# Source paper archive

- Source title: Ultrasonic emissions of sunflower and buckwheat carry species identity in their spectrum and drought in their rate
- Author(s): Saha, B., Haque, M. I., Strozecki, M., Rastogi, A.
- Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.17.767468v1
- PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.17.767468v1.full.pdf
- PDF SHA-256: dd03f786c245f4c8999382f23688b0f9878cff16ba9176b9857529f15389a507
- Converter: pymupdf4llm 0.0.27
- Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version.

---

&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)

# 1 Ultrasonic emissions of sunflower and buckwheat carry species 2 identity in their spectrum and drought in their rate


3 **Running title:** Ultrasonic emissions of sunflower and buckwheat


4 Bedabrata Saha &#91;1,*&#93;, Md. Intesaful Haque &#91;1&#93;, Marcin Stró ż ecki &#91;1&#93; and Anshu Rastogi &#91;1,*&#93;


5 1 Department of Bioclimatology, Faculty of Environmental and Mechanical Engineering, Pozna ń


6 University of Life Sciences, Pi ą tkowska 94, 60-649 Pozna ń, Poland


7   - Correspondence: anshu.rastogi@up.poznan.pl or bedabrata.saha@up.poznan.pl


1


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


8 **Highlight**


9 Drought raised airborne ultrasonic event rates in sunflower and buckwheat. In a proof-of

10 concept test on eight plants, the share of event energy above 40 kHz identified the species.


11 **Abstract**


12 Plants under water deficit produce ultrasonic acoustic emissions that can be recorded in air.


13 Species have been compared mainly by peak frequency. We report, to our knowledge, the first


14 airborne recordings of sunflower ( _Helianthus annuus_ ) and buckwheat ( _Fagopyrum_


15 _esculentum_ ), with gas exchange and chlorophyll fluorescence on the same plants. Across 119


16 recordings of 30 min each, made on two drought and two watered end plants per species,


17 every drought plant emitted more events over the series than the watered plants. Rates rose


18 107 fold in sunflower and 5.3 fold in buckwheat, coincided with low stomatal conductance,


19 and later fell in single plants. In sunflower, emission was already raised when maximum


20 Photosystem II quantum yield was similar between treatments. At equal loudness, sunflower


21 events carried a median of 24% of their energy at 40-80 kHz, against 2% in buckwheat, in


22 drought and watered plants alike, although both species peaked near 30 kHz. In a proof of


23 concept on four plants per species, this share identified a held-out plant's species in 96% of


24 draws of five events. Airborne ultrasound thus carries species identity in its spectrum and


25 drought in its event rate, two signals a non-contact sensor could combine.


26 **Keywords: Acoustic emission, chlorophyll fluorescence, drought,** _**Fagopyrum**_


27 _**esculentum**_ **,** _**Helianthus annuus**_ **, stomatal conductance, xylem embolism**


28 **Abbreviations:** AE, acoustic emission; AUC, area under the receiver operating characteristic


29 curve; CI, confidence interval; DAS, days after sowing; dBFS, decibels relative to full scale;


30 _F_ v / _F_ m, maximum quantum yield of PSII; _g_ s, stomatal conductance to water vapour; PI abs,


31 performance index on absorption basis


32 **Introduction**


33 Water deficit is a major cause of yield loss in crops, and early signs of plant water stress are


34 useful for irrigation management (Saha and Rastogi, 2026). Acoustic emissions (AE) from the


35 plant itself are one candidate signal. They are produced in the water-conducting tissue as it


36 dries (SharathKumar et al., 2026). In a greenhouse comparison of 10 sensor types on tomato,


37 AE recorded with an ultrasound microphone near the stem, stomatal conductance and stem


38 diameter all changed within two days of withholding water (Dutta et al., 2025), and contact


2


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


39 AE sensors detected osmotic stress in hydroponic lettuce before visible wilting (Hase and


40 Ogihara, 2026).


41 Ultrasonic AE from drying xylem has been studied for decades with sensors attached to the


42 stem (Milburn and Johnson, 1966; Tyree and Dixon, 1983). Many events arise when


43 cavitation breaks the water column under tension, and their acoustic energy increases with


44 conduit size (Ponomarenko et al., 2014). The settling time and peak frequency of ultrasonic


45 pulses were related to xylem vessel radius in 10 species (Dutta et al., 2022). In the field,


46 emissions coincided with stem shrinkage that tracks xylem tension (Hölttä et al., 2005), and in


47 potted wheat events became frequent once the leaves curled (Jia et al., 2006). Some events


48 come from other sources, such as cell wall shrinkage and cell damage late in dehydration


49 (Vergeynst et al., 2016; Lamacque et al., 2022; Boccato et al., 2024; Nardini et al., 2024).


50 Event counts are now used to build vulnerability curves in woody plants (De Boeck and


51 Steppe, 2025). Stomata usually close at higher water potentials than those that start embolism


52 (Martin-StPaul et al., 2017), so emissions would be expected mainly once stomata have


53 closed.


54 Khait et al. (2023) showed that tomato and tobacco emit airborne ultrasonic sounds when


55 dried or cut. Drought stressed tomato emitted about 35 sounds per hour, watered plants fewer


56 than one and pots with soil alone none. Mean peak frequency differed between the two


57 species, and a classifier separated species and treatments. Bonisoli et al. (2024; 2025)


58 recorded two to three events per 30 min from bean and tomato with the low-cost microphone


59 model used here. Microphones next to the stems of tree saplings detected induced embolism


60 outdoors (Klaminder et al., 2025).


61 A recent review in this journal described airborne ultrasound as a source of acoustic


62 phenotypes that can indicate plant water status, with xylem cavitation as the main proposed


63 source (Font Bordera et al., 2026). It concluded that emissions are specific to both species and


64 stress, that their acoustic signature probably reflects vessel diameter and tissue anatomy, and


65 that machine-learning classification could support precision agriculture. It also noted that not


66 every emission need come from cavitation, that current datasets are small and from controlled


67 conditions, and that more species are needed. Airborne recordings exist for few crops, and it


68 is unclear whether species differ in the spectral content of their emissions when loudness is


69 matched. Event counts depend on detection thresholds and electrical noise, and protocols are


70 not standardised (SharathKumar et al., 2026). When several plants share one chamber, it is


3


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


71 also unclear which plant produced an event, and some authors have questioned how much


72 information airborne plant sounds carry (Son et al., 2024).


73 Here we recorded drought and watered plants of sunflower ( _Helianthus annuus_ L.), a tall


74 oilseed crop with a thick stem, and common buckwheat ( _Fagopyrum esculentum_ Moench), a


75 short-season pseudocereal described as highly sensitive to drought (Hou et al., 2019), with the


76 same chamber, microphones and detection thresholds. To our knowledge, airborne emissions


77 of neither species have been reported. We addressed three points raised by Font Bordera et al.


78 (2026). First, we tested whether the event rate rises under drought and how it relates to


79 substrate drying, stomatal conductance and PSII function measured on the same plants.


80 Second, we compared the spectral content of events of the two species at equal loudness, in


81 drought and watered plants. Third, we tested, as a proof of concept, whether events identify


82 the species of plants not used to train a classifier. We show that the share of event energy at


83 40-80 kHz differs about 10 fold between the two species, although their events peak at a


84 similar frequency, and that this share identifies the species. Drought raised the event rate,


85 which coincided with low stomatal conductance and fell again in single plants as they dried


86 further.


87 **Materials and methods**


88 _Plant material and drought treatment_

89 Sunflower ( _Helianthus annuus_ L.) and common buckwheat ( _Fagopyrum esculentum_ Moench)


90 were sown individually in 1 L pots (18.77 g empty) filled with 150 g of commercial peat


91 substrate (pH 5.5-6.5; 47.7 g dry mass). Plants grew in a glasshouse at Pozna ń University of


92 Life Sciences, Poland, under natural light (midday photosynthetic photon flux density 400

93 1100 µmol m &#91;−2&#93; s &#91;−1&#93; ), at 28-35 °C by day, 15-20 °C at night and 60-70% relative humidity.


94 They were watered to container capacity every two days and fertilised with a complete


95 nutrient solution. At 37 days after sowing, six uniform plants of each species were selected


96 and their pots brought to similar weights. Three became drought plants, and three remained


97 watered controls that were watered every two days until water dripped from the pot.


98 Watering of the drought plants stopped at 37 days after sowing (DAS). Sunflower was


99 recorded once the plants began to wilt, at 42, 44 and 47 DAS (5-10 days without water), and


100 buckwheat at 44, 48, 51, 55 and 58 DAS (7-21 days without water). Plants were always


101 recorded before watering. After recording, drought pots were rewatered to an estimated 40

102 50% of field capacity, allowing for the estimated plant mass.


