In modern taxi networks, large amount of real-time taxi occupancy and location data are collected from networked in-vehicle sensors. They provide knowledge of system models on passenger demand and taxi supply for efficient dispatch control and coordinating strategies. Such dispatch approaches face a new challenge: how to deal with future customer demand uncertainties while fulfilling system's performance requirements, such as balancing service across the whole city and minimizing taxis' total idle cruising distance. To address this problem, we present a novel robust optimization method for taxis dispatch problems to consider polytope model uncertainties of highly spatiotemporally correlated demand and supply models. An objective function concave over the uncertain demand parameters and convex over the variables is formulated according to the design requirements. We transform the robust optimization problem to an equivalent convex optimization form by strong duality and minimax theorem, and computational tractability is guaranteed. By Monte-Carlo simulations, we show that the robust taxi dispatch solutions in this work are less probable to get large costs compared with non-robust results.