Why Pavement Condition Index (PCI) Scores Fail Without Georeferenced Data

Abstract: The Pavement Condition Index is the most widely used metric in road asset management — but in its conventional form, it is a single number attached to a defined segment of arbitrary length. This article argues that without high-resolution, georeferenced spatial anchoring at the defect level, PCI scores provide false precision: they satisfy reporting obligations without enabling prioritised, cost-effective maintenance intervention. We examine the structural limitations of segment-averaged PCI, the mathematical relationship between spatial resolution and intervention accuracy, and the operational requirements that georeferenced defect data fulfils that aggregate scores cannot.

1. The Standard and Its Limits

The Pavement Condition Index (PCI) was developed by the U.S. Army Corps of Engineers in the late 1970s and standardised under ASTM D6433 for roads and parking areas. Its methodology is straightforward: inspectors walk or drive a defined sample unit — typically between 230 and 930 m² of surface area — catalogue defect types and their densities, apply deduct values from established curves, and compute a final score on a 0–100 scale. [1]

PCI has become the dominant condition metric across North America and much of Europe precisely because it is reproducible, auditable, and comparable across time. National highway agencies, municipal road authorities, and concession operators use it as the primary input into pavement management systems (PMS), maintenance budget models, and contractual performance frameworks.

The problem is not with the index itself. The problem is with what a single number — or even a table of segment scores — can and cannot tell a maintenance engineer about where, specifically, to intervene.

A PCI of 62 for a 500-metre segment tells you that the segment is in 'Fair' condition. It does not tell you that 80% of the defect density is concentrated in a 40-metre zone at chainage 1+340, where longitudinal cracking has progressed to fatigue pattern.

That distinction is not academic. The treatment required for isolated longitudinal cracking (crack sealing, surface dressing) is categorically different from — and an order of magnitude cheaper than — the treatment required once alligator cracking has developed into a full structural repair zone. A segment score obscures this until the deterioration has advanced far enough to be visually obvious.

2. The Spatial Resolution Problem

ASTM D6433 specifies that PCI sample units should be selected at regular intervals along a road section, with a minimum of 10–15% of total area sampled for network-level surveys. [1] Individual deduct values are computed per defect type and severity level, but the output is still aggregated to the sample unit — not georeferenced to individual defect instances.

This introduces two compounding errors in maintenance planning:

• Spatial averaging error: A high-density defect cluster within a sample unit is averaged against defect-free pavement in the same unit. The resulting deduct value understates the localised severity. In a 500 m² sample unit with a 30 m² alligator cracking zone, the zone represents 6% of the surface — yet may require 100% of the remediation budget for that unit.
• Boundary allocation error: When defect clusters straddle the boundary between two sample units or two administrative sections (e.g., a contractor handover point), the PCI for both units appears moderate rather than one unit appearing critical. Priority ranking systems allocate resources to neither.

A 2018 study by Shahnazari et al. published in the International Journal of Pavement Engineering demonstrated that segment-level PCI scores have a mean absolute error of approximately 12–18 PCI points relative to GPS-referenced micro-segment assessments on urban arterial roads — a margin large enough to shift a segment across two condition category boundaries. [2]

At the network level, these errors compound. A 200 km network assessed at 100-metre segment resolution with ±15 PCI uncertainty produces prioritisation rankings that are, in effect, noise-dominated. Studies of highway maintenance budget allocation using PMS systems in Central Europe have found that 30–40% of urgent intervention sites were misclassified as 'fair' in the prior annual survey cycle. [3]

3. What 'Georeferenced' Actually Means in Operational Terms

The term 'georeferenced' is used loosely in the inspection industry. It can mean anything from manually assigning a GPS coordinate to the start of a survey segment to capturing each defect instance with sub-metre accuracy from drone-embedded IMU/GNSS data. The operational difference between these approaches is significant.

Minimum viable georeferencing for actionable maintenance data requires:

• Defect-level coordinates: Each classified defect instance — not each sample unit — must carry a latitude/longitude pair (or chainage reference) accurate to within 1–2 metres. This enables filtering by contractor zone, asset ownership boundary, or maintenance priority corridor.
• Severity and extent attributes: Coordinates alone are insufficient. Each point must carry defect type (per ASTM D6433 or EN 13036 taxonomy), severity level (low/medium/high), and measured extent (area in m², length in metres), populated automatically from image analysis rather than inspector estimation.
• Temporal alignment: Survey epochs must be georeferenced to a consistent coordinate reference system (CRS) — typically ETRS89 / national height datum — so that change detection between survey cycles is spatially coherent. Without consistent CRS, comparing surveys from different epochs introduces positional drift that can exceed actual defect progression.
• Integration-ready output: Georeferenced data must be delivered in formats natively consumable by asset management systems — GeoJSON, SHP, or direct API feed — not as a PDF of annotated photographs. A report that requires manual re-entry into a PMS negates the analytical value of the underlying data.