4


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


103 _Pot weights and substrate water_

104 Pots were weighed after recording on each recording day and at 43 DAS (sunflower), and


105 drought pots also after rewatering. Field capacity was determined on three 100 g substrate


106 samples that were saturated, drained until dripping stopped, weighed and dried at 60 °C for 36


107 h. The substrate held 5.41 ± 0.11 g water g &#91;−1&#93; dry mass at field capacity, or 258 g per pot.


108 Substrate water (% of field capacity) was calculated as (pot weight − 66.5 g) ÷ 258 g × 100,


109 where 66.5 g is the pot plus dry substrate. Plant fresh mass was not subtracted, so these values


110 are upper limits. Daily water loss of the drought pots was calculated between successive


111 weighing, including the unweighed additions.


112 _Leaf gas exchange and chlorophyll fluorescence_

113 Leaf gas exchange and chlorophyll fluorescence were measured with a GFS-3000 system with


114 a fluorescence module (Heinz Walz GmbH, Effeltrich, Germany) on fully expanded leaves of


115 the plants used for the recordings, at 44 and 47 DAS for sunflower and at 48 and 51 DAS for


116 buckwheat, between 11:40 and 15:45 h. The measurement program was the same as in Haque


117 et al. (2026). The cuvette was set to a photosynthetic photon flux density of 1000 µmol m &#91;−2&#93;


118 s &#91;−1&#93; from the integrated LED source, a reference CO 2 concentration of 400 µmol mol &#91;−1&#93;, a water


119 vapour concentration of 17,000 µmol mol &#91;−1&#93; and a flow rate of 750 µmol s &#91;−1&#93;, and leaf


120 temperature followed ambient conditions. Two leaves were measured on each plant, each


121 once after the readings had reached steady state. We recorded stomatal conductance to water


122 vapour ( _g_ s ), transpiration ( _E_ ), intercellular CO 2 concentration ( _c_ i ) and the effective quantum


123 yield of PSII ( Φ PSII).


124 Chlorophyll _a_ fluorescence transients (OJIP) were recorded with a FluorPen (Photon Systems


125 Instruments, Drásov, Czech Republic) on two leaves of each recorded plant after 20 min of


126 dark adaptation in leaf clips, as in Haque et al. (2026), at 42, 44 and 47 DAS for sunflower


127 and at 44, 51, 55 and 58 DAS for buckwheat. From each transient we took the maximum


128 quantum yield of PSII ( _F_ v / _F_ m ), the performance index on absorption basis (PI abs ) and the


129 relative variable fluorescence at the J (2 ms) and I (30 ms) steps ( _V_ J, _V_ I ) (Strasser et al., 2000).


130 Transients were plotted as fluorescence relative to _F_ o and as the difference in relative variable


131 fluorescence between drought and watered plants. At 42 and 44 DAS the FluorPen recorded a


132 background (Bckg) of 139 instead of 422 for some sunflower leaves. These readings had


133 lower minimal and maximal fluorescence, and gave a higher PI abs and a lower _V_ J than the


134 other leaf of the same plant. Fluorescence variables were therefore compared only between


135 leaves with the same background value, so 42 DAS was not tested and 44 DAS had two


5


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


136 drought plants. Two drought buckwheat leaves at 51 DAS had an _F_ v / _F_ m of 0.20 and 0.32.


137 Their _F_ o lay within the range of the other leaves measured that day, while their _F_ m was 2-3


138 times lower than in any other leaf, so the low values came from a lack of variable


139 fluorescence. Leaves with an _F_ v / _F_ m of 0.47-0.54 also occurred in drought plants at 55 DAS.


140 These readings were retained, and results are also reported without them.


141 _Recording set-up_

142 Recordings were made in a wooden chamber lined with pyramidal acoustic foam (62×62×90


143 cm, inner width × depth × height; Fig. 1A) with two ultrasonic microphones (Ultramic 384K


144 BLE, Dodotronic, Castel Gandolfo, Italy) powered by mains USB adapters and recording at


145 384 kHz and 16 bit. The three plants of one treatment stood in a row between the


146 microphones, with M1 facing the plant at one end and M2 the plant at the other end (Fig. 1B

147 D; 5 cm from the nearest stem). The layout was the same for both species and treatments.


148 Each end plant was recorded in four consecutive 30-min recordings per day. On most days the


149 drought plants were recorded in the morning (10:10-13:15 h) and the controls in the early


150 afternoon (12:40-16:20 h); Supplementary Table S1 gives the time window of each group.


151 The thresholds were set on empty-chamber recordings.


152 _Spike repair and event detection_

153 The mains adapters introduced one-sample electrical spikes, broadband up to 192 kHz


154 (Supplementary Fig. S1). A sample was replaced by the five-point running median when it


155 differed from that median by more than six robust standard deviations and the energy at 150

156 184 kHz exceeded 9% of that at 20-100 kHz. In six test recordings, 705 of 709 events were


157 detected again after repair, with unchanged peak amplitude (ratio 0.999-1.002).


158 The repaired signal was split into four bands (20-40, 40-60, 60-80 and 80-100 kHz; fourth

159 order zero-phase Butterworth filters). Each band threshold was set 1 dB above the highest


160 amplitude in 16 empty-chamber recordings (479 min), giving −50.8, −59.8, −60.7 and −60.5


161 dBFS (decibels relative to full scale). An event began at a threshold crossing and ended after


162 10 ms without one. Overlapping events in different bands were merged, and the event level


163 was the highest band peak. A band event was kept when its peak exceeded eight times the


164 robust standard deviation of the band signal within ±50 ms. Recordings with a background


165 above 1.1 times the empty-chamber level were classed as having raised electrical background.


166 Only events above the highest threshold (−50.8 dBFS), called primary events, were analysed,


167 so that every band had the same sensitivity. Events recorded by both microphones were


6


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


168 counted once, at the louder microphone, and buckwheat events classed as electrical impulses


169 were removed.


170 _Waveform check and spectral features_

171 Clips of ±3 ms around each event (3,483 sunflower and 358 buckwheat clips) were spike

172 repaired and band-pass filtered at 15-150 kHz. The Hilbert envelope gave the event peak and


173 the duration above 25% of the peak. The spectrum was computed over 0.5 ms (Hann window)


174 from 0.1 ms before the peak, minus the spectrum of an equal background window 1.8 ms


175 earlier. From it we took the peak frequency, spectral centroid, bandwidth at −10 dB and the


176 shares of energy at 15-40, 40-80 and 80-150 kHz. The decay time was the time for the 20-40


177 kHz envelope to fall to 1/e of its peak. Spectrograms used 0.125 ms Hann windows.


178 Buckwheat clips were classed as ringing bursts ( ≥ 0.06 ms, stable frequency, most energy


179 below 80 kHz), electrical impulses (&lt;0.04 ms, much energy above 80 kHz) or unclear events.


180 All sunflower recordings had normal background. Species were compared at equal loudness,


181 with primary events between −50.8 and −44 dBFS (915 sunflower events from 37 recordings


182 and 90 buckwheat events from 38 recordings). For these events we also noted the bands in


183 which each event exceeded the threshold, in recordings with normal background.


184 _Events recorded by both microphones_

185 The microphones ran on separate clocks. For each pair of simultaneous recordings, the clock


186 offset was the most frequent time difference between M1 and M2 events within ±5 s, and a


187 line fitted to the candidate pairs corrected for drift. Event pairs within ±3 ms of this line were


188 classed as shared events. For the shared sunflower events of 42 DAS, the arrival-time


189 difference between events louder at M1 and at M2 was estimated by least squares, and the


190 spectra of the near (louder) and far copies were compared.


191 _Event size, amplitude index and timing_

192 For the sunflower drought plant at M1, which emitted the most events, the level distribution


193 on each day was described by the share of events above −35 dBFS and by the maximum

194 likelihood power-law exponent β = n ÷ Σ ln(A i ÷ A 0 ), where A i is the peak amplitude of event


195 i and A 0 that at −50.8 dBFS (a smaller β means more loud events). The within-recording


196 relation between level and the share of energy at 40-80 kHz was estimated by splitting each


197 event level into its recording mean and the deviation from it. As a relative index of emission


198 intensity, the squared peak amplitudes of the primary events, relative to full scale, were


199 summed per 30 min. This summed squared peak amplitude is not calibrated to sound pressure


200 and is not the energy of the whole waveform.


7


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


201 Event timing was tested in recordings with at least 30 primary events, by the coefficient of


202 variation of the intervals between events (1 for random, Poisson, timing) and by the Fano


203 factor of counts in 1, 10 and 60 s bins, compared with 2,000 random placements of the events.


204 Excess short intervals (&lt;0.1, 0.25 and 0.5 s) were tested with a Poisson test. Buckwheat


205 intervals from recordings with at least three events were scaled by their recording mean,


206 pooled and compared with the exponential distribution, and buckwheat counts were tested for


207 overdispersion among the recordings of each plant and day.


