Industries / Food, farming and forestry / Agriculture and farming / Irrigation
Scheduling and sensors
Deciding when to water and how much, from weather data, soil sensors and signs from the plants themselves.
7
Milestones achieved, from 1948 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1948
Evaporation from weather data
Howard Penman combined two ways of working out evaporation, one from the turbulent movement of moist air and one from the energy balance of sunlight, into a single equation. In his words it gave "for the first time an opportunity to make theoretical estimates of evaporation rates from standard meteorological data", and he tested it on open water, bare soil and turf. Versions of it still drive weather-based irrigation: California's CIMIS network computes reference evapotranspiration with a modified Penman equation and with the later Penman-Monteith equation.
Credit
Howard Latimer Penman
1951
Tensiometers reach the farm
A tensiometer is a water-filled tube with a porous ceramic cup buried in the soil; as the soil dries it pulls water out through the cup, and a vacuum gauge reads that pull, called soil water tension. L. A. Richards and Willard Gardner described tensiometers for measuring soil water tension in 1936. In 1951 T. W. Prosser, a citrus grower in Arlington, California, who had seen research tensiometers at the University of California, Riverside, brought a commercial version to market; his company, Irrometer, says Glen Larson developed its Watermark granular matrix sensor, an electrical-resistance block, from 1978.
Credit
L. A. Richards and Willard Gardner (design); T. W. Prosser (Irrometer)
1952
Neutrons count soil water
Wilford Gardner and Don Kirkham published a way to measure soil water by neutron scattering. A radioactive source sends fast neutrons into the soil, hydrogen in the water slows them, and a detector counts the slow ones. The IAEA says the neutron moisture meter was used throughout the world and dominated in the 1970s and 1980s, and its 2008 expert group still found a field-calibrated meter the most accurate way to measure water down a soil profile.
LessonAccuracy did not keep it on farms: users need radiation training, dose monitoring, leak tests every six months and transport papers, cheaper electronic sensors arrived, and the IAEA says radiation methods "are being discouraged in many countries".
Credit
Wilford Gardner and Don Kirkham
1981
Leaf heat shows thirst
Leaves cool themselves by evaporating water, so a crop short of water runs warmer. Ray Jackson, Sherwood Idso, R. J. Reginato and Paul Pinter read canopy temperature with infrared thermometers and, with air temperature, humidity and an estimate of net radiation, calculated a crop water stress index (CWSI), equal to one minus the ratio of actual to potential evapotranspiration. On four wheat plots irrigated at different times, the index "closely paralleled" the extractable water in the top 1.1 m of soil. Idso and colleagues published the matching method for setting a crop's no-stress baseline the same year.
Credit
Ray D. Jackson, Sherwood B. Idso, R. J. Reginato, Paul J. Pinter and Jerry L. Hatfield. 90032-7)
1982
A weather network for irrigators
California's Department of Water Resources (DWR) and the University of California, Davis, set up the California Irrigation Management Information System (CIMIS) in 1982. Its automated weather stations report reference evapotranspiration (ETo), the water a well-watered grass surface loses in a day, which growers multiply by a crop coefficient to estimate their own crop's use. DWR took over running it in 1985; a 2010 DWR presentation counted more than 130 stations, some owned by local water agencies, universities and federal agencies, and a 2026 study calls it "widely used by growers, consultants, water agencies, and researchers".
Credit
California Department of Water Resources and UC Davis
2008
Sensors fail an outside test
Over five years an expert group contracted by the International Atomic Energy Agency (IAEA) compared soil water sensors in the laboratory and the field. Capacitance sensors, the electronic probes sold widely since the 1990s, varied far more than direct measurements and were thrown off by salt and temperature; factory calibrations "were inaccurate in most soils studied"; and apart from, possibly, tensiometers and granular matrix sensors, none of the sensors studied was practical for on-farm scheduling, being either too inaccurate or too costly and hard to use.
LessonA factory calibration shows the best a sensor can do in uniform, repacked soil, not what it will read in a farmer's field.
Credit
IAEA expert group on soil water sensing
2024
A scheduling tool beats habit
The Nebraska On-Farm Research Network ran four replicated trials on center pivots in Platte and Butler counties, Nebraska, each comparing the grower's own irrigation decisions with those of a commercial scheduling tool on halves of one pivot, with pivot speed control setting each section's amount. Over the 2024 season the grower-managed sections got 5.2 inches (132 mm) of irrigation on average and the tool-managed sections 1.2 inches (30 mm) less, about 23% less by this file's arithmetic; yields and net returns were statistically the same.
