AbstractAccurately estimating the extent of damage after an earthquake requires labor-intensive reconnaissance surveys, which may take months to cover the entire building inventory in an impacted region. This paper provides a data-driven framework to guide a survey team efficiently through a reconnaissance mission and estimate regionwide damage by inspecting only a fraction of buildings. First, it is shown that by considering a relatively small set of representative buildings in the training data, the necessity of inspecting the entire building inventory is diminished, and accurate estimation of the regional damage is made possible within 2 weeks after the earthquake. Second, to develop a cost-effective solution, the problem of prioritizing buildings and designing efficient inspection routes is formulated as an orienteering problem. The results of the sparse field observations obtained by the end of each inspection day are used to retrain a Gaussian process regression model, which is applied to estimate damage for the uninspected buildings. A regional earthquake simulation testbed was used to validate and evaluate the performance of the proposed method.