208 _Classification of species and treatment_

209 Logistic regression classifiers were trained on primary events between −50.8 and −44 dBFS,


210 using either the log share of energy at 40-80 kHz alone or eight features (the three energy


211 shares, spectral centroid, peak frequency, bandwidth, duration and decay time), standardised


212 and with balanced class weights. Each of the eight end plants was left out in turn and


213 classified by a model trained on the other seven, and the area under the receiver operating


214 characteristic curve (AUC) was computed from these predictions. To estimate how many


215 events identify a plant, 1-10 events were drawn without replacement 2,000 times from each


216 left-out plant and their predicted log-odds averaged. The smallest plant had 10 events, so the


217 10 event draws of the two smallest plants contained the same or nearly the same events.


218 Drought and watered plants were classified in the same way within each species and


219 compared with the AUC of the event rate of single plant-days (mean of the four recordings).


220 _Statistical analysis_

221 The end plant was the experimental unit for treatment effects on emission, with two drought


222 and two watered end plants per species. The four consecutive recordings of a plant on one day


223 were repeated measurements of that plant-day and were not treated as independent replicates.


224 With two plants per treatment, the model-based confidence intervals and p values are


225 approximate, and the treatment comparisons are exploratory. Event counts of each species


226 were analysed with a generalised linear mixed model (log link; glmmTMB) with treatment,


227 recording day and treatment × day as fixed effects, random intercepts for plant and plant-day,


228 and an offset for recording length. A negative binomial model was used for sunflower and a


229 Poisson model for buckwheat, whose counts were not overdispersed, without the 58 DAS


230 recordings, which had no controls. Day-specific rate ratios came from linear combinations of


231 the coefficients with Wald 95% confidence intervals (CI), and the treatment × day interaction


232 was tested with a likelihood-ratio test. The buckwheat model was refitted without the 55 DAS


233 recordings, with normal-background recordings only and with the background ratio as a


8


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


234 covariate. At plant level, the mean rate of each end plant over the days with controls was


235 compared between treatments with an exact permutation test stratified by species (36


236 allocations). Recording background and repaired spike samples were compared between


237 treatments with Mann-Whitney U tests on recordings.


238 Event levels and spectral features were compared with linear mixed models with plant and


239 recording as random effects and event level as a covariate, with log-transformed energy share


240 at 40-80 kHz and decay time and with Kenward-Roger degrees of freedom. Treatment and


241 species × treatment terms were tested in the same models. Band detection was compared


242 between species with plants as units, as percentage-point differences with 95% CI from 5,000


243 bootstrap samples of plants within species and permutation tests over the 70 allocations of the


244 eight end plants. For plant-level comparisons, events recorded at each microphone were


245 attributed to the nearest end plant. Gas-exchange and fluorescence variables were averaged


246 over the two leaves of each plant and compared between treatments on each day with two

247 sided Welch's t-tests (n = 3 plants). Plants were compared with exact Mann-Whitney U tests,


248 and near and far copies of shared events with Wilcoxon signed-rank tests. All tests were two

249 sided. Detection was run in R&#91;4.6.2&#93; (tuneR, signal, seewave, readr, dplyr, stringr), mixed


250 models in R (glmmTMB, lme4, lmerTest) and all other analyses in Python 3 (NumPy, SciPy,


251 statsmodels, scikit-learn).


252 **Results**


253 _Recording quality and event checks_

254 All sunflower recordings had a background of 1.00-1.09 times the empty-chamber level, and


255 drought and watered recordings did not differ in the number of repaired spike samples (p =


256 0.13 in sunflower and 0.12 in buckwheat). In buckwheat, all recordings at 55 and 58 DAS and


257 the M2 recordings at 44 DAS had a raised background (1.10-1.33 times). Buckwheat events


258 above −50.8 dBFS were mostly ringing bursts with a spectral peak near 28-30 kHz, and


259 weaker events were mostly background peaks or single impulses. The final data set held 1,698


260 primary events from drought and 198 from watered sunflower, and 72 from drought and 26


261 from watered buckwheat.


262 _Drought raised event rates in both species_

263 Each end plant was followed over the recording days (Fig. 2A-D; Table 1). At 42 DAS, both


264 sunflower drought end plants emitted far more events than the controls (103 and 52 against


265 2.0 and 0.25 events per 30 min). The two drought end plants then followed different courses.


9


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


266 The plant at M1 emitted 166 events per 30 min at 44 DAS and 9.5 at 47 DAS (Fig. 2A), and


267 the plant at M2 emitted 0.25 and 94 (Fig. 2B). The control at M2 was also nearly silent at 44


268 DAS (0.5 events per 30 min), and control rates rose to 13.5-19.5 events per 30 min at 47


269 DAS.


270 Buckwheat emitted far fewer events (Fig. 2C, D). Drought and watered plants both emitted


271 0.25 events per 30 min at 44 DAS, and controls emitted more at 48 DAS (2.0 against 0.5


272 events per 30 min, 8 against 2 events). Both drought end plants then rose to 3.0 events per 30


273 min at 51 DAS (controls 0.5-0.67) and to 2.8-4.8 at 55 DAS (controls 0-1.0), and fell to 1.3

274 1.8 at 58 DAS, when no controls were recorded.


275 In the mixed model, sunflower drought plants emitted 107 times more events than controls at


276 42 DAS (95% CI 4.5-2533, p = 0.004; Fig. 2E; Table 1). At 44 and 47 DAS, when only one


277 drought end plant exceeded the controls, the effect was not significant (rate ratios 3.2 and 1.9,


278 p = 0.47 and 0.69). In buckwheat, controls emitted more at 48 DAS (rate ratio 0.25, 95% CI


279 0.08-0.75, p = 0.013), and drought plants emitted more at 51 DAS (5.3, 1.8-15.1, p = 0.002)


280 and 55 DAS (7.5, 2.6-21.3, p &lt; 0.001). Over the days with controls, both drought end plants


281 of each species emitted more events than both watered end plants (sunflower 93 and 49


282 against 12 and 4.8 events per 30 min; buckwheat 2.1 and 1.6 against 0.7 and 0.9; Table 1).


283 With two plants per treatment, this complete separation in both species is the most extreme


284 ordering possible (exact permutation test, p = 0.056).


285 The buckwheat result did not depend on the recordings with raised background


286 (Supplementary Table S2). Without 55 DAS, drought plants still emitted 5.3 times more


287 events than controls at 51 DAS (95% CI 1.8-15.1, p = 0.002), and normal-background


288 recordings alone gave the same value. The local background check removed most detections


289 at 55 DAS in both treatments, so the result of that day is less certain than that of 51 DAS.


290 _Sunflower events carried more energy above 40 kHz_

291 Events of drought sunflower had a higher median level than those of watered sunflower


292 (−43.1 against −46.9 dBFS), but with two plants per treatment the difference was not


293 significant (3.0 dB, p = 0.35). Buckwheat events lay close to the counting level in both


294 treatments (median −49.5 and −47.9 dBFS, p = 0.34). Single events of both species were short


295 ringing bursts with most of their energy near 30 kHz (Fig. 3A). At equal loudness, the median


296 spectrum of sunflower stayed high above 40 kHz, and that of buckwheat fell steeply above 35


297 kHz (Fig. 3B). Sunflower events carried a median of 24% of their energy at 40-80 kHz,


10


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


298 against 2% in buckwheat (Fig. 3C; Table 2). With plants and recordings as random effects,


299 this share was 9.5 times higher in sunflower (95% CI 6.1-14.7, p &lt; 0.001). Sunflower events


300 also had a higher spectral centroid, were shorter and decayed faster (Table 2). The median


301 peak frequency was similar (30.2 against 28.8 kHz). Although 30% of sunflower events


302 peaked above 35 kHz, against 4% in buckwheat, the mean peak frequency did not differ


303 significantly (4.7 kHz higher in sunflower, 95% CI −1.4 to 10.8, p = 0.11).


304 The difference in the share at 40-80 kHz held for every plant and recording day (Fig. 3C).


305 Plant medians ranged from 13.9 to 36.3% in sunflower and from 1.5 to 2.2% in buckwheat


306 (exact Mann-Whitney test, four plants per species, p = 0.029), and all nine sunflower plant

307 days with at least five events exceeded all seven buckwheat plant-days. Peak frequency did


308 not separate the plants (plant medians 29.1-31.7 against 28.2-29.5 kHz, p = 0.11). Within each


309 species the share did not differ between drought and watered plants (sunflower 23.5% against


310 33.6%, p = 0.67; buckwheat 2.0% against 1.8%, p = 0.33; species × treatment, p = 0.20). The


311 difference also held in the narrowest level class (30% against 2% between −50.8 and −49.5


312 dBFS). Across all single events, the share at 40-80 kHz separated the species with an AUC of


313 0.92, against 0.67 for peak frequency.


