The emergence of social media provides a rich source of posts to facilitate situational awareness and management during crisis events. Since interactions within social media users are frequently used to establish social networks, previous studies have explored how social media contributes to the information diffusion about crisis events. However, most studies have concentrated on the dynamics of the online social network, whereas the information diffused within social space and physical world during different event evolution stages tends to be overlooked. To address this problem, this study introduces a methodology for modeling and analyzing the information diffusion of crisis events based on the retweet patterns of social media users at city level. The proposed methodology consists of two parts: 1) Fisher ordinal clustering method is applied to divide the evolution stages of crisis events based on social media user activities. 2) To explore the information diffusion of crisis events in the physical world, the social-physical network (SPnet) is constructed by extending the retweet network to the geographical space for each evolution stage. The Weibo data related to the Kunming terrorist attack in 2014 are analyzed to demonstrate the methodology. By identifying the geo-location of users (90.8% of the total active users), two types of SPnet are constructed based on the retweet patterns of ordinary users and media users, respectively. The results show that the information diffusion in the SPnet of ordinary users are mainly from Kunming, while in the SPnet of media users, major developed cities are the key nodes for information diffusion. This study provides useful insights for understanding how human activities react to crisis events through the social-physical network at city level.