Averaging of profiles and retrieval of distributions

A data exploration tool for averaging and accessing large data sets of snow stratigraphy profiles useful for avalanche forecasting

This vignette creates a link between the methods described in the paper

Herla, F., Haegeli, P., and Mair, P.: A data exploration tool for averaging and accessing large data sets of snow stratigraphy profiles useful for avalanche forecasting, The Cryosphere, https://doi.org/10.5194/tc-2022-29, accepted.

and this companion R package. While the basic workflow of matching layers between pairs of snow stratigraphy profiles and thereby aligning them is described in the vignette Basic workflow, this vignette describes how larger groups of profiles can be aggregated into a representative summary profile (average profile) and how underlying distributions of layer or profile properties can be queried through the average profile.

#> Loading required package: sarp.snowprofile

To demonstrate how useful aggregations and distributions of snow profiles can be computed with our R package, we use two very simplistic and brief example data sets of snow profiles: SPgroup2 and SPspacetime. While the results of these simplified examples will not look overly interesting, these examples teach you how to apply our tools to your own more challenging data sets.

As detailed in the reference paper, this package includes two different averaging algorithms for snow profiles, one that can be applied to any group of profiles (default algorithm) and one that can be applied to seasonal data sets of profiles (timeseries algorithm).

1. Default algorithm

1.1 Averaging of profiles

The default averaging algorithm can be accessed conveniently through the function averageSP and is implemented exactly as described in the reference paper. You can also consult the documentation (?averageSP) for more details.

## compute the average profile (here with default settings)
avgSP <- averageSP(SPgroup2)
## plot profile set and resulting average profile
plot(SPgroup2, SortMethod = 'hs', xticklabels = 'originalIndices')

Note: To speed up the computation in this vignette, the above figure was not computed with default settings in the background, but with a limited number of initial conditions. The displayed result might not represent the optimal average profile, so recompute it yourself with the default settings displayed above, or experiment with other settings!

It might be worthwhile to reflect on which layers you consider weak layers. With the function argument classifyPWLs you can make the averaging algorithm consider your perspective on weak layers and make them more likely to get included in the resulting profile. One approach (the default) could be to consider all surface/depth hoar layers as weak layers. Another approach could be to consider all persistent grain types that have one particular stability index above or below a certain threshold. Or you can also consider multiple different stability indices. Make sure to consult the documentation for some more details and tips.

1.2 Retrieval of distributions: Backtracking of layers

Since all layers of all individual profiles are matched against the layers of the average profile, we can backtrack each layer of the average profile to obtain all underlying layers that were matched against it. This allows us to compute various layer distributions or profile distributions as visualized in Figure 2 of the reference paper: