Assessment of 21st-century planetary albedo changes in CMIP6 models

Planetary albedo is a key parameter of the planet’s energy balance, and its sustained decline contributes appreciably to present-day warming. Using CERES EBAF Ed4.2.1 satellite data and the ERA5 reanalysis for March 2000 – December 2014, we evaluate the representation of planetary albedo by 51 CMIP6 models. According to CERES, global albedo declined by (1.20 ± 0.61)×10⁻³ per decade, of which (0.85 ± 0.48)×10⁻³ occurred under clear-sky conditions; the ERA5 trend is statistically indistinguishable from zero. The models systematically overestimate albedo (by 1.3% on average) and underestimate the observed darkening: over 2000–2014 the difference remains within internal variability, but over the extended 2000–2025 record it becomes statistically significant (−(1.05 ± 0.15) versus −(2.46 ± 0.33)×10⁻³ per decade). An objective selection procedure based on median-normalized climatological metrics and model genealogy identifies a nine-model ensemble that reduces the mean-albedo bias to + 0.1% while preserving the trend magnitude. In the selected ensemble’s projections, albedo decline continues under all ScenarioMIP scenarios: by 2091–2100, global albedo falls 0.0046–0.0116 below its 2015–2024 level, equivalent to an additional ~ 1.6–3.9 W m⁻² of absorbed solar radiation; the largest changes occur at polar latitudes, and only under SSP1-2.6 does the decline level off. Over 2015–2025 the scenarios remain statistically indistinguishable in albedo, while the observed decline lies at the lower edge of the ensemble spread: only two of the 36 scenario members darken at least as fast as observed. A faithful representation of cloud and cryospheric processes remains the primary prerequisite for reliable projections of Earth's radiation budget.