Credit
Nebraska On-Farm Research Network, University of Nebraska–Lincoln Extension
Watch for
Deep-learning crop water forecasts
Adarsha Neupane and Vidya Samadi describe NeuralFAO56, open Python software that forecasts reference evapotranspiration up to 7 days ahead with neural networks (LSTM and Transformer models), and report "strong short-term predictive skill" across 867 US weather stations. A published review of the preprint notes that the forecasts were compared only with FAO-56 values computed from the same stations' data, with no simple baselines and no independent measurement of evapotranspiration.
Achieved whena peer-reviewed version shows the forecasts beating persistence and climatology at each lead time, checked against measured evapotranspiration such as eddy-covariance towers.
Claimed by
Neupane and Samadi, EGUsphere preprint, July 2026
Open
01
Soil moisture from a photo
A phone app that estimates how much water is left in the root zone from a photo of a soil sample taken with an auger, the same handful a grower would squeeze for the feel test, and says whether the field needs water. Test it on working farms against samples dried in an oven.
The briefCredit · unclaimed02
Listen for thirsty plants
A field test of whether ultrasonic microphones can tell when a crop needs water. Recorders sit among tomato plants for a season, a small AI model counts the clicks the plants make, and the counts are compared with soil moisture and with plant water status measured by hand.
The briefCredit · unclaimed
Open 01 · the brief
Soil moisture from a photo
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app that estimates how much water is left in the root zone from a photo of a soil sample taken with an auger, the same handful a grower would squeeze for the feel test, and says whether the field needs water. Test it on working farms against samples dried in an oven.
Why it is still open
In USDA's 2023 Irrigation and Water Management Survey, 85,093 of 212,714 US irrigating farms reported deciding when to water by the feel of the soil. Published phone methods were trained on photos of prepared samples: 629 photos of 38 soil samples from seven places around Sydney (Hossain and Kabir, 2023), and 238 photos of samples from 14 places in India (Remote Sensing Applications: Society and Environment, 2025, best R² 0.78, R² being the share of variation explained). No test on working farms, at root depth, scored on the watering decision was found.
Built with what exists
A phone camera and a printed colour card in every photo, to correct for light; a soil auger; a small image model or an open vision-language model, run on the phone or a laptop (see the Note "A first stable number for local AI"); a kitchen oven and scale for the oven-dry checks.
First step
Take 50 auger samples from one field through one drying cycle, photograph each beside a colour card, and dry and weigh them.
Done looks like
20 farms on at least four soil textures and 500 samples over one season, each photographed and oven-dried, with the app's error in percent water by weight and the share of samples on which it agrees with the oven about "water now" or "wait", against a threshold set from each soil's texture, published with the photos.
Open 02 · the brief
Listen for thirsty plants
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A field test of whether ultrasonic microphones can tell when a crop needs water. Recorders sit among tomato plants for a season, a small AI model counts the clicks the plants make, and the counts are compared with soil moisture and with plant water status measured by hand.
Why it is still open
Tel Aviv University researchers (Khait and colleagues, Cell, 2023) recorded airborne ultrasonic sounds from tomato and tobacco plants, which made 30 to 50 pops an hour when stressed, heard up to 5 m away, in an acoustic chamber and a greenhouse. In 2025 Bonisoli and colleagues picked out the same kind of sounds from water-stressed pinto bean and tomato plants outdoors, despite wind, rain and insects. Neither study tested the sounds as a signal for when to water through a growing season, and no such test was found.
Built with what exists
AudioMoth recorders, open-hardware sound loggers that sample up to 384,000 times a second; tensiometers or other soil moisture sensors; a pressure chamber for weekly plant water readings; an open audio classifier run on a low-power AI chip such as Google's Coral NPU (see the Note "An open compiler").
First step
Put two AudioMoths beside ten potted tomatoes, stop watering five of them, and count clicks for two weeks.
Done looks like
One season with at least 40 plants in watered and dried plots, hourly click counts, daily soil moisture and weekly plant water status, and a published score of how well click counts flag stress days (share of stress days caught and share of false alarms), with the audio released.