (Fig. 2a, b, d). In the second experiment, the PID was stabilized at 233 three different temperatures for a full day each, at constant relative humidity, and 234 isobutylene dilutions were applied every few hours (Fig. 2c, e, f).

235
PID baseline signal was negatively sensitive to relative humidity of the measured 236 air stream (Fig. 2a), positively sensitive to enclosure temperature (Fig. 2b), and 237 negatively sensitive to instrument run time (Fig. 2c). PID responsivity to calibration gas 238 was negatively sensitive to relative humidity (Fig. 2d), negatively sensitive to 239 temperature (Fig. 2e), and not sensitive to instrument run time (Fig. 2f). Extrapolating the 240 modeled sensitivities (linear regressions, dashed lines in Fig. 2) across a 50 % relative 241 humidity range, and a 10 ˚C temperature range, the relative importance of each factor can 242 be assessed in terms of typical ambient environmental variation encountered during a day 243 of outdoor field sampling. Relative humidity emerges as the most important driver of PID 244 drift, causing a reduction in baseline signal equivalent to 300 ppb (Fig. 2a), and a 30 % 245 reduction in PID responsivity (Fig. 2d) across a 50 % range of relative humidity. This 246 source of measurement error is especially relevant to sampling gases from leaf 247 enclosures, during which transpiration can rapidly increase sample humidity.

265
The importance of PID measurement error for a given sampling objective depends 266 on the measurement conditions and the lower detection limits and precision required. To 267 facilitate such an evaluation, Fig. 3 shows modeled responses of a PID with a traditional 268 on-board zero and span calibration to gas concentrations between 0 and 2000 ppb. 269 Deviations of PID readings from true concentrations are modeled based on the responses 270 of both baseline signal and responsivity to humidity, temperature, and time shown in Fig.  271 2. For example, if sample humidity is 50 % greater than calibration humidity, a false zero 272 will be observed up to true concentrations of 436 ppb (the 'blind spot'), and 2000 ppb 273 will be interpreted as 1109 ppb (Fig. 3a). An instrument temperature 10 ˚C higher than 274 during calibration will impose a false positive of 148 ppb in purified air, but will 275 underestimate 2000 ppb as 1715 ppb (Fig. 3b). Note especially that the standard 276 calibration technique necessarily imposes a mix of environmental conditions: purified 277 ambient air (at ambient T and RH) is used for the zero, while the span is provided by 278 direct application of calibration gas from the bottle. The calibration gas is both dry (near 279 0 % RH) and cold due to decompression-the cooling of the sensor causes the typical 280 downward drift in signal observed after completing the span and before disconnecting the 281 PID from the gas bottle. Note that Fig. 3 should be treated only as a rough guide, as 282 sensitivities may vary considerably between sensors and calibration configurations.

293
and inaccuracies in non-zero gas readings (sensor responsivity). RH will be the most important 294 concern for most users, though direct sun exposure can cause large changes in T. Controlling or 295 correcting for PID measurement error will be most important for applications requiring lower 296 detection limits, precision, and accuracy better than approximately 500-1000 ppb. This should be 297 used only as a rough guide, as PID sensitivities likely differ between sensors.

299
While these analyses demonstrate the potential to measure the sources of PID 300 measurement error and apply corrections in data post-processing, the best results are 301 achieved by first maximizing environmental control. For example, temperature 302 measurement of the sample air cannot be used to correct for thermal sensitivity unless 303 sample and instrument are in thermal equilibrium. In PORCO, PID temperature 304 sensitivity is mitigated by controlling instrument and sample-gas temperatures. The PID 305 is enclosed in an insulated case, in which the air is heated to an above-ambient constant 306 temperature (± 0.5 ˚C) with a thermostat-controlled Peltier device (Fig. 4). Sample (and 307 zero or calibration) gas is equilibrated to the enclosure temperature by routing through 308 coils of stainless steel tubing upstream of the PID (Figs. 1, 4). Relative humidity is held 309 low and constant (approximately 10 % with precision ± 0.5 % at the PID inlet) by 310 routing sample air through tubing embedded in ice and water in an insulated enclosure 311 (Figs. 1, S1), which maintains a saturated air space at constant temperature. This method 312 produces constant humidity over 16 hours of continuous run time in hot (30-40˚C air) 313 conditions. A small chamber at the bottom of a V-shaped tubing path in the dehumidifier 314 catches condensed water and allows sample gas to pass above any liquid instead of 315 traveling through it (Fig. S1). Isoprene and other common isoprenoids emitted by plants 316 such as alpha-pinene (a monoterpene) have minimal water-solubility (Kim et al., 2020; 317 Martins et al., 2017), so interference by liquid water is negligible. While some PIDs offer 318 an optional on-board correction for humidity and temperature (toggle in the software; 319 RAE Systems Inc., 2013a), this should be disabled as the correction is small relative to 320 the magnitude of sensitivities, and introduces unaccountable noise into the pseudo-raw 321 data. We find that humidity removal by condensation is the most effective strategy, as our

