By Tanaya Gondhalekar, Deborah Khider & Julien Emile-Geay
Climate field reconstruction is the task of estimating variations in one or more climate fields (e.g. surface temperature or precipitation) from a collection of paleoclimate observations (a.k.a "proxies"). Many statistical methods are available for doing so; a relatively new and impactful one has been offline data assimilation, as implemented in the Last Millennium Reanalysis (see Hakim et al. (2016), Tardif et al. (2019)). Part of PReSto's mission is to democratize these tools and enable a wider variety of actors, from seasoned researchers to students or citizen scientists, to generate their own reconstructions based on available code and data.
The purpose of this repository is to showcase how to use tools from the LinkedEarth Python research ecosystem (and broader scientific Python stack) to expand on the Last Millennium Reanalysis (version 2.1) which used the offline data assimilation method of Hakim et al. (2016) together with the PAGES 2k database (PAGES 2k Consortium, 2017). Namely, we showcase the addition of the Iso2k, and CoralHydro2k databases.
This PaleoBook is an update on, and continuation of, Reproducing LMRv2.1 with the cfr package, which is referenced in some chapters.
The reconstruction workflow is broken down into 3 major steps, some of which have variants:
- Data assembly: gathering, selection and cleaning
- Data assimilation, which blends proxy observations with calibration data and the model prior
- Validation and comparison to other relevant reconstructions
Here we offer two different ways to carry out Step 1, all of which result in a netCDF file that can be used in Step 2:
- Step 1a illustrates how to generate the
ProxyDatabaseobject from the LiPDVerse. - Step 1b illustrates how to generate the
ProxyDatabaseobject from the LiPD Graph. Both notebooks make heavy use of the pyLipd package, whose usage is explained in the pyLipd tutorials.
Step 2 can be common to all workflows, depending on the provided metadata. In this PaleoBook we showcase one way, but another (using class-based seasonality) can be found in here
- Step 2 uns the DA with metadata-based seasonality. Once again, only one instance is illustrated in this notebook.
Step 3 focuses on validating both our results from Step 2, as well as comparing different proxy databases from Step 1 for forensics purposes. This is specific to PReSto2k, which are the newer reconstructions based on updates to PAGES 2k, and validates the three reconstructions run with this proxy data, using very similar methodologies to earlier chapters.