314 The detection bands showed the same difference (Fig. 3D; Table 2). Of 915 sunflower events


315 between −50.8 and −44 dBFS, 81% were detected in the 40-60 kHz band and 49% in the 60

316 80 kHz band. Of 47 buckwheat events from recordings with normal background, 4% reached


317 the 40-60 kHz band and none reached higher bands. Detection at 40-60 kHz was 77


318 percentage points higher in sunflower (95% CI 71-81 from resampling plants), and every


319 sunflower plant exceeded every buckwheat plant in each band measure (plant-level


320 permutation test, p = 0.029, the lowest possible with four plants per species; Supplementary


321 Table S3). In sunflower, detection at 40-60 kHz was similar in drought and watered plants


322 (81-83% against 67-81%), and the 80-100 kHz band was reached by 3% and 19% of events of


323 the two watered plants, against 0.3% and 0.9% of the drought plants.


324 _Events recorded by both microphones_

325 Shared events were rare. At 42 DAS, 35 events of the sunflower drought plants were recorded


326 by both microphones, 6% of the 621 primary events of that day. Each was 12-27 dB louder at


327 one microphone (Fig. 4A), more than the other shared events (median 20.7 against 5.7 dB, p &lt;


328 0.001). After correction for clock offset and drift, the two groups differed in arrival time by


329 2.24 ms (95% CI 2.20-2.27), and single events lay within 0.1 ms of their group median (Fig.


330 4B). Only six other shared events were found in sunflower and three in buckwheat. The far


11


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


331 copy of an event kept its share of energy at 40-80 kHz when the event was louder at M2


332 (median 81% at both microphones, n = 21), and lost more than half of it when the event was


333 louder at M1 (9.1% against 2.3%, p = 0.011, n = 14; Fig. 4C).


334 _Substrate drying in the drought pots_

335 The drought pots lost water steadily (Fig. 5A, B; Supplementary Table S4). Before


336 rewatering, the sunflower drought pots held at most 50-61% of field capacity at 42 DAS and


337 36-38% at 47 DAS. At 44 DAS the pots of the end plants at M1 and M2 held at most 49% and


338 46% of field capacity, and by 47 DAS both had lost 63-64 g of water over three days. The


339 buckwheat drought pots held at most 72-78% of field capacity at 44 DAS, 51-55% at 51 DAS


340 and 34-38% at 58 DAS. Their event rates rose to 3.0-4.8 per 30 min at 42-55% of field


341 capacity (51 and 55 DAS) and fell to 1.3-1.8 at 34-38% (58 DAS).


342 _Stomatal conductance and PSII function of drought plants_

343 Drought plants of both species had much lower stomatal conductance and transpiration than


344 watered plants (Fig. 6A, B; Supplementary Table S5). In sunflower, _g_ s of drought plants was


345 33 and 17 mmol m &#91;−2&#93; s &#91;−1&#93; at 44 and 47 DAS, against 237 and 235 mmol m &#91;−2&#93; s &#91;−1&#93; in watered


346 plants (p &lt; 0.001 and p = 0.014), and their _c_ i rose to 477 µmol mol &#91;−1&#93; at 47 DAS. In


347 buckwheat, _g_ s of drought plants fell from 55 mmol m &#91;−2&#93; s &#91;−1&#93; at 48 DAS to 21 mmol m &#91;−2&#93; s &#91;−1&#93; at


348 51 DAS, against 179 and 132 mmol m &#91;−2&#93; s &#91;−1&#93; in watered plants (p &lt; 0.001 and p = 0.019).


349 Within each day, the two sunflower drought end plants had similar _g_ s (30 and 25 mmol m &#91;−2&#93;


350 s &#91;−1&#93; at 44 DAS, 19 and 17 mmol m &#91;−2&#93; s &#91;−1&#93; at 47 DAS), although their event rates differed widely


351 (166 against 0.25 and 9.5 against 94 events per 30 min; Supplementary Table S6). Φ PSII


352 differed between treatments only in buckwheat at 48 DAS (0.17 in drought against 0.07 in


353 watered plants, p = 0.039).


354 Drought lowered PSII performance in both species (Fig. 6C-F; Supplementary Table S7;


355 Supplementary Fig. S2). In sunflower, _F_ v / _F_ m of drought plants stayed at 0.83-0.85. At 42


356 DAS it was numerically similar to that of watered plants (0.845 against 0.855, not tested), and


357 at 44 and 47 DAS slightly lower (p = 0.001 and 0.009). Their PI abs was 30% lower than in


358 watered plants at 44 DAS (8.2 against 11.7, n = 2 drought plants, p = 0.031) and half as high


359 at 47 DAS (4.7 against 9.3, p = 0.011). In buckwheat, PI abs of drought plants was similar to


360 that of watered plants at 44 DAS (3.6 against 3.9, p = 0.85) and fell to a quarter at 51 DAS


361 (0.63 against 2.38, p &lt; 0.001), when their _F_ v / _F_ m was 0.57 against 0.80 (p = 0.096). Without


362 two leaves with an _F_ v / _F_ m of 0.20 and 0.32, PI abs was 0.97 (p = 0.001) and _F_ v / _F_ m 0.74 (p =


363 0.009). PI abs stayed low at 55 and 58 DAS, and watered buckwheat kept an _F_ v / _F_ m of 0.80

12


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


364 0.83. Over the series, raised event rates coincided with low _g_ s in both species (Fig. 6G, H). In


365 sunflower, event rates were already raised at 42 DAS, and a lower PI abs was detected from 44


366 DAS. In buckwheat, event rates rose when PI abs fell.


367 _Event size and spectrum during drying_

368 The sunflower drought plant at M1 changed its sound as it dried (Fig. 5C, D). The share of its


369 events above −35 dBFS rose from 13% at 42 DAS to 41% at 44 DAS, the day of its highest


370 rate, when its events were 5.7 dB louder (p &lt; 0.001). The power law exponent fell from 1.04


371 to 0.62, so loud events became more common. At 47 DAS, 5% of its 38 events exceeded −35


372 dBFS (exponent 1.29). At equal loudness, its share of energy at 40-80 kHz was 24%, 10% and


373 36% at 42, 44 and 47 DAS (44 against 42 DAS, ratio 0.57, p &lt; 0.001; 47 against 42 DAS,


374 ratio 1.39, p = 0.026), while the watered plant at the same microphone stayed at 36-37%.


375 Within recordings, each 10 dB increase in level reduced this share by a factor of 0.76 (95% CI


376 0.71-0.82). The drought plant at M2 did not show this relation (factor 1.07, p = 0.49). The


377 summed squared peak amplitude of the plant at M1 rose 9.7 fold from 42 to 44 DAS while its


378 count rose 1.6 fold. Over all drought recordings it was about 710 times higher in sunflower


379 than in buckwheat, against 39 times more events.


380 The events of both buckwheat drought end plants were pooled (Fig. 5E). They stayed close to


381 the counting level at 51 and 55 DAS (median −49.8 and −49.6 dBFS) and were 3.8 dB louder


382 at 58 DAS (95% CI 1.3-6.4, p = 0.003). Against 55 DAS alone, which had a similar


383 background, the difference was 3.2 dB (p = 0.08).


384 _Timing of events_

385 Events occurred at random in time (Fig. 5F). In the 15 recordings with at least 30 primary


386 events, all from sunflower drought plants (1,632 events), the coefficient of variation of the


387 intervals between events was 0.85-1.14 (median 0.96), close to the value of 1 expected for


388 random timing. In 14 of the 15 recordings, the Fano factor lay within the range of random


389 simulations at 1, 10 and 60 s. Intervals shorter than 0.25 s were slightly more frequent than


390 expected (38 against 27.8, p = 0.04). The 50 intervals pooled from 17 buckwheat recordings


391 agreed with random timing (Kolmogorov-Smirnov test, p = 0.24).


392 _Identifying species and drought from the sounds_

393 As a proof of concept, the share of energy at 40-80 kHz alone identified the species of single


394 events from plants not used for training (Fig. 7A, B). Its AUC was 0.91, and 80% of single


395 events were assigned to the correct species. Eight spectral features together did not do better


13


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


396 (AUC 0.90). With three events from a left-out plant, the species was correct in 93% of draws,


397 and with five events in 96%, with 85% for the least well classified plant. The 99.6% reached


398 with 10 events is less informative, because the two watered buckwheat plants had only 10 and


399 13 events. Within each species, the spectrum separated drought and watered plants no better


400 than chance (AUC 0.44 in sunflower and 0.40 in buckwheat), and the event rate of single


401 plant-days separated them with AUCs of 0.76 and 0.70.


402 **Discussion**


403 Our results bear on three points raised by Font Bordera et al. (2026), namely how emission


404 relates to plant water status, whether emissions differ between species and whether


405 classification can support plant monitoring.


