A hierarchy of hierarchies

Collected posts on merging datasets.

In the last post, I got as many ensemble members as I feel I’m likely to get any time soon. I reordered the hierarchy a little bit to accommodate the datasets I have. The branching is based on similarity of SST datasets. It looks like this.

Alternative mergings have been suggested which would group the datasets by land surface air temperature datasets (LSAT) or by interpolation method. Here’s my first go at grouping by LSAT.

There’s one more primary branch – Berkeley Earth is separated from HadCRUT now as they have their own LSAT analysis, but GloSAT is grouped with HadCRUT because they share CRUTEM (different versions, at least for the moment). I grouped reanalyses together again.

I’m a little more stumped for interpolation. I broke it down like this initially, but NOAAv6 is partly pattern-based and partly neural network based for now. GISTEMP is kriging like but also uses spatial patterns for SSTs. HadCRUT and Berkeley Earth use patterns of a sort. This is not a well-defined hierarchy.

So, there we go: a hierarchy of hierarchies for generating an ensemble of ensembles of ensembles.

#climate #climateChange #globalMeanTemperature