Weather Intelligence for Kenyan Farmers

From Open-Meteo, an open weather data service that serves global numerical weather models (the same kind of computer models national services run, combined and open). It is not an official government weather service, and the page says so wherever it matters. The dashboard shows the time the data was retrieved, and you can refresh it manually. The service's public endpoints need no key, so nothing about you is tied to the request except the coordinates you choose.

The forecast is calculated for a point — the town you select, or your GPS position when you use it. For most farms that point is close enough to be useful, because weather systems in Kenya are rarely smaller than a few kilometres; but a farm on a hillside or in a valley can differ from the town by a degree or two, and microclimates (a windward slope, a valley floor that fogs) are not captured by any grid forecast. The honest answer: the pattern is reliable, the exact number is indicative. That is why the guidance on this page is about checking the field, not trusting the screen.

It means that across the model's own runs for that hour or day, rain of a measurable amount appeared in roughly 6 out of 10 runs. It is a measure of the model's confidence, not a guarantee about your plot — 60% can miss, and 40% can hit. Read it relatively: 15% and 80% are very different days, while 55% and 65% are the same kind of day. The dashboard pairs the probability with the expected amount in millimetres for a reason — a 60% chance of 0.5 mm is a different decision from a 60% chance of 12 mm.

It is a deterministic set of rules — no statistics invented, no black box — that weighs three weather variables: the rain probability and amount for the next 24 hours, the reference evapotranspiration (how hard heat, sun and wind are pulling water out of the soil), and whether a dry spell is building across the 7-day window. The thresholds live in one published configuration file, so the same weather always produces the same outlook. The output is an attention level — LOW, MODERATE or HIGH — about how much thought the weather is asking for, not a command. It will never tell you "do not irrigate today"; it will tell you that rain makes checking the soil before your normal cycle the sensible move.

Reference evapotranspiration (ET₀) is a weather-only number: how much water a hot, sunny, windy day pulls from an open surface, in millimetres per day. It does not measure your crop's need — that also depends on soil, stage, cover and variety — but it is the closest thing weather data has to "how thirsty is the air today". A high ET day after no rain is when dry spells quietly become crop stress, which is exactly the situation an irrigation outlook should catch. Where the provider does not supply ET for a location, the dashboard says so and works from rain and temperature instead.

Use it the way farmers use any 7-day view: as the shape of the week, not the script for it. The first 1–3 days are where the numbers earn their keep; by day 5–7 the forecast is a planning signal — "a wet spell is coming" or "a dry stretch is ahead" — and the right response is to stage the decision (prepare beds now, wait for the window, harvest the crop that won't wait rather than the one that will) instead of betting on a specific number seven days out. For season-scale decisions, the rainfall outlook for your zone plus the soil in your hand outrank any single week's forecast.

Because the models themselves change. They are re-run several times a day, and every new run revises the previous one — a system can strengthen, weaken or shift overnight. To keep the dashboard from flickering, identical requests are served from a short local cache (about 30 minutes) instead of hitting the service again; the Refresh forecast button bypasses that cache and fetches the latest. If a refresh cannot reach the service, the page shows the last available data and labels it as such rather than pretending it is current.

No — and it should never do that. The forecast is a property of the location, full stop. Selecting a crop changes only the interpretation: what the same rain, wind and temperature mean for that crop's sensitive windows (disease pressure for tomatoes, wet feet for onions, water stress at flowering for passion fruit). The same 8 mm day is read differently for a watermelon at fill than for kale in a shadehouse — the numbers stay identical, the advice doesn't.

They are the two sides of the water balance the irrigation outlook weighs. Rain adds water to the soil; evapotranspiration removes it (directly from the surface and, as the soil dries, from the root zone). A day of 15 mm under a high-ET sky is a different soil state than 15 mm under a cool, overcast one — the first can evaporate a meaningful share before it soaks in, the second mostly sticks. The outlook looks at both together, which is why it can flag "check before your cycle" even on days with modest rain, or flag a dry spell that the rain numbers alone would not show until it is already hurting the crop.

That is exactly what a proper drip design does — but farm-specific. This page is a free daily tool: it tells you what the weather is asking for today. A farm assessment goes further: it measures your water source and soil, works out your crop's real requirement, and sizes the system so that the dry spells the forecast keeps pointing at are handled by storage and scheduling instead of by guesswork. If the weather page is changing how you water, the quote page is where the farm-specific version starts.