avalanche advisory data questions
So I’ve recently started a personal project to apply my kung-fu data scientist skills to snow science and am wondering a few things. The overall goal of the project is to explore the use of machine learning in snow science and potentially providing new tools for people to use. Obviously a computer will never replace in person observations and analysis but I think that exploring the use of modern AI and data analysis tools could provide useful insights for snow science. I am wondering a few things that maybe people here could provide insight on – especially if you work at or have inside knowledge of any of the regional forecast organizations.
- Databases: behind the scenes are daily advisories stored in tabular format in a database and assembled automatically into the familiar paragraph form or are these hand created every day? For example, is the little pinwheel graphic that show color coded rating by aspect and above/below tree line created automatically or by hand every day?
- Do the centers consider their advisories to be proprietary information that they would not want accessible via other channels besides their web site or do they consider the info “open source” that can be repeated (with proper citation) in other places? I am thinking of why there is no centralized US (or even world) scale resource for avy advisories since they are now in (mostly) a uniform format. I am thinking of writing a package in the “R” statistical language that lets you pull and parse the daily advisory from a variety of regional centers and making that publicly available for use in research. You could do the same thing with a public database/web site of course (although I probably wouldn’t do that).
Where I am at right now I can use text mining to pull out the specifics of an advisory for the Sierra Avalanche Center from their paragraph advisory for any day in their archives and load it as categorical tabular data in R. I have messed around with pulling corresponding data from automated weather and snowpack sensors from CDEC plots and seeing if machine learning models can be built that take automated sensor data and infer daily rating. Nothing serious yet. Future goals would be to do things like see if you can actually infer snow pack structure from automated sensor data or compare the structure of models between different regions to see if the differences in the behind-the-scenes math can tell you anything about the differences between snow packs (maritime, continental, hybrid, etc).
Thoughts? Anybody know of anyone attempting something similar? This is just in my free time so will go slowly.
