Load tests are one of the most powerful tools for bridge assessment, revealing true system behavior, calibrating models, and building public confidence. However, temperature and thermal gradient effects, especially in sub-arctic climates, pose a significant challenge, as they can dominate the structural response and make performance assessments harder to interpret. This paper addresses this challenge by analyzing a unique dataset from a prestressed concrete bridge, comprising five years of multi-sensor SHM data and a final, controlled load test before its demolition. We present a climate-aware, hybrid-modelling workflow that separates temperature-driven responses from vehicle-induced effects. By systematically normalizing the predicted responses to a defined reference climate, the methodology demonstrates tangible benefits, as reflected in discernible variations in owner-facing performance metrics. Furthermore, an “error vs https://www.w3.org/1998/Math/MathML" display="inline">ΔT ” analysis is proposed to highlight the relationship between thermal gradients and prediction accuracy, offering additional insights into the robustness and applicability of the climate-conditioned model.