327
The PID is fixed inside an insulated enclosure with a battery-powered Peltier heater. A 328 thermocouple-equipped thermostat regulates the heater to maintain a fixed above-ambient 329 temperature inside the enclosure. Sample gas enters the enclosure by tubing connection to a 330 bulkhead union. Sample gas is pre-heated to match the enclosure temperature via coils of 331 stainless steel (ss) tubing. Gas temperature and humidity are measured with an iButton in a PTFE-332 insulated chamber upstream of the PID. An acrylic window enables viewing of the PID display.

333
Wooden pins at the box interface provide access to the buttons on the PID without opening the 334 enclosure. PID sample exhaust is piped to an outlet port, which can be connected to alternative 335 sampling media (e.g., bags, adsorption cartridges).

337
PIDs have different responsivities to different gases due to variation in ionization 338 energy among compounds. Measurements of different compounds can be scaled by the 339 relative responsivity of the PID to isobutylene calibration gas using published correction 340 factors (CF) specific to different UV lamp voltages (RAE Systems Inc., 2013a). 341 However, empirical CF may differ between sensors or sample configurations. We find 342 that the PID is 2.25 times more sensitive to isoprene than isobutylene (CF=0.44, Fig. S2), 343 compared to 1.59 times (CF=0.63) reported by Rae Systems (RAE Systems Inc., 2013a). 344 Rarely, some compounds can interfere with PID responsivity (RAE Systems Inc., 2013c).

345
Volatiles from silicone lubricants in valves, for example, can cause chronic signal 346 reduction ('sensor poisoning') requiring cleaning of the sensor and sample path to 347 recover responsivity (data not shown). These considerations are particularly important 348 when sample gas contains variable mixtures of compounds, and when choosing materials 349 that contact the sample gas. PORCO employs stainless steel and PTFE or equivalent 350 tubing. Other materials are minimized and tested to ensure non-interference. calibrations to the FIS showed the inline PID had no effect on FIS calibration linearity or 364 signal-to-noise ratio, but the PID did reduce isoprene concentrations at the FIS by 8.2% 365 (Fig. S3). By calibrating and measuring leaves under the same instrument configuration, 366 signal degradation by the PID was held constant and thereby nullified. 367 The PID and FIS showed strong agreement in the time-series of isoprene standard 368 (Fig. 5) and leaf emission measurements (Figs. 6, S4). Compared to the FIS, the PID 369 showed less signal noise at 1 Hz, slightly poorer linearity of response to isoprene 370 standard dilutions, and less drift in responsivity (calibration slope) over time (Fig. 5). The 371 PID showed more drift in baseline signal (not shown, but see Fig. 2c). Both instruments 372 captured the subtle signal of a post-illumination isoprene emission burst (Li & Sharkey,373 2013) from an oak leaf (Fig. 6a). Averaged emission rates estimated by the two 374 instruments from the final two minutes of each leaf measurement strongly agreed (linear 375 regression, r 2 = 0.999, Fig. 7). The PID underestimated emission rates relative to the    adaptations can improve real-time field sampling precision from a diversity of cuvettes, 424 PORCO employs custom leaf cuvettes with design features that optimize precision and 425 detection limits relative to the sensitivities of the PID. The cuvettes also provide a low-426 cost solution for quantitative leaf emission sampling using other gas-detection devices. 427 Leaf illumination can be provided from ambient light, or using the custom modular LED 428 panels described in Section 4: Leaf light environment. 429 430