406 _Drought raised emission rates in both species_

407 In sunflower both drought end plants emitted far more than the controls at 42 DAS, and on


408 each later day one of them did (Fig. 2A, B, E; Table 1). In buckwheat both drought plants


409 emitted more from 51 DAS, 14 days after watering stopped (Fig. 2C-E). Absolute rates are


410 hard to compare because thresholds, microphone distance, plant size and counting rules differ


411 (Supplementary Table S8).


412 We regard 51 DAS as the main evidence for a drought effect in buckwheat. On that day the


413 background was normal and the effect held in every sensitivity analysis (Supplementary Table


414 S2), while the larger ratio at 55 DAS came from recordings with raised background. With two


415 drought and two watered end plants per species, the model estimates are exploratory and their


416 intervals wide (Table 1). The drought effect was clear only on days when both drought plants


417 emitted strongly, and the rate results need confirmation on more plants. Drought plants were


418 also usually recorded in the morning and controls in the early afternoon (Supplementary


419 Table S1), and plant water status changes over the day. The morning slot alone did not


420 produce high rates, because at 48 DAS the afternoon controls of buckwheat emitted more than


421 the morning drought plants, and each sunflower drought plant emitted at or below control


422 level on one morning. Recording the treatments together or alternately would remove this


423 confound. Drought pots were partly rewatered after each recording day, and recovery was not


424 recorded. Sunflower does not refill embolised conduits after rewatering (Stewart et al., 2026),


425 so if events mark embolism, its later events came from conduits not yet embolised.


426 Emission rate did not simply follow the water content of the substrate (Fig. 5A, B;


427 Supplementary Table S4). The sunflower drought pots dried in parallel, yet the plant at M1


14


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


428 peaked at 44 DAS and nearly stopped by 47 DAS while its pot kept losing water, and the


429 plant at M2 peaked at 47 DAS (Fig. 2A, B). Part of the low at M2 at 44 DAS probably came


430 from the microphone, because the control at M2 was also nearly silent and M2 recorded none


431 of the 275 loud events of the plant at M1. Differences between the plants, in size or xylem


432 vulnerability, are a more likely cause than differences in drying. In both species, emission was


433 highest at intermediate substrate water and lower in the driest pots, as in drying tomato (Khait


434 et al., 2023). A single low rate can therefore come from an unstressed plant or from one past


435 its peak. Control rates in sunflower rose over the series (Fig. 2A, B), probably because the


436 plants grew and their pots dried between waterings, and this narrowed the difference between


437 treatments.


438 _Emission rose on days of low stomatal conductance_

439 On the days with gas-exchange data, emission was raised only when stomatal conductance of


440 drought plants was low (Fig. 6). In sunflower, at least one drought plant emitted many events


441 on both measurement days, when _g_ s of drought plants was 17-33 mmol m &#91;−2&#93; s &#91;−1&#93;, or 7-14% of


442 that of watered plants. In buckwheat, the event rate of drought plants rose from 0.5 to 3.0


443 events per 30 min between 48 and 51 DAS, while their _g_ s fell from 55 to 21 mmol m &#91;−2&#93; s &#91;−1&#93; .


444 This association over two days per species does not show that stomatal closure causes the


445 emissions. A simpler explanation is that both changes follow falling water potential. Stomata


446 close at higher water potentials than those that start embolism (Martin-StPaul et al., 2017),


447 and direct observations in leaves showed embolism starting only after stomatal closure (Creek


448 et al., 2020). Closure slows the further rise in xylem tension, but water is still lost through the


449 cuticle and incompletely closed stomata (Duursma et al., 2019), so tension keeps rising as the


450 substrate dries and the most vulnerable conduits cavitate. Emissions are therefore expected


451 mainly after stomatal closure, although two measurement days cannot show exactly when


452 emission began.


453 In sunflower, emission was already raised on the first recording day, when Fv/Fm was


454 numerically similar between treatments, and a lower PIabs was detected from 44 DAS (Fig.


455 6C, E, G). The fluorescence of that first day could not be tested, only two drought plants


456 could be compared at 44 DAS, and PI abs is more sensitive to drought than _F_ v / _F_ m (Oukarroum


457 et al., 2007). Our data therefore do not show that emission rose before every change in PSII


458 function. In greenhouse tomato, PSII yield likewise showed no early response while emission


459 did (Dutta et al., 2025). In buckwheat, PI abs had already fallen to less than half of the control


460 value when emission rose at 51 DAS, with or without the two leaves with very low _F_ v / _F_ m


15


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


461 (Fig. 6F, H), which shows that the low emission rate of buckwheat did not come from milder


462 stress.


463 We propose a conceptual hydraulic model that links these observations (Fig. 8). It is a


464 hypothesis, because we did not measure water potential or embolism. If each event marks one


465 embolising conduit, the event rate is the slope of the vulnerability curve times the rate at


466 which water potential falls. Cumulative emission energy follows the loss of conductivity in


467 drying wood (Mayr and Rosner, 2011), and after stomatal closure water potential falls


468 through residual water loss (Duursma et al., 2019). The model predicts few events before


469 closure, a peak on the steep part of the curve and a decline once the vulnerable conduits are


470 used up. The sunflower plant at M1 followed this course (Fig. 2A), which is consistent with


471 the model without testing it. Sunflower stems lose half their conductivity near −3 MPa


472 (Ahmad et al., 2018), so the events at moderate drying probably came from the most


473 vulnerable conduits. The model could also explain why the two sunflower end plants emitted


474 at rates 10 fold apart at 47 DAS, although both had a _g_ s near 18 mmol m &#91;−2&#93; s &#91;−1&#93; (Supplementary


475 Table S6). Near its minimum, _g_ s no longer tracks water potential, and the two plants could sit


476 at different points on their curves. The plant at M1, past its emission peak, also had the lower


477 PI abs (3.8 against 5.6). Event rate may therefore carry information that gas exchange does not,


478 and calibrating it against leaf water potential, as proposed by Font Bordera et al. (2026),


479 would test the model.


480 _The two species differ in the spectral content of their events_

481 At equal loudness, sunflower events carried about 10 times more of their energy at 40-80 kHz


482 than buckwheat events (Fig. 3C; Table 2), and about half of them reached the 60-80 kHz


483 band, against none in buckwheat (Fig. 3D; Supplementary Table S3). The difference held in


484 every check and did not differ between drought and watered plants (Fig. 3C, D). Most events


485 of both species peaked near 29-30 kHz (Fig. 3B; Table 2). A peak shared by two species


486 probably reflects the response of the microphone and chamber, because the waveform


487 changes on its way from the source to the sensor (Vergeynst et al., 2015), and microphones


488 near tree saplings also recorded about 30 kHz on average (Klaminder et al., 2025). Khait et al.


489 (2023) separated tomato and tobacco by mean peak frequency. In our data peak frequency


490 was a weak marker, and the energy above 40 kHz and band specific detection, which a shared


491 resonance affects less, were much stronger. Font Bordera et al. (2026) described plant


492 emissions as specific to both species and stress, and our data support species specificity for


16


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


493 sunflower and buckwheat. At equal loudness, the spectrum did not differ between drought and


494 watered plants over all days, although it shifted during drying in one plant (Fig. 5D).


495 The difference is a property of the plants and the recording geometry together, and our data


496 cannot say which plant property produces it. The microphones, chamber, thresholds and


497 microphone-to-stem distance were the same for both species, and the matched loudness


498 removes differences in event level. The species were, however, recorded in separate series,


499 buckwheat in the two weeks after sunflower, so a change in the recording chain between the


500 series cannot be excluded, although the difference held on every recording day of both series.


501 Three plant-related sources remain. The first is the source itself. Font Bordera et al. (2026)


502 attributed species signatures to vessel diameter, tissue anatomy or wood density, and in the


503 vessel model of Dutta et al. (2022) resonance frequency falls and settling time rises with


504 vessel radius. A source origin would therefore require narrower vessels in sunflower, whose


505 events were shorter and richer in high frequencies than those of buckwheat. The second is the


506 transmission of the event through stem and leaf tissue, which depends on stem diameter and


507 on tissue properties, because lignified xylem walls carry high frequencies and softer tissues


508 damp them (Font Bordera et al., 2026). The third is radiation into the air, which depends on


509 plant size and on where the emitting tissue sits relative to the microphone. The taller


510 sunflower plants would have placed more sources further from the microphone, which would


511 lower their high-frequency share. Vessel diameters, a contact sensor on the stem recording


512 with the microphone, and recordings at several distances would separate these sources.


