Item

Abstract

Climate change, global warming, and increasing weather extremes, especially in Sub-Arctic and Arctic regions with unusual freeze-thaw cycles, can cause more and more challenges to the infrastructure such as road networks. The maintenance and repair of road network can be time consuming and expensive. Better targeted and proactively planned maintenance could have economical benefits and increase the safeness of the roads. To tackle this, artificial intelligence (AI) and machine learning (ML) techniques with the availability of digitalised diverse historical and real-time data, can be utilised, on one hand, to better understand the causes of the thaw damages and frost heave affecting the roads, and on the other hand, to build more advanced forecasting models for short- and long-term road conditions and thaw damage risks. In this work, as a first step, for building data-driven ML approaches to Arctic road damage forecasting, the possibilities of applying different multi-source are analysed. To this end, we are applying multi-source data sets of historical weather observations, in situ and mobile measurements of road surface and ground, and response variables of thaw damage and road wearing. As a result, we are showing 1) the benefits of different data sources using explanatory analysis, 2) the importance of different observations explaining the road conditions, and 3) the guideline of building explainable AI and ML approaches to combine digitalised information to forecast road conditions such as the thaw damage probability on road network in Northern Finland.