Natural　disasters　cause　extensive　damages　to　urban　cities,　demanding　authorities　to　take　urgent　and　effective　measures　to　restore　normalcy　on　transportation.　During　a　disaster,　restoring　road　transportation　in　a　timely　manner　for　rescue,　supply　and　also　prevent　the　risk　of　road　accidents　due　to　obstacles　is　of　vital　importance.　Traditional　post-disaster　road　obstacle　work　relies　on　a　manual　investigation　which　is　time-consuming　and　labor-intensive.　Predicting　road　risks　can　provide　decision　support　for　emergency　management　departments,　reducing　the　damage　caused　by　disasters.　In　this　paper,　we　propose　a　three-phase　framework　for　predicting　road　risks　post-disaster　leveraging　heterogeneous　urban　data.　Firstly,　We　use　a　clustering　algorithm　to　extract　and　classify　urban　road　networks　based　on　the　floating　car　data.　Then　we　extract　the　spatiotemporal　features　of　the　urban　roads.　Through　social　network　data,　we　collect　historical　risk-prone　data　using　the　crowdsensing　method.　To　address　the　challenges　of　the　small　amount　of　labeled　data,　we　train　our　model　based　on　self-training.　We　verified　the　validity　of　this　model　by　using　a　real　dataset　in　Xiamen　island　which　proves　that　our　model　accurately　predicts　road　risk　with　precision　and　recall　both　more　than　85%　respectively.　©　2022,　China　Computer　Federation　(CCF).