MICE mice˙575 Dispatch: October 13, 2008 CE: AFL Journal MSP No. No. of pages: 10 PE: Edward 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 Computer-Aided Civil and Infrastructure Engineering 24 (2009) 1–10 Recognizing Patterns in Seasonal Variation of Pavement Roughness Using Minimum Message Length Inference M. Byrne Department of Civil Engineering, Monash University, Clayton Campus, Victoria, Australia D. Albrecht Clayton School of Information Technology, Monash University, Clayton Campus, Victoria, Australia & J.G. Sanjayan ∗ & J. Kodikara Department of Civil Engineering, Monash University, Clayton Campus, Victoria, Australia Abstract: Pavement roughness is a common measure of pavement condition regularly measured by road au- thorities. An approach to recognize patterns of seasonal variation in rural sealed granular pavement roughness by minimum message length (MML) inference is demon- strated in this article. MML solves two fundamental ques- tions: First, is the seasonal variation a systematic pattern or merely the result of random scatter? Second, given ev- idence of seasonal variation to what level of complex- ity should the seasonal trend be modeled? The MML technique developed does not require user input rather will identify in a quantitative and consistent manner any patterns evident in the data. The patterns identified with MML can be used to remove seasonal variation effects. The analysis utilized 104,188 roughness values obtained from a particular region in Australia over 15 years. MML inference recognized patterns of seasonal variation and demonstrated that these are not merely due to random scatter. The optimum model selected by MML inference has four separate segments of variation. These segments correspond to changes in climatic conditions that support the inference. ∗ To whom correspondence should be addressed. E-mail: jay. sanjayan@eng.monash.edu.au. 1 INTRODUCTION One of the generally accepted measures to define over- all pavement condition is road roughness. Roughness is a measure of irregularities in the longitudinal road sur- face. The advantage of roughness as the dependent vari- able is that it is an objective measure of pavement per- formance that also relates to user costs (Martin, 1998). In this article, a hypothesis that roughness measure- ments have a pattern of seasonal variation is tested with the use of an innovative inference method called mini- mum message length. Although it is common practice to repeat roughness measurements on pavement sections during the same yearly period, alternate sections of pavements are com- pared with measurements taken at various stages dur- ing the year. The true performance of alternate pave- ment sections may be masked by seasonal effects. It has already been concluded elsewhere that “interpre- tation of annual roughness measurements typical of network-level pavement monitoring must account for possible seasonal...changes in roughness” (Karamihas et al., 2000). The fundamental issue in inferring a model of sea- sonal variation in roughness data is in demonstrating C  2009 Computer-Aided Civil and Infrastructure Engineering. Published by Blackwell Publishing, 350 Main Street, Malden, MA 02148, USA, and 9600 Garsington Road, Oxford OX4 2DQ, UK.