The study asks whether low-cost printed sensors, whose resistor is a biodegradable PHBV/carbon composite that soil microbes digest, can serve as an in-situ proxy for biological soil health. In the first multi-plot field deployment, 44 sensors were buried for 50 days (summer 2023) in a long-term grassland restoration experiment at Colt Park Meadows, UK, logging resistance every 30 minutes; 31 passed quality filtering. The rate of resistance rise, and in a separate analysis the number of detected changepoints, were compared with laboratory measurements from soil cores. All three statistical appr

A sensor built to be eaten

Researchers printed an electrical wire from ink that soil microbes can digest, buried 44 of them in a Yorkshire meadow, and logged them every 30 minutes for 50 days. The faster a wire’s resistance climbed, the more microbial life lab tests found in that soil.

Could a field one day report on its own underground life — continuously, cheaply, from thousands of points?

That is where the authors want to go: low-cost printed sensors for “spatially dense data collection” of soil biology, which could help “track and verify the rehabilitation of degraded land”.

Live model: a printed sensor being digested in meadow soil

Predict, then try: slide toward teeming and bury a fresh sensor. Does the line climb sooner or later?

Illustrative model of the sensing idea — the real field data comes further down. In the team’s earlier lab tests, sensors in sterilized compost tea, autoclaved sand or pure water barely changed. Each run uses its own random weak spots, so the moment a wire breaks is noisy, while the steady climb tracks the life in the soil.

Why go to this trouble? Because soil life is remarkably hard to watch.

01 The problem

Soil life is hard to watch

Underfoot, bacteria and fungi break down dead plants, recycle nitrogen and phosphorus, and steer what happens to carbon in soil — in the authors’ words, “the largest terrestrial carbon pool on the planet”.

Yet when soil health is checked, biological measures get less attention than physical or chemical ones — partly because they are hard to measure.

The usual route is to dig soil cores and take them to a lab. That gives one snapshot , it costs money, and by the time it is analysed the sample may no longer reflect the field. Burying tea bags, cotton strips or bags of leaf litter captures decomposition where it happens — but you have to dig them up, dry them and weigh them, and you get one number at the end .

8 of 8 methods pass

Adapted from Table 1 of Sharpe et al. (2026). Each card’s strip sketches 50 days: dots are one-off samples, a line is a continuous record.

So what does a bite actually do to the current?

02 How eating becomes a number

Inside the trace: be the microbe

The ink is carbon flakes stirred into PHBV, a biodegradable bioplastic that microbes can digest with or without oxygen. Pushed through a stencil onto glass, it forms a resistor — a loop 3 mm wide.

Current crosses it by hopping between flakes that touch. As microbes digest the plastic, the trace swells, flakes lose touch, fewer paths remain, and resistance rises . When the last connected path breaks, current stops — the sharp spike that ends every sensor’s life. And resistance is easy to read with cheap electronics.

Predict: where does one bite raise resistance most — in a crowded patch of flakes, or at a thin bottleneck? Drag across the strip (or press and hold) to feed.

Bites at bottlenecks count most. That’s also why single traces fail unpredictably — and why the team plans several traces per sensor.

Illustrative model of a conductive composite; not a simulation from the paper. Node voltages are solved with Kirchhoff’s laws after every change; the moving dots follow the computed currents.

That is the idea. Here is what happened when 44 of these wires went into a real meadow.

03 The real field Real sensor records

Forty-four wires in a Yorkshire meadow

Colt Park Meadows lies 350 m above sea level in the Ingleborough National Nature Reserve, northern England. Since 1989 it has hosted a grassland restoration experiment: 16 combinations of treatments, repeated across 3 blocks — 48 plots, each 3 × 3 m.

The treatments are inorganic fertilizer (since 1990), farmyard manure (since 1998), seeds of 19 meadow species (since 1990) and red clover (sown in 2004 and 2011). Decades of different care have left the plots with different microbial biomass, enzyme activity and soil carbon — a ready-made test bed.

The team placed one sensor at the centre of 44 plots. Battery-powered loggers, six sensors each, read them every 30 minutes and wrote to SD cards — there is no internet out there. Sensors went in on 19 June 2023, at the end of a UK drought, and logged until 7 August. In mid-July, soil cores taken about 30 cm from each sensor went to the lab.

Day 0 · 21 Jun 2023 · 0 sensors past R/R₀ 2

Real sensor records from the open dataset (Sharpe et al., OSF doi:10.17605/OSF.IO/K98Q4), resampled to 3-hour medians. Day 0 is 21 June 2023, the first day used in the analysis; the sensors went into the ground two days earlier.

But not every wire told a readable story.

04 Quality control Real sensor records

Which trails can we trust?

Not every wire survived. Some were cracked when pushed into the soil, some never responded, some failed within days.

The team’s rule: fit a smooth curve — a second-order polynomial — to each record. If the curve explains less than half of the record’s variation (R² below 0.5), the record is too erratic to read. Sensors with an abnormal starting resistance were dropped too. That left 31 of 44 (70.5%) .

They chose 0.5 because it best matched their own visual inspection, and they note that the results are sensitive to that choice. Slide it and see.

31 of 44 sensors kept

Each thumbnail: R/R₀ over the whole record, clipped to 0.5–2.5. Labels show sensor id and the fit R² from the dataset. A record is kept when R² is above the threshold and the starting resistance is normal (recorded, below 5 kΩ).

Now the big question: did the climb track the life in the soil?

05 The result Recomputed from real data

Does a steeper climb mean more life?

Each kept sensor gives one number: how steeply its resistance climbed in a shared decomposition window — after the wires settled into the soil and before they began failing. The team located that window with changepoint detection on the fleet’s average signal, trimmed by 3-day buffers: roughly 27 June to 6 July 2023.

That slope (change in R/R₀ per day) is compared with the lab’s microbial biomass carbon (MBC) from the same plot — how much carbon is held in living microbes, a standard, slow-changing indicator of soil biological health.

Paper reports r = 0.70 for this window. Live numbers are recomputed here from the open data.

The authors tested shifting the window by 3 days or shrinking/stretching it: the correlation stayed significant but weaker.

They stress this is a pattern, not a calibration curve.

When the moment of failure was compared instead, it did not track the lab measurements.

Slopes are ordinary least-squares fits of R/R₀ against time inside the window, for the 31 records the paper kept; two of them have no lab values, so n = 29 plots. Correlation is Pearson’s r with a two-tailed p-value.

One window is one choice. Did other ways of reading the trails agree?

06 Cross-checks Paper’s reported table

Three ways to read a trail

Pearson r from the paper’s Table 3; all shown values have p < 0.05; n.s. = not significant. MBC = microbial biomass carbon, GLC/NAG/PHO = carbon/nitrogen/phosphorus enzymes, SOC = soil organic carbon.

Every approach linked the wires to microbial biomass carbon and soil organic carbon.

One more worry: the weather in the soil moves the needle too.

07 Telling signals apart

Wiggle or bite?

Resistance also responds to the weather in the soil. In lab tests with fresh sensors, varying one thing at a time: temperature from 5.4 to 23.5 °C moved R/R₀ from 1.06 to 0.86 (warmer, lower); soil moisture from 18.9% to 46.4% water by volume moved it from 0.83 to 1.44 (wetter, higher); soil conductivity from 437 to 1120 µS/cm moved it from 1.0 to 0.6 (saltier, lower).

Those changes were predictable, temporary and reversible. And because all the sensors shared one field, rain and daily temperature swings form a background common to the whole fleet.

Playing a day–night temperature swing — touch any control to take over.

Weather pushes the needle and lets go; digestion only ratchets one way.

Endpoints are the paper’s reported values; the straight line between them is our simplification. “Start conditions” are where each line crosses R/R₀ = 1. The weekly feeding step (+0.04) is illustrative.

Which brings us back to the question we started with.

08 What this brings closer

A meadow that reports on itself

The authors describe the sensor as showing “initial promise” — a first field test, not a finished instrument. Here is what they say it could open up, and what has to come next.

What it could enable

Tracking and verifying the recovery of degraded land.

Watching how soil microbes respond to, and recover from, shocks such as drought or pest fumigation.

Following microbial biomass at high resolution in space and time, at low cost.

Other digestible binders — lignin, cellulose, local plant litter — to target other decomposition pathways.

A signal that should come only from living microbes — avoiding the “relic DNA” that dead cells leave behind in genetic tests.

What has to come next

Several traces per sensor, so one break doesn’t end the record (13 of 44 were dropped here).

Co-located temperature, moisture, pH and respiration sensors to separate biology from weather.

Genetic analysis of which microbes actually do the eating.

More soils and more seasons — this was one field, one summer.

Radio or satellite telemetry instead of collecting SD cards by hand.

One summer, one meadow, 31 readable wires — a first step toward soils that can be read as they live.

09 Source & credits

Source & credits

Taylor J. Sharpe, Madhur Atreya, Shangshi Liu, Mengyi Gong, Nicole Luna, Noah Smock, Jessica Davies, John N. Quinton, Richard D. Bardgett, Jason C. Neff, Rebecca Killick, Gregory L. Whiting. In-situ decomposition sensor output correlates with soil health indicators. Computers and Electronics in Agriculture 244 (2026) 111427. https://doi.org/10.1016/j.compag.2026.111427

Field records and lab values: open dataset, Sharpe et al., OSF https://doi.org/10.17605/OSF.IO/K98Q4 — resampled to 3-hour medians and replotted; quality filter and slopes recomputed here following the paper’s described method. The table of correlations in “Three ways to read a trail” is the paper’s reported Table 3; the method shelf is adapted from its Table 1.

All drawings, simulations and charts on this page are original illustrations by iPaper.ai; simulations are illustrative models, not the authors’ models. No figures from the article are reproduced.

Fonts: Fraunces, Instrument Sans and JetBrains Mono (SIL Open Font License).