Closed tomr-stargazer closed 9 years ago
tomr-stargazer
commented
9 years ago the actionable thing:
What you did is ok provided you adopted a masking level which is 5 times the noise at |b|>1; in that case you will miss some weaker real emission in the plane, but that's probably ok.
tomr-stargazer
commented
9 years ago Here's what I did:
I re-made the hybrid "interpolated-then-momentmasked" cube for the cube where instead of choosing the noise in the |b|<1 region, I chose twice that noise (assuming that the interpolated data outside that have 4 pixels for every 1 pixel inside that, and taking sqrt(4) = 2) and used that as the basis for my moment masking algorithm (which then takes 5 times that noise level for the masking).
I created it and now it's called DHT36_Quad4_interp_mom.fits and it's used in make_fourthquad_stub.py
TMD: Why don't you use the moment masked survey in the 4th quadrant, as you do in the other 3 quadrants?
TSR: Essentially what I am doing here is creating a moment-masked cube of my own, from the interpolated 4th quadrant survey with higher latitude coverage. The reason has to do with the limited latitude coverage of the 4th quadrant moment-masked survey, as-provided. It only spans -1<b<1, which makes it much different from the otherwise analogous 1st quadrant survey (spanning roughly -4<b<4). From the point of view of a dendrogram analysis, I think it's much better to preserve the (low-resolution) interpolated data at the high latitudes rather than throw that data away, since it is difficult to reason about how to convert emission into clouds when the emission straddles an edge of the data and is "cut off".
To do the above, I actually implemented the moment-masking algorithm in my own Python code based on the Dame (2011) arXiv paper, so that I could create my own moment-masked cubes.
TMD: The data at |b| > 1-deg is sampled every 1/4-deg, compared with 1/8-deg sampling closer to the plane. If you interpolate the data at |b| > 1 to 1/8-deg sampling the noise in that region will be correlated spatially and the spatial smoothing that you need to do for the moment masking will not lower the noise as expected (and as it will at lower latitudes). Depending on the masking level you choose, you will either miss significant emission in the plane or allow noise features through the masking at higher latitudes. Proper moment masking really requires uniform sampling and noise; that's why the higher latitudes were blanked. What you did is ok provided you adopted a masking level which is 5 times the noise at |b|>1; in that case you will miss some weaker real emission in the plane, but that's probably ok.