Working With Downscaled Datasets

This post originally appeared on May 20, 2013 and is part of our throw-back series.

One of the projects I’m involved with for the Southeast Climate Science Center involves synthesizing the information available from several datasets created from downscaled climate change projections.  In the process of working on this project I was also asked to inventory the metadata from every downscaled dataset I could find.  From this combination of projects, it is evident and important that in any application which uses downscaled data should consider the accuracy of the data (when compared to past history) and the metadata of the dataset itself.

As far as the inventory goes, so far I’ve counted over 30 downscaled datasets that are either available now or will be available in the future.  When you think about the differences in metadata alone between all these datasets, you realize that this is incredibly daunting mountain of information.  Before I go further, here is what I consider as basic metadata of a downscaled dataset:

• Global Climate models (GCMs) used
• Downscaling Method used
• Spatial Domain / Spatial Resolution
• Time period(s) / Temporal Resolution
• Available variables (i.e. temperature, precipitation, etc.)

From here think about every downscaled dataset you may have heard of, whether created by an individual or a group project like NARCCAP.  Then realize one thing; there is no consistency between these datasets.  That is, as yet there is no set standards for some of this metadata with regards to how downscaling should be done (consistent spatial resolution, domain, time periods, etc.).  If you need to use this data and its associated metadata for your application or impact assessment, here’s the million dollar question: which downscaled dataset do you use?

As part of this synthesis project, I hosted a workshop last week which brought together climate scientists with the hydrologists, ecologists, and biologists in the Southeast CSC to discuss this question among many others.  Whether you use this or not in your work, here are a few tips from these discussions on the information a climate scientist would like you to remember even though you may never have asked for it!

1. Think about what about climate (temperature, precipitation, etc.) is most strongly related to your species or ecosystem – the downscaling method and GCMs used all have strengths and weaknesses in which variables of climate they can represent, and which ones they represent accurately.
2. How fine is fine enough? – I’ve seen several papers where a really fine spatial resolution (< 4km) of downscaled data is recommended for use in ecology.  There are several reasons why most downscaled datasets are coarser than 4km (including computing power among other things), so while a resolution that fine may be possible, think about if something a little coarser will suffice.
3. Confused? Go find a climate scientist! – At the workshop last week there were several folks who walked up and asked me if I could help them figure out what is best to use.  I highly recommend this!  If you are unsure of what to use that is best for your work, find a climate scientist and talk with them about it.  It will make your research more robust, and it will help you build a relationship with scientists that can help you in the future.  My final tip is related to all the above:
4. Don’t ask a climate scientist which dataset is best, ask which dataset is best for what you need it for! – The metadata and data differences between individual downscaled datasets make them all unique, and they are all unique because they were potentially created for very different initial applications.  As such, asking which dataset is best is like asking if a Ford Focus is better than a Toyota Tacoma.  Just like these two cars are designed for two different applications, downscaled datasets are the same way.  Discuss your application with a climate scientist first and work with them to help you find the best dataset for use with your application.

I hope these little tips of mine prove helpful the next time you need to use downscaled data.  Most important of all, come talk to a climate scientist when you need data, and don’t just go a use anything that’s available.  Given the differences in metadata and the output data between downscaled datasets, choosing a dataset to use without help can be really confusing.  Therein is a benefit of collaboration with a climate scientist before you use a downscaled dataset, we’ll help you navigate the mountain associated with these datasets.

The cartoon at the beginning is courtesy of the R-bloggers blog post on big data.