Statistical Downscaling of Last Glacial Maximum and mid-Holocene climate simululations over the Continental United States

Abstract:

We document the creation of new high-resolution temperature and precipitation fields over the continental United States during the Last Glacial Maximum (LGM) and mid-Holocene intended for hind-casting species distributions and other biotic scenarios. Global climate simulations do not have the resolution to capture local climate variability that is needed to model ecological and biological variability. To this end, we use a recently developed statistical downscaling method, Equidistant CDF Matching (EDCDFm), developed by Li et al. (2010) [1] to create synthetic high-resolution estimates of the LGM and mid-Holocene climate over the continental United States. We find that this method works well for temperature but performs poorly for precipitation. This required processing over 1.5 billion time series. To do this, we wrote cluster-computing routines in MATLAB and implemented them on Amazon Elastic Compute Cloud.