Confirm validity of impacts

[rod-glover]

It is surprising how similar the impacts results all are, across regions and and across time periods. I felt suspicious, so I diffed a bunch of the files the nginx-p2a-impacts server is serving, and found that indeed, at least in the sample I "randomly" chose, this seems to be the case. I still feel suspicious, but it will take an expert to evaluate the answers we are displaying.

This is not really a frontend issue.

[rod-glover]

@jameshiebert and @corviday , this issue may or may not have been resolved by the work recently done on validating and recomputing backend data. Any opinions?

Something that would interest me would be a human-readable comparison of the impacts across regions and time periods. Depending on how much time I have, I may do that for myself. I have a pretty shrewd idea of how to do it using the P2A frontend infrastructure. OTOH, if this is already being contemplated in some form, let me know.

[corviday]

It is definitely the case that we had - and fixed - a bug that made results for some rules identical for all regions. We had a bug where in order to determine whether a watershed was snow-dominated or rain dominated, the rules engine looked at the highest temperature for the entire year, instead of the highest temperature for winter.

Previously, all area had true values for the rules affected by the bug:

• rule_3c-i-hydro (snow-dominated watershed convert to rain-dominated watershed)
• rule_3c-ii-hydro (decline in snow melt)
• rule_3a-i-for (shorter winter logging season)

We now have more diversity in results (between one third and two thirds true) for those three rules.

---

I have not done any other plausibility checks on the rules results as a group, though I don't think it would be hard to do so.

[jameshiebert]

I think that our scientific reviewers are more qualified to answer this question. As Lee said, the issue that was raised has been fixed. If our scientific reviews haven't raised any other issues, it's reasonable to take the rule output at face value.