Temperature converges, precipitation diverges: a systematic evaluation of three Quaternary paleoclimate reconstructions over the last 800,000 years

Alexander Gamisch 

Aim
Spatially explicit paleoclimate reconstructions are widely used in macroecology, biogeography, and archaeology to infer past species ranges, refugia, and biodiversity dynamics. Three products, Oscillayers, Krapp et al. (2021), and PALEO-PGEM-Series, provide global bioclimatic variables across the Quaternary, yet rest on fundamentally different methodological foundations, and their mutual consistency remains unquantified.
Methods
This study provides the first joint comparison of all three reconstructions on a common bioclimatic grid across 79 time slices spanning 20–800 ka BP, focusing on mean annual temperature (BIO1) and total annual precipitation (BIO12). Agreement was assessed via raw-value and anomaly correlations, partial correlations controlling for latitude, root mean squared error, threshold-based agreement rates, and validation against 19 independent terrestrial proxy archives, including the first continuous time-series validation of Oscillayers. Inter-dataset consistency rather than absolute accuracy is quantified; proxy archives serve as the single external benchmark.
Main conclusions
The results reveal a clear contrast between variables. For temperature, all dataset pairs showed high spatial correlations (r = 0.98–0.99) and consistently positive cell-wise temporal correlations across virtually the entire terrestrial realm, with disagreement concentrated in high-latitude regions (northern North America, Siberia). For precipitation, spatial patterns remained correlated (r = 0.94–0.95), but anomaly correlations were near zero (r = 0.05–0.19), with large areas of negative temporal correlation across the Amazon Basin, Congo Basin, and South and Southeast Asia. Proxy validation yielded comparable performance across all three datasets (mean |r| ≈ 0.50). The three reconstructions are largely interchangeable for studies centred on broad-scale temperature gradients, but should not be treated as equivalent for analyses requiring past precipitation histories, particularly in tropical and monsoonal regions. Dataset choice should be treated as explicit uncertainty rather than an arbitrary background decision.