The study focuses on mapping spatiotemporal patterns and detecting the potential drivers of traffic congestion with multi-source data. First, based on real-time traffic data retrieved from an online map, the k-means clustering algorithm was applied to classify the spatiotemporal distribution of congested roads. Then, we applied a geographical detector (Geo-detector) to mine the potential factors for each spatiotemporal pattern. The results showed six congestion patterns for intra-regional roads and inter-regional roads on weekdays. On both intraregional and inter-regional roads, congestion density reflected by building height was the strongest indicator during the morning peak period. Public facilities such as hospitals, tourist sites and green spaces located near areas of employment or residential areas contributed to congestion during and off-peak hours. On intra-regional roads, the sparse road network and greater distance from the city center contribute to congestion during peak hours. On inter-regional roads, the number of bus stops contributed most to the early evening peak congestion, while the design of the entrances to large buildings in mixed business areas and public service areas increased the level of congestion. The results suggest that land use should be more mixed in high-density areas as this would reduce the number of trips made to the city center. However, mixed land-use planning should also be combined with a detailed design of the microenvironment to improve accessibility for different travel modes in order to increase the efficiency of traffic and reduce congestion. The innovative approach can be potentially applied in traffic congestion and land use planning studies elsewhere based on real-time multi-source data.