Can you hear a plant scream?

The mechanism

Why a thirsty plant pops

Water climbs a stem under tension, pulled upward by evaporation at the leaves, the same physics that lets you draw water up a straw. Drought pushes that tension past a breaking point: air is sucked into a water-carrying channel and the column snaps from liquid into a gas-filled void. That rupture, called cavitation, releases a burst of sound. Cutting a stem causes the same failure faster, by exposing the tensioned water straight to open air.

The evidence

How researchers captured it

1

Ultrasonic microphones. The pops sit far above human hearing, so Khait, Hadany et al. (Tel Aviv University) recorded across 20 to 250 kHz, about 10 cm from each plant.

2

A soundproofed chamber first. Tomato and tobacco recorded three ways: untouched, five days without water, and freshly cut.

3

Then a working greenhouse. Repeated amid ordinary background noise to prove it wasn't a laboratory artifact.

4

Machine learning. A classifier trained on many acoustic features (not one number) separated stressed from healthy, drought from cutting, and tomato from tobacco.

Listen

The recordings

Press play on any card, then hover a dot to inspect that individual click.

Pitch-shifted down into audible range, with irregular gaps so each pop stands alone. Every dot on a graph is one real click: height is its pitch, size is how loud it was, and it appears as you hear it.

The seconds on each graph are just how far you are into this short recording, not how long the plant was stressed. Drought played out over five days and a cut stem ruptures within moments; either way, what you're hearing is only a short excerpt of that.

From the paper itself

Tomato, continuous recording

36 seconds of supplementary audio released by the research team with the Cell paper. No graph: we don't have verified per-click measurements for this file.

Tomato

Drought

0/30

clicks played

Five days without water. Tension in the water column climbs until it snaps.

bigger dot = louder click

View 30 clicks as a table

Pitch, loudness and timing of every click in the Tomato Drought clip
Click #	Pitch (kHz)	Loudness (%)	Time (s)
1	77	100	0.3
2	46	55	1.4
3	26	17	2.8
4	53	56	3.8
5	81	50	5.0
6	32	14	5.7
7	62	54	8.9
8	39	19	9.4
9	30	48	12.2
10	60	19	13.2
11	74	44	14.0
12	42	21	14.9
13	74	38	15.7
14	50	40	16.9
15	32	20	18.3
16	33	23	20.3
17	68	39	21.0
18	34	15	24.1
19	64	27	25.2
20	40	22	26.0
21	26	17	27.2
22	36	29	27.8
23	42	18	28.4
24	30	25	29.9
25	40	17	31.0
26	34	42	33.3
27	44	54	34.2
28	33	18	35.1
29	39	28	36.6
30	26	31	37.6

Tomato

Cut stem

0/30

clicks played

Stem severed, exposing the tensioned water column straight to open air.

bigger dot = louder click

View 30 clicks as a table

Pitch, loudness and timing of every click in the Tomato Cut stem clip
Click #	Pitch (kHz)	Loudness (%)	Time (s)
1	58	25	0.3
2	58	14	1.1
3	48	24	2.4
4	51	38	3.5
5	90	16	5.1
6	42	56	5.6
7	48	18	6.9
8	43	26	7.6
9	84	12	8.4
10	41	19	11.4
11	48	64	13.0
12	50	38	16.6
13	53	56	18.5
14	35	13	20.8
15	53	23	21.5
16	80	13	22.2
17	48	59	23.0
18	50	88	24.4
19	50	54	25.9
20	48	99	26.5
21	71	34	27.4
22	69	65	28.7
23	56	100	29.3
24	67	16	29.9
25	40	35	31.1
26	42	98	33.2
27	50	12	35.8
28	53	15	36.9
29	54	29	38.5
30	73	52	39.6

Tobacco

Drought

0/27

clicks played

Same drought treatment, different species and vascular anatomy.

bigger dot = louder click

View 27 clicks as a table

Pitch, loudness and timing of every click in the Tobacco Drought clip
Click #	Pitch (kHz)	Loudness (%)	Time (s)
1	39	57	0.3
2	55	44	1.0
3	44	98	3.5
4	58	46	4.7
5	56	71	6.0
6	53	74	6.5
7	58	98	7.9
8	80	31	9.3
9	54	17	10.6
10	75	31	11.6
11	43	41	13.2
12	42	11	15.0
13	53	20	16.1
14	59	58	16.9
15	44	100	18.7
16	44	23	19.8
17	58	25	20.9
18	61	59	21.5
19	40	39	22.3
20	41	57	24.4
21	46	15	25.5
22	43	93	26.1
23	46	25	27.6
24	59	71	28.6
25	42	97	31.4
26	52	36	32.6
27	58	45	33.9

Tobacco

Cut stem

0/30

clicks played

Same cut treatment, different species and vascular anatomy.

bigger dot = louder click

View 30 clicks as a table

Pitch, loudness and timing of every click in the Tobacco Cut stem clip
Click #	Pitch (kHz)	Loudness (%)	Time (s)
1	55	51	0.3
2	63	33	0.9
3	79	30	5.7
4	52	29	6.6
5	57	16	8.8
6	65	17	9.4
7	63	30	12.2
8	69	16	12.7
9	58	51	13.9
10	52	30	14.5
11	54	56	15.2
12	48	25	17.0
13	66	25	19.6
14	53	19	20.7
15	56	72	22.4
16	44	41	23.7
17	62	41	24.6
18	56	49	27.0
19	56	47	28.9
20	68	25	29.7
21	69	39	30.8
22	95	44	32.3
23	42	40	33.7
24	48	27	34.3
25	65	14	39.1
26	69	56	39.6
27	63	35	41.2
28	65	31	42.8
29	76	100	43.5
30	38	18	44.6

The honest comparison

Can you tell the stresses apart?

Peak frequency of ~2,500 clicks measured straight from the original recordings, grouped into five conditions. Look at the two shaded zones below: a healthy plant's clicks stay low and separate (gray), while every stressed plant (drought or cut, tomato or tobacco) clicks noticeably higher (orange). Telling those two zones apart is easy. Telling the four stressed conditions apart from each other is the hard part.

Shape = how that condition's clicks are spread across pitch. A taller point means more clicks landed near that pitch. Dot = the median: the "typical" click. Exactly half of that condition's clicks were higher-pitched than this, half were lower.

The gray and orange zones barely touch, but the four orange shapes sit almost on top of each other.

Greenhouse background Tomato, drought Tomato, cut Tobacco, drought Tobacco, cut

That overlap is the point. Scored for separability, background noise splits from plant clicks cleanly (AUC 0.88), while the four stress conditions score just 0.53 to 0.65 against each other, barely above the 0.50 of a coin flip. That is exactly why the study needed a machine-learning model trained on many features at once, not a threshold on one number. What gives a stressed plant away isn't the sound of any single pop. It's how often they come.

Plant clicks vs. background noise

0.88

Any two stress conditions vs. each other

0.53 to 0.65

0.50 · coin flip1.00 · perfectly separable

The real signal

Watch an hour go by

Every clip above holds click count constant so they're comparable by ear. In life, rate is what gives stress away. Pick a condition and watch one simulated hour at its published rate.

0 clicks appeared; published rate: pick one above