Cluster analysis is a useful tool to explore underlying structures of a stochastic process and relations between observations by grouping items into same categories according to their similarity. Of particular interest is to cluster spatial-temporal units with elevated risks with accounting for potential risk factors and spatial-temporal correlation simultaneously. A motivation for this paper is from the study and surveillance of dengue fever (DF) infection in Taiwan. Since there is no effective vaccine or specific medicine to treat the dengue infection, a surveillance system, which can map clusters of cases, identify virus serotypes, and evaluate impact of environmental factors, is thus essential to prevent DF epidemic. We first develop an integrated cluster-temporal model and related parameter estimation. Then, an iterative procedure for identification of spatial-temporal clusters is devised. We adapt a deviance criteria for model selection such that the identified clusters can reflect different diffusion patterns. The proposed method is applied to the DF data of Taiwan for illustration.