The satellite through the maintenance crew's eyes

From the weekly snapshot to the field job capture. A maintenance lead's daily practice — and everything the satellite doesn't replace.

01Monday morning — a weekly snapshot

Our real system: vegetation risk map (Warsaw East, 2025) — live satellite imagery, ML height prediction, dynamic ±15 m buffer zone, per-span hazard class.

The maintenance team opens the interface coffee in hand on a Monday morning. A map, the network, a few sections shifted to yellow, two turned red since last week. None of them is an alarm. Each is just a question: do we need to go out, or can it wait?

This is the weekly snapshot the Sentinel-2 satellite provides. It revisits any given point on Earth roughly every five days and produces 10-metre resolution imagery of the line corridors. In itself, that's still just a picture. What the system does is translate it into a language a maintenance lead can act on at 8 a.m. over one coffee: which sections have shifted since last week — and in which direction.

That's the point. A red section doesn't mean trouble. It means look here first. And green isn't a guarantee — only that nothing material has changed since the last pass. That's what the satellite gives you on its own. In exchange, it gives it for the whole network, every week, for the price of a sandwich per kilometre.

But we deliver much more than that

Ten metres is just the entry point — the Sentinel-2 native resolution. Under Talamone's management we turbocharge it further: with super-resolution methods, fusion processing and grid-specific calibration, we work at near-metre resolution. Continuous testing, continuous improvement — what was 10 m a year ago is much finer today, and finer again next week. This is not a sealed system but a maintained tool.

The weekly frequency — not about growth, about change

The maintenance lead has a fair point: vegetation doesn't grow into a line in a week. Correct. But change can happen at any time. Maybe someone cut the tree last week. Maybe a storm broke it. Maybe a June heatwave dried out the surroundings. The weekly snapshot doesn't mean vegetation changes weekly — it means every week, we notice if anything has changed. In any direction.

And we don't wait for Monday morning

Here is the most important part. The weekly report is just the default — the automatic snapshot. But the system is interactive, not batch. At any time, for any period, you can query it:

"What changed over the past three months on corridor 47?" — answered in seconds. "When was the Maule section last trimmed, and how much has grown back since?" — one query. "What's the current fire risk in the southern region, and where is it highest?" — one map. You just have to ask the system. That's the task.

This means the maintenance team isn't waiting for someone else to tell them what matters. They query their own network themselves, by their own criteria, on their own time horizon. The weekly snapshot is just the base rhythm — the system behind it is always fresh.

02What the satellite doesn't do for you

This part doesn't get said often enough. There's too much magical thinking around satellite monitoring. Let's be precise about what it's not good for:

It doesn't identify a hairline crack on an insulator. Ten-metre resolution is built to evaluate a corridor section, not a component. Drones and vehicle cameras handle that.

It doesn't tell you how to cut. Tree species, growth rate, local specifics, access route — these are the technologist's decisions and the field engineer's job. The satellite doesn't know these better than the operator.

And it doesn't replace the site survey. Before any intervention there's always a field check. The only question is where, with what, and how precisely the team goes out.

And that's exactly what the satellite is for: it makes that decision precise.

03The weekly rhythm — what this means in maintenance

Ten, twenty, fifty thousand kilometres of network. Under the traditional cycle, the crew gets to every section once every three to five years.

The traditional vegetation-management cycle on distribution networks is three to five years. The crew gets to any given section once every three to five years; the rest of the time, events drive the work — after a storm, after a complaint, after an outage. The problem is that what started growing toward a conductor in year 2 is an outage by year 4.

The weekly satellite snapshot removes that blind spot. Not by detecting something everywhere — but by providing daily ranking. The maintenance lead doesn't discover a tree growing into the line five years later. They see it the week the environment around it started changing, in the satellite's view.

No more than that. The rest, the maintenance lead knows better.

The whole MV network in one view, colour-coded per span. BioGuardNet-H height inference (Sentinel-2, year-over-year delta) decides which span to visit first — yellow and orange mark the approaching cases, red the next intervention.

04A typical case — from red section to closed job

On the office side the ordered package is broken down into work: scheduling, crew and responsible person, permits (outage, road closure, consumer notice) and cost in one place. The map shows the full span, pole by pole, with the task area.

Take a routine situation. Early June, the weekly report flags a corridor section red — the satellite shows an NDWI drop, meaning the vegetation's water content is declining. Heatwave in June, faster drying, increased fire risk on an otherwise normally-cycled section.

The maintenance lead takes a look. The map automatically offers context: which span, what vegetation type was there at the last pass, how close to the conductor. Pulling this used to take a GIS query — now it's one click.

That's the satellite. From here it's the field engineer.

The field engineer receives the job — the satellite signal and the environmental context already on their phone. From here it's the engineer's job: validate, intervene, close.

The job is dispatched. On the engineer's phone, the section appears with the satellite's reason for the flag, and an empty job sheet with the few questions only they can answer on site: is the risk actually present, which tree specifically, accessible by road or on foot, does the owner need to be notified, what type of vegetation is the issue (branch, trunk, brush), and what's the recommendation — go now, or combine with the next scheduled visit.

They go out the next day, photograph, fill it in. A week later, when the satellite passes again, the next image confirms it: the intervention happened, and the risk class has dropped. The section is reclassified — from red to green. Not because the vegetation has "greened up" (on a dry, fire-prone section, NDVI greening would itself be the risk), but because the risk model, based on the intervention and the new measurement, assigns a lower class.

That's one week. That's the rhythm.

