The credibility centerpiece

The water balance is the measuring instrument.

Sprinkler Gnome doesn’t guess and it doesn’t just read a sensor. It keeps a running water budget for each zone — and waters to refill exactly what’s missing. Here’s the whole mechanism, in plain language.

Why a timer wastes water

A fixed timer waters the same minutes regardless of weather, season, or what your soil is actually holding. On a cool, wet week it overwaters; in a dry spell it can underwater. Roughly half of outdoor water use is lost to overwatering. The fix isn’t a bigger schedule — it’s watering to need.

Evapotranspiration: the daily water budget

Plants and soil lose water every day to evaporation and transpiration — together, evapotranspiration. Reference ET (ET₀) is how fast a standard grass surface loses water under today’s sun, heat, wind, and humidity. Sprinkler Gnome pulls ET₀ from an online weather service, then scales it to each zone’s actual planting with crop coefficients:

ETc = (Ks·Kcb + Kr·(KC_MAX − Kcb)) × ET₀

In words: how much water this zone’s plants want today (their growth-state coefficient Kcb, tapered by how dry the soil is via Ks), plus what evaporates off the soil surface (Kr). Bermuda, fescue, shrubs, beds and gardens each get their own coefficients and growth state.

From deficit to dose

The model tracks a single number per zone: the root-zone deficit (how many millimeters below “full” the soil is). Each day ET draws the deficit down; rain and irrigation fill it back up. When the deficit crosses a refill trigger — a configurable fraction of the soil’s available water — the zone is due. Then it waters a refill-to-full dose: just enough to close the deficit, converted to a runtime from your system’s output rate. Forecast rain in the next 24 hours cancels the run.

It learns your soil — and your sprinklers

This is where Sprinkler Gnome leaves weather-only controllers behind. It doesn’t just apply a generic model — it learns your specific yard. Cheap soil sensors are trend-faithful but distorted in absolute terms, so we never read “how much to water” straight off the sensor. Instead, the controller learns your soil from natural rain events: it anchors the wet end (saturation, and the drainage “knee” that marks field capacity) and infers the dry end from soil physics. It also learns each zone’s real application rate — how fast your actual sprinklers put water down — so the dose is tuned to your system, not a nameplate number.

That’s genuine on-device learning — call it AI if you like — and it compounds: the controller gets more accurate the longer it runs, and it’s tuned to your soil and your heads, not a one-size-fits-all schedule. The weather-based incumbents don’t do this. Day to day, a gated filter uses the sensor to correct the model’s deficit and to verify the soil actually responded after watering; display and control share the exact same computation.

Why it runs locally

Every step above runs on the controller, on your network. The only outside call is fetching the weather forecast; if that’s unavailable, the model keeps going. There’s no vendor cloud deciding your schedule, no account, and nothing about your yard leaving your LAN. That’s the difference between a controller that’s connected and one that’s dependent.

What you see

A dashboard on your own network shows live zone status, the 7-day soil-moisture projection (produced by the same model that makes the decision, so the forecast and the schedule can’t disagree), watering history with calibrated moisture and real soil temperature, and mowing windows. Manual control is always one tap away.

Questions

Does it need the internet to work?

It uses an online weather forecast when one is available, but the watering decision is computed on the device. If the connection drops, it keeps running on its model rather than stopping.

Where does the weather data come from?

Open-Meteo (free, no API key) supplies reference evapotranspiration (ET₀) and the rain forecast. That’s the only outside data the model needs, and it’s pulled directly by the controller — not routed through a vendor cloud.

What if I don’t install a soil sensor?

The model is the dose authority on its own, so a sensorless zone still waters correctly. A sensor adds a correction-and-verification layer — it nudges the model toward what your soil actually did.