What Is a Quality Grade in Pavement Condition Assessment?

Surface Quality Grading (1-5 Scale) for Pavement Condition

1-5 Quality Scale Definitions

A quality grade system maps the visual condition of a pavement surface onto an ordinal scale from 1 to 5, where 1 represents an Excellent surface and 5 represents a Very Bad surface. This five-level classification follows the ordered categorical ratings found in established pavement management practice. It is a coarse visual rating, not a distress-based protocol like the Pavement Condition Index, and agencies can map its five levels to their own maintenance and rehabilitation trigger levels.

TarmacView does not produce condition grades. This page explains the concept and how agencies and researchers use it.

The five grades are usually described as follows.

Grade 1, Excellent. No visible defects of any kind. The surface texture is uniform and there is no cracking, raveling, potholes, patching, or flushing. The surface looks freshly constructed or recently rehabilitated, and joints, if present, are tight and sealed.

Grade 2, Good. Minor wear on a structurally sound surface. Very light cracking may be present but cracks are narrow (hairline, less than 1/8 inch wide). There is no spalling, no significant raveling, and no patching. Slight oxidation or color change is possible. Routine preventive maintenance such as crack sealing or fog seals is enough.

Grade 3, Fair. Visible distress is present but the surface remains serviceable. Moderate transverse, longitudinal, or block cracking with widths up to about 1/4 inch, light raveling, and patching over a small share of the surface (typically under 10%). Preservation treatments such as slurry seals, microsurfacing, or thin overlays are candidates.

Grade 4, Poor. Significant distress that affects ride quality. Moderate-to-severe cracking including alligator (fatigue) cracking in wheel paths, cracks wider than 1/4 inch with spalling at the edges, moderate to advanced raveling, patching over 10-30% of the surface, and possible rutting. The pavement needs rehabilitation such as a structural overlay or mill-and-fill.

Grade 5, Very Bad. The worst category. Extensive severe cracking, advanced alligator cracking with loose pieces, potholes, advanced raveling, patching over more than 30% of the surface or failing patches, and possibly deep rutting. The surface may pose safety concerns and needs reconstruction or major rehabilitation.

This 1-5 scale is intentionally coarser than the 0-100 continuous PCI scale. The coarser granularity matches the precision of visual rating and typical network-level decision tiers.

How Agencies Grade Sections

Agencies have long assigned condition grades by manual survey. Trained raters drive or walk the network, often at 15-25 mph, judge each section by eye or from recorded video, and record a grade or a set of distress ratings. The process is slow and subject to inter-rater variability. Even with training and certification, two raters can assign different ratings to the same section, and the FHWA Practical Guide for Quality Management of Pavement Condition Data Collection documents reproducibility as one of the hardest problems in manual surveys.

Image-based collection changes how the images are gathered, with vehicles driving at traffic speed with roof-mounted cameras, but the grade still has to be assigned, either by people looking at the images or by a model. Models are attractive because they are repeatable: the same image gives the same grade on every run.

Quality Grading as Embedding and kNN Research

One research approach avoids a traditional classification network with a softmax output layer. Each pavement image tile is passed through a vision transformer that was pretrained with self-supervised learning on a very large image collection, such as the DINO family of models. The output is an embedding vector, a compact numerical fingerprint of the visual content of the tile that captures local texture such as surface grain, crack patterns, and aggregate exposure as well as global structure such as joints and lane geometry.

The embedding is compared against a curated reference set, a collection of tile embeddings whose grades were set by expert human raters. The usual metric is cosine similarity:

cosine_similarity(A, B) = (A · B) / (||A|| × ||B||)

where A and B are the two embedding vectors. It ranges from -1 (opposite) through 0 (unrelated) to +1 (identical direction).

The k nearest neighbors of the tile in the reference set each vote for their expert grade, and the majority vote is assigned to the tile. A neighborhood of about five is a common choice. Ties can be broken by weighting each vote by its similarity. Compared with a trained classifier head, a nearest-neighbors vote has several practical benefits:

• A surface that looks like both Grade 3 and Grade 4 references gets a grade that reflects that ambiguity instead of being forced across a single decision boundary.
• The reference set can be extended with new expert-graded tiles without retraining the backbone.
• A tile that is not similar to any reference can be flagged as out-of-distribution, which signals a surface type or condition the reference set does not cover.

The embedding encoder can also be fine-tuned with supervised contrastive learning, which pulls together tiles that share a grade and pushes apart tiles that do not.

