Performance Overview

We compared the performance of different subsets of LTPP data, cross-classifying on dimensions expected to correlate with different long-term pavement performance trends. These dimensions include: region, KESAL (traffic loading in thousands of equivalent single-axle loads), and pavement subgrade type (treated or untreated). For each dimension, we examine the performance of Asphalt Concrete (AC), Jointed Plain Concrete Pavement (JPCP), Jointed Reinforced Concrete Pavement (JRCP) and Continuously Reinforced Concrete Pavement (CRCP) separately.

The LTPP program is organized geographically into four administrative regions: North Atlantic (NA), North Central (NC), Western (WR), and Southern (SR). This cross-classification was performed as a first-order approach to examining interactions between geographical region, pavement type, and roughness.

Traffic load is a major factor in pavement degradation. Hence, we cross-classify sections by traffic load in thousands of single-axle equivalent loads per year to see if there is any discernible correlation between loading, pavement type, and roughness over time.

The third cross-classification involves categories based on whether the sections were constructed on a treated or untreated subgrade. Treated subgrade types in the database include bituminous treated bases, non-bituminous treated bases, and a catchall treated subgrade type. Untreated subgrade types include granular bases, unbound bases, and a catchall untreated subgrade type.

Results by Region

To evaluate the effects of traffic loading we grouped the sections into quartiles based on their mean yearly ESAL loading, with the first quartile representing the lowest loading scenarios and the fourth quartile representing the highest loading scenarios.
Test Subset
  • AC sections in Western (WR) and North Atlantic (NA) states saw maintenance occur sooner than in the Southern (SR) regions of the country, with p = 0.03 and p=0.02 respectively.
  • The initial roughness of AC sections is lower (p < 0.001) but increases faster (p < 0.001) than for CRCP sections (Figures 1 & 2).
  • AC sections in the NA and WR saw the first intervention, at 10.8 and 11.8 years respectively, compared to the SR and NC regions where the maintenance occurred at the 15.2 and 15.5 year marks respectively. (NA versus NC, p = 0.01; NA versus SR, p = 0.002; NC versus WR, p = 0.02; SR versus WR, p = 0.005).
Figure 1: Regional performance profile for AC in the test dataset.
Figure 2: Regional performance profile for CRCP.
LTPP Dataset
  • AC sections in the NA and WR saw the first intervention, at 10.8 and 11.8 years respectively, compared to the SR and NC regions where the maintenance occurred at the 15.2 and 15.5 year marks respectively. (NA versus NC, p = 0.01; NA versus SR, p = 0.002; NC versus WR, p = 0.02; SR versus WR, p = 0.005).
  • Figure 3, illustrates the initial IRI and IRI at first intervention for AC, JPCP, JRCP and CRCP sections. AC sections had a statistically significant lower initial IRI akkcross all regions (with each comparison meeting the criteria of p < 0.05 for a Students’s t-test), while having a statistically significant higher rate of change over time compared to all the other concrete pavement types across all regions (each comparison meeting the criteria of p < 0.05 for a Students’s t-test). CRCP pavements tended to have the slowest rate of IRI change.
  • Fit statistics for each of the sections for each region were developed. Figure 4, illustrates the metrics developed for AC sections in the SR. Similar plots were developed for each pavement type in each region. However, they are not being presented on this page, in order to preserve focus on broad trends.
Figure 3: Initial IRI and IRI at first intervention by region (IRI in m/km)
Figure 4: IRI metric distributions for AC sections in the southern region
Figure 5: Regional performance profile for AC using full LTPP dataset.

