Large spectral differences in paleo data assimilation reconstructions

Climate variability exists at all timescales. However, most paleo data assimilation (PDA) reconstructions are validated against relatively short instrumental measurements. This validation process fails to reveal PDA variability information at long timescales. Here, we compare the variability over the Common Era of near surface air temperature, hydroclimate, and sea level pressure of different PDA reconstructions: 1) the Last Millennium Reanalysis (LMR) project, 2) the Paleo Hydrodynamics Data Assimilation product (PHYDA), and 3) a global monthly paleoreanalysis of the modern era (ModE-RA). Although all PDA reconstructions use a similar version of the ensemble Kalman filter, they differ in important methodological choices. We assess differences in loss of variance, reconstruction uncertainty, spectral features, and the power-scaling exponent ß, globally and regionally. For clarity, higher ß-values would indicate a stronger dominance of longer timescales in the power spectrum. We find that ModE-RA has the largest uncertainty levels and the lowest ß-values across all climate indices globally and regionally; it also shows, in most cases, the largest changes in uncertainty and the greatest loss of variability. Furthermore, reconstructions that differ only by prior models show systematically higher or lower ß-values, globally and regionally, across all climate indices. Indeed, when reconstructions differ, they tend to differ systematically, having universally higher or lower ß-values. Differences in spectral power and ß-values between offline and online reconstructions are surprisingly small but vary across regions. Our results show that different, reasonable methodological choices substantially affect the variability of PDA reconstructions.