Glossary and references#


1D Analysis#

1D analysis or spectral analysis where data are reduced to a simple 1D geometry along the reconstructed energy axis. In Cherenkov astronomy, this is classically performed with an OFF background measurement [Piron2001].

3D Analysis#

3D analysis or cube analysis, where data are reduced to a 3D cube with spatial coordinates and energy axes [Stewart2009]. In Gammapy, these cube are represented by Map objects (see Sky maps (DL4)) and contained in a MapDataset object.


Short for “effective area”: it is the IRF representing the detector collection area. See Instrument Response Functions (DL3) and Effective area.


Short for “background”: it is the IRF representing the residual background rate per solid angle. See Instrument Response Functions (DL3) and Background.


The cash statistic is a Poisson fit statistic usually used when signal and background can be modeled [Cash1979]. It is defined as \(2 \times log(L)\). See Cash : Poisson data with background model in fit statistics.


In Gammapy a dataset bundles the data, IRFs, model and a likelihood function. Based on the model and IRFs the predicted number of counts are computed and compared to the measured counts using the likelihood. See Datasets (DL4)


Short for “data level 3”: it is used to mention or describe science-ready data, ie the event list (with basically a Time, an arrival direction and an energy) and the four IRFs (Aeff, EDisp, PSF, Bkg). See data flow.


Short for “data level 4”: it is used to mention or describe binned-science data, ie N-dim maps (multi-dimensional histograms) of the spatial, temporal and/or spectral components of the DL3 data in instrumental units (e.g. counts). See Sky maps (DL4) and ref:data flow.


Short for “data level 5”: it is used to mention or describe advanced-science data, ie N-dim maps of the spatial, temporal and/or spectral astrophysical quantities in physical units (e.g. cm-2 s-1). See data flow.


Short for “data level 6”: it is used to mention or describe catalogs of sources, with their spectral and spatial properties. See data flow.


Short for “energy dispersion”: it is the IRF that represents the probability of measuring a given reconstructed energy as a function of the true photon energy. See Instrument Response Functions (DL3) and Energy Dispersion.


Gammapy utility classes to compute astrophysical quantities based on a model. They are used for the computation of DL5 products from the DL4 ones. See Modeling and Fitting (DL4 to DL5) and the data flow.


Short for “field of view”: it indicates the angular aperture (sometimes also the solid angle) on the sky that is visible by the instrument with a single pointing.

FoV Background#

Background estimation method typically used for the 3D analysis. See Data reduction (DL3 to DL4).


Short for “high level interface: this API aims to realize most of the analysis types using YAML configuration files. See High Level Analysis Interface


Short for “Gamma Astro Data Format”. The open initiative GADF provides a Data Format for gamma-ray data that is currently used by many IACT experiments and by HAWC. Gammapy I/O functions are compliant with this format.


Short for “good time interval”: it indicates a continuous time interval of data acquisition. In CTA, it also represents a time interval in which the IRFs are supposed to be constant.


Short for “instrument response function”: they are used to model the probability to detect a photon with a number of measured characteristics. See Instrument Response Functions (DL3).

Joint Analysis#

A joint fit across multiple datasets implies that each dataset is handled independently during the data reduction stage, and the statistics combined during the likelihood fit. The likelihood is computed for each dataset and summed to get the total fit statistic. See Joint Analysis


Gammapy utility classes performing data reduction of the DL3 data to binned datasets (DL4). See Data reduction (DL3 to DL4) and the data flow.


Short for “mission elapsed time”. see also Mission elapsed times (MET) in Time handling in Gammapy.


Short for “point spread function”: it is the IRF representing the probability density of the angular separation between true and reconstructed directions. See Instrument Response Functions (DL3) and Point Spread Function.

Reco Energy#

The reconstructed (or measured) energy (often written e_reco) is the energy of the measured photon by contrast with its actual true energy. Measured quantities such as counts are represented along a reco energy axis.

Reflected Background#

Background estimation method typically used for spectral analysis. See Data reduction (DL3 to DL4).

Ring Background#

Background estimation method typically used for image analysis. See Data reduction (DL3 to DL4).


Short for “region of interest”: it indicates the spatial region in which the data are analyzed. In practice, at each energy it corresponds with the sky region in which the dataset mask is True.

Stacked Analysis#

In a stacked analysis individual observations are reduced to datasets which are then stacked to produce a single reduced dataset. The latter is then used to obtain physical information through model fitting. Some approximations must be made to perform dataset stacking (e.g. loss of individual background normalization, averaging of instrument responses, loss of information outside region of interest etc), but this can reduce very significantly the computing and memory cost. For details, see Stacking Multiple Datasets

True Energy#

The true energy (often written e_true) is the energy of the incident photon by contrast with the energy reconstructed by the instrument. Instrument response functions are represented along a true energy axis.


Short for “test statistics”. See Estimating TS and Fit statistics.


The WStat is a Poisson fit statistic usually used for ON-OFF analysis. It is based on the profile likelihood method where the unknown background parameters are marginalized. See WStat : Poisson data with background measurement in fit statistics.


This is the bibliography containing the literature references for the implemented methods referenced from the Gammapy docs.


H.E.S.S. Collaboration (2018), “The H.E.S.S. Galactic plane survey”


Albert et al. (2007), “Unfolding of differential energy spectra in the MAGIC experiment”,


Berge et al. (2007), “Background modelling in very-high-energy gamma-ray astronomy”


Cash (1979), “Parameter estimation in astronomy through application of the likelihood ratio”


Cousins et al. (2007), “Evaluation of three methods for calculating statistical significance when incorporating a systematic uncertainty into a test of the background-only hypothesis for a Poisson process”


Feldman & Cousins (1998), “Unified approach to the classical statistical analysis of small signals”


Lafferty & Wyatt (1994), “Where to stick your data points: The treatment of measurements within wide bins”


Li & Ma (1983), “Analysis methods for results in gamma-ray astronomy”


Meyer et al. (2010), “The Crab Nebula as a standard candle in very high-energy astrophysics”


Mohrmann et al. (2019), “Validation of open-source science tools and background model construction in γ-ray astronomy”


de Naurois (2012), “Very High Energy astronomy from H.E.S.S. to CTA. Opening of a new astronomical window on the non-thermal Universe”,


Piron et al. (2001), “Temporal and spectral gamma-ray properties of Mkn 421 above 250 GeV from CAT observations between 1996 and 2000”,


Rolke et al. (2005), “Limits and confidence intervals in the presence of nuisance parameters”,


Stewart (2009), “Maximum-likelihood detection of sources among Poissonian noise”