Model Accuracy

A quality grade model is evaluated with three metrics. Exact-match accuracy is the share of tiles that receive exactly the correct grade. Off-by-one accuracy is the share of predictions within one grade level of the ground truth, which is the more meaningful metric on an ordinal scale because a Grade 2 tile predicted as Grade 3 is far less consequential than one predicted as Grade 5. The Mean Absolute Error (MAE) is the average of |predicted_grade - true_grade| across all tiles. A confusion matrix shows which neighboring grades the model mixes up. Results are usually validated with k-fold cross-validation against the reference set, and inter-rater agreement between the expert graders sets an upper bound on achievable model performance.

Quality and Surface Type Independence

A grade is meant to describe condition regardless of whether the surface is asphalt, concrete, composite pavement, or a surface treatment such as chip seal. Research systems try to achieve this by pretraining the encoder on a diverse image corpus and by including every surface type in the graded reference set, so that the embedding responds to condition-relevant features such as cracks, raveling, patches, and texture loss rather than to surface color or joint pattern. Whether a given model really is surface-type independent has to be checked on held-out surfaces, because a model trained mostly on one surface type can grade another one badly.

Aggregating Tile Grades to a Section

Per-tile grades are not directly useful for management, which works on sections. Tile grades are commonly aggregated to the section level with the median, which is robust to outliers. A single tile misclassified by two grades would pull the mean but not the median, and transient artifacts such as oil drips, tire marks, puddles, or debris are filtered out in the same way.

A confidence measure can accompany the aggregated grade. The interquartile range (IQR) of the tile grades within a section shows how uniform the surface is. A section with median grade 2 and IQR of 0 is uniformly good. A section with median grade 2 and IQR of 2 has localized defects that the median alone would hide, and an individual grade 4 or 5 tile inside an otherwise good section is a hotspot worth a field look.

Quality Grade in Condition Assessment

A quality grade can serve as an input to an overall pavement condition assessment. At the simplest level, the condition of a section is its median grade, and agencies set trigger thresholds that move a section from one management category to another.

These triggers are not fixed. Agencies calibrate them to their budget cycles, performance targets, and risk tolerance. A well-funded agency might trigger rehabilitation at Grade 3, while a budget-constrained one might wait until Grade 4.

The share of the network in each grade gives a high-level health indicator, and the change in grade from one survey cycle to the next shows the deterioration rate. Sections that decline by one or more grades between cycles deserve a closer look.

Quality Grade Stability

For year-over-year pavement management, a grade has to be stable under changes that are not real changes in the pavement. Temporal stability means a section that has not deteriorated gets the same grade in consecutive surveys. Lighting stability means direct sun, overcast sky, dawn, and shade do not change the grade, which brightness, contrast, saturation, and hue augmentation during training helps to achieve. Camera stability means different sensors, lenses, resolutions, and color profiles give consistent grades, which depends on a standardized image processing pipeline. Images taken under poor lighting, such as at night or in heavy shadow, should be flagged and excluded.

Quality Grade vs PCI and Other Indices

A 1-5 grade coexists with established pavement condition indices and plays a different role.

Pavement Condition Index (PCI) is a continuous index from 0 to 100, defined by ASTM D6433 for roads and parking lots and by ASTM D5340 for airports. It is computed from a detailed distress survey. The inspector identifies each distress type present (19 for asphalt and 15 for concrete in D6433), rates its severity as low, medium, or high, quantifies its extent, and applies deduct values from the standard. PCI is the most widely used pavement condition index in North America and many agencies require it.

A quality grade correlates with PCI but does not map linearly to it. A section with extensive low-severity cracking (PCI around 45) and one with a single severe pothole but an otherwise sound surface (PCI around 50) could both be graded 3. A quality grade is a holistic visual impression, not a distress-engineering computation, and an agency that wants PCI-compatible outputs from grades has to calibrate them against PCI measured on the same sections.

Pavement Serviceability Rating (PSR) and Present Serviceability Index (PSI) date from the AASHO Road Test (1958-1960). PSR is a subjective 0-5 rating given by a panel riding the pavement, and PSI is computed from roughness, cracking, patching, and rutting. A quality grade differs in that it is based on how the surface looks, not on ride quality.

International Roughness Index (IRI) measures longitudinal profile roughness and is a functional measure of ride quality. The two are complementary but not interchangeable. A smooth but badly cracked pavement has a poor quality grade and an acceptable IRI, while a pavement with frost heave and an intact surface has the opposite. A complete condition picture needs both.

Pavement Quality Index (PQI) is a composite of IRI, rutting, and surface distress that some agencies use as a simpler alternative to PCI. A quality grade can serve as its surface-distress component.