Scenario-Conditioned Actual Meteorological Years (sAMY): A Stochastic Weather Generator Using Multi-Decadal Observations

Building energy simulation (BES) requires hourly, point-scale weather files, yet climate models typically provide daily outputs at coarse spatial resolutions ( ∼ 25–200 km). Existing workflows, including TMY-based, morphing, and precomputed time-slice approaches, address parts of this gap but typically remain tied to transformed baseline years or discrete representative periods, limiting continuous multi-year or probabilistic assessment. This paper presents the scenario-conditioned Actual Meteorological Year (sAMY) framework: an observation-based statistical weather generator that explicitly learns hourly covariance structures from multi-decadal station observations (20–68 years) and transfers this structure to future periods conditioned on daily climate projections. The core innovation is the separation of observation-trained hourly structure from climate-signal forcing through regime-switching vector autoregression, with solar radiation anchored to satellite-derived observations (NSRDB, 4 km) before applying climate change signals. Validation against independent observations (2011–2020) demonstrates strong performance, including a 79% reduction in temperature RMSE (0.79 ∘ C) and a 64% reduction in solar radiation bias relative to the Annex 80 benchmark used here. Multi-city testing confirms robustness across Mediterranean, tropical, and continental climates, while multi-model testing across CORDEX and CMIP6 shows limited sensitivity to the forcing source: validation metrics change by only 11–26% despite 4–8 times coarsening of climate model input. By explicitly modeling the resolution gap through statistical learning, sAMY enables generation of continuous 90-year stochastic ensembles (2011–2100), facilitating multi-year extreme event assessment using publicly available data. The framework provides a reproducible pathway for climate-informed weather files that preserve observational realism and climate signal fidelity.