Understanding users of level-1 satellite data

FIDUCEO aims to bring the rigour of metrology to the field of Earth Observation, in particular where time series are long enough to be relevant to climate.  It is therefore very much a project focussed on the needs of its users.  But who are those users of climate-relevant Earth observations from space — and what do they need?  Those are the questions the user requirements survey sets out to answer.

My own experience with reading measurements from the Advanced Very High Resolution Radiometer (AVHRR), from the days that I was a PhD student, taught me two things: firstly, that reading and understanding meteorological satellite data is not necessarily easy.  Secondly, that using level-1 data (i.e.\ calibrated radiances) is not as typical as I thought.  I ended up writing my own routine for reading and calibrating the data.

The FIDUCEO User Requirement's Survey aims to learn where level-1 users are, and what experiences and needs they have.  The survey consists of a part with general questions, and of a part with specific questions for each sensor series used.  Some of the topics addressed are:

  • Experiences with obtaining the data
  • Experiences with reading the data
  • How they inform their users of uncertainties in the level-2 data they produce
  • ... and other topics

So far, I have interviewed ten users based in six countries on two continents.  Some are based at universities, others at meteorological institutes or national research laboratories, others yet are employed at private companies.  We aim to interview around twenty people in total, but halfway through we already have a wealth of valuable information.

Defining user requirements

There are many different ways that CDR producers think about CDR users and their requirements.  Pragmatically, user requirements tend to be adapted to be close to what can be achieved — or, to put it differently: CDR producers mostly know the users whose needs they may be able to address.  A breakthrough in what can be achieved could then lead to reaching new users that are not part of the existing community.

A common source of user requirements comes from the Global Climate Observing System (GCOS), a project by the World Meteorological Organisation and others to provide a full description of the systematic observation needs for climate.  However, few CDR producers take the GCOS recommendations as they are.  Some consider the GCOS requirements unfeasible and use more relaxed criteria, whereas others ore told by their users that the ambition level set by GCOS is insufficient.  For many, GCOS is just a starting point, or one among several sources of user requirement documents.

Dealing with uncertainty

CDR producers also differ in their level of ambition as to what uncertainty estimates they provide to their users.  With some exceptions, the level-1 data that they rely on provides very little uncertainty information (or none at all), which renders formal error propagation impossible.

Despite this, a majority of CDR producers still aim to provide some uncertainty estimates. All the groups that I have spoken to so far include flags to warn their users of doubtful pixels, and most also include a characterisation of the total uncertainty and information about stability.  Despite the absence of uncertainty information in the level-1 data, about half do provide a pixel-level uncertainty estimate to their users. 

Even where the level-1 data does contain uncertainty information, CDR producers typically do not trust those estimates as they are, but rather take additional steps to verify provided estimates and calculate their own.

Downloading and using level-1 data

User experiences of downloading and reading level-1 data vary considerably.  Some users, in particular those at national meteorological institutes, are close to the data and so both data and reading routines may be already available to them, making it easy to start using the data straight away. Others need to obtain data more or less easily from an archive, with an interface that might not be designed to transfer the complete set of measurements, which may be many terabytes large. 

User experiences of reading the data range from very easy (reading routines made by someone else) to very difficult (requiring a dedicated effort to write non-trivial software to read the data and apply calibration coefficients, in some cases with incomplete or insufficient documentation).

Although not everybody needs it, an effort to make this process easier is generally appreciated.

… any other topics?

From the entire survey, the single question whose responses may ultimately lead to the most discussion is: Did you encounter any problems with the data?" 

For some users, the reply begins with: "Oh yes, where do I start?" while others answer: "No, not really" - even when both are using the same sensor series.  That is less strange than it may seem: some problems are only apparent with a close look at the data.  If you don't carefully look orbit by orbit, scanline by scanline, problems may go unnoticed and even enter gridded monthly means (known as level 3 data), propagating all the way through — but a problem that is clear when looking at individual scanlines may not be clear when looking at a gridded product.  And a problem that is clear and perhaps expected for one user, might not be obvious for another.

As one user put it, we need a complete and thorough description of the instrument event history.

Next Steps

Within the next weeks, I will be interviewing around ten more level-1 users to learn about their experiences and needs.  Subsequently, we will write up a document summarising the results. This document will inform work in FIDUCEO.

If you wish to participate in the survey or be informed when the results are available, please contact me (Gerrit Holl) at g.holl@reading.ac.uk.

Add new comment