How to catch a failing sprinkler zone from its water flow

Zones fail quietly

Most sprinkler problems do not announce themselves. A riser snaps below the mulch and the head geysers on the next cycle. A valve diaphragm wears and the zone weeps long after the schedule ends. A drip emitter clogs and a row of plants slowly goes dry. None of these trip an alarm. You usually find out weeks later, from a brown patch, a soggy corner, or a water bill that jumped for no obvious reason.

The frustrating part is that the evidence was there the whole time, in the water. A broken head moves more water per minute than a healthy one. A clogged zone moves less. If something is watching each zone's flow, the problem shows up on the very next run, not on the next bill.

Why this is not leak detection

It is worth being precise, because "leak detection" already means something. Whole-home leak detection, including the alerts your Flume monitor sends, watches your total household water for flow that should not be happening: a continuous trickle at 3 a.m., a fixture left running, a slab leak. That is a great safety net and we do not try to duplicate it.

Flow monitoring asks a different question. It only looks at water used while a specific zone is running, and it asks whether that zone is behaving normally for itself. A zone that waters at its usual rate is fine even if the whole house is using a lot of water that day. A zone that suddenly waters fifty percent harder than it ever has is suspicious even if the household total looks ordinary. The two approaches catch different failures, and they work best together.

Every zone has a normal flow rate

Barranca Verde already attributes gallons to each Rachio zone run by lining up Rachio's start and stop events with your Flume flow data (we covered that mechanism in our Rachio and Flume integration guide). Divide a run's gallons by its minutes and you get its flow rate in gallons per minute. Do that across a zone's run history and a pattern emerges fast: most zones are remarkably consistent. The same heads, the same pipe, the same pressure produce nearly the same rate every time.

That consistency is the whole opportunity. If a lawn zone has run at right around 15 gallons per minute on every one of its last thirty cycles, then a run at 28 gallons per minute is not weather or chance. Something changed at that zone. The flip side matters too: a drip zone that normally moves 2 gallons per minute and suddenly moves 1 has very likely clogged.

Because every zone has its own normal, we compare each zone only to itself. A fixed company-wide threshold would be hopeless here. Set it low and the big rotor zones alarm constantly; set it high and you would never catch a problem on a small drip zone. Each zone gets its own yardstick.

Why we use the median, not the average

To describe a zone's normal we need a center (its typical rate) and a spread (how much it normally wiggles). The obvious choice is the average and the standard deviation. We deliberately do not use them, for one important reason: they are easily fooled by the very problems we are hunting for.

Imagine a zone whose head broke two weeks ago and has been running high ever since. Those bad runs drag the average up and inflate the spread. Now the zone's "normal" has quietly absorbed the malfunction, and the next bad run looks perfectly ordinary. The detector has been trained to ignore the problem.

The fix is to use robust statistics: the median for the center and the median absolute deviation for the spread. The median is the middle value, so a handful of extreme runs barely move it. If twenty-six of a zone's runs sit near 15 and six recent ones spike to 18, the median stays at 15 and those six spikes stand out instead of hiding. A past problem cannot desensitize the detector to the next one. That property is the difference between an alert that fires when it should and one that has been lulled to sleep.

How an alert is decided

With a robust center and spread in hand, judging a new run is straightforward. We measure how far it sits from the zone's median, counted in units of that zone's normal wiggle. A run that is a fraction of a wiggle away is business as usual. A run that is many wiggles away is the zone shouting.

Two guardrails keep the alerts honest:

• Persistence. A single odd run can happen for harmless reasons: a soak-cycle split, a co-running zone, a sensor hiccup. Real hardware problems persist, because a broken head stays broken. So an ordinary alert waits for two runs in a row to deviate the same direction before it speaks up. One-off wobble stays quiet.
• Severity. If a single run is extreme enough, far outside anything the zone has ever done, we do not wait. That fires immediately as a higher-severity alert, on the theory that a truly large jump is worth interrupting you for right away.

The result is a detector that stays silent on a consistent, healthy system and speaks up only when a zone genuinely changes behavior, in either direction. When it does, you get a banner on your dashboard, a badge in the event log on the offending runs, and at most one email per zone per day so an ongoing problem does not turn into a flood of notifications.

A real zone we caught

What it is, and what it is not

A few honest boundaries, because no detector is magic:

• It is retrospective, not real time. Flume's flow data arrives a few hours after a run, so an alert lands after the cycle finishes, not while a head is spraying. It is a "go check that zone" signal, not an emergency shutoff.
• It only sees water used during a zone run. A leak that flows when no zone is scheduled, or that continues long after the schedule ends, is outside its view. That is leak-detection territory, and it is why your Flume's own alerts remain the right tool for standing leaks.
• It needs history. A zone has to log enough runs before it has a trustworthy normal. New zones simply wait until there is enough to judge against.
• It is a beta. We are still tuning the sensitivity against real outcomes, so treat an alert as a prompt to look, not a diagnosis.

Turning it on

Flow monitoring rides on the same data you already have once Rachio and Flume are connected, so there is nothing new to install. It is rolling out gradually while we tune it. If you already track your zones with Barranca Verde and want in, let us know at [email protected]. If you are not set up yet, connecting both devices takes about five minutes and gets you per-zone gallons and cost right away, with flow monitoring layered on top as it expands.