AbstractPerformance forecasts of highway bridge networks are informative inputs for stakeholders for cost-effective decision-making of funds allocation and asset management. Developing performance forecast models requires establishing a target performance measure (PM) and developing deterioration models accordingly. In the United States, the Federal Highway Administration (FHWA) and state departments of transportation use FHWA’s national PMs to assess highway bridge performance. Incorporating the national PMs for forecast modeling is the focus of this paper. Existing approaches commonly utilize stochastic processes for deterioration modeling, introducing uncertainty into forecasts and creating challenges for decision makers to capture the complete information. Numerical estimation through Monte Carlo simulations is achievable but inefficient. This paper presents an application of the generalized Poisson–binomial distribution as an efficient statistical tool with closed-form solutions to characterize uncertain network PMs based on stochastic forecasts of individual bridges within a network. The study demonstrated and validated the modeling tool with a case study on highway bridges in the United States using historical data (1993–2019). The results showed that this modeling tool achieved a robust and low prediction error. Further, the study demonstrated the feasibility to infer maintenance, repair, and rehabilitation (MR&R) effects when specific MR&R data were not available.Practical ApplicationsThis study demonstrated and validated an application of the generalized Poisson–binomial distribution for bridge network performance forecasts in terms of FHWA bridge performance measures. The study presented how to develop probabilistic forecasts of a network or a group of bridges. The primary advantage of this study is the theoretical formulations that characterize the network forecasts’ statistics. For practical applications, selections of bridges in a network can vary among multiple users. The theoretical formulation features near-real-time computations. For applications that are sensitive to computations, e.g., online analysis tools, the presented study can be a well-suited candidate. The study also showed the capability of modeling bridge MR&R effects in network performances through a calibration process. The capability can advocate potential applications for developing bridge asset management systems that consider bridge MR&R plaining. The third potential use of the study is to develop a fast or rough estimation of network forecasts with budget constraints. However, readers should refer to the “Limitation” section before starting their own applications.