I've updated the notebook, and wanted to share some results which I think are promising! Thanks for putting this code together in the first place Ian. Using the xarray datasets, it's now pretty trivial to add or remove features, so I've added in the elevation and slope angle features as well.
I've got it all running now, and my very initial results are that I found some improvements over the SCL mask from using a random forest classifier with these features. The ROC went up to 0.96 (not necessarily a good thing since this just means the result is closer to the SCL labels) but the visualisation is interesting.
I've only looked at one year of data from one test set snow patch so far, here are some examples:
@ipoole @SimonFisher92
I've updated the notebook, and wanted to share some results which I think are promising! Thanks for putting this code together in the first place Ian. Using the xarray datasets, it's now pretty trivial to add or remove features, so I've added in the elevation and slope angle features as well.
I've got it all running now, and my very initial results are that I found some improvements over the SCL mask from using a random forest classifier with these features. The ROC went up to 0.96 (not necessarily a good thing since this just means the result is closer to the SCL labels) but the visualisation is interesting.
I've only looked at one year of data from one test set snow patch so far, here are some examples: