Soil moisture sensors to recommend#

Which soil probes a farm on fieldmesh should be pointed to, how each reaches the log, and what fieldmesh's own evidence says about choosing and installing them. It is the soil side of weather stations. None of the makes below has been run on a fieldmesh farm yet (UNTESTED).

What a probe has to give#

• Volumetric water content (cm3/cm3), the quantity the soil advisory works in. A probe reports a raw number; the calibration that turns it into water content is its own record (§3.2), versioned, so a recalibration never rewrites history.
• Soil temperature from the same body. The frozen-soil check needs it: a frozen probe reads falsely dry. So does the depth check, which reads a probe's depth from its daily temperature wave (§4.1, siting-real-soil).
• A depth and a cover, recorded on its installation, so the advice weights it right and the depth check compares like with like.
• A second probe in the bed. One probe cannot see its own drift. The cross-check between two probes on one feature can (§4.2: "a feature with one sensor has no drift detection, and drift is the commonest real failure").
• A way into the log: MQTT, LoRaWAN through ChirpStack, Modbus, or a node.

What fieldmesh's checks were proven on#

The health checks were scored against ISMN's independent quality flags on two networks' probes in and around Tennessee (ismn-scoring), both Stevens Hydra Probes:

• NRCS's SCAN, 2.2 million good readings;
• NOAA's Climate Reference Network, 1.5 million.

In the top 20 cm the checks flagged 0.02% to 0.29% of the readings ISMN rates good.

That is evidence for the checks on that kind of probe. On a cheap capacitive probe, their thresholds are untested.

The candidates#

The way to SDI-12 probes is a node. Most research probes speak SDI-12, and fieldmesh has no SDI-12 driver. EnviroDIY's Arduino-SDI-12 library (LGPL-2.1) talks to them from an ESP32 with no extra hardware. That node would publish through ESPHome or MQTT, which the bridge already reads.

What I'd recommend#

What fieldmesh's own evidence says about choosing and installing#

• A probe that reads low is forgiven; one that reads high is not. On the advisory's backtest bed, McAllister Farm's 2013, a true probe's advice used 208 mm (the simulator of 2026-10-16).

  • A probe reading 0.05 low, unlearned, called for 1,045 mm. Learning its own field capacity from its drain-downs, which are 0.05 low too, it used 422 mm.
  • A probe reading 0.05 high says the bed is fuller than it is, so the bed is watered late and little, 127 mm. The probe then sees too few soakings to learn from, and the bed went dry: 293 hours of stress.
  • Learning cannot fix this, because the bias stops the very watering it would learn from. A second probe can.

  (TestAProbeThatReadsOff)

• Learn the bed's own field capacity rather than trusting the survey's. In a landscaped yard the survey's values drowned the bed. The bed's own drain-downs did not (the bed's own soil).

• Probes in two soils say so. Two probes whose learned capacities differ by 0.08 or more raise a note asking for each layer's values.

• Calibrate a set point to the soil, never to a default. A soil sensor's set point left at the Florida guide's middle starved a silt-loam lawn (lawn sprinklers).

• Put the probe where the roots are. About 3 inches deep for turf (UF/IFAS AE437); for beds, the middle of the root zone, with a second probe below it to show water leaving.

• Record the cover. Mulch and sod read as extra depth to the depth check, unless the cover is on the installation.

• Avoid a probe whose readings only reach its maker's cloud, for the same reason as a weather station.

What fieldmesh would build next for probes#

Temperature, tension, salinity and drainage, the other things worth putting in the soil, are in soil sensors beyond moisture; soil temperature is built, as the sowing advisory.

1. An SDI-12 node: an ESP32 with Arduino-SDI-12, publishing through ESPHome. It reaches the HydraProbe, the TEROS and most research probes.

2. A HydraProbe Modbus map in the Modbus collector, from Stevens' register documentation.

3. A "wet too long" note: a probe that reads wetter than field capacity for days with no rain and no watering is reading high, or the bed is waterlogged. Either needs a person.

4. A term for soil water potential, for the Watermark and tensiometers. Built 2026-10-12 (beyond moisture).

5. A calibration by known water. A measured amount poured over a measured square around the probe, say 10 litres on one square metre, is 10 mm. The probe's rise against the rise those 10 mm should make measures its gain, and its drain-down afterwards measures the level it settles at.

   • Built 2026-10-12 as the soak test, a regimen that leads a person through it: pour this much here, wait, check the rise, and two days later read the level it drained back to (soil-moisture advisory).
   • The sprinkler doing it on its own is a command, and waits for §6.

   It would also catch a probe reading high, which learning cannot.

Testing without a probe#

openSenseMap's public stations include soil probes, TRUEBNER SMT50s and Dragino LSE01s among them, readable without a key: tools/opensensemap_collect.py brings one station's soil series into a log through the checks (test data sources).

Sources#

• Stevens HydraProbe: Fondriest (interfaces, probe-to-probe figure); USCRN's use: NOAA NCEI, Stevens Hydra Probe SDI-12.
• METER TEROS 11/12: tech specs.
• Dragino LSE01: user manual; Milesight EM500-SMTC: product docs.
• A. A. Abdelmoneim et al., "Calibration of low-cost capacitive soil moisture sensors for irrigation management applications", Sensors 25(2), 343, 2025. doi:10.3390/s25020343
• J. D. González-Teruel et al., "Design and calibration of a low-cost SDI-12 soil moisture sensor", Sensors 19(3), 491, 2019. doi:10.3390/s19030491
• EnviroDIY Arduino-SDI-12: github.com/EnviroDIY/Arduino-SDI-12.
• Watermark: Oregon State University, granular matrix sensors.
• UF/IFAS AE437 (probe depth for turf): EDIS.