Last-millennium energy-budget reconstruction from proxies (850–2000 CE): data and code for Stiller & Hakim (2026)

When using this dataset, please cite: Stiller, D. and Hakim, G. J. (2026). "Top-of-atmosphere radiation over the last millennium reconstructed from proxies." Journal of Climate. https://doi.org/10.1175/JCLI-D-25-0568.1 Brief dataset description: Proxy-based reconstruction over the last millennium of near-surface air temperature (SAT), sea surface temperature (SST), top-of-atmosphere (TOA) energy imbalance/net radiative flux (EEI), TOA reflected shortwave radiation (RSR), TOA outgoing longwave radiation (OLR), ocean heat content over 0–300 m (OHC300), and Arctic/Antarctic sea-ice concentration (SIC). The reconstruction uses online data assimilation with priors from linear inverse models trained on five CMIP6 past1000 simulations (the "model priors"). The assimilated proxies primarily include tree rings and corals. Surface fields validate well against instrumental observations and TOA radiation is generally skillful, but volcanic forcing is greatly underestimated and anthropogenic forcing is not included. The reconstruction is most reliable after 1600, with considerable but reduced skill over 1100–1600. The code is available through this Zenodo record or on GitHub at https://github.com/DominikStiller/energy-budget-from-proxies Temporal extent: 850–2000 CETemporal resolution: seasonal (DJF, MAM, JJA, SON)Spatial resolution: 2°×2°Ensemble size: 2000 members (400 per model prior) Caution: Most variables are provided as anomalies relative to 1961–1990, but some represent absolute values. Refer to the netCDF metadata for details. The EEI only includes natural forcing and the radiative response, but not anthropogenic forcing. Reconstruction members using the CESM2-WACCM-FV2 model prior may be less reliable (spurious Northern Hemisphere signals, excessive variability) due to an issue with the specified forcings in the CESM2 past1000 simulation. Omit these when in doubt. Antarctic sea-ice concentration with the MIROC-ES2L model prior is biased low and was removed for the analysis in Stiller & Hakim (2026). File descriptions: globalmean_ensmembers_850-2000.nc: Global means of SAT, SST, absolute EEI, RSR, OLR, OHC300. Additionally includes absolute Arctic/Antarctic sea-ice area. Full ensemble (2000 members). gridded_ensmean_850-2000.nc: Gridded (2°×2°) fields of SAT, SST, EEI, RSR, OLR, OHC300, absolute SIC. Ensemble mean only (averaged over 400 members for each model prior). The ensemble mean is the best state estimate but its variability is damped compared to true ensemble members; the ensemble members should thus be used when analyzing time variability. gridded_ensmembers_{tas,tos,eei,rsr,olr,ohc300,siconc}_850-2000.nc: Gridded (2°×2°) fields of SAT, SST, EEI, RSR, OLR, OHC300, SIC. Subset of 50 ensemble members (10 members per model prior). Use these files instead of the ensemble mean to analyze time variability (e.g., ENSO power spectrum or spatial trend uncertainty). proxies.nc: Temperature-sensitive proxy records (n = 382) assimilated in the reconstruction, compiled from the PAGES 2k database (https://doi.org/10.1038/sdata.2017.88), the CoralHydro2k database (https://doi.org/10.5194/essd-15-2081-2023; dataset: https://doi.org/10.25921/yp94-v135), and the NOAA/NCEI Paleoclimatology archives. The combined database is filtered as described in the companion paper. Proxy values are standardized (zero mean, standard deviation of one), although metadata carry the original unit. lmrecon-code.zip: Code for reconstruction and analysis. See the README inside for further details.