1 INTRODUCTION Flexible pavements are sensitive to changes in cli- matic factors including temperature and moisture. Under a changing climate, temperature and moisture profiles may change gradually over a pavement’s life and thus cause additional damage to accumulate. In regions where temperature, precipitation, groundwa- ter level and the frequency of extreme weather events will likely increase significantly in the future, flexible pavements are at greater risk in terms of ad- ditional damage due to climate change (Daniel, Jacobs et al. 2014). Temperature is a significant parameter for asphalt layer modulus and pavement deterioration e.g. rut- ting and roughness (Tighe, Smith et al. 2008, Mills, Tighe et al. 2009, Li, Mills et al. 2011, Qiao, Flintsch et al. 2013). An increase in temperature can reduce the moduli of asphalt layers and the reduction can be significant at extreme high temperatures (Monismith, Secor et al. 1965, Buttlar and Roque 1996). Increases in precipitation and groundwater level may increase the moisture content of unbound materials and subgrade. Resilient moduli of the un- bound layers and subgrade can reduce significantly when the layers are exposed to high moisture content or become saturated (Hicks and Monismith 1971, Drumm, Reeves et al. 1997, Lekarp, Isacsson et al. 2000, Dawson 2009, Salour 2015). In cold climates, freeze-thaw can cause significant road damage when roads are open to traffic in the thawing period. In some regions, ordinary freeze-thaw cycles can be af- fected by climate change and, in a worst case, thaw- ing period may be extended, inducing additional road damage. For a flexible pavement, deterioration rate of rut- ting, roughness, and cracking is usually greater when layer moduli are significantly lower (Timm, Newcomb et al. 1999, AASHTO 2009). When the modulus decreases, the load spreading ability of the road decreases and stress and strain in the pavement layers will increase. Increases of strain, especially at critical positions i.e. horizontal strain under asphalt layers and vertical strain on top of subgrade, is unde- sirable, because they are critical for damage related to fatigue and rutting, respectively. Modelling responses of pavement performance to changes in climatic factors is complicated and diffi- cult. Typically, such quantitative assessments rely on comprehensive mechanistic-empirical modeling of the pavement (e.g. (Tighe, Smith et al. 2008, Li, Mills et al. 2011, Qiao, Dawson et al. 2015). A sig- nificant amount of information is needed to perform such assessment, especially at a local scale. The in- formation usually includes traffic volume, truck per- centage, pavement structure (thickness and various material parameters), and (historical and future) cli- A Method to Assess Climate Change Induced Damage on Flexible Pavements with Machine Learning Y. Qiao University of New Hampshire Y. Zhang University of Nottingham M. Elshaer University of New Hampshire J. S. Daniel University of New Hampshire ABSTRACT: Environmental conditions including temperature and moisture vary over the life of flexible pavements. Their combined effects can have long-term impacts on layer stiffness, stress and strain responses, and damage of the pavements. Climate change can change the temperature and moisture profiles of pave- ments, accelerating damage and reducing the service life. This research aims to introduce a method to assess the impacts of climate change on damage of flexible pavements. A supervised machine learning algorithm was adopted to train Artificial Neural Networks (ANN), using data including climatic factors, traffic, mainte- nance, and Falling Weight Deflectometer (FWD) back-calculated layer stiffness. The trained ANN was used to predict layer stiffness under future climate. Critical strain, the number of load repetition to failure and ac- cumulation of damage was evaluated using layered elastic analysis to determine the additional damage that can be attributed to climate change. A case study was performed based on a Long Term Pavement Perfor- mance (LTPP) road section in Minnesota to demonstrate the method. For this case study, projected changes in climate will double the speed of rutting damage accumulation and correspondingly reduces pavement service life up to 50% under 2100 climate.