Sensing the spatial strain field of prefabricated girder bridges remains a major challenge in structural health monitoring because sensors are typically installed on only a subset of individual girders, making it difficult to capture full-bridge field information. To address this issue, in this study, a distributed fiber-optic strain-field reconstruction method for prefabricated girder bridges that is based on a physics-guided physics-informed neural network (PINN) is proposed. The core idea is as follows. First, vehicle-borne monitoring data are used to develop a nonuniform composite Poisson traffic load model, enabling the quantitative characterization of the spatial correlation of strains among multiple girders under realistic traffic flow. This spatial correlation—which represents the lateral load distribution behavior of the bridges—is then embedded into the physics-guided PINN framework as a physics-based constraint. Consequently, the network is guided to extrapolate strain responses from instrumented girders equipped with distributed optical fibers to uninstrumented girders, thereby achieving full-bridge strain-field reconstruction. Both scaled model tests and field monitoring data demonstrated that the proposed method can effectively reconstruct physically consistent spatial strain fields, which can serve as the basis for subsequent structural condition assessment.