Unsafe SafeMaskMaker

Gammapy version The dev version

Bug description The issue happens during the Data Reduction step, without the use of a cutout. When using a large FoV of the final stacked MapDataset, the processing by the SafeMaskMaker of an observation with a pointing position far from the center of the final MapDataset (e.g. >2.5deg for HESS) leads to a python error. The mask_safe can not be computed (crash in L176 of makers/safe.py). If a cutout centered on the pointing position is used to make an empty MapDataset from the final stacked MapDataset, the SafeMaskMaker works correctly.

It seems that there are two issues:

• the handling of the position parameter of SafeMastMaker: the observation center is not used for a standard treatment of a run (in make_mask_energy_aeff_max and also in make_mask_energy_edisp_bias)
• if no threshold is found in make_mask_energy_aeff_max, one should return an empty mask

Expected behavior If an observation is at the very edge of a MapDataset, the worst case should be that an empty mask is return. In all other case, no crash should happen and the mask_safe should be correctly computed.

To Reproduce

import astropy.units as u
from gammapy.maps import Map, MapAxis, WcsGeom
from gammapy.datasets import MapDataset
from gammapy.makers import MapDatasetMaker, SafeMaskMaker
data_store = DataStore.from_dir("$GAMMAPY_DATA/hess-dl3-dr1")
selection = dict(
    type="sky_circle",
    frame="icrs",
    lon="83.633 deg",
    lat="22.014 deg",
    radius="5 deg",
)
selected_obs_table = data_store.obs_table.select_observations(selection)
observations = data_store.get_observations(selected_obs_table["OBS_ID"])
energy_axis = MapAxis.from_energy_bounds(1.0, 10.0, 4, unit="TeV")

geom = WcsGeom.create(
    skydir=(85.633, 22.014),
    binsz=0.02,
    width=(10, 10),
    frame="icrs",
    proj="CAR",
    axes=[energy_axis],
)

# Reduced IRFs are defined in true energy (i.e. not measured energy).
energy_axis_true = MapAxis.from_energy_bounds(
    0.5, 20, 10, unit="TeV", name="energy_true"
)

stacked = MapDataset.create(
    geom=geom, energy_axis_true=energy_axis_true, name="crab-stacked"
)
offset_max = 2.5 * u.deg
maker = MapDatasetMaker()
maker_safe_mask = SafeMaskMaker(
    methods=["offset-max", "aeff-max"], offset_max=offset_max
)

for obs in observations:
    empty = stacked.copy(name=f"obs-{obs.obs_id}")
    # A MapDataset is filled in this cutout geometry
    dataset = maker.run(empty, obs)
    # The data quality cut is applied
    dataset = maker_safe_mask.run(dataset, obs)

Other information Any other information you think will be useful for us to fix the issue can go here.

[bkhelifi]

With the current code, the example returns:

Effective area is all zero at [85d37m58.8s 22d00m50.4s]. No safe energy band can be defined for the dataset 'obs-23592': setting mask_safeto all False. So, I think that we are safe now!

A second point is the improvement of the documentation about the Data Reduction as mentioned @AtreyeeS . One should use: empty = stacked.cutout(obs.pointing_radec, width=2 * offset_max, name=f"obs-{obs.obs_id}" ) instead of empty = stacked.copy(name=f"obs-{obs.obs_id}") I think that this is well described in this tutorial https://docs.gammapy.org/0.20.1/tutorials/starting/analysis_2.html.

So, I propose to close this issue.

[adonath]

Addressed in #3987.