More is not always better: downscaling climate model outputs from 30 to 5-minute resolution has minimal impact on coherence with Late Quaternary proxies

Abstract

Abstract. Both proxies and models provide key resources to explore how palaeoenvironmental changes may have impacted diverse biotic communities and cultural processes. Whilst proxies provide the gold standard in reconstructing the local environment, they only provide point estimates for a limited number of locations; on the other hand, models have the potential to afford more extensive and standardised geographic coverage. A key decision when using model outputs is the appropriate geographic resolution to adopt; models are coarse scale, in the order of several arc degrees, and so their outputs are usually downscaled to a higher resolution. Most publicly available model time-series have been downscaled to 30 or 60 arc-minutes, but it is unclear whether such resolution is sufficient, or whether this may homogenise environments and mask the spatial variability that is often the primary subject of analysis. Here, we explore the impact of further downscaling model outputs from 30 to 5 arc-minutes using the delta method, which uses the difference between past and present model data sets to increase spatial resolution of simulations, in order to determine to what extent further downscaling captures climatic trends at the site-level, through direct comparison with proxy reconstructions. We use the output from the HadCM3 Global Circulation model for annual temperature, mean temperature of the warmest quarter, and annual precipitation, which we evaluated against a large empirical dataset of pollen-based reconstructions from across the Northern Hemisphere. Our results demonstrate that, overall, models tend to provide broadly similar accounts of past climate to that obtained from proxy reconstructions, with coherence tending to decline with age. However, our results imply that downscaling to a very fine scale has minimal to no effect on the coherence of model data with pollen records. Optimal spatial resolution is therefore likely to be highly dependent on specific research contexts and questions, with careful consideration required regarding the trade-off between highlighting local-scale variation and increasing potential error.