Condition-based pavement management systems accounting for model uncertainty and facility heterogeneity with belief updates

•Jointly optimizing inspection, maintenance, and reconstruction for pavement systems.•The framework is defined as a Partially Observable Markov Decision Process.•A bottom-up approach is proposed to consider facility-specific heterogeneity.•We use a statistical learning method for belief updates to reduce model uncertainty.•Condition-based strategies are found and validated through a numerical example.