Learn 6 common drone thermal inspection mistakes that can reduce solar inspection accuracy, from poor flight conditions to data analysis and reporting.
Karthik Mekala
CMO
Published on
Drone thermal inspection can cover a utility-scale solar site far faster than a manual walkthrough. But collecting thermal imagery quickly does not automatically mean the inspection is accurate. Poor flight conditions, incorrect flight parameters, inadequate sensor calibration, weak image quality, and shallow data analysis can all lead to missed or misclassified defects.
Accuracy in solar thermal inspection is not a single step. It depends on the entire workflow: how the flight is planned, how the data is captured, how it is analyzed, how defects are classified and mapped to the asset, and how findings are turned into remediation. This article covers 6 mistakes that show up across each stage of that workflow, and what to do instead.
An inaccurate thermal scan does not just mean a missed hotspot. It creates a chain of downstream problems:
Missed thermal anomalies (false negatives) that continue degrading undetected until the next scan cycle
False positives that send field teams chasing shadows, reflections, or soiling instead of real defects
Incorrect defect classification, which can lead to delays in correcting the root cause of the anomaly
Incorrect temperature differential (ΔT) readings, which throw off severity scoring for every defect on the report
Missed warranty claims, since most manufacturers require defensible, standards-aligned evidence within a claim window
Difficulty tracking recurring defects, since inconsistent capture conditions make it hard to tell if an anomaly is new or the same one reappearing
Here are the 6 mistakes that most commonly undermine solar thermal inspection accuracy, and how to fix each one.
Problem: Flight altitude directly determines Ground Sample Distance (GSD), the real-world size each pixel represents in the captured thermal imaging output.
Why it skews accuracy: Choosing too high a GSD causes small anomalies like single-cell hotspots to blend into the surrounding module and go undetected. Choosing too low a GSD, on the other hand, drives up flight time, battery consumption, and cost quickly, since the drone needs far more passes to cover the same site. Effort needed at all stages, drone flight time, data upload, and data processing time, increases exponentially as GSD decreases. For example, the flight time and number of images at 3cm GSD is 4 times the flight time and image count for a 6cm GSD scan. 6cm GSD is fine enough to highlight cell-level defects. A cutting-edge 700Wp module typically has cells sized 20cm x 20cm, which ensures 2-3 pixels per cell at that resolution, so performing a 3cm GSD scan in most cases leads to increased effort for minimal marginal benefit.
Fix: Match altitude and GSD to the panel size and defect type being targeted. A scan meant to catch cell-level hotspots needs a tighter GSD than one meant only to flag string-level or module-level anomalies. At the same time, do not over-specify the GSD, as that leads to a waste of time and effort.
Problem: Thermal contrast on a PV module depends on how much energy it is absorbing and converting at the moment of capture. Flying outside the irradiance conditions the inspection methodology calls for produces a weak or misleading thermal signal. Even a passing cloud can drop irradiance below the recommended 600 W/m² threshold.
Why it skews accuracy: Low or inconsistent irradiance flattens the temperature difference between a healthy cell and a defective one, so real anomalies show up faint or not at all, and normal variation gets misread as a defect.
Fix: Use reliable weather forecast services to schedule flights within the irradiance conditions defined by the inspection methodology, and factor in wind speed and cloud cover alongside irradiance rather than checking irradiance in isolation. In case of transient conditions like passing clouds, pause the drone flights and resume once the situation is resolved.
Problem: In infrared inspection, a thermal sensor does not measure temperature directly. It measures radiated energy and converts that into a temperature reading using settings like emissivity and reflected temperature that the operator configures.
Why it skews accuracy: Incorrect emissivity or reflected-temperature assumptions distort the absolute temperature reading and, more importantly, the ΔT between a suspect area and its surroundings.
Fix: Establish the correct sensor settings for the module surface being scanned and run the required calibration and reference checks before data collection starts, not after a flight has already been logged.
Problem: Front and side overlap between consecutive images, along with a consistent altitude and flight path, is what allows individual thermal frames to stitch into one continuous, georeferenced orthomap of the site.
Why it skews accuracy: Inadequate overlap or an irregular flight path creates gaps in coverage, stitching artifacts, and sections of the site that never get a clean thermal read, so defects in those gaps simply don't appear in the final output.
Fix: Use a predefined flight plan with the overlap settings the methodology calls for, and hold altitude and flight parameters consistent for the full mission.
Problem: A single utility-scale flight can generate thousands of thermal images. Reviewing that volume manually, image by image, is slow and dependent on whoever happens to be looking at the screen that day.
Why it skews accuracy: Manual review introduces reviewer-to-reviewer inconsistency: the same anomaly can get classified differently depending on who reviewed it, and fatigue over a large dataset increases the chance of a missed defect.
Fix: Use AI-assisted analysis to detect and classify thermal anomalies consistently across the full dataset, freeing reviewers to focus on validating flagged findings instead of scanning every image from scratch.