513 _Classification as a proof of concept_

514 The share at 40-80 kHz identified the species of plants not used for training (Fig. 7A). Single


515 events were assigned correctly in 80% of cases, similar to the 70-80% reported for single


516 tomato and tobacco sounds (Khait et al., 2023), and five events from a plant gave at least 85%


517 accuracy. Font Bordera et al. (2026) presented classification of plant ultrasound as a route to


518 precision agriculture. Our result comes from four plants per species of one cultivar in one


519 chamber, and is a proof of concept. A diagnostic method would need validation on many


520 more plants, growth stages and recording systems. It does show that one interpretable spectral


521 feature can carry most of the species’ information. The spectrum did not separate drought and


522 watered plants, while the event rate did (Fig. 7B). A sensor could use the spectrum to identify


523 the crop and the course of the event rate to detect drought, because a plant past its emission


524 peak can emit as little as an unstressed one. A monitoring system would also need a


17


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


525 sensitivity check for each microphone, because a failing microphone looks like a quiet plant;


526 the events recorded by two microphones in common provide one.


527 _Shared events suggest that each microphone recorded mainly the nearest plant_
528 At 343 m s &#91;−1&#93;, the 2.24 ms difference in arrival time between the two groups of shared events


529 (Fig. 4B) corresponds to a path difference of about 0.38 m, which fits a source near one end


530 of the row. With the near microphone 5 cm from the stem, spherical spreading to the far


531 microphone, about 0.43 m away, predicts a loss of 19 dB, close to the observed 21 dB (Fig.


532 4A). Events from the middle plant would arrive about 14 dB weaker, and only 27% of


533 sunflower and 1% of buckwheat events would still be detected. Shared events were few,


534 however (35 of 621 events at 42 DAS) and came mostly from one day, so events from the


535 middle plant cannot be excluded, and we treat M1 and M2 as positions dominated by the


536 nearest plant. The far copies of events from the plant at M1 lost more than half of their high


537 frequency share (Fig. 4C), although the extra distance was the same in both directions. This


538 points to the microphones or their orientation, which studies comparing spectra should keep


539 fixed.


540 _The size and spectrum of events changed as plants dried_

541 In the sunflower plant at M1, events at the emission peak were louder and carried less energy


542 above 40 kHz than two days earlier, and after the peak the few remaining events were quiet


543 and rich in high frequencies (Fig. 5C, D). Within recordings, louder events also had less high

544 frequency energy. More distant sources would give quieter events with less high-frequency


545 energy, the opposite relation, and the watered plant at the same microphone had stable


546 spectra, which places the relation in the plant. If loud, low-pitched events come from large


547 conduits (Mayr and Rosner, 2011; Ponomarenko et al., 2014; Khait et al., 2023), this plant


548 would have lost mainly large conduits at its emission peak and mainly small ones later. The


549 amplitude and frequency of emissions do not always reflect embolism size, however


550 (Vergeynst et al., 2015; Lamacque et al., 2022; Font Bordera et al., 2026), and this


551 interpretation rests on one plant. The summed squared peak amplitude amplified the changes


552 in count, as expected if emission energy follows the loss of conductivity (Mayr and Rosner,


553 2011).


554 _Events occurred at random in time_

555 Fast cascades of embolism spreading between conduits would cluster events at short intervals,


556 but events occurred at random in time (Fig. 5F). We found only about 10 extra pairs shorter


557 than 0.25 s among 1,617 intervals, so most detected events were independent. With random


18


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


558 timing, a count of 100 events gives the rate within about ±20%, which would take about 25 h


559 at the buckwheat rate of about two events per 30 min.


560 **Conclusions**


561 Airborne ultrasonic emissions of sunflower and buckwheat carried species identity in their


562 spectrum and drought in their event rate. At equal loudness, sunflower events carried about 10


563 times more energy at 40-80 kHz than buckwheat events in drought and watered plants alike,


564 although events of both species peaked near 30 kHz. As a proof of concept on four plants per


565 species and one cultivar, this share identified the species from a few events. Every drought


566 plant emitted more events than the watered plants of its species. In single plants the rate rose


567 and then fell, so a low rate can come from an unstressed plant or from one past its emission


568 peak. The rise coincided with low stomatal conductance, as expected if emissions arise as


569 xylem tension keeps rising after stomata close, and in sunflower emission was already raised


570 when maximum PSII quantum yield was still similar between treatments. To our knowledge,


571 these are the first airborne recordings of either species. Recordings of more plants, with both


572 treatments at the same time of day and with leaf water potential, stem anatomy and contact


573 sensors on the same plants, will be needed to link these signals to plant hydraulics.


574 **Author contributions**


575 BS: conceptualization, investigation, data curation, formal analysis, methodology, writing 

576 original draft, writing- review and editing; MIH: investigation, data curation, formal analysis,


577 writing- review and editing; MS: investigation; AR: conceptualization, supervision, writing 

578 review and editing, resources.


579 **Use of generative AI**


580 During the preparation of this work, the author used the AI-assisted tool Claude (Anthropic)


581 to improve the readability and language of the manuscript and to assist with drafting figures.


582 The author reviewed and edited all content and takes full responsibility for the publication.


583 **Conflict of interest**


584 No conflict of interest declared.


585 **Funding**


586 This work was supported by the Polish National Agency for Academic Exchange (NAWA)


587 through an Ulam postdoctoral fellowship to BS (BNI/ULM/2024/1/00145).


19


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


588 **Data availability**


589 The recordings, event tables and analysis scripts (R and Python) are available from the


590 corresponding author upon request.


591 **References**


592 Ahmad HB, Lens F, Capdeville G, Burlett R, Lamarque LJ, Delzon S. 2018. Intraspecific variation in
593 embolism resistance and stem anatomy across four sunflower ( _Helianthus annuus_ L.) accessions.
594 _Physiologia Plantarum_ 163, 59-72.


595 Boccato E, Petruzzellis F, Bordenave CD, Nardini A, Tretiach M, Mayr S, Candotto Carniel F. 2024.
596 The sound of lichens: ultrasonic acoustic emissions during desiccation question cavitation events
597 in the hyphae. _Journal of Experimental Botany_ 75, 6579-6592.


598 Bonisoli L, Pelicella I, Arru L. 2024. Plant ultrasound detection: a cost-effective method for

599 identifying plant ultrasonic emissions. _Plant Signaling &amp; Behavior_ 19, e2310974.


600 Bonisoli L, Forti L, Arru L. 2025. Outdoor detection of plant ultrasonic emissions using a contactless
601 microphone. _Physiologia_ 5, 9.


602 Creek D, Lamarque LJ, Torres-Ruiz JM, Parise C, Burlett R, Tissue DT, Delzon S. 2020. Xylem
603 embolism in leaves does not occur with open stomata: evidence from direct observations using the
604 optical visualization technique. _Journal of Experimental Botany_ 71, 1151-1159.


605 De Boeck K, Steppe K. 2025. Revised method for constructing acoustic vulnerability curves in trees.
606 _Tree Physiology_ 45, tpaf001.


607 Dutta S, Chen Z, Kaiser E, Matamoros PM, Steeneken PG, Verbiest GJ. 2022. Ultrasound pulse
608 emission spectroscopy method to characterize xylem conduits in plant stems. _Research_ 2022,

609 9790438.


610 Dutta S, van den Berg T, Koning M, et al. 2025. Comparison of multiple plant sensors aimed at early
611 detection of drought stress in the greenhouse. _Agricultural Water Management_ 315, 109535.


612 Duursma RA, Blackman CJ, López R, Martin-StPaul NK, Cochard H, Medlyn BE. 2019. On the
613 minimum leaf conductance: its role in models of plant water use, and ecological and
614 environmental controls. _New Phytologist_ 221, 693-705.


615 Font Bordera P, Crespo Martinez E, Ombrosi D, Forment J, Brumos J. 2026. Plant bioacoustics:
616 decoding ultrasonic emissions. _Journal of Experimental Botany_, doi: 10.1093/jxb/erag398.


617 Haque MI, Saha B, Juszczak R, Rastogi A. 2026. Foliar silicon enhances drought resilience and
618 productivity in buckwheat by stabilizing photosynthetic performance. _Scientific Reports_ 16,

619 23271.


620 Hase A, Ogihara G. 2026. Application of acoustic emission sensing for detecting water stress in
621 lettuce grown in plant factories. _Next Research_ 12, 102250.


622 Hölttä T, Vesala T, Nikinmaa E, Perämäki M, Siivola E, Mencuccini M. 2005. Field measurements of

623 ultrasonic acoustic emissions and stem diameter variations. New insight into the relationship
624 between xylem tensions and embolism. _Tree Physiology_ 25, 237-243.


20


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


625 Hou Z, Yin J, Lu Y, Song J, Wang S, Wei S, Liu Z, Zhang Y, Fang Z. 2019. Transcriptomic analysis
626 reveals the temporal and spatial changes in physiological process and gene expression in common
627 buckwheat ( _Fagopyrum esculentum_ Moench) grown under drought stress. _Agronomy_ 9, 569.


628 Jia X, Zhang L, Ma R, Wang Z, Zhang Q, Yao Y. 2006. Ultrasonic acoustic emissions from leaf
629 xylem of potted wheat subject to a soil drought and rewatering cycle. _Agricultural Sciences in_

630 _China_ 5, 346-355.


