Skip to content
Drone Data

How are drone datasets delivered to a PV analytics platform?

Published on Updated on 10 min read

deliver drone datasets to PV analytics platform

Drone datasets are delivered to a PV analytics platform as complete, unmodified image sets — radiometric JPEGs plus RGB frames, EXIF and GPS intact, organised in the folder structure the platform asks for, and pushed either through the platform’s own uploader or to a cloud link. For parks up to 20 MW, Drone Data (Bucharest / Ilfov, missions anywhere in Romania) delivers within 5 working days.

What exactly does a PV analytics platform receive after a flight?

A platform receives three things: the raw radiometric thermal images, the matching RGB images, and the flight metadata that ties each frame to a position on the site. Nothing is resized, re-encoded or stripped. The dataset arrives as it came off the sensor.

The thermal frames from the DJI Mavic 3T are radiometric JPEGs. Every pixel carries a temperature value, not just a colour, so the platform can recalculate ΔT, apply its own emissivity and reflected-temperature settings, and classify anomalies against its own thresholds.

RGB frames at 48 MP are included for visual confirmation. A hotspot in the thermal image needs a visual twin — soiling, a cracked glass, a shaded cell, a junction box — before an O&M team sends a technician to the string.

Alongside the images, the delivery includes a flight log and a short acquisition note: date, start and end time, measured irradiance, ambient temperature, wind, cloud cover, flight altitude, GSD and overlap values actually flown. That note is what lets an analyst defend a report six months later.

When the scope includes reporting rather than raw acquisition only, we run the anomaly analysis and the output is a PDF report with annotated images, anomaly classes, string-level localisation and an orthomosaic of the site. Platforms that run their own analytics usually want the raw set only — which is the faster, cheaper path for them.

Why does flying to the platform’s own specification matter?

Because a platform’s detection algorithms are tuned to a fixed input. If the GSD, overlap or flight geometry drifts from the specification, cell-level anomalies get smeared across too few pixels and the automatic classification degrades. The acquisition specification is part of the product.

A typical platform brief reads something like 6 cm GSD with 85% front and 65% side overlap, nadir camera, single-grid pattern, constant altitude above the module plane. Drone Data flies that brief as written, including altitude recalculation on sloped terrain or on trackers that tilt during the pass.

Overlap is not decoration. At 85% front overlap, each module appears in several consecutive frames, so a thermal signature can be confirmed rather than guessed from one noisy capture. Side overlap of 65% keeps the mosaic geometrically clean for stitching.

Conditions matter as much as geometry. Following the recommendations of IEC TS 62446-3, thermography is flown at a minimum irradiance of 600 W/m², with low wind and stable cloud conditions, so module temperatures are high enough for a real ΔT to appear.

Coverage is checked on site before the crew leaves. Image count against planned waypoints, gaps in the flight path, battery swap overlaps and sensor focus are verified in the field. A return visit for a 40-hectare park costs days — a ten-minute check does not.

Flying to your platform's specification

Drone Data flies the brief as written — GSD, overlap, irradiance window — and checks coverage on site before the crew leaves. See how the acquisition partnership works across sites in Romania.

See partner details

How are drone datasets delivered to a PV analytics platform in practice?

Three delivery routes cover almost every case: the platform’s own uploader or desktop ingestion client, a cloud transfer in the platform’s folder structure, or a shipped physical drive for very large sets. The route is agreed in the brief, before the flight, not after.

Route one — the platform’s uploader. Many PV analytics providers supply a web uploader or a desktop tool that ingests a folder and validates EXIF on the way in. Drone Data runs the upload directly, with platform credentials supplied by the client, and reports the validation result back.

Route two — cloud transfer. The set is uploaded to a shared location as a structured archive, with a checksum list so the receiving team can confirm nothing was lost in transit. Folder naming follows the platform convention exactly: site, block, inverter, string, flight, sensor.

Route three — physical media. For multi-site campaigns running into hundreds of gigabytes on a site with poor connectivity, an encrypted drive shipped the same week is often faster than a week of broken uploads. The cloud copy still goes up in the background.

A standard folder layout looks like this: a top-level folder named by site and flight date, then subfolders for THERMAL and RGB, then one subfolder per flight or per block, with the original file names preserved. Renaming breaks the link between frame and log, so file names stay untouched.

Every delivery includes a manifest: file counts per folder, total size, sensor model, firmware version, flight dates and the measured conditions. One technical contact handles the delivery from brief to upload confirmation, so there is no handover gap if the platform’s ingestion team asks a question.

Why must EXIF and GPS data stay intact in a drone dataset?

EXIF and GPS tags are how a platform turns 4,000 loose images into a map of your plant. They carry latitude, longitude, altitude, timestamp, yaw, camera model and the radiometric parameters needed to read temperature. Strip them and the images become unanalysable.

Without GPS, no anomaly can be located. A hotspot is only useful if it resolves to block 7, inverter 3, string 12 — and that resolution comes from the geotag matched against the plant layout. An anomaly without coordinates is a photo, not a work order.

Without the radiometric tags, temperature values are gone. A re-saved or compressed thermal JPEG looks identical on screen but no longer carries per-pixel data, so ΔT cannot be computed and the platform rejects or misclassifies the frame.

Three common ways EXIF gets destroyed: opening and re-saving images in a photo editor, sending them through a messaging app or email client that recompresses, and running a batch rename or format conversion. Drone Data copies card to drive to delivery with no intermediate editing step, which is why datasets arrive with EXIF and GPS intact.

Integrity is also an audit question. Investors and asset managers reviewing a thermal campaign want to see unmodified source data behind the report, with timestamps that match the stated flight window and conditions. An EXIF-intact dataset is the evidence trail.

Need an acquisition crew in Romania?

