Planning and managing an energy system in a district require a comprehensive understanding and accurate modelling of people's occupancy and circulation among multiple buildings. Due to the lack of occupancy modelling tools for district scale analysis, energy models still use simplified occupancy patterns provided in building codes and standards. However, the simplified information restricts the reflection of complex occupancy patterns driven by urban heterogeneity. This paper fills this research gap and presents a hazard-based model combined with nested copula dependence to describe the complex occupants' interactions between buildings in a district, enabling the characterisation of irregular occupancy patterns in special cases. The proposed model is calibrated using Wi-Fi authentication data from the Imperial College London (UK) South Kensington campus and is validated using the following days of the same data by evaluating the performance of predicted occupancy patterns both on average and day by day. The validation results demonstrate that the model can accurately capture the effects of the urban environment on occupancy duration and choice of transition within a district. Mean Absolute Percentage Errors (MAPEs) of average-pattern predictions are between 7% and 16% for most buildings, though a bit lower in accuracy for the Library and Food Hall predictions with MAPEs of 32%–36%. We also discuss the contributions of the proposed occupancy model to potential future applications, including efficient building space use, local energy planning and management.