The Invisible Elephant in the Room

 Mar 7, 2016    by Adrienne Wootten

A few weeks ago I had the opportunity to speak with Dr. Peter Thorne when he visited the Southeast CSC and the NCSU Department of Marine, Earth, and Atmospheric Sciences.  Dr. Thorne is one of the lead authors of Chapter 2 of the National Climate Assessment (“Our Changing Climate”) and a lead section author for the IPCC Fifth Assessment Report.  During his time visiting with us, Dr. Thorne discussed much of the challenges of working with observed data and particularly with station data.  Without working with station data, as I and other climatologists do, you may not realize some simple, subtle, but very important issues that can happen in a station’s record.  The urban heat island, station moves, instrument failure, and measurement error are a short list of things that can affect the records of a station that mask the influence of climate change at any individual station.  Discussing these with Dr. Thorne a few weeks ago reminded me of two particular problems associated with station data in recent history: missing data and quality control.

As time rolls on, the atmospheric science world has moved more toward automated weather stations.  The abundance of data that results from an increasing number of automated stations and accessibility leaves questions on the best way to assess when a station may have suffered instrument failure or measurement errors.  That is, what is the best way to check the quality of the data measured by a station?  In terms of how to check data quality, there are numerous algorithms developed and published in literature for this, but there is one thing that has bothered me about this.  As more stations have become automated, so too has the quality control for the station data.  What will happen to the station data that many researchers and decision makers use should those who maintain the station data rely only on automated quality control algorithms?  There are many instances where something indicated to be poor quality data is actually accurate, or values that the algorithms indicate to be correct are actually wrong.  In those cases, human review is required.  As such there is no way to entirely remove the need for humans to review station measurements.

Missing data is an issue before you even get to handling quality control.  Missing data is something that all climatologists understand happens regardless of how good a station is.  Interestingly, people that use the station data can be inclined to think that missing data doesn’t happens, or that we have a consistent set of ways to estimate missing data.  Some of my colleagues put in a proposal a few years ago to find the best techniques to estimate missing data with regards to use in crop modeling.  The proposal was ultimately rejected, but the comment from one of the reviewers was “Surely the atmospherics scientists have solved this problem by now!”  For as many techniques as there are to estimate missing data, the reply to that comment is no, we don’t have that figured out yet!  A technique that works well to estimate missing data in one part of the country is not the best in other parts of the country.  What works well for one parameter doesn’t work well for another parameter (precipitation vs. temperature is a good example of that).  The challenge of missing data comes before quality control and well before using the data for any analysis of relationships between climate and any other sector.  Missing data and quality control is the invisible elephant in the room; the atmospheric scientists see it all the time.  Therefore all I have to say is before you use station data (or any climate dataset really) for an impact assessment or to research the relationship between local climate and something in ecology or another sector, make sure you understand the caveats with that dataset and the quality of the data.  I really recommend it; otherwise that invisible elephant in the room is going to hit you hard later!

This post originally appeared in April 2013 and is part of our throwback series.

Adrienne Wootten is a PhD Candidate at North Carolina State University.

NOTE: Comments will be visible to the public. Before commenting for the first time, please review the ECCF's Editorial Policy.

Add comment

Log in to post comments