92 Transportation Research Record: Journal of the Transportation Research Board, No. 2524, Transportation Research Board, Washington, D.C., 2015, pp. 92–99. DOI: 10.3141/2524-09 The development of field transverse cracking prediction models is highly complicated because of several factors, including the difficulty in differ- entiating thermal cracking from reflective cracking in the field, the high variability of field conditions, and the potential variability in crack initia- tion and crack propagation mechanisms. As a result, a statistical-based approach is preferred to a mechanical-based prediction model. In this study, statistical methods, partial least squares regression, and binary logistic regression were used to establish prediction models for field trans- verse cracking. Results indicated that crack initiation and crack propaga- tion were controlled by predictor variables. Material properties (mixture creep compliance, work density, and percentage passing the No. 200 sieve), pavement structure (overlay thickness), climate (low temperature hour), and traffic (average annual daily truck traffic) were found to be key indica- tors for transverse crack propagation. Low temperature hour, percentage passing No. 200 sieve, indirect tensile strength, and service life were critical predictor variables for crack initiation. In particular, the crack initiation model, developed by the binary logistic regression, predicted the prob- ability of crack initiation. Both models show good predictability and are well validated. These models appear to work for hot-mix and warm-mix asphalt pavements. Transverse cracking is a major distress type for asphalt pavements. It can be caused by the accumulation of thermal stress either from the sudden drop of pavement temperature or from repeated daily temperature fluctuations; this process is referred to as thermal crack- ing. The thermal crack usually initiates at the pavement surface and propagates downward under additional low temperature cycles. Another type of transverse cracking, reflective cracking, is associ- ated with existing pavement cracks and joints and usually extends from the bottom (where existing cracks and joints are located) to the surface of pavements under traffic load or thermal expansion and contraction. Aging of the asphalt binder in the pavement mix- ture can accelerate both types of cracks. Transverse cracks impair smoothness and ride quality and reduce the structural integrity and load bearing capacity of pavements. They also allow water to pen- etrate into the pavement structure and further deteriorate pavement performance. Significant effort has been put forward to evaluate the transverse cracking performance of asphalt pavements and establish prediction models to correlate field transverse cracking with material or other properties. Roque and Ruth developed a physical model to predict thermal cracking on the basis of nine variables: (a) pavement tem- perature, (b) pavement geometry, (c) asphalt constant power viscos- ity, (d ) layer moduli, (e) Poisson’s ratio, ( f ) truck load, (g) coefficient of thermal contraction, (h) asphalt content, and (i) air voids (1). In SHRP Contract A-005, a performance model called the Superpave ® thermal crack model (TCMODEL) was developed to predict field thermal crack by using asphalt and asphalt mixture properties (2). The TCMODEL was further modified and recalibrated by using more field specimens and thermal crack observations and was incorporated into the 2002 Mechanistic–Empirical Pavement Design Guide program (3). Although it is one of the most popular and thoroughly validated field thermal cracking prediction models, the TCMODEL sometimes can give extremely high or low predicted results (crack length). This deviation could be related to one assumption in the model: a crack is not counted as a crack until it propagates through the entire depth of the asphalt pavement surface layer. Dave et al. made some improve- ments to the current TCMODEL by using a cohesive zone–based finite element analysis and developed the accompanying software, IlliTC (4). Despite the prediction quality, these existing models have some common deficiencies for predicting field transverse cracking, including the following: 1. Most existing models focused on only one mechanism, such as thermal cracking. However, field transverse cracking may be thermal or reflective cracking and could be caused by a combina- tion effect of thermal and traffic variables. Clear separation between these two crack types in the field is extremely difficult. 2. Most existing models targeted only one of two situations: (a) on crack initiation to predict when, or at what temperature, pavement will crack or (b) on crack propagation to estimate the amount of crack in the pavement. Very few models considered crack initiation and propagation together. In fact, it is very possible that pavement may follow different mechanisms and trends for crack initiation and propagation. Development of Predictive Models for Initiation and Propagation of Field Transverse Cracking Weiguang Zhang, Shihui Shen, Prasanta Basak, Haifang Wen, Shenghua Wu, Ahmed Faheem, and Louay N. Mohammad W. Zhang, H. Wen, and S. Wu, Department of Civil and Environmental Engineering, Washington State University, 405 Spokane Street, Sloan 101, Pullman, WA 99164. S. Shen, Division of Business and Engineering, 103B Sheetz Family Center, and P. Basak, Division of Mathematics and Natural Sciences, Depart- ment of Statistics, 149 LRC, Pennsylvania State University, Altoona, PA 16601. A. Faheem, Department of Civil and Environmental Engineering, University of Wisconsin–Platteville, Platteville, WI 53818. Current affiliation: College of Engineering, Temple University, 1947 North 12th Street, Philadel- phia, PA 19122. L. N. Mohammad, Department of Civil and Environmental Engineering, Louisiana Transportation Research Center, Louisiana State University, 3520B Patrick F. Taylor Hall, Baton Rouge, LA 70803. Corresponding author: S. Shen, szs20@psu.edu.