I assume that everyone has seen the post on our website discussing the changes that NSIDC has instituted to make our sea ice data available again. I don’t want to repeat that, but I thought I would respond to some of the more general issues that came up in Anthony’s posts and accompanying comments. I thank Anthony for giving me this opportunity. I write here from my personal viewpoint and not in an official capacity as a representative of NSIDC or the University of Colorado.
I apologize for the error in our data and for the relative slowness in responding to it. I’m glad that so many people are interested in the data and I understand that seeing errors is frustrating and can undermine confidence in the data. Anthony is correct that many people do now pay close attention to our website and we do have a responsibility to attend to errors as fast as we can. We will try to do better in the future. There are two major points that I hope everyone can take away from this event:
(1) The error in no changed any of our conclusions about the long-term changes in Arctic sea ice. The ice extent is declining significantly and the ice is thinning.
(2) Errors like the one that occurred are part of the normal course of dealing with satellite data. We hope that they are rare, but they are not unexpected.
On the first point, there is no doubt; it is verified by numerous independent observations and is well-discussed in numerous places, including in the entries on our analysis web page.
On the second point, I think it is worth providing some background on satellite data and how it is processed, stored, and used by scientists, including those at NSIDC. In doing so, I’m not making excuses for the error in NSIDC’s data, but I hope I can help people understand how such errors are part of the scientific process of quality controlling and fine-tuning data and techniques to ultimately provide the best information possible to track climate change.
Climate science is focused on understanding long-term changes and the mechanisms that drive them. In terms of satellite data, this means taking great care and making the data as good as it can possibly be. The focus is on assuring a time series good enough to track potentially subtle trends. This involves careful quality control of data and developing and fine-tuning algorithms to convert raw satellite data into a useful climate parameter (such as sea ice extent). Like all of science this has traditionally been done slowly, methodically, and privately. And up until about ten years ago, there was no other choice but to move slowly because of severe constraints on computer processing speeds, limited data storage capacities, and difficulties in simply sharing data. One of the earliest papers to note the long-term decline in Arctic sea ice was published in 1999 (Parkinson et al., J. Geophysical Research); it was based on data only through 1996. It simply took that long to collect and carefully analyze the data, make sure algorithms were robust and stable, and get a paper through scientific peer-review.
Data distribution was also limited because of similar computational, storage, and distribution constraints. For example, NSIDC used to received updates every five years or so of final quality-controlled sea ice products. We would then distribute the data by mail on CD-ROM only to registered users.
Immediate data analysis was solely the province of operational centers, like the National Weather Service, who had special access to near-real-time data. Their focus was on getting only what was needed of any data before moving on to the next analysis or forecast cycle. Quality control was focused on catching major errors; smaller errors that didn’t significantly impact a short-term analysis were not caught or were ignored. There was no consideration given to the long-term context of the data, which were often not even saved.
There was a very clear delineation between science and operations.
Science is still done slowly and methodically, with final results disseminated the way they always have been – through peer-reviewed scientific journals. It still takes time to do final quality control on climate products. NSIDC now receives final sea ice data about once a year. But in the past ten years or so, access to data has changed dramatically. Computer processing power and data storage capacities have increased exponentially and high-speed internet has allowed near instantaneous distribution of data to a broad community. Satellite data that used to require days or weeks of processing and required dozens of tapes or CDs to store can now be processed in minutes, stored on a portable hard drive or even a memory stick, and distributed over the internet. This has been a boon to scientists who now have much faster and easier access to large amounts of data.
At the same time algorithms have matured and become more stable. This means that significant adjustments to the algorithms are not regularly needed and they can be run confidently on near-real-time data, with the understanding that the results may change during final quality-control. This has allowed to NSIDC implement a near-real-time version of the sea ice data. For the past several years this data has been freely distributed online for anyone who wished to use it, though the primary audience was scientists who would be familiar with associated caveats of using near-real-time data.
