The layer everything else is built on
Neither the risk models, the water balance, nor the advisories ever read a station directly; all 3 read this dataset instead. 29 variables get computed every hour from the station’s own readings through 1 shared set of formulas, which is exactly why the dashboard tile showing a spray window and the model deciding whether an infection period has started are always looking at the same number.
Field by field
6 fields carry the hourly means that make a row complete on its own. Dew point is the temperature at which the air’s own water would condense; dew-point depression is air temperature minus dew point, and the smaller that gap runs, the closer the canopy sits to wetting. Vapour-pressure deficit, in kilopascals, is the drying power of the air itself, the quantity plants respond to directly. The 6-hour pressure trend flags a front approaching or already passing through. Reference evapotranspiration for the hour, in millimetres, is the water a reference crop lost in that hour, and the water balance subtracts exactly that from the root zone.
Delta-T is dry-bulb minus wet-bulb temperature, the number sprayer operators work from: below 2 °C droplets don’t dry and drift, above 8 °C they evaporate before they ever land. The spray-quality index runs 0 to 100 for the hour, with a label from ideal to unsuitable and the single factor that limited it named alongside: wind, delta-T, inversion or rain. Inversion risk reads low, moderate or high, since a night-time inversion can carry fine droplets sideways for kilometres. The leaf-wetness hour reads 1 the moment the canopy is judged wet, with the reason and the length of the current wet run attached; the fungal models count these wet hours at given temperatures straight into disease severity. 10 degree-day fields give the hour’s own contribution above each organism’s base temperature, 1 for the crop and 9 for named pests: aphid, whitefly, potato tuber moth, fall armyworm, armyworm, diamondback moth, Tuta absoluta, thrips and coffee berry borer.
Calibration tiers
Every field carries 1 of 5 tiers: measured straight from a sensor, physics from a closed formula on measured inputs, validated against an instrument on this exact station type, literature for a published model applied without local validation yet, or proxy for an inference standing in for a sensor the station simply doesn’t have. Leaf wetness on a station without a wetness sensor stays a proxy, built from dew-point depression and humidity, until an actual sensor gets fitted, at which point it becomes measured.
Reading the derived weather variables samples
48 hours at Naivasha: nights ruled out by inversion risk, afternoons limited by delta-T once humidity drops below 40 %, and usable windows confined to the 2 or 3 hours after sunrise and before sunset. The same 48 hours at a coffee station in Uasin Gishu read poor right through the afternoon, with delta-T above 10 °C at a drier, lower-altitude site, and its degree-day column tracks coffee berry borer rather than potato entirely.