LiDAR-sampled QA: what a flight can grade, and what it leaves for the field
A LiDAR flight measures one attribute: the distance between a return classified as conductor and a return classified as vegetation at the instant of acquisition. Over the flown corridor that is a census, so the remaining field sample exists for every defect class the sensor cannot see. A finding needs both dates on its face.
On May 20, 2020, the Big Creek 4 circuit near Springville, California was LiDAR inspected for vegetation clearance. The contractor was GIS Surveyors. The circuit runs 82.4 miles, 81.4 of them in high fire risk area.
Energy Safety's audit of Southern California Edison's substantial vegetation management work in 2020 records that flight in four sentences, as proof that SCE did the thing its wildfire mitigation plan said it would do. One day, 82.4 miles, nobody under a conductor. That is the offer LiDAR makes to a QA/QC program, and everything below is about the terms attached to it.
The flight is a census of one attribute
A field audit samples because walking costs time per span, so the cost of evidence rises with coverage. A flight inverts that. Once the aircraft is in the air over a circuit, the marginal span costs nothing, and the dataset covers the corridor end to end whether the program wanted 5 percent of it or all of it.
For the one attribute LiDAR measures, that means the audit is no longer a sample. Over the flown extent the sample size equals the population. Acceptance sampling has nothing left to do on that attribute, because every span in the corridor was measured.
The attribute is narrow, and it is worth saying exactly what it is. A point cloud yields the distance between a return the vendor classified as conductor and a return the vendor classified as vegetation, at the instant of acquisition. Every other question in a QA/QC program sits outside it.
The consequence for the rest of the program is the useful part. The field sample stops being a proxy for the clearance question and becomes the sole evidence for everything else. The lot it is drawn from, the decision rule it is accepted under, and the randomization that keeps it honest are unchanged in kind, and how a sample plan is defined and defended is the same work it always was. What changes is the defect list the sample is sized against, because one class has left it.
SCE's own plan shows the two living side by side. Quality control of completed routine work was a separate initiative from the LiDAR initiatives, and it was measured in walked miles: Energy Safety's analysis validated that 6,307 high fire risk area circuit miles of QC inspection were completed in 2020. In the same year, 45 transmission circuits comprising 1,700 miles were inspected for vegetation clearance by LiDAR. Two initiatives, two kinds of evidence, audited separately.
Fourteen percent of the points were on wire that owes no clearance
SCE's response to Energy Safety's data request DR-061 put a number on that decision:
SCE piloted LiDAR in 2020 to determine the use of LiDAR for future use related to vegetation-to-conductor clearance. Data reviewed from the pilot flight identified challenges with distinguishing conductors with varying voltage levels. Fourteen percent of LiDAR points were identified on communication lines, secondaries, guy wires, and aerial cable, for which SCE is not required to maintain radial clearance distances.
A sensor returns geometry. Voltage, ownership and obligation arrive from somewhere else. Somebody decides which returns are phase conductor and which are messenger, guy, service drop or cable television, and that decision is made in software by the firm that processed the cloud. Roughly one in seven of the points SCE reviewed named a wire that carries no radial clearance obligation at all.
The response does not say whether "points" means returns in the cloud or findings on the exception list, and the difference matters to anyone budgeting adjudication time. That ambiguity belongs to the source, and the person who can resolve it is the utility that commissioned the flight.
Either way the operational consequence is the same, and it has a direction. A secondary misread as a primary produces a false exception: an arborist opens it, looks at the circuit, closes it, and the cost is an hour. A primary misread as a communication line produces a pass that stays closed.
The expensive error is not the false exception, it is the false pass, and it is invisible by construction. The secondary conductor's absence from the clearance tables is what makes both errors possible in the first place, because the two wires look alike to a sensor and are graded differently by the rule.
The acquisition date is the audit date
A flight is an instant. What it grades is the condition of the corridor on one morning, which is the crew's work plus everything that happened to the corridor since the crew left. Attribution therefore needs two dates.
