Grenoble, France, 2 - 5 April 2013
By Melody Sandells SSI Fellow and Research Fellow at University of Reading
New ideas, and clarification on how to quantify not only the distribution of solid material in the snow matrix, but also the voids. Also, how these might be modelled with physics.
The workshop included a field trip to see how different instruments quantify snow grain size, to understand (to some degree) how the data are processed, and how real-time field data user interfaces can be used to identify poor data.
I met a range of new people e.g. Christian Matzler, retired Godfather of snow microwave emission modelling; Charles Fierz, president elect of International Association of Cryospheric Sciences and Henning Lowe, rising star of grain-scale energy balance modelling.
At this conference, two colleagues (Nick Rutter and Richard Essery) and I sketched out a NERC proposal to develop the community microwave emission model CMEM further (details still to be nailed down), but they bought in to the concept of collaborating with a computer scientists, although neither had a clue about where to begin to initialise such a collaboration.
The focus of the workshop was to investigate the relationship between difference definitions of grain size, different techniques of measuring grain size, and how to use grain size information in different models. Possibly considered the most accurate method of quantifying grain size (specific surface area) is the gas absorption method, or ‘BET’, which stands for Brunaer, Emmet and Teller, whose 1938 paper detailed the methodology. Two facilities have the capability of doing these measurements in Europe: LGGE in Grenoble, and SLF, Switzerland. However, neither currently do so as no-one is experienced and/or willing to carry out these measurements. KIT, Japan are developing a field instrument, but it may not be ready for a formal grain size measurement intercomparison to be carried out March 2014. The next accurate method is tomography, where SLF are the only group known to have cryo-facilities, and is the only direct method of correlation length measurement, possibly the best parameter that describes microwave behaviour. Charge per slice (commercial rates) is EUR1, and small samples of a few cm are possible. Optical methods come next, either relying on the spectral reflectance, or reflectance at a single wavelength and also give specific surface area. However, these methods are sensitive to illumination conditions, so these need to be controlled carefully. The resolution of these instruments is of the order of a few cm, so cannot resolve mm scale thin layers in the snow. The final method presented was the snow micropenetrometer, a rod that slides down through the snow and measures the force required to break snow crystal bonds, which has been empirically related to the correlation length.
Part of the problem is that the microwave emission and snow energy balance models all tend to have different representation of snow crystals, and snow crystal growth. Energy balance models tend to have an effective optical grain diameter, possibly with some form of shape taken into account. Provenance is an issue, with an unknown origin to some of the equations behind MOSES, the precursor to the widely used community land surface model JULES. Different representations of the grain size in the microwave model include correlation length, maximum extent and ‘stickiness factor’. It is common to scale grain sizes in all these models, and it is not clear how these parameters relate to each other. A sub-working group was formed as an outcome of this meeting, and I will be co-ordinating it. There will be a workshop next summer on snow-microwave interactions, and I would be keen to have some focus on code sustainability. Colleagues were intrigued by the notion of the ‘bus factor’, and one colleague commented that Seymour Laxon (killed in a tragic incident on New Year’s Day) was the only person who knew the intricacies of the Cryosat algorithms. This, I do not believe, as the European Space Agency document their code very well, and involve computer scientist partners. No satellite is launched without the Algorithm Theoretical Baseline Document in place, but it is a perception, and is very poignant with another tragic, accidental loss of a UCL Cryosat team member since this meeting, who was herself trying to deal with the fallout from the first death.
Although this is not a leading community in terms of code sustainability, I think the conditions may be right for it to become one. People are usually more than happy to share code on an ‘ask me and I’ll email it to you’ basis, the only model that I know that was deliberately put out into the community (in this particular research field) was the DMRT-ML microwave emission model. The author, Ghislain Picard, spent a week developing the model (he claims!), but a lot longer when he later came to tidy and document the code for his release. He wanted to do this for the community to draw a line under the research (this I took to mean to stop him from fiddling with it). Its release was by word of mouth through selected email distribution. To my knowledge there has been no formal release under git hub or similar, but it can be downloaded .
There was some talk about a data repository, where people could download data to test their algorithms. There was also a mention that we are an open-source community, where we are trying to solve these issues in an open, collaborative way. However, there was no mention of code sharing or a code repository, and indeed it seems that one person was using their code in a circular way i.e. use a reflectance model to retrieve grain size, then using a similar model to predict reflectance for the retrieved grain sizes for comparison with the observations. Unsurprisingly, the model did a good job. So it seems we have some way to go. On the microwave side, at least we have all models available to us if we ask for them, but we need to understand how the different parameters relate to each other. On a personal note, I’ve been offered code to run ‘JIM’, a stripped-down version of the community land surface model JULES. It has already been used for Ensemble Kalman Filter data assimilation of snow mass, and I intend to incorporate it into the particle filter data assimilation of microwave radiance. As a direct outcome of the future snow-microwave interaction meeting, I would like to see a central resource for all these models and data. I would also like to increase community awareness of the resources that are out there, and to tweak the way we conduct the science. To this end, I am planning on submitting a Town Hall session at the American Geophysical Union Fall Meeting to discuss collaborative research and code sustainability.