If your platform handles the analytics and you need raw radiometric sets delivered through your own uploader, we can cover single parks or multi-site campaigns anywhere in Romania.

Read about partnering

How large is the dataset and how fast does it arrive?

A 10 MW park typically produces a few thousand thermal frames plus the matching RGB set, in the range of tens of gigabytes. Flight time is the smaller part of the schedule: a roughly 10 MW park is flown on site in about 2 h 15 min, and reports for parks up to 20 MW are delivered within 5 working days.

Size scales with array area, flight altitude and whether RGB is captured at full 48 MP alongside the thermal pass. A dual-sensor pass over a dense fixed-tilt layout generates considerably more data per MW than a sparse tracker site with wide row spacing.

Daily capacity on a day of good irradiance reaches up to about 50 MW. The largest mission in 2026 covered about 60 MWp over about 70 ha and was completed in one or two flight days. Total PV capacity flown in 2026 passed 400 MW.

Timing in practice: the raw set is copied and checked the same evening, so a platform that only needs acquisition can usually start ingesting within one to two working days of the flight. For a detailed breakdown of a typical site day, see our guide to how long a thermal inspection of a 10 MW solar park takes.

Scheduling is the other half of the answer. Mobilization anywhere in Romania takes 24 to 72 hours, and quotes come back within 24 hours on working days — which matters when a platform has an ingestion window to hit or an asset manager needs data before a board review.

How should an international platform brief a local acquisition partner?

Send the acquisition specification, the plant layout and the delivery target. With those three, a flight plan can be built and a slot booked without a single call. Everything else — altitude, battery plan, crew size — follows from the specification.

The acquisition specification should state GSD, front and side overlap, camera angle, minimum irradiance, maximum wind, acceptable cloud cover and whether RGB is required in the same pass. Drone Data flies to that document rather than to a house standard.

The plant layout matters more than most briefs assume. A block, inverter and string naming map lets anomalies be localised in the delivered structure, so the platform’s own layout import lines up with the folder names without manual remapping.

The delivery target defines the last mile: uploader credentials, folder convention, file naming rule, maximum archive size, checksum requirement. Agreeing it before the flight removes the most common cause of re-delivery — a correct dataset in the wrong shape.

Compliance is part of the brief too. Flights are run by pilots holding EASA open category A1/A3 and A2 certificates, with airspace clearance handled in advance, and thermography flown in compliance with the recommendations of IEC TS 62446-3. For asset managers in 2026 building a repeatable annual inspection programme across several Romanian sites, that combination — fixed specification, fixed folder structure, one technical contact — is what makes the data comparable year on year.

Where the scope goes beyond thermography, the same acquisition discipline applies to photogrammetry: orthomosaics, 3D models and point clouds processed in RealityCapture and CloudCompare, with PPK corrections through ROMPOS for centimetre-level accuracy.

Frequently asked questions

Why must EXIF and GPS data stay intact?

EXIF and GPS tags carry the position, timestamp and radiometric parameters that let a PV analytics platform locate each anomaly and calculate temperature. Without them, a thermal image cannot be matched to a block, inverter or string, and ΔT cannot be computed. Re-saving images in a photo editor or sending them through email compression destroys this data, which is why datasets are copied card to drive to delivery with no editing step.

What file formats do PV analytics platforms accept?

Most PV analytics platforms accept radiometric JPEG for thermal frames and standard JPEG for RGB frames, with EXIF and GPS metadata unchanged. Some also accept R-JPEG variants or TIFF exports where per-pixel temperature is preserved. Deliverables beyond the raw set include PDF reports, orthomosaics, 3D models and point clouds, depending on the agreed scope.

How large is a thermal dataset for a 10 MW park?

A 10 MW park typically yields a few thousand radiometric thermal frames plus the matching 48 MP RGB set, which lands in the range of tens of gigabytes. The exact size depends on array density, flight altitude and whether RGB is captured in the same pass. Flight time on site is short by comparison — a roughly 10 MW park is flown in about 2 h 15 min.

Can you upload through the platform’s own uploader?

Yes. Drone Data uploads directly through the platform’s web uploader or desktop ingestion client using credentials supplied by the client, and reports the validation result back. Where no uploader exists, the dataset is delivered by cloud transfer in the platform’s folder structure with a checksum list, or on an encrypted drive for very large multi-site campaigns.

How long after the flight is the dataset available?

The raw set is copied and checked the same evening as the flight, so a platform needing acquisition only can usually start ingesting within one to two working days. Full reports for parks up to 20 MW are delivered within 5 working days. Mobilization anywhere in Romania takes 24 to 72 hours, and quotes come back within 24 hours on working days.

What irradiance is required for a thermal inspection of PV modules?

Thermography of PV modules requires a minimum irradiance of 600 W/m², following the recommendations of IEC TS 62446-3. Below that level, module temperatures are too low for a reliable temperature difference to appear between a healthy cell and a faulty one. Low wind and stable cloud conditions are also needed, which is why flight slots are planned around the forecast rather than the calendar.

Related services

  • Solar farm thermal inspection

    Radiometric thermal and RGB survey of PV plants, with every anomaly classified and located by string and module. Report in 5 working days for plants up to 20 MW.

    See the service

  • Thermal and visual drone inspections

    Roofs, facades, industrial halls, power lines and wind turbines inspected from the air, with an annotated report and a ranked defect list.

    See the service

Tell us about your site

Send the location, the size and the deliverable you need. We reply with a written quote and a proposed flight window.

  • Reply within 24 h on working days
  • Written quote, no obligation
  • Your data stays confidential

We work across Romania, with mobilization in 24–72 h.

  • Reply within 24 h on working days
  • Written quote, no obligation
  • Your data stays confidential
CallWhatsAppQuote