Fix Issues reported by Yang

[jsignell]

Working towards resolving issues. Experimenting with Canadian fire in new notebook 14_Hulls.ipynb

Things are mostly resolved. I am just trying to figure out how somethings like this can still happen. Red is first timestep (cluster hulls 2023-06-24 12:00:00), blue is second (2023-06-25 00:00:00).

Here is s zipped version of the allfires object since it takes a while to generate: allfires_20230812_PM.zip

[mccabete]

I'm gathering that you already solved this, but yes that looks like you aren't including the previous perimeter in your alpha hull calculations, so the algorithm is drawing an outline around just the new observations.

[mccabete]

I am also concerned about those little those little fire outlines that aren't getting merged into the fire and aren't changing. I took a quick look the old version and it doesn't have those.

It looked more like this:

^Those perimeters are coming from the plotting of a largefire file. Are you showing a largefire file, or are you plotting all the fires in that bounding box?

[jsignell]

I am also concerned about those little those little fire outlines that aren't getting merged into the fire and aren't changing

Yeah sorry should have specified those are the pre-merge step outlines. The merged version incorporates the little blobs into the big blob as expected.

[jsignell]

looks like you aren't including the previous perimeter in your alpha hull calculations, so the algorithm is drawing an outline around just the new observations

I have mostly solved it, but am still seeing a little strangeness particularly at that one timestep. For a given fire at t_4 that is composed of cluster_1 and cluster_2 (last seen at t_2) the intention of the v3 code is to generate the hull by:

1) calculating a hull using:

• exterior pixels for cluster_2 at t_2
• exterior pixels for cluster_1 at t_3
• all new pixels for cluster_1 at t_4 2) unioning that hull with the hull of cluster_1 at t_3

I just found a possible issue with the existing code, so I'm testing that theory, but if that doesn't work I might experiment with adding the hull of cluster_2 at t_2 as an additional input to the hull.

Having written that out I'm wondering if there shouldn't be an additional step of unioning the hull of cluster_2 at t_2.

[jsignell]

Ok things are looking good! Here is the allfires.zip and the largefire outputs for the big fire.