05The language of indices — briefly, from the technician's side

A satellite image by itself says nothing. The spectral indices do — and it's worth saying a few words about them, because these are the actual language of the weekly snapshot. A quick round, from the maintenance perspective:

NDVI — the classic vegetation index. Tells you whether there's anything to look at in the corridor: is the vegetation alive, dense. What the crew can see by eye — except the satellite does it every week, for the whole network.

NDWI — water content. This is the fire filter: where vegetation dries out, flammability rises before visible symptoms appear. In a June-July heatwave, this is worth watching daily.

NDRE — early stress, chlorophyll level. What still looks green to the eye, but the plant is already stressed — this is the 2-4 week early warning that "something's coming".

SAVI — soil correction. Where vegetation is sparse, it stops the system from raising false alarms. So it's strong on the where there's definitely no issue side.

No single index gives the answer alone. Several layers together filter where real risk lies — the rest stay green, and the maintenance crew doesn't burn kilometres for nothing.

06What stays with the engineer — the field job capture

The real field/office tool: preparing the job — scheduling, team, permits. Only the fields the engineer actually has to answer appear.

The surveyor in the field: for the marked point they pick a work type — pruning, tree felling, pole-base cleaning, substation, scrub control, right-of-way, stump removal. The device works offline, with the point and the safety corridor shown on the map.

The form for the chosen work type asks only what matters: quantity, terrain type, and — where needed — a mandatory photo. Field capture stays fast, consistent and traceable.

So far the satellite has been doing the work. But the whole point of the chain is that the maintenance team makes the decision — the satellite only saves time and kilometres. This is where the field tool comes in, the one the engineer uses.

The field tool is not a form, and not an alarm list. It's a job capture. The logic is simple: every section (span) is a worksite, identified in the network registry. The engineer doesn't have to type "a tree next to pole 47" — the system already knows which span, and what vegetation type was there at the last pass.

Context menu in the buffer zone

The engineer taps on the map at the point in question — within the safety strip around the conductor (the buffer zone). The system offers what can be recorded there: is this a worksite tag, an observation, or an owner to be notified? One tap, and the right data lands in the right place — no code to type, no field to pick, no dropdown to browse.

Two-step span → worksite

Step one: which span are we talking about. The system itself proposes this from GPS and network data. Step two: what's the action there. This opens a short, context-aware form that shows only the fields that actually matter.

Conditional fields — only what's relevant

If the vegetation type is "trunk", the system asks about access (road, on foot, lift needed). If "branch", it doesn't. If the owner can be named, a notification block appears — needs notifying, when, how. If it's public land, this is skipped. The engineer only sees what needs to be filled.

Closure = save

When the engineer is done capturing, they close the worksite. That's the save. From there, the back-office workflow starts: the dispatcher approves, assigns, schedules. The engineer doesn't have to deal with the connecting processes.

When the satellite passes again a week later and the risk class has dropped — that's not an interesting data point on its own. That's the re-measurement that closes the entire loop. The weekly snapshot, the maintenance lead's decision, the field job capture, the closure, the next snapshot. That's one week. That's the rhythm.

07Back and forward — both directions of the timeline

The same network on a time axis: past satellite measurement on the left, now in the centre, forecast on the right. Drag the slider to replay how a span moved from watch to intervention — and what's expected over the coming quarters.

A single span on the time axis: the blue line is the history of real satellite measurements, the vertical marker is now, and the cone opening to the right is the forecast uncertainty band. The dashed red line is the critical threshold — where intervention becomes necessary.

Any earlier date can be recalled: the top sliders pick a year and day, and below it the state of that moment appears per hazard class (HC0–HC3), together with the measured coverage.

A ten-year forecast with a simple, transparent model: assuming half a metre of growth per year, the system shows how vegetation migrates toward higher hazard classes — and how many kilometres slip into the critical band.

Comparing two dates: pick an earlier and a later moment, and the system computes the actual change per hazard class — how much the vegetation grew between the two measurements.

So far we've talked about the weekly cycle. But the system doesn't begin on Monday morning. It reaches back, and it looks forward. That's what turns this from scattered alerts into a process.

Back: 3–4 years of timeline, already yours

When a DSO first puts their network on the system, they don't start from zero. The Sentinel-2 archive imagery is immediately available for the past 3–4 years — the system pulls the historical vegetation trajectory for every section out of it. "What did this span do since 2023? When was it last trimmed? How fast did it grow back?" — answered with a query.

This isn't archaeology. It's the calibration base: your network's past trains the model of its future. A generic satellite snapshot says nothing. Your 4 years on your sections — that says a lot.

A span's trajectory: 2022-2026 the historical record (archive Sentinel-2), 2026 the present state, 2026-2028 the modelled future — two scenarios: with scheduled intervention or without.

Forward: 1–2 year medium-term modelling

If a new measurement arrives every week, and there's a 3-4 year timeline behind you, the system builds a trend for every section. Not prediction — calibrated, network-specific extrapolation: the past seasonal patterns + the current growth rate + the species parameters.

The horizon is 1-2 years, depending on the species. For slow-growing oak this is a stable scheduling window. For black locust the intervention window may close within a year — there, months matter. The species parameter is built in.

The system isn't a sequence of weekly snapshots. It's a continuous timeline that reaches back into the past and looks forward into the future — and every week, one new measurement adds to it. That's the full system-level capability that makes the weekly rhythm meaningful.

TO BE CONTINUED

The next piece is about the drone and the vehicle camera — what happens when the satellite signal sends you in to look at a specific span up close. And why our own AI isn't ChatGPT.