In this context, let me now move on to NSIDC and its Arctic Sea Ice News and Analysis web site. NSIDC is a science institution. Our mission is science and science support, not operational support for any kind of critical operational decisions (e.g., what regions are free enough of sea ice to be safely navigated). Because we must focus on science, the resources we can devote to near-real-time data production and analysis are limited. Nonetheless, as climate change became an important topic, it was clear that Arctic sea ice was a leading indicator of the observed changes. Since NSIDC stores and distributes the sea ice data, many people started to come to NSIDC scientists to ask about sea ice conditions and the implications for the climate. When 2005 set a record low summer extent, there was a lot of media attention; in response we issued a press release. Through summer 2006 we received many requests asking about how the ice was looking, both from the media and fellow scientists. As the summer wore on it started to feel a bit like being on a family road trip and having the kids in the back continually asking “are we there yet?” As summer 2007 started, it was a clear that a new record low was quite possible. The questions began again in earnest.
In the sense that science ultimately serves society, it was becoming apparent that scientists and the public were coming to expect a near-real-time analysis of Arctic sea ice conditions. In response, we decided to develop the website so that we could post occasional data updates and science-based discussion of the conditions. This worked quite well, but the summer of 2007 was so remarkable and Arctic sea ice had become such a big story both scientifically and in the public consciousness that we realized there would be the expectation to do even more during 2008. In response, in addition to our occasional summer posts of data and analysis, we decided to provide daily data updates and at least monthly analyses throughout the year. This decision was possible only because the products are mature and stable and further quality control to produce final data results in only minor changes. This was an added burden on NSIDC resources, but with automated processing the day-to-day impacts could be managed.
This all evolved in an ad hoc manner, with improvements made when we had resources available. Remember, none of this is NSIDC’s primary mission, which is to archive hundreds of cryospheric datasets and support peer-reviewed research. The sea ice analysis website is one of dozens of research and data management projects at NSIDC. People working on the web site often have to fit it in where and when they can amid other duties. There is no single person at NSIDC who works only or even primarily on the sea ice analysis page. This is not an ideal situation, but it is the only way we can support the analysis while still fulfilling all of our responsibilities.
This is one reason why we appeared slow to address the error last week. We have a group at NSIDC whose responsibility is to respond to user questions and comments on any of our hundreds of datasets. NSIDC’s standard is to provide a response to user inquiries within 24 hours during the business week. This is very quick for a science institution and NSIDC’s user services works very hard to always meet that standard. However, it is not particularly fast compared an operational center that works on a 24/7 schedule. We will work to put into place better QC measures and more streamlined procedures to catch future errors more quickly, but we simply do not have the resources to work in an operational environment.
This of course begs the question: why don’t operational centers do this instead of NSIDC? Operational centers do indeed provide near-real-time sea ice data. However, I believe there are a couple reasons why operational centers are not poised to provide the kind of science-based support found at NSIDC. First, their only priority is on supporting critical users with the most useful operational information about sea ice – e.g., ships sailing in and near ice-infested waters; their data is not well-suited for easy understanding by a general audience. Second, operational centers are focused on near-real-time support, not on climate issues. Thus their expertise in putting near-real-time data in the context of long-term climate is limited.
NSIDC and other climate data/research centers (e.g., NASA GISS) do have that expertise. And that is crucial. It is only through evaluation of the near-real-time data in the context of the long-term climate that one can properly assess the relevance to climate change. This mixture of climate science and near-real-time data analysis is perhaps not optimal, but I think it is worthwhile.
The easy access to climate data has been a boon for scientists and I would argue it has also been a great benefit for society. Science ultimately serves society and quick and easy access to data provides quality up-to-date information on important issues, such as climate change. The problem is that such data can come to be viewed by journalists and other members of the public as completely routine and reliable. When small changes or errors occur, they may be given greater import than they deserve in terms of what they imply about climate change. This means there is a responsibility for places like NSIDC distributing data to thoroughly explain the data and respond quickly to any issues. I believe NSIDC does an excellent job in explaining the data through considerable documentation on all aspects of the sea ice data. However, in terms of responding to data issues, NSIDC and like centers have been slow to realize that the audience for such data has expanded beyond fellow scientists and informed journalists and that any issues need to be addressed as soon as possible lest they confuse or mislead the public. This is a difficult task for places like NSIDC, whose resources are limited and whose primary mission is not operational support. The recent data error has been a learning experience for those of us at NSIDC and we will try to do better.
I hope that this information gives people a greater appreciation for the hard-work done by my colleagues at NSIDC and an understanding of the difficulties inherent in supporting near-real-time data with limited resources amid myriad other responsibilities. Finally, I hope that people come away with a better sense of what goes into analyzing satellite data and how such data is so beneficial to our understanding of climate. Thank you.