Traffic Load

To evaluate the effects of traffic loading we grouped the sections into quartiles based on their mean yearly ESAL loading, with the first quartile representing the lowest loading scenarios and the fourth quartile representing the highest loading scenarios.
Test Subset
  • AC is used more in low-traffic scenarios, with only four AC sections found in the fourth (e.g. highest loading) ESAL quartile. CRCP and JPCP are used more in high-traffic scenarios. (No hypothesis test performed.)
  • There is no evidence in the test dataset that IRI performance over time is sensitive to loading within pavement groups. However, there is evidence to suggest that JPCP may undergo earlier maintenance in the second (p = 0.035) and third (p = 0.049) loading quartiles respectively.
  • AC undergoes repair earlier than JPCP in the first loading quartile (p = 0.005) and earlier than both CRCP (p = 0.006) and JPCP (p = 0.034) in the second loading quartile.
  • The roughness of CRCP sections increases at a slower rate in the third and fourth loading quartiles than both AC (p = 0.009, p < 0.001) and JPCP (p = 0.012, p = 0.009).
LTPP Dataset
  • There is no evidence that IRI performance over time is sensitive to different levels of loading. Of course, it is worth considering that this is a conditional independence given that the designs of pavements catering to different loads may vary significantly.
  • Consistent with trends from the test data set, AC undergoes repair earlier (difference of approximately 7 years on average) than JPCP in the first, third and forth loading quartiles (p = 0.001, p = 0.03 and p = 0.002) and earlier than both CRCP (p = 0.02) in the forth quartile.
  • Figure 6, illustrates the initial IRI and IRI at first intervention for AC, JPCP, JRCP and CRCP sections. AC sections had a statistically significant lower initial IRI than the concrete pavement types in the third loading quartile (with each comparison meeting the criteria of p < 0.05 for a Students’s t-test), while having a generally higher rate of change over time compared to all the other concrete pavements (statistically significant in loading quartiles 3 and4). CRCP pavements tended to have the slowest rate of IRI change.
Figure 6: Comparison of Initial IRI and IRI at First Intervention by Traffic Quartiles (IRI in m/km)

Subgrade Treatment

Test Set

  • AC has lower initial IRI with an untreated base (p = 0.018).
  • CRCP is not sensitive to base type – both treated and untreated bases perform equally.

LTPP Dataset

  • AC has lower initial IRI for treated bases than for untreated bases (p = 0.03).
  • None of the concrete pavements showed any significant sensitivity to the type of base.
  • For sections with both treated and untreated bases AC had significantly lower initial IRI compared to all the concrete pavements.
  • For sections with treated bases AC had significantly higher time rate change of IRI compared to all the concrete pavements. Figure 7, illustrates a comparison of each pavement type’s initial IRI and IRI at first intervention for treated and untreated bases.
Figure 7:Comparison of Initial IRI and IRI at First Intervention by base treatement(IRI in m/km)

Summary

Generalizability

Caution must be exercised when generalizing the results of LTPP data analysis. As the above results illustrate, the choice of the data subset can lead to different answers to the same question. For example, an analysis of performance of AC sections in the NA region, for the test subset and the entire LTPP dataset, produce different trends. Similarly, analysis of the initial IRI performance for AC on treated or untreated bases is different depending on whether the test dataset is used, or the entire LTPP dataset is used. Cross-classification of the data also reveals differences in trends dependent on the choice of the data subset.

These inconsistencies illustrate that the results of performance analysis have very narrow applicability to the context of the dataset used. On the same note, when a particular trend is repeatedly and consistently observed across multiple contexts, its scope of application becomes broader and therefore more general. For example, across all datasets analyzed, it was observed that the initial IRI for AC sections tend to be statistically significantly lower than concrete sections, while the time rate change of IRI for concrete sections also tend to be statistically significantly lower than those for AC sections, resulting in longer times to first intervention for concrete sections when compared to AC sections.

No Free Lunch
When broad trends can be observed across multiple datasets they often highlight trade-offs in performance. For example, there is evidence of tradeoffs between initial roughness (favoring AC) and the rate of change of roughness over time (favoring CRCP). Identifying such trade-offs should be the objective of analyzing historical datasets as they help decision-makers develop deeper insights regarding their networks and the maximize the level of service they can provide. It also veers the conversation away from biasing the pavement selection process by presenting a partial view of the data analysis.

Context is King
The above two points segue into the third and final takeaway that emphasizes the need for context sensitive decision-making. Analysis of LTPP data to support decision-making should be encouraged, however, data subsets for analysis should be constructed to accurately reflect the context (location, use, functional class) of the decision question. As a corollary, the results of the analysis should narrowly apply to the scope of the decision question at hand, and not be generalized to apply to other derivative decisions.