SCE said as much about its pilot:
The results did successfully identify encroachment conditions requiring mitigation when the flight timing correlated with the annual inspection and trim cycle. In 2021, LiDAR flights showed positive results that when used at times well-coordinated with inspection cycles, LiDAR can identify additional points with a high degree of accuracy.
Take a distribution unit whose crew closed it on April 9 and a flight that acquired the circuit on May 20. Forty-one days separate them. A cottonwood that was cut to clearance on April 9 and measured at less than clearance on May 20 has produced a finding that is true about May 20 and silent about the crew.
What the growth rate between the trim and the reading buys and the difference between clearance at time of trim and clearance maintained are what decide whether that finding is a defect or arithmetic.
So a LiDAR finding is gradeable against completed work only when it carries the acquisition date and the completion date of the unit on its face. A finding carrying one date belongs to the calendar rather than to a contractor. Which event the completion date actually names is the half of that pair most programs already get wrong with a tape in hand.
A flight that lands before the crew is a different product with a different buyer. It is an input to scoping, and it belongs to the pre-inspection side of the program rather than to QA/QC.
The conductor was where the weather put it
A point cloud records the wire in the position that morning's air temperature and that hour's load put it in. The criteria are written against a different position. GO 95 Rule 35 asks for the Table 1 clearances under normal conditions, Public Resources Code section 4293 asks for the clearance to hold at any position of the conductor up to 120 degrees Fahrenheit, and FAC-003-5 asks about the conductor at its Rating.
How a field reading is graded against those conditions and what conductor movement to Rating means for a transmission measurement are each settled ground.
What LiDAR adds is a second number, and the audit record has to tell them apart. Every corridor clearance report contains a measured distance at acquisition and a modeled distance at the graded condition. The second one comes out of a sag and blowout model run against conductor type, span geometry, tension and structure coordinates that the utility supplied to the vendor. It is an engineering calculation dressed as a measurement, and it is worth exactly what the parameters behind it are worth.
Three practical consequences follow. A report that labels neither distance leaves the auditor guessing, so the first question to the vendor is which one printed. The modeling assumptions are the utility's to define and to write down, because the rules leave them there. And a span that clears the number as-flown on a cool, still morning can fail once the model puts the conductor where the rule requires it to be measured.
What a point cloud cannot grade
The defect list an audit works from is longer than one distance. Set the two against each other and the shape of the residual field sample falls out:
What the audit has to grade | What settles it | Available from a flight |
Measured distance, modeled to the graded condition | Yes, subject to classification and timing | |
The prescription, and what was actually cut | No | |
The cut face, the branch collar, the stub | No | |
Brush, debris and ground disturbance at the site | No | |
Herbicide application record | The application record itself | No |
Decay, cavity, root plate, lean | No | |
The attestation and the contact record | No | |
A signature tied to its evidence | No |
The prescription row is the one that catches people, because it is the document that says what the work was supposed to be. PacifiCorp's program makes the dependency explicit, in the words California Energy Safety's Independent Evaluator recorded: "Forester signing off work as it is complete. PacifiCorp notes it reviews pre-inspections (prescriptions) at the time of conducting audits of completed work." That document lives in the work management system, out of the sensor's reach, and a clearance that passes leaves the ordered prune entirely open.
The published QC standard does not contemplate a flight
PG&E's SW-0003, Quality Control Vegetation Management for Routine Electric Distribution, revision 03, publication date May 31, 2024, is a QC standard aimed squarely at completed work. Its scope says so:
Distribution Routine VM locations in High Fire Threat District (HFTD) that have been completed by the VM Inspector (VMI), and tree crews (TC) as documented in the system of record.
Its associated tools are three: Field Maps, the Survey123 application, and measuring tools. Its first process step is Access Constraint, in which the QC Specialist determines whether the location can be reviewed at all and documents the reason when it cannot. The words LiDAR, aerial, remote, imagery, sample and percent do not appear anywhere in the document.
The absence is a gap in the mapping rather than a prohibition. Somebody still has to write down what a point-cloud measurement is worth against those acceptance criteria, and the utility that owns the standard is the only party who can do it. Which document grades your span is the question, and the answer is a numbered revision with a date on it.