631 Khait I, Lewin-Epstein O, Sharon R, et al. 2023. Sounds emitted by plants under stress are airborne

632 and informative. _Cell_ 186, 1328-1336.


633 Klaminder J, Rollo MM Jr, Näsholm T, Viklund J, Henriksson N, Jämtgård S. 2025. Ultrasonic
634 acoustic emissions as indicators of tree drought stress in outdoor forest settings: testing the
635 concept using cut saplings. _Ecological Indicators_ 176, 113580.


636 Lamacque L, Sabin F, Améglio T, Herbette S, Charrier G. 2022. Detection of acoustic events in
637 lavender for measuring xylem vulnerability to embolism and cellular damage. _Journal of_
638 _Experimental Botany_ 73, 3699-3710.


639 Martin-StPaul N, Delzon S, Cochard H. 2017. Plant resistance to drought depends on timely stomatal
640 closure. _Ecology Letters_ 20, 1437-1447.


641 Mayr S, Rosner S. 2011. Cavitation in dehydrating xylem of _Picea abies_ : energy properties of
642 ultrasonic emissions reflect tracheid dimensions. _Tree Physiology_ 31, 59-67.


643 Milburn JA, Johnson RPC. 1966. The conduction of sap. II. Detection of vibrations produced by sap
644 cavitation in _Ricinus_ xylem. _Planta_ 69, 43-52.


645 Nardini A, Cochard H, Mayr S. 2024. Talk is cheap: rediscovering sounds made by plants. _Trends in_

646 _Plant Science_ 29, 662-667.


647 Oukarroum A, El Madidi S, Schansker G, Strasser RJ. 2007. Probing the responses of barley cultivars
648 ( _Hordeum vulgare_ L.) by chlorophyll _a_ fluorescence OLKJIP under drought stress and re649 watering. _Environmental and Experimental Botany_ 60, 438-446.


650 Ponomarenko A, Vincent O, Pietriga A, Cochard H, Badel É, Marmottant P. 2014. Ultrasonic
651 emissions reveal individual cavitation bubbles in water-stressed wood. _Journal of the Royal_
652 _Society Interface_ 11, 20140480.


653 Saha B, Rastogi A. 2026. Plant acoustic emission as early stress signals: towards remote integrated
654 monitoring for sustainable agriculture. _European Journal of Agronomy_ 177, 128100.


655 SharathKumar M, Bieling TJ, de Klerk BB, Dutta S, Verbiest GJ. 2026. Tuning into plant ultrasonics:
656 can acoustics enable smart agriculture? _Computers and Electronics in Agriculture_ 254, 112274.


657 Son J-S, Jang S, Mathevon N, Ryu C-M. 2024. Is plant acoustic communication fact or fiction? _New_
658 _Phytologist_ 242, 1876-1880.


659 Stewart JJ, Allen BS, Polutchko SK, Hunter C, Ocheltree TW, Gleason SM. 2026. No evidence of

660 xylem embolism refilling during recovery from drought stress in intact sunflowers. _Journal of_
661 _Experimental Botany_ 77, 5890-5900.


21


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


662 Strasser RJ, Srivastava A, Tsimilli-Michael M. 2000. The fluorescence transient as a tool to

663 characterize and screen photosynthetic samples. In: Yunus M, Pathre U, Mohanty P, eds. Probing
664 photosynthesis: mechanisms, regulation and adaptation. London: Taylor and Francis, 445-483.


665 Tyree MT, Dixon MA. 1983. Cavitation events in _Thuja occidentalis_ L.? Ultrasonic acoustic
666 emissions from the sapwood can be measured. _Plant Physiology_ 72, 1094-1099.


667 Vergeynst LL, Sause MGR, Hamstad MA, Steppe K. 2015. Deciphering acoustic emission signals in
668 drought stressed branches: the missing link between source and sensor. _Frontiers in Plant Science_

669 6, 494.


670 Vergeynst LL, Sause MGR, De Baerdemaeker NJF, De Roo L, Steppe K. 2016. Clustering reveals
671 cavitation-related acoustic emission signals from dehydrating branches. _Tree Physiology_ 36, 786
672 796.


673


22


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


674 **Table 1.** Ultrasonic event rates of sunflower and buckwheat drought plants and watered


675 controls by recording day and microphone position


**Species** **DAS** **Drought M1** **Drought M2** **Watered M1** **Watered M2** **Rate ratio (95% CI)** **p**


Sunflower 42 103.1 ± 11.6 52.3 ± 13.5 2.0 ± 1.1 0.25 ± 0.25 107 (4.5-2533) 0.004

44 165.9 ± 12.4 0.25 ± 0.25 13.8 ± 2.5 0.50 ± 0.50 3.2 (0.13-75) 0.47


47 9.5 ± 0.6 93.9 ± 1.9 19.5 ± 0.3 13.5 ± 2.6 1.9 (0.09-38) 0.69


Mean &#91;c&#93; 92.8 48.8 11.8 4.8          -          

Buckwheat 44 0.25 ± 0.25 0.25 ± 0.25 0.25 ± 0.25 0.25 ± 0.25 1.0 (0.14-7.1) 1.00

48 0.50 ± 0.29 0.50 ± 0.29 2.0 ± 0.6 2.0 ± 0.7 0.25 (0.08-0.75) 0.013


51 3.0 ± 0.9 3.0 ± 0.8 0.67 ± 0.67 &#91;a&#93; 0.50 ± 0.29 5.3 (1.8-15.1) 0.002


55 &#91;b&#93; 4.8 ± 1.3 2.8 ± 0.9 0.00 ± 0.00 1.0 ± 0.7 7.5 (2.6-21.3) &lt; 0.001


58 &#91;b&#93; 1.8 ± 0.9 1.3 ± 0.5            -            -            -            

Mean &#91;c&#93; 2.1 1.6 0.73 0.94          -          

676 Values are primary events per 30 min (mean ± SE of four 30-min recordings of one end plant). M1 and M2, end
677 plants of a row of three, next to microphone M1 or M2; DAS, days after sowing. Rate ratios (drought/watered)
678 with 95% CI come from generalised linear mixed models with treatment × day and random intercepts for plant
679 and plant-day (sunflower, negative binomial, 12 plant-days and 48 recordings; buckwheat, Poisson, 16 plant680 days and 63 recordings). &#91;a&#93; Three recordings. &#91;b&#93; Raised electrical background in all recordings; the buckwheat M2
681 recordings at 44 DAS also had raised background. &#91;c&#93; Mean of the recording days with watered controls (events
682 per 30 min). -, not recorded or not estimated.


683


23


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


684 **Table 2.** Spectral features and band-specific detection of sunflower and buckwheat events at


685 equal loudness


**Difference, sunflower −**
**Feature** **Sunflower** **Buckwheat** **buckwheat (95% CI)** **p**


**Spectral features from raw-waveform clips, all recording days (sunflower n = 915 events from 37 recordings, buckwheat n**
**= 90 events from 38 recordings)**


Event level (dBFS) −48.1 (−49.4 to −46.2) −49.5 (−50.4 to −48.0) +1.0 dB (0.1 to 1.9) 0.035


Peak frequency (kHz) 30.2 (28.6-37.3) 28.8 (27.2-30.0) +4.7 kHz (−1.4 to 10.8) 0.11


Spectral centroid (kHz) 36.3 (32.8-42.2) 31.2 (29.3-33.5) +7.6 kHz (2.6 to 12.5) 0.009


Bandwidth at −10 dB (kHz) 7.9 (5.8-10.7) 6.9 (6.2-8.3) +0.8 kHz (−0.2 to 1.8) 0.10


Energy at 15-40 kHz (%) 73.3 (54.4-86.7) 95.8 (90.8-97.2) −27.5 points (−39.0 to &lt; 0.001

−16.0)


Energy at 40-80 kHz (%) 24.4 (10.8-43.8) 1.9 (1.1-4.1) × 9.5 (6.1 to 14.7) &lt; 0.001


Energy at 80-150 kHz (%) 1.5 (0.8-2.6) 2.2 (1.2-4.2) +0.5 points (−1.7 to 2.8) 0.60


Duration above 25% of peak (ms) 0.06 (0.04-0.08) 0.08 (0.06-0.12) −0.04 ms (−0.06 to −0.03) &lt; 0.001


Decay time of 20-40 kHz ringing to 1/e 0.07 (0.05-0.10) 0.12 (0.08-0.15) × 0.65 (0.57 to 0.74) &lt; 0.001
(ms)


**Band-specific detection in all recordings with normal background (sunflower n = 915 events from 37 recordings,**
**buckwheat n = 47 events from 23 recordings)**


Event level (dBFS) −48.1 (−49.4 to −46.2) −49.6 (−50.5 to −48.4) +1.3 dB (0.4 to 2.1) 0.006


Events detected in the 20-40 kHz band −12.3 points (−17.4 to
87.7 100 0.029
(%) −7.8)


Events detected in the 40-60 kHz band
81.1 4.3 +76.8 points (70.6 to 81.4) 0.029
(%)


Events detected in the 60-80 kHz band
48.5 0 +48.5 points (33.9 to 60.1) 0.029
(%)


Events detected in the 80-100 kHz band
1.7 0 +1.7 points (0.5 to 8.0) 0.43
(%)


Events strongest in a band above 40 18.7 0 +18.7 points (12.9 to 29.0) 0.029
kHz (%)

686 Values are medians (interquartile range) or percentages of primary events between −50.8 and −44 dBFS.
687 Differences in event level and spectral features come from linear mixed models with plant and recording as
688 random effects and event level as a covariate, with Kenward-Roger degrees of freedom; × marks ratios for log689 transformed features. Differences in band detection are percentage points, with 95% CI from bootstrap samples
690 of plants within species and two-sided permutation p values with plants as units (70 allocations).


