Cellular automaton (CA) is a widely-used tool for modeling land-use changes that can enhance our understanding of past and future urban development. Previous studies have demonstrated that patch-based CA models are superior to cell-based CA because the spatial homogeneity of urban growth at local scales can be considered. However, there still exists a major limitation that traditional patch-based CA did not incorporate any information about landscape pattern into land-use change modeling. To alleviate this issue, we present a novel landscape-driven patch-based CA (LP-CA) model that can simultaneously consider landscape similarity and cell-by-cell agreement. We have examined this new model by simulating and predicting the urban expansion in a fast-growing city. Results indicate that our method performs better than stand-alone traditional patch-based CA and landscape-driven cell-based CA in terms of the combined error. The capability to characterize and replicate landscape pattern is greatly important for urban planners to explore the potential influences of urban expansion under different land-use planning scenarios. Therefore, this proposed model has the potential to provide valuable support for urban land-use planning and policy-making processes. •A landscape-driven patch-based CA (LP-CA) is proposed for urban growth modeling.•Landscape similarity and cell-based agreement are considered during CA calibration.•LP-CA outperforms traditional patch-based CA and landscape-driven cell-based CA.•LP-CA has the capability to characterize and replicate landscape pattern.•This capability is useful for analyzing potential impact of future urban expansion.