Agriculture Data by NuaSense Get a free keyFree key

Weather station data for farms: raw readings, roll-ups, spray windows and disease risk

A station on the field reports about every 10 minutes. The platform keeps the raw uplinks, rolls them up hourly and daily, computes 29 derived variables every hour, and runs the risk models on top of them. 5 datasets, 70 fields, served only to the account that owns the station.

Catalogue group: Weather data, sensor data. Your weather station and everything computed from it.

Pages in this group: Weather station data from your own field, raw to daily, Spray window, leaf wetness and degree-day data, hourly and Disease and pest risk data for potato, onion and coffee.

The 3 dataset pages in weather data

Weather station data from your own field, raw to daily →farm.weather.raw · farm.weather.hourly · farm.weather.daily Raw uplinks about every 10 minutes plus hourly and daily roll-ups, 33 fields across 3 datasets: temperature, humidity, pressure, light, rainfall, wind, position, battery and radio telemetry. ~10 min · hourly · daily own account Spray window, leaf wetness and degree-day data, hourly →farm.derived.hourly 29 variables computed every hour from the station's own readings: dew point, vapour-pressure deficit, evapotranspiration, delta-T, spray quality with its limiting factor, inversion risk, leaf wetness and degree-days for the crop and 9 pests. Every field carries a calibration tier. hourly own account Disease and pest risk data for potato, onion and coffee →farm.disease_risk Risk per threat for the station's crop, every hour: percent, level, threat score, accumulated disease severity, the conducive hours behind the score and how much input data backed it. 22 models across potato, onion and coffee. hourly own account
ideal good marginal poor unsuitable
48 hours of spray quality at the Naivasha station, 1 cell per hour, from the derived variables dataset.

From station reading to disease risk

Step Dataset Cadence What it gives
1 Weather station data ~10 min, hourly, daily Readings as sent, and hourly and daily means for charts over a week or a season
2 Derived variables hourly Dew point, vapour-pressure deficit, evapotranspiration, delta-T, spray quality with its limiting factor, inversion risk, leaf wetness, pest degree-days
3 Disease and pest risk hourly, per threat Risk percent and level for each threat in the crop’s model set, with the hours and the data completeness behind the score

Where spray windows live

The spray-quality index, its label and whatever factor is limiting it all sit as fields on the derived variables dataset, recomputed every hour straight from the station’s own readings. A spray-window advisory only goes out when a sprayable disease is elevated in the risk dataset and the hours ahead are suitable; soil-borne threats never trigger spray advice at all. Whether it fires or not, that advisory gets recorded in Farm records either way.

Calibration tiers

Every derived field carries 1 of 5 tiers: measured, physics, validated, literature or proxy. A dew point counts as physics, built off 2 measured inputs. A leaf-wetness hour inferred from dew-point depression and humidity stays a proxy until an actual leaf-wetness sensor sits on the station. The tier travels with the field itself, so an integrator can decide on the spot how much weight to put on it.

Which crop’s risk models run

A station’s crop, whether set on the farm profile or read straight off a sensor’s own crop label, picks the model set: 11 potato threats, 6 onion threats or 5 coffee threats. Potato is the calibrated set, built at 1,900 m in Naivasha; onion and coffee still run on literature defaults until they’re tuned the same way. A crop with no matching models gets logged and skipped, never guessed at.

Access

Readings, and everything computed from them, go only to keys on the account that owns the station. A new station arrives as pending and only answers over the API once it’s approved. Stations NuaSense didn’t install itself join through Sensor Ingest instead, and get the exact same derived variables and risk models once they’re in.