One asymmetry is worth naming because it produces evidence nobody planned for. The access constraint step exists because a QC Specialist can be kept off a property. An aircraft overflies the gate. A span that closed in the field record as unable to access, behind a locked gate and a past refusal, has a clearance measurement sitting in the point cloud.
The program now holds data about a location its own standard recorded as unreviewed, and what notice and a refusal do and do not entitle a utility to do governs what happens next.
Somebody has to walk the passes
A program that verifies LiDAR findings almost always verifies the exceptions. A crew is dispatched to the flagged trees, and a tally comes back: 1,200 exceptions issued, 1,050 confirmed on the ground. That number measures how often the flight is right when it says fail.
The same tally stays silent about how often the flight is right when it says pass. The passing population is where the false negatives live, it dwarfs the exception list, and crews are dispatched elsewhere. A program in that position has tested one half of its instrument and reported the result as though it covered both.
The correction is a second, much smaller sample, drawn from the spans the flight cleared and walked with a tape. Its purpose differs from the audit sample's: it calibrates the sensor and the classifier against a field measurement. Fifty passing spans a quarter, drawn at random across vendors, circuits and crews, will tell a program more about its data than another thousand confirmed exceptions.
Two conditions make that sample worth running. The walk has to be done by somebody who did not produce the point cloud, for the same reason the grader's employer matters at all. And the disagreements have to be recorded and reconciled rather than closed, which is the same rater calibration discipline a QA program already owes applied to a machine instead of a person.
What arrives, and what has to be on the face of it
The deliverable from a corridor flight is usually three things: a classified point cloud, a table of vegetation-to-conductor distances keyed to circuit and structure, and an exception list filtered to some threshold. The cloud is archival. The table is the evidence. The exception list is somebody's decision about the table, and the threshold behind it belongs in the record.
Accuracy has to be stated, not assumed, and the public benchmark shows why. The U.S. Geological Survey's 3D Elevation Program quality levels are the specification most publicly available lidar in the United States is collected to:
Quality level | Aggregate nominal pulse density | Vertical accuracy |
QL0 | 8 or more points per square meter | 5 cm |
QL1 | 8 or more points per square meter | 10 cm |
QL2 | 2 or more points per square meter | 10 cm |
QL3 | 0.5 or more points per square meter | 20 cm |
Those are topographic mapping levels, written for bare earth rather than for conductors, and a utility corridor capture is far denser and tighter than any of them. They are here as a reminder of scale.
QL2, the 3DEP baseline, states 10 cm of vertical accuracy, which is close to four inches, against a six inch criterion that PG&E's QC standard applies to a major woody stem. A dataset can be entirely fit for the purpose it was specified for and useless for grading a span, and the only way to know is for the stated accuracy of the actual capture to be written on the finding.
Put together, a defensible LiDAR finding on one valley oak carries eleven entries:
Field on the record | Entry |
Unit and tree | Unit 4412, pole 118762, valley oak |
Unit completion date | May 6, 2026, closed by crew |
Acquisition date | May 20, 2026, corridor capture |
Stated vertical accuracy of the capture | 0.08 ft |
Measured distance at acquisition | 3.1 ft, return classified high vegetation to return classified phase conductor |
Position in the span | Center phase, 0.4 of the span south of the pole |
Conductor state at acquisition | 71 degrees Fahrenheit, 260 amps |
Modeled distance at the graded condition | 2.2 ft, normal-conditions assumption in the transmission clearance procedure |
Criterion applied | |
Adjudication | Defect, J. Ruiz, ISA Certified Arborist, June 2, 2026 |
Field verification | June 9, 2026 |
Nothing in that table comes out of the sensor except the 3.1 feet. The dates, the classification, the conductor state, the model, the criterion and the name are what turn a measurement into a finding, and a person supplies all of them. A point cloud cannot accept a span, because acceptance is an attestation by somebody who can be asked about it later.
Where the state reads the flight
No California statute or general order requires LiDAR. It enters the compliance record as a promise. SCE's 2020 plan made one:
SCE utilizes LiDAR technology to inspect select transmission and sub-transmission lines with respect to FAC 003-4, GO 95-Rule 35 and PRC section 4293, to maintain appropriate clearances between SCE's lines and vegetation.
That sentence is what Energy Safety audited against, and the determination is one sentence long:
Based on the analysis above, Energy Safety finds SCE compliant with its 2020 WMP initiative 5.3.5.8: LiDAR inspections of vegetation around transmission electric lines and equipment.
What satisfied the auditor was a data request response naming 45 circuits and 1,700 miles, plus one flight on one circuit with a date, a contractor and a mileage. Documents matching a representation, nothing more exotic. The same audit reached the distribution LiDAR initiative, where SCE had promised only that it was "evaluating whether this [LiDAR] data can be incorporated into its routine vegetation management inspection process," and found SCE compliant with that too. A modest promise is audited modestly, and a specific one is audited against the specifics.
The second reading is harder. The Independent Evaluator over PacifiCorp sets out to "verify WMP activities performed by PacifiCorp, control (QA/QC) programs," which puts the method itself inside the review rather than only its output. A LiDAR program presented as the QA/QC of completed work will be read as a control program. Control programs are judged on four things:
What the method grades
What it leaves to the field
How its own error rate is measured
Who signed the finding
The flight covers 82.4 miles before lunch. The two dates, the classification decision, the modeling assumption and the name of the arborist who adjudicated the exception are what make those miles admissible. None of them are in the sensor.
Further reading
Sampling versus full census: Defining the lot and the decision rule the residual field sample is drawn under.
Was the clearance achieved?: The distance, the point in the span, and the conditions a reading is graded against.
What a QA/QC audit covers: The defect classes, the acceptance criteria, and the evidence file behind a unit.
Frequently asked questions
Does a LiDAR flight replace the field sample in a QA/QC program?
No. It replaces the sample for one attribute only. A point cloud yields the distance between a return classified as conductor and a return classified as vegetation at the moment of acquisition. Trim type against the prescription, cut quality under ANSI A300, brush and debris, herbicide application records, decay and root condition in a hazard tree, refusals, and the signature accepting the span are all outside it. The field sample stops being a proxy for clearance and becomes the only evidence for everything else.
Can a LiDAR finding fail a span on its own?
Only if the program has written down how a point-cloud measurement maps to its acceptance criteria. PG&E's published QC standard for routine distribution vegetation management, SW-0003, revision 03 of May 31, 2024, lists its associated tools as Field Maps, the Survey123 application and measuring tools, and contains no reference to LiDAR, aerial inspection, remote sensing or imagery. A utility whose acceptance standard describes a specialist at the location has to state, in its own document, what a remote measurement is worth against that standard.
How much of a LiDAR clearance report is on wire the utility owes no clearance from?
Southern California Edison's 2020 pilot put the figure at fourteen percent. Energy Safety's audit records SCE's response: fourteen percent of LiDAR points were identified on communication lines, secondaries, guy wires and aerial cable, for which SCE is not required to maintain radial clearance distances. The response also reported challenges distinguishing conductors with varying voltage levels. That figure is SCE's pilot on SCE's system and should not be carried onto another utility's data, but the class of error is general.
Does the measured distance in a point cloud satisfy GO 95 or the MVCD?
Not by itself. The flight records the conductor where that morning's temperature and that hour's load put it. GO 95 Rule 35 requires the Table 1 clearances to hold under normal conditions, and FAC-003-5 measures against the conductor's position at its Rating. Converting as-flown geometry to the graded condition is a modeling step performed against conductor and structure parameters the utility supplied. A LiDAR clearance report carries two distances, and a report that prints one number without saying which it is cannot be graded.
Does California require a utility to use LiDAR for vegetation inspection?
No statute or general order requires it. LiDAR appears in the compliance record as a wildfire mitigation plan commitment. In its 2020 WMP, SCE identified 20 vegetation management initiatives, two of which were LiDAR inspections of vegetation around distribution lines and around transmission lines. Energy Safety audited both against what SCE had promised and found SCE compliant with each. The obligation is the representation the utility made, not the technology.