691


24


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


692 **Figure legends and alt text**


693 **Fig. 1.** Recording chamber and plant layout. (A) Empty chamber lined with acoustic foam,


694 with the two ultrasonic microphones. (B) Watered and (C) drought plants in a row between


695 the microphones. (D) Layout used in all recordings; orange arrows, direct path from each end


696 plant to its microphone; dashed arrows, path from the middle plant. Not to scale.


697 **Alt text:** Photographs of a foam-lined wooden recording chamber with two microphones,


698 empty, with three upright watered plants and with three wilted drought plants in a row, and a


699 schematic of the plant row between microphones M1 and M2.


700 **Fig. 2.** Drought increased ultrasonic event rates in sunflower and buckwheat. (A-D) Each


701 panel follows one drought and one watered end plant over the recording days. Large symbols,


702 mean ± SE of four 30-min recordings (three for the watered buckwheat plant at M1 at 51


703 DAS); small dots, single recordings; watered buckwheat was not recorded at 58 DAS. (E)


704 Rate ratios with 95% CI from generalised linear mixed models with plant and plant-day as


705 random effects (sunflower, negative binomial, 12 plant-days; buckwheat, Poisson, 16 plant

706 days); open symbol, raised electrical background in all recordings. M1 and M2, microphones;


707 DAS, days after sowing.


708 **Alt text:** Four dot plots of event rates per 30 minutes over the recording days, showing higher


709 rates in drought than watered plants of both species, and a plot of rate ratios with confidence


710 intervals for each day.


711 **Fig. 3.** Sunflower events carried more energy above 40 kHz than buckwheat events at equal


712 loudness. Primary events between −50.8 and −44 dBFS from 37 sunflower and 38 buckwheat


713 recordings. (A) Example event and mean normalised spectrogram of each species. (B) Median


714 power spectrum with interquartile range (shading). (C) Boxes, median and interquartile range;


715 whiskers, 1.5 × interquartile range; dots, single events; open circles, plant medians; p, linear


716 mixed models with plant and recording as random effects and event level as covariate. (D)


717 Bars, pooled share of events in recordings with normal background (47 buckwheat events);


718 open circles, single plants; p, two-sided permutation tests with plants as units. Dashed lines,


719 40, 60 and 80 kHz; ns, p ≥ 0.05.


720 **Alt text:** Example waveforms and mean spectrograms of sunflower and buckwheat events,


721 their median spectra, box plots of the share of energy at 40 to 80 kHz, and bar charts of


722 detection in four frequency bands, all showing more high-frequency energy in sunflower.


25


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


723 **Fig. 4.** Events recorded by both microphones were much louder at one of them. (A) Level of


724 each shared event at microphones M1 and M2; dashed line, equal levels; dotted lines, 6 dB


725 difference; p, two-sided Mann-Whitney U test. (B) Arrival delay at M2 relative to M1 of the


726 shared sunflower drought events, after removal of clock offset and drift; bars, medians;


727 bracket, difference with 95% CI. (C) Share of energy at 40-80 kHz of the same events at the


728 near (louder) and far microphone; lines join the two copies of each event. DAS, days after


729 sowing.


730 **Alt text:** A scatter plot of event levels at the two microphones with points far from the line of


731 equal level, a dot plot of arrival delays in two separate groups, and paired lines comparing


732 high-frequency energy at the near and far microphone.


733 **Fig. 5.** Substrate drying, event size, spectrum and timing in drought plants. (A, B) Substrate


734 water of the three drought pots; small open symbols, after rewatering; numbers, mean events


735 per 30 min of the end plants. (C) Share of events louder than a given level. (D) Median share


736 of energy at 40-80 kHz (interquartile range) in level classes with at least six events; the


737 within-recording factor comes from a linear mixed model. (E) Level distribution of


738 buckwheat drought events; recordings at 55 and 58 DAS had raised electrical background;


739 level differences from linear mixed models. (F) Intervals between events relative to the mean


740 interval of each recording, for recordings with at least 30 (sunflower) or three (buckwheat)


741 events; dashed line, exponential distribution expected for random timing. M1 and M2,


742 microphones; CV, coefficient of variation; p, two-sided Kolmogorov-Smirnov test; DAS,


743 days after sowing.


744 **Alt text:** Line plots of substrate water falling over the recording days in both species, curves


745 of event level distributions and high-frequency energy by level for one sunflower plant, level


746 distributions of buckwheat events on three days, and interval distributions close to the random


747 expectation.


748 **Fig. 6.** Stomatal conductance ( _g_ s ), PSII function and event rates of drought and watered


749 plants. (A-F) Large symbols, mean ± SE of three plants; small dots, single plants, each the


750 mean of two leaves (one gas-exchange reading for one watered buckwheat plant at 51 DAS);


751 p, two-sided Welch's t-test. Open circle, sunflower drought leaves at 42 DAS with a different


752 background value, not tested (nt); at 44 DAS only leaves with the same background value are


753 included (two drought plants). (G, H) Large symbols, mean of the two end plants; small dots,


754 single end plants; open symbols, raised electrical background; zero rates plotted at 0.1; nr, not


26


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


755 recorded. _F_ v / _F_ m, maximum quantum yield of PSII; PI abs, performance index on absorption


756 basis; DAS, days after sowing.


757 **Alt text:** Eight dot plots over the recording days of both species: stomatal conductance far


758 lower in drought plants; maximum PSII yield and performance index of drought sunflower


759 falling only slightly and late, and of drought buckwheat falling strongly from 51 days after


760 sowing; and event rates on a logarithmic scale, higher in drought plants.


761 **Fig. 7.** The spectrum identified the species, and the event rate separated drought from watered


762 plants. Logistic classifiers were trained on events between −50.8 and −44 dBFS from seven of


763 the eight end plants and tested on the eighth, for each plant in turn. (A) Mean share of 2,000


764 random draws of 1-10 events that gave the correct species; shading, range down to the least


765 well classified plant (40-80 kHz share); dotted line, chance. (B) Area under the receiver


766 operating characteristic (ROC) curve for species and treatment from the spectrum of single


767 events, and for treatment from the event rate of single plant-days.


768 **Alt text:** A line plot of species identification accuracy rising with the number of events per


769 plant, and a dot plot of the area under the ROC curve for species and treatment classification.


770 **Fig. 8.** Conceptual hydraulic model, proposed as a hypothesis, linking stomatal closure,


771 xylem embolism, acoustic events and PSII function. (A) Event rate, loss of conductivity and


772 stomatal conductance as xylem water potential falls, if each event marks one embolising


773 conduit. Circles, two plants with the same low stomatal conductance at different points of the


774 curve. P 12, P 50 and P 88, water potentials at 12, 50 and 88% loss of conductivity. (B, C)


775 Hypothesised order of the signals during drying; vertical lines, recording days. Stomatal


776 closure is inferred from wilting at 42 DAS and low _g_ s at 44 DAS in sunflower and from _g_ s at


777 48 DAS in buckwheat. Second plant, sunflower end plant at M2. Curves are schematic. DAS,


778 days after sowing; _g_ s, stomatal conductance; PI abs, performance index on absorption basis;


779 _F_ v / _F_ m, maximum quantum yield of PSII.


780 **Alt text:** Schematic diagrams: a sigmoid vulnerability curve with a bell-shaped event-rate


781 curve that peaks on its steep part, with stomatal closure marked before embolism starts; and


782 two timelines in which stomatal conductance falls first and the event rate then rises, before the


783 PSII performance index falls in sunflower and together with its fall in buckwheat.


27


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


784

785 **Fig. 1.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


786

787 **Fig. 2.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


788

789 **Fig. 3.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


790

791 **Fig. 4.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


792

793 **Fig. 5.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


794

795 **Fig. 6.**


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


796

797 **Fig. 7.**


34


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.17.767468; this version posted November 20, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.17.767468)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)


798

799 **Fig. 8.**


35