Cuvette design 431
To quantify emission rates, sampled trace gas concentrations must exceed the lower 432 detection limit (LDL) of the detector. Emitted gases accumulate inside a cuvette at a rate 433 dependent on the leaf size, emission rate, cuvette volume, and the replacement rate of 434 cuvette air. The PORCO cuvettes were designed to achieve a LDL equal to a leaf area-435 based emission rate (ER) of 1 nmol m -2 s -1 or better within 60 sec of measurement from a 436 majority of leaf types found among tropical rainforest trees. A 60 sec LDL of ER = 1 is a 437 conservative goal that allows for confident distinction of weak emissions from non-438 emission over slightly longer (minutes) measurement times under non-ideal conditions 439 such as sub-optimally sized leaves, reduced physiological activity of cut branches, or 440 environmental instability causing PID measurement error. We evaluated cuvette designs 441 in terms of the ratio of ER to the concentration of the emitted gas accumulated in the 442 cuvette after 60 s, which we term the 'signal amplification factor' (SAF). Given typical 443 signal noise of up to ± 2 ppb of isoprene, a confident LDL for the PORCO-PID is 444 approximately 5 ppb. Therefore, a minimum SAF = 5 is optimal for detection of ER = 1 445 (i.e. 5 ppb is accumulated in 60 s). For comparison, for the LI-6400 leaf cuvette, SAF = 446 1.5 due to the small cuvette volume relative to a standard flow rate (550 sccm), resulting 447 in LDL = 3.3 nmol m -2 s -1 . Moreover, at a given emission rate, gas concentrations in the 448 LI-6400 cuvette reach steady state in less than 1 min, while concentrations continue to 449 increase in larger cuvettes beyond 1 min, giving a greater potential to exceed lower 450 detection limits. 451 Because the flow rate through the cuvette must exceed the sampling rate of the 452 PID (300 or 500 mLpm), increasing SAF requires increasing the amount of leaf area 453 enclosed, while minimizing the increase in cuvette volume. We designed three cuvettes 454 that provide SAF > 5 by fully enclosing entire leaves or leaflets (of compound leaves), 455 with distinct shapes that accommodate leaf types of a large majority of tropical tree 456 species. 457 458

Cuvette construction 459
Custom PORCO leaf cuvettes (Fig. 8) are rigid acrylic boxes that enclose entire leaves. 460 The petiole or twig traverses a port by which leaves maintain connection to their branch.

461
Operation is a flow-through system where pure air is pumped through an inlet port, and 462 sample air is drawn by the pump in the PID via an outlet port (diagram in Fig. 1). Excess 463 flow is exhausted through a partial-seal around the leaf petiole ('petiole gasket', Fig. 8).

