Multivariate bias correction of ERA5 using in-situ observations for planning and engineering

Abstract Climate risk analyses for infrastructure and energy systems typically require a point-based frame of reference. However, the surface weather data informing these decisions often originate from gridded reanalysis products like ERA5, which represent area-averaged estimates rather than conditions at specific point locations. This representativeness discrepancy, alongside structural modeling errors and data assimilation biases, can lead to systematic discrepancies in the representation of both mean climate and extremes, reducing the utility of gridded data for applications demanding local accuracy. To bridge this gap, I developed SCOPE-ERA5 (Station-Calibrated Outputs for Planning & Engineering), a post-processed daily dataset spanning 1979–2024 at 7115 global weather stations. Using the Multivariate Bias Correction in N-dimensions algorithm trained on in-situ observations (MBCn; Cannon (2018 Clim. Dyn. 50 31–49)), I adjusted both marginal distributions and inter-variable dependencies to produce physically consistent time series of daily mean, minimum, and maximum temperature; relative humidity; 10 m surface wind speed; and surface pressure. The adjustment eliminates mean bias across all stations and reduces root mean square error at 52%–96% of sites, with the most significant improvements observed for wind speed. Out-of-sample testing suggests the transfer function is temporally robust, with only modest performance degradation outside the training period. Compared to raw ERA5, SCOPE-ERA5 provides a higher-fidelity representation of local near-surface conditions and serves as a temporally complete alternative to fragmented station records. It is well-suited for applications requiring long, multivariate climate time series, such as infrastructure design, energy system modeling, and climate model bias adjustment.