The road to assimilating climate hindcasts into paleo-reanalyses

Analysing longer time periods than what is covered by modern instruments helps to gain further insights into the climate system, its variability and understanding of historical climate events. However, to study the period before the availability of state-of-the-art instrumental measurements, e.g., to analyse the intra-annual dynamics of past climate changes, datasets with a high temporal resolution are required. For this, paleo-reanalyses are an essential solution.

By assimilating observational, documentary and proxy climate information into atmospheric general circulation model simulations, a full 4D field of reality-informed gridded climate information is produced. However, data for assimilation becomes increasingly sparse as we move into the past, particularly during boreal winter seasons and in unpopulated regions.

Here we present a new idea for alleviating this issue: The assimilation of full-field seasonal climate hindcasts for boreal winter months, initiated during high-confidence boreal summer months of the reanalysis. These machine-learning driven hindcasts provide a multi-member output, which helps to map uncertainties accordingly.

In our presentation, we will discuss initial results regarding the skill of these hindcasts and a way forward in assimilating these data for creating a new paleo-reanalysis data set.