464
An internal fan ensures even air mixing, necessary for non-steady-state flux calculations. 465 All internal parts are acrylic, stainless steel, or PTFE to minimize adsorption and release 466 of trace gases, except for the plastic mixing fan, which does not produce significant 467 interference. The acrylic initially produced significant background concentrations of 468 ionizable gases, but this was successfully eliminated by baking cuvettes in a drying oven 469 at 60 ˚C for five days. A roof mounting system accommodates modular LED panels and a 470 light diffusor. Reflective siding ensures even light distribution and minimal interference 471 from ambient light. 472 473 can be validated by simulating leaf emissions by conveying calibration gas to the cuvette 489 inlet (Fig. 9). Insufficient air mixing would reduce the effective mixing volume, 490 producing an initial increase in concentration that is faster than expected. Our simulated 491 emissions show an initial increase that is slightly slower than expected. Insufficient 492 excess flow out of the petiole gasket may produce a diluted curve if the seal is poor, due 493 to mixing with outside ambient air. However, tightening the petiole gasket seal or 494 increasing the flow rate to the cuvette does not alter the results (data not shown), 495 suggesting that excess flow through the loosely sealed petiole gasket under normal 496 sampling conditions is sufficient to mitigate dilution. 497 Simulated emissions show that the largest measurement bias arises from pressure 498 differences caused by different plumbing configurations. The balance of pressure from 499 the zero-air pump (positive) and PID pump (negative) differ between calibration and 500 sample modes (Fig. 1), and between different sampling configurations. During 501 calibration, positive pressure is determined by the total flow rate and the resistance of the 502 excess-flow exhaust path (coiled 1/8" tubing). During sampling, the PID pump pulls air 503 from the cuvette through a relatively long pathway via the dehumidifier and thermal 504 equilibration tubing (Fig. 1), causing a pressure drop. We have not attempted to measure 505 pressure in order to quantify this effect. Pressure effects are particularly evident when 506 changing flow modes by turning valves, which inevitably causes a transient spike in PID 507 readings (Fig. S5). If valves are turned too quickly, or the flow path is momentarily 508 blocked, this reading spike can be extremely large and persist in the form of an 509 asymptotic recovery toward the previous baseline signal over a half hour or more. Strong 510 negative pressure could also exacerbate the effect of small leaks in sample path 511 connections, though we note that sample dilution caused by leaks is inconsistent with the 512 observation of reduced signal at the PID (after switching from calibration to sample 513 mode) relative to an independent instrument sampling directly from the PID exhaust (i.e. 514 the independent sensor should have seen the same reduction in signal, see Section 2.2 515 PORCO-PID validation against Fast Isoprene Sensor). Pressure effects are plausibly 516 related to the fact that the PID pump, ionization bulb, and sensor are all powered by the 517 same battery, and therefore a change in pump effort caused by a change in flow pressure 518 causes a change in sensor voltage. Different calibration and sample configurations 519 therefore produce a corresponding offset of observation from expectation during 520 simulated emissions. The offset is consistent within configurations (Fig. 9c), allowing 521 calculation of a configuration-specific correction factor for leaf emission estimates. 522 Calibrating exclusively via simulated emissions would obviate the need for correction 523 factors, but this requires much more calibration gas, which is in limited supply in the 524 field due to the logistical challenge of transporting compressed gas cannisters. were designed to mimic the peak light wavelengths and ratios employed in the LI-6400 547 leaf chamber: 90% red light with peak output at 670 nm, and 10% blue light at 465 nm. 548 The PORCO light panels use Cree Xlamp "Photo Red" (peak output at 650-670 nm) and 549 "Royal Blue" (450-465 nm) LEDs (Cree Inc., Durham, NC, USA). Light ratios and total 550 quantity are regulated based on a calibrated relationship between measured PAR output 551 and electrical current to the bulbs (Fig. S6). During measurements, electrical current to 552 each bulb color on each panel is adjusted to target values via dimmer dials, based on 553 voltage read across precision inline resistors (yellow box in Fig. 1). Electrical current-554 controllers maintain a steady photon flux density at the leaf plane. 555 The modular LED panels can be adapted to a diversity of cuvettes. In PORCO, up to 556 three panels are used at a time depending on cuvette size. The LED panels could be easily 557 applied to custom cuvettes adapted for other purposes, such as moss or algal growth 558 chambers. 559 560 5 Field calibration and data post-processing 561 The continuous pseudo-raw data logged at 1 Hz from the PID is processed with custom 562 code in R (R Core Team, 2020; Wickham, 2016). Fig. 10 illustrates the processing flow 563 from a day of field measurements. During measurements, the user records start and end 564 times for each type of sample, such as zero-air, calibration air, cuvette blanks, and leaf 565 samples. Data processing is applied to specific time brackets in the time-series output 566 based on the type of measurement recorded. 567 Linear calibration curves are made by measuring isobutylene standard diluted in 568 purified ('zero') air. The air streams are regulated at desired flow rates by mass-based 569 flow controllers (Alicat Scientific, Tucson, AZ, USA) and joined before exhausting 570 through coiled eighth-inch tubing to minimize ambient interference, while the PID 571 samples the mixture from a tee junction. Calibrations are performed at the beginning of 572 each day of measurements, and frequently at the end of each day to ensure responsivity 573 stability. Periodic zero-air readings can be used to interpolate and subtract baseline drift 574 (Fig. 2c). 575 For standard leaf measurements with PORCO cuvettes, the stabilized signal from 576 the cuvette with the leaf installed and lights off ('cuvette blank') is used as the baseline 577 for emission calculations. Leaf emissions are calculated as a non-steady state flux from 578 the rate of change in concentration relative to the blank, determined by linear regressions 579 on a 40 sec sliding window of data following the moment the cuvette lights are turned on 580 (Fig. 10).    conditions suitable for photosynthesis (Fig. 11). Measurement preparation begins with 609 choosing a cuvette for the leaf that maximizes the ratio of leaf area to horizontal cuvette 610 area without bending the leaf edges. Purified ambient air is conveyed to the cuvette inlet 611 port via a mass-flow controller, while the PID draws air from the outlet port. The LED 612 panels are attached to the cuvette, and electrical current is adjusted to produce the desired 613 light output (1000 PAR for standardized measurements). The lights are then switched off. 614 The leaf is installed by carefully positioning the cuvette to enclose the leaf, using 615 an articulating and locking arm mount to closely match the cuvette orientation to leaf 616 orientation, minimizing leaf disturbance. The cuvette faceplate is installed and the petiole 617 loosely sealed by the petiole gasket. Purified air is conveyed to the cuvette at a rate ≥ 150 618 sccm above the PID sampling rate (either approximately 500 mLpm or 300 mLpm 619 depending on PID pump-speed setting). Excess flow is exhausted from a loose seal 620 around the leaf petiole to resist diffusion of ambient air into the cuvette. The leaf is left in 621 darkness inside the closed cuvette for sufficient time to clear ambient air and achieve a 622 stable background signal (6-9 min depending on cuvette volume). The LED lights are 623 then switched on, driving photosynthesis and VI emissions if the leaf is an emitter. 624 For standardized measurements at 1000 PAR, leaves are illuminated for 3.5 min.

