This study proposes an innovative approach employing physics-informed neural networks (PINNs) to reconstruct airflow patterns around buildings while concurrently assessing wind pressure exerted on their surfaces, leveraging sparsely recorded airflow velocity data in wind tunnel experiments. To validate the effectiveness of the proposed method, a numerical experiment was conducted focusing on the flow inside and outside of a two-dimensional urban street canyon with an aspect ratio of one. High-resolution data about airflow and pressure distributions obtained from large-eddy simulation (LES) served as the foundation for generating simulated average flow velocity measurements at select sensor locations. We devised a PINN model to infer average flow velocity and associated parameters at any designated point within the target domain. Findings showed that feasible reconstruction of the average velocity distribution at an elevated resolution was achieved by spacing sensor measurements at a resolution of approximately H/6 (H: the canyon height), yielding a root-mean-square error of 0.023Uref ( \(U_{ref}\) : the mean flow velocity along the principal flow direction at a height of 2H). Furthermore, successful estimation of average pressure exerted on building surfaces was achieved, demonstrating a correlation coefficient of 0.73 compared to LES-derived data. The proposed method exhibited promising potential in facilitating the efficient assessment of average flow distributions and building surface pressure within wind tunnel experimentation settings.