Here's a good suggestion from @fjaviersanchez , following up on the SSim discussion today:
"Right now, thanks to the dither strategy most of the observational effects are pretty homogeneous across the survey area (so not using the observing conditions won't affect too much the object catalogs). However, some of the interesting ones (dust, stars) have structure and variation at large scales. For example, in regions with higher dust extinction you expect to have slightly lower number density and providing information about dust to the network will make easier for it to find the correlations and correctly predict this behavior. In addition, I think that using OpSim's inputs or MAF maps with the mean/median/variance of the observing conditions would be great and very useful for Weak Lensing and LSS people. This is because, in most cases, we assume that the variation in the number density of objects is linear with the observing conditions (so n_galaxies_true = n_galaxies_observed + alpha*value_of_observing_condition). This assumption is tested with different types of checks and is usually good for regions where the deviation from the mean value of the observing condition is small. However, using the DNNs we can get directly a model/kernel for how these observing conditions change the number density. "
I guess both an analytic model and a machine learning model could make good use of such sky maps! Labeling as an enhancement, for someone to take on.
Here's a good suggestion from @fjaviersanchez , following up on the SSim discussion today:
I guess both an analytic model and a machine learning model could make good use of such sky maps! Labeling as an enhancement, for someone to take on.