625
Longer illumination periods at high light levels are avoided due to the potential to 626 overheat the cuvette. Continued monitoring following lights-off allows confirmation of 627 weak emission peaks, and analysis of species variation in post-illumination emission 628 dynamics (e.g., Fig. 6a). Live emission information can be qualitatively determined 629 during a measurement by monitoring the numerical output on the PID display (a single 630 number varying at 1 Hz, equivalent to the datalog) through the enclosure window (Fig.  631  4). Careful observation allows the distinction of even weak emitters from non-emitters in The primary components of PORCO, excluding leaf cuvettes, are mounted on an 646 external-frame backpack for portability. The system can be carried by one person through 647 a forest, up a canopy tower, or hauled by ropes into trees or canopy walkways (Fig. 12). 648 All of the plastic cases can be closed during transport with all air-flow components still 649 running to maintain PID stability. For tall trees, branches can be cut, carefully lowered, 650 re-cut under water to maintain the transpiration stream, and then measured on the ground. 651 Where accessible, the portability of PORCO allows in-situ measurements of tree leaves, 652 which eliminates the potential for undesired physiological responses to branch cutting. 653 654

660
We used our uniquely portable detection system to quantify leaf VI emissions from 661 tropical trees at a field site in the eastern Amazon, the "k67" eddy flux tower site in the 662 Tapajos National Forest near Santarém, PA, Brazil (Rice et al., 2004). In 27 days of field 663 sampling, we obtained high quality emission data from 138 leaves from 67 trees and 51 664 species (Table 1) Aug. 22, 2020, Table S2). 668 This emissions inventory increases tropical species coverage by approximately 669 10% relative to the literature compilation by Taylor  have shown that careful measurement of highly aromatic leaves that lack the capacity for 683 light-dependent VI emissions (e.g., Retrophyllum, Myrcia, and Zanthoxylum spp.) do not 684 produce detectable volatiles during a standard PORCO measurement (data not shown).

685
Emission rates >1 nmol m -2 s -1 are more likely to be predominantly composed of 686 isoprene, and less commonly, smaller amounts of light-dependent monoterpenes 687 ( Table 1. Emission rates ranged from 694 consistently less than 1 nmol m -2 s -1 in the Lauraceae family, to frequently exceeding 20 695 nmol m -2 s -1 in the genus Protium. The highest observed emission rate was from an 696 Eschweilera species, 28.5 nmol m -2 s -1 . 697 698

