Early results: as floods get rarer, rain seems to matter more
The title of this project asks what pathways lead to a compound rain-on-snow flood. A year in, there is something to report — though not yet something to conclude. Rather than assume that relationships can be governed by a few parameters — maybe that runoff responds to its drivers through a linear mean with normally distributed error, or that named copula families are enough to store the dependence — I’m letting machine learning do what it does best and find the patterns itself. The model learns the whole conditional distribution: give it a rainfall and a snowmelt, and it returns a full distribution over runoff. You can then turn the question around and ask which combinations of rain and snow most likely produced an event of a given return period.