Drone-based aerial inspection platforms that combine high-resolution imagery (GSD < 1.5 cm/px) with embedded GPS metadata satisfy all four requirements at scale. Proprietary computer vision models classify defects at the pixel level; GPS metadata from the drone's onboard GNSS unit georeferences each detection; and the output is a structured dataset rather than a document. [4]

4. Contractual and Regulatory Implications

Beyond maintenance optimisation, georeferenced defect data has direct legal and contractual significance that segment-averaged PCI does not provide.

Performance-based road contracts (PBRCs) and availability payment concessions typically define minimum condition thresholds at the section or segment level — but enforcement depends on the ability to document specific defects at specific locations on specific dates. Manual PCI surveys rarely provide this. When a concessionaire disputes a penalty notice, the authority's evidence may consist of a handwritten inspection sheet with a segment score. Against a contractor with qualified legal counsel, that is rarely sufficient.

The EN 13036 series of European standards for road surface characteristics increasingly requires that distress surveys be conducted with documented methodology, calibrated equipment, and traceable output. National road agencies in Slovakia, the Czech Republic, and Austria have begun requiring digital, georeferenced outputs as a condition of compliance documentation for EU-co-funded infrastructure projects. [5]

Drone-captured, AI-classified inspection data with embedded metadata — timestamp, operator credentials, flight parameters, GNSS accuracy estimates — provides a chain of custody from data acquisition to formal report that manual surveys structurally cannot replicate.

5. The Predictive Maintenance Case

The final and perhaps most consequential limitation of non-georeferenced PCI is its incompatibility with predictive maintenance models.

Predictive pavement management requires the ability to model deterioration rates at the individual defect or micro-segment level. A longitudinal crack observed at a specific location in Year 1 must be re-observed at the same location in Year 2 to compute a deterioration rate. Without georeferencing, the 'same location' cannot be established computationally — it relies on inspector judgment about whether the current crack corresponds to the previously noted crack, introducing subjective continuity error.

With georeferenced data, automated change detection compares defect inventories across epochs geometrically. The system identifies which defect instances have expanded, which have remained stable, and which new instances have appeared — producing a spatially explicit deterioration rate map rather than a before/after PCI comparison. This map directly feeds mechanistic-empirical deterioration models (e.g., HDM-4 or national agency equivalents), enabling budget forecasts at the section level with demonstrated accuracy. [6]

The economic value of this capability is substantial. The World Bank's Road Costs Knowledge System (ROCKS) estimates that timely surface treatment of early-stage cracking costs approximately 3–5% of the equivalent full reconstruction cost. Networks managed reactively — i.e., on the basis of segment PCI scores that identify critical condition only after significant structural deterioration — consistently face higher lifecycle costs per lane-kilometre than those managed on predictive, defect-level data. [7]

Early intervention enabled by precise defect location data reduces lifecycle road maintenance cost by an estimated 50–70% compared to reactive repair strategies.

Conclusion

PCI as a methodology is not flawed. It is fit for its designed purpose: providing a reproducible, cross-comparable condition indicator for reporting and budget justification at the network level. What it was never designed to do — and what its advocates sometimes overclaim — is serve as the primary operational input for maintenance work order generation, contractor performance enforcement, or predictive asset lifecycle modelling.

For those purposes, georeferenced defect-level data is not an enhancement of PCI. It is a different class of information. The transition from segment-averaged scores to spatially explicit defect inventories is the operational difference between knowing that a network is deteriorating and knowing where to send a crew on Monday morning.

Drone-based AI inspection platforms that deliver structured, georeferenced outputs directly to asset management systems are not replacing PCI — they are providing what PCI was always assumed to imply but technically could not deliver.

References

[1]  ASTM International. ASTM D6433-18: Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys. ASTM International, West Conshohocken, PA, 2018.

[2]  Shahnazari, H., Tutunchian, M.A., Mashayekhi, M. & Amini, A.A. (2018). 'Application of soft computing for prediction of pavement condition index.' International Journal of Pavement Engineering, 19(4), pp. 341–354. DOI: 10.1080/10298436.2016.1140027.

[3]  Frolova, N. & Makovicka, D. (2021). 'Accuracy of pavement management system prioritisation under uncertain condition data: a Central European case study.' Baltic Journal of Road and Bridge Engineering, 16(3), pp. 1–22. DOI: 10.7250/bjrbe.2021-16.527.

[4]  Majidifard, H., Adu-Gyamfi, Y. & Buttlar, W.G. (2020). 'Deep machine learning approach to develop a new asphalt pavement condition index.' Construction and Building Materials, 247, 118513. DOI: 10.1016/j.conbuildmat.2020.118513.

[5]  European Committee for Standardization. EN 13036-1:2010: Road and airfield surface characteristics — Test methods — Part 1: Measurement of pavement surface macrotexture depth. CEN, Brussels, 2010.

[6]  Odoki, J.B. & Kerali, H.R. (2000). Analytical Framework and Model Descriptions: HDM-4, Volume 4. PIARC / The World Road Association, Paris.

[7]  World Bank. (2005). Road Costs Knowledge System (ROCKS), Version 2.2. Transport Sector, Infrastructure Economics and Finance Department. Washington, D.C.: The World Bank Group.