How to estimate feeder-level PV penetration without AMI data

AMI rollouts don't line up with feeder boundaries. A utility might have smart meters on 70% of its customer base and still have entire rural or legacy circuits running on electromechanical meters that report nothing about generation behind the meter. If your DER forecast depends on AMI net-export signatures to spot rooftop PV, those feeders are a blind spot, and they're often the ones with the fastest adoption curves because that's where incentive programs hit hardest.

So the question planners ask isn't "how accurate is our AMI data." It's "what do we do for the feeders where we don't have any."

Why AMI alone undercounts PV by feeder

Even on feeders with full AMI coverage, net-export detection has gaps that matter for load forecasting. A system sized to offset most of a household's own load may never show a clean export signature, so it doesn't trip the algorithms that flag "this meter has solar." Add in meters that get swapped mid-cycle, interval data that drops out during outages, and customers on legacy tariffs that don't require interval reporting at all, and the count you pull from AMI is a floor, not a total.

Interconnection queue data fills some of that gap, but it has its own lag. An application gets filed, inspected, and approved on a timeline that runs weeks to months behind the install itself. For a feeder where adoption is accelerating, that lag means your queue-based count is always chasing the real number instead of matching it.

Building a feeder-level estimate without smart meter signatures

The practical workaround most planning teams land on is to combine three things:

Interconnection queue records, for a baseline count with known lag. Permit and contractor-filing data where the jurisdiction makes it available, which closes some of the queue lag but varies wildly by county and often isn't geocoded to a feeder boundary at all. And a physical check of the roofs themselves, because rooftop PV is the one thing in this whole chain you can actually go see.

That third piece used to mean a windshield survey or a GIS tech pulling permits parcel by parcel and manually joining them to a feeder shapefile. It's slow, it's labor-intensive, and it still misses installs that never got permitted or where the permit record is missing a parcel ID.

The alternative is to work from imagery instead of paperwork. Satellite or aerial imagery at sub-meter resolution can resolve individual roof planes, score each one for azimuth, tilt and shading loss, and flag which planes have panels on them today. Run that same scan again a year later over the same footprint and you get a count of roofs that went from bare to panelled in the interval, independent of whatever AMI or permit data existed for that feeder. Join the roof footprints to your feeder shapefile once and you have a penetration estimate that doesn't depend on meter type, tariff class, or permit backlog.

A roof-plane scan run on an annual cadence gives you that adoption count as a map layer by feeder, alongside the usable roof area still unclaimed, which is the number most hosting capacity studies want next.

What this estimate is good for, and what it isn't

A roof-level imagery count tells you what's installed and roughly how much headroom remains on roofs that haven't gone solar yet. It doesn't tell you system size, inverter rating, or export behavior, so it's not a replacement for AMI where AMI exists and works. Think of it as the layer that covers the feeders AMI doesn't reach, and as a cross-check on the feeders where it does, since a queue count and an imagery count that disagree by a wide margin usually means one of your data sources has a gap worth finding before it shows up in a load forecast.

If your DER team is staring at a feeder with no AMI, a thin interconnection queue, and a hosting capacity study due next quarter, an annual roof-plane pass over that circuit is worth a look.