Validation of field measurements 713
Visual analysis of processed leaf-emissions data demonstrate background signal noise to 714 rarely exceed ± 0.2 nmol m -2 s -1 . All emission curves exceeding 0.4 nmol m -2 s -1 were 715 distinctly different from background fluctuations by 210 sec of illumination (Fig. 13). We 716 therefore assign a conservative uniform detection limit of 0.4 nmol m -2 s -1 to the 717 inventory dataset (Table 1) observed that where within-species variation was highest (Table 1), species data was 730 amalgamated from measurements of multiple distinct trees or branches from different 731 light environments. We tested whether PORCO measurements are sufficiently consistent 732 to detect ecologically driven variation at the branch scale, as distinguished from leaf level 733 variation and detection precision. We found that between-branch variation (standard 734 deviation) in mean emission rates exceeded within-branch (between-leaf) variation by a 735 factor of 1.6 (t-test, p << 0.001, Fig. S7), demonstrating that PORCO is able to detect 736 ecologically driven variation in leaf emission rates. 737 Emission measurements using PORCO show good agreement with other data 738 sources for the same species and genera (Table 1). Independent data was available at the 739 species-level for ten of our measured species. These included our own validation 740 measurements from several of the same individual trees, in which leaves were acclimated 741 to 1000 PAR and 30˚C until photosynthetic stability was achieved (approximately 2 min) 742 in a Walz photosynthesis chamber head (Heinz Walz GmbH, Effeltrich, Germany), 743 followed independent data sources for all ten shared species in terms of detectable emissions (5 749 species) versus emissions below detection limit (BDL, 5 species), except E. uncinatum, 750 reported as BDL by Harley et al. (2004), but confirmed as an emitter by independent 751 measurements on multiple individuals during this campaign. PORCO measurements also 752 agreed well with published data at the genus-level (Table 1) (Table S2). For species with weak or undetected emissions by PORCO, a low proportion 755 of species from the corresponding genus ("congeners") were reported as emitters in the 756 literature. Conversely, stronger emissions corresponded to a higher proportion of 757 congeners reported as emitters in the literature (Table 1).

758
Likely due to the low detection limit achieved with PORCO, we report a higher 759 proportion of species with light-dependent VI emissions (67% above detection limit; 59% 760 > 1 nmol m -2 s -1 ) compared the inventory at the same site by Harley  emission capacities (Fig. 14a) and the greatest proportion of emitting species (Fig. 14b) 790 found in the mid-canopy region, and surprisingly high emissions found among the 791 smallest trees. We cannot attribute this trend to an over-sampling of canopy species at 792 sub-canopy positions, because we sampled most trees at sizes within the maximum size 793 class for their species. For example, Protium species are mid-canopy dominants, never 794 exceeding 45 cm diameter at breast height at our site, and were among the strongest 795 emitters. Two species in the Violaceae family showed consistently high emissions, and 796 they are exclusively understory specialists, never exceeding 10 cm diameter and often 797 found in shaded environments. The strongest observed emission rate was from an 18.8 m 798 tall Eschweilera sp. (less than half the height of the tallest trees in this forest) in an open, 799 highly sun exposed crown environment, while similarly exposed canopy-dominant 800 species (Erisma and Tachigali) showed moderate emissions (Table 1). 801 This vertical distribution of emission capacities suggests a more nuanced view of 802 the adaptive significance of VI emissions for tropical trees. As an on-demand trait, VI 803 emission may be adaptive for temporally dynamic thermal environments (Sharkey et al.,804 2008), rather than chronically hot ones. In the mid-canopy, leaves experience alternating 805 shade and direct sun exposure as the sun traverses overhead canopy material and gaps. 806 This view presents a challenge for modeling emissions from tropical forests.  combined with an ability to distinguish compounds by simultaneously sampling onto 865 adsorbent cartridges from the partially ionized PID exhaust (see Fig. S3), or separately 866 from the leaf cuvettes.

868
Conclusion 869 Despite the well demonstrated critical roles of volatile isoprenoid emissions in plant 870 physiology and atmospheric chemistry, the field research required to bridge the scales 871 from leaf to atmosphere has been held back by methodological limitations. The 872 adaptations presented in PORCO resolve key limitations, allowing an increased rate of 873 species sampling and the in situ measurement precision required to access exciting new 874 questions in ecology, evolution, and atmospheric science. For example, modeling forest 875 emissions requires that we understand the landscape distributions of emitting species (A. Field campaigns will enable targeted hypothesis testing, and species inventories 886 such as reported here (Table 1)

909
PORCO adaptations could also benefit applied research. VI-emitting crops may 910 be more resistant to drought and ozone, but at a cost to growth rate ( previously in the Ph.D. dissertation of Tyeen C. Taylor (Taylor, 2017). The submitted 932 form of this article was posted as a pre-print on bioRxiv. 933 934