Clustering Pavement Condition Data Based on Spatiotemporal Patterns

In pavement management, classic statistical methods are commonly used for filtering pavement condition data for modeling purposes. In these methods, condition data are grouped into broad groups (e.g., pavement type, traffic loading level, climatic zone, etc.). Outlier values within each group are detected based on various statistical criteria, such as points deviating by more than two or three standard deviations from the mean. However, pavement condition is influenced by local factors, such as drainage, subgrade, and construction and repair history. Also, the condition data for sections in close proximity are normally collected by the same inspector and equipment. This paper describes a new concept for clustering pavement condition data that takes into consideration the geographic proximity of pavement sections. This proximity-based approach allows for detecting extreme observations or outliers in the condition data of neighboring pavement sections. 2013 American Society of Civil Engineers.

Clustering Pavement Condition Data Based on Spatiotemporal Patterns Conference Paper