AI-assisted analysis can help standardize anomaly detection across large datasets, while tools such as SenseHawk Therm can automate detection and classification of common thermal issues like hotspots, bypass diode failures, and string faults.
Problem: A simple list of thermal anomalies provided in the form of an Excel/PDF report is good for desktop analysis. However, for field technicians trying to locate defects in the field, referring to a flat list while walking the site is inconvenient and difficult.
Why it skews accuracy: Without accurate asset context, field teams waste time relocating the defect on site, duplicate findings can get logged as new issues, and there's no reliable way to tell if an anomaly is recurring.
Fix: Link every thermal finding to a georeferenced site layout or digital asset model, down to the module, table, string, and inverter level, so each finding is addressable rather than just visible. A good mobile application with a map view of the defects will allow technicians to navigate to the exact location without them having to orient themselves in the field.
Plan altitude, GSD, overlap, and flight path around the site layout and the defect type you're targeting.
Verify irradiance, wind, and weather conditions against the inspection methodology before every flight.
Calibrate sensors and confirm camera and emissivity settings ahead of data collection.
Capture consistent, complete thermal imagery across the full site.
Analyze, map, prioritize, and track findings through to remediation.
If available, use a map-based digital twin for reporting defects to make the lives of field teams easier and to quicken the remediation process.
Most of the mistakes above happen before the data ever reaches a platform: during flight planning, capture, and classification. Better capture leads to better thermal interpretation, which leads to better classification, prioritization, and remediation. What SenseHawk Therm, part of the TaskMapper platform, adds is context and a connected path to action once that upstream data is accurate.
Therm uses AI to detect and classify thermal issues, including hotspots, string faults, and bypass diode failures, directly from drone-captured IR imagery, replacing manual, reviewer-dependent review with consistent classification across the dataset.
Every classified finding is mapped to the specific module, string, table, and inverter it belongs to inside TaskMapper's digital twin. A thermal anomaly is never left as just a coordinate on an isolated image. It is tied to an addressable location on the site.
Therm prioritizes anomalies by temperature differential (ΔT), potential power loss, and defect severity, so teams work by impact instead of a flat, unranked list.
Findings move through one flow: detect, classify, prioritize, assign, remediate, close. Each becomes a task inside TaskMapper, assignable to a field team and trackable on mobile through to resolution.
A standalone thermal inspection finds anomalies and stops there. A connected workflow finds anomalies, maps them to the asset, prioritizes them by impact, and tracks the work to resolution.
That distinction matters for accuracy specifically. The digital twin and remediation workflow don't make the thermal scan itself more accurate, better flight planning and classification do that. What they do is make sure an accurate finding doesn't get lost, misfiled, or left unaddressed once it's detected. That's why SenseHawk builds thermal analytics as part of TaskMapper, alongside solar asset data, drone imagery, and field workflows, rather than as an isolated tool.
Solar thermal inspection accuracy starts with capture: altitude, irradiance, calibration, timing, overlap, and site conditions. Consistent AI-assisted classification and a prioritization system carry that accuracy through to the field. Fix the 6 mistakes above, and both the detection and the response improve together.
What is the ideal altitude for drone thermal inspection of solar panels?
There's no single correct altitude. It depends on the GSD needed to resolve the defect type you're targeting: catching single-cell hotspots requires tighter resolution than a scan meant only to flag string- or module-level anomalies, so altitude should be set relative to panel size and inspection goal, not a fixed number. As a general rule of thumb, for current state-of-the-art PV modules, 10cm GSD is sufficient for module and string level defect identification, and 6cm for cell level defect identification.
What irradiance level is needed for accurate solar thermography drone scans?
Accurate scans require irradiance levels within the range defined by the inspection methodology being followed, commonly referenced against IEC 62446-3 for PV thermography (which specifies a minimum threshold of 600 W/m²). Flying outside that range reduces thermal contrast and increases the risk of missed or misread anomalies.
Can drone thermal inspections give false positives?
Yes. Common causes include inconsistent irradiance during capture, uneven cooling from wind, shadows from structures or vegetation, soiling, and reflections, all of which can mimic a genuine thermal anomaly if site conditions aren't recorded and factored into the analysis.
How often should solar sites undergo drone thermal inspection?
Most utility-scale sites run a baseline scan at commissioning, then a semi-annual or annual scan during operations, with additional on-demand scans after events like module cleaning, storms, or an unexplained performance dip.
Does AI improve drone thermal inspection accuracy?
Yes. AI-assisted classification applies consistent detection logic across an entire dataset instead of relying on manual, reviewer-dependent judgment, which reduces the inconsistency and fatigue-driven errors that come with reviewing large volumes of thermal imagery by hand.
An inaccurate thermal scan does not just mean a missed hotspot. It creates a chain of downstream problems.
Karthik Mekala
CMO
