Reliable micrometeorological forcing data are essential for process-based climate impact models. Open-source flux-tower products such as ICOS and FLUXNET provide highly valuable standardized datasets for ecosystem research, but their continuous reprocessing, version updates, and multi-step gap-filling may introduce forcing-data sensitivities that are rarely quantified from a modelling perspective. In this study, we applied the subdaily version of the physically based one-dimensional soil-plant-atmosphere model BROOK90 at the spruce forest ecosystem monitoring site DE-Tha in Tharandt, Germany, to assess how dataset versioning, short-term forcing discontinuities, and different gap-filling strategies affect simulated water and energy fluxes. Successive dataset versions led to substantial version-dependent differences in historical time series of air temperature, solar radiation, precipitation, wind speed, and vapor pressure deficit, which altered simulated energy fluxes and the partitioning of evaporation components. Short-term discontinuities associated with sensor gaps and transitions between measured, MDS-filled, and reanalysis-filled values propagated into physically unlikely flux components, including negative interception and soil evaporation as well as reduced transpiration in selected periods. Controlled forcing-substitution experiments showed that model uncertainty increased with both the proportion and duration of substituted data sequences. Latent heat flux was comparatively less sensitive to ERA data substitution, whereas sensible heat flux showed substantial and systematic underestimation. Multivariate forcing substitution amplified these effects, and the partitioning of latent heat flux revealed a consistent overestimation of interception and soil evaporation associated with an underestimation of transpiration. Although BROOK90 remained numerically stable, forcing-data inconsistencies affected the simulated representation of evapotranspiration processes. These findings underline the value of explicit dataset version tracking, diagnostic checks of gap-filled forcing periods, and continued development of user guidance for reproducible and physically interpretable impact modelling.