This file describes the maps and catalogs for the H-ATLAS Data Release 1. Details are 
contained in the publications Valiante et al. (2016, Paper I) and 
Bourne at al. (2016, Paper II). Please refer to these papers every time you use these 
products in publications.

----------------------------------------------------------------
----------------------------------------------------------------
1. FILES 
----------------------------------------------------------------

We release 1 catalogue file including sources from all the GAMA fields. For each field a 
total of 18 files are released: 6 map files for each of the 3 bands. Moreover, we release 
the PSFs for the SPIRE instrument derived from the Neptune map (see Griffin et al. 2013 
for more details) and the filters, derived from the PSFs, used to smooth the maps and 
reduce the S/N of point sources (see Paper I for more details).

catalogue from filtered
backsub map:				HATLAS_DR1_CATALOGUE_<version> 
catalogue from filtered
backsub map with all IDs 
entries:				HATLAS_DR1_CATALOGUE_ALLIDS_<version>
column description:			HATLAS_DR1_CATALOGUE.COLUMNS

SPIRE MAPS
raw maps:      				HATLAS_GAMA<FIELD>_DR1_RAW<BAND>.FITS
masks:                      		HATLAS_GAMA<FIELD>_DR1_MASK<BAND>.FITS 
background subtracted maps: 		HATLAS_GAMA<FIELD>_DR1_BACKSUB<BAND>.FITS
noise (excluding confusion) 
maps:   				HATLAS_GAMA<FIELD>_DR1_SIGMA<BAND>.FITS
filtered background 
subtracted maps:      	   		HATLAS_GAMA<FIELD>_DR1_FILT_BACKSUB<BAND>.FITS
noise (including confusion) 
for filtered maps: 		   	HATLAS_GAMA<FIELD>_DR1_FILT_SIGMA<BAND>.FITS

PACS MAPS
background subtracted maps: 		HATLAS_GAMA<FIELD>_DR1_BACKSUB<BAND>.FITS
nscan maps:				HATLAS_GAMA<FIELD>_DR1_NSCAN<BAND>.FITS

MORE FILES
SPIRE point spread 
functions:				PSF<BAND>_fromNeptune_wcs.fits
SPIRE matched filters:			MF<BAND>_instr<instr.noise>_conf<conf.noise>_wcs.fits
PACS encircled energy 
functions:				EEF<BAND>.dat
						
----------------------------------------------------------------
----------------------------------------------------------------
2. NOTES ON SPIRE MAPS (see Paper I for further details)
----------------------------------------------------------------

The 'raw' map files contain the image extension of the HIPE product with the masks applied,
in order to exclude data where Herschel slows and turns around at the end of each scan leg.
The map maker transfer function has not been deconvolved in the current release.

The 'mask' files limit the map to the region where there is good data from at least two 
cross-scans; this is the region that should be used for science analysis.  All masks have 
the same dimensions as the maps they mask and consist of 1s and 0s.

The 'backsub' maps have had a local background subtracted from the raw map using the 
nebuliser function from the CASU package
(see http://casu.ast.cam.ac.uk/surveys-projects/software-release/background-filtering
for further details). These maps might introduce a bias if used to measure fluxes of 
extended sources with an aperture radius larger than 1.5 arcmin (see below) and need to 
have the mean set to zero when performing aperture photometry. 

The 'sigma' noise maps give an estimate of the instrumental noise each pixel in the raw 
data map. Use this in preference to the error map contained within the HIPE product. The 
map file error extension is not consistently reliable, because in some areas there are not 
enough scans to accurately determine the noise from the variance in the data. Please use 
instead the 'sigma' noise map supplied which is created as described in the section about
"Noise maps" below. 
THESE MAPS DO NOT INCLUDE THE CONFUSION NOISE.

The 'fbacksub' files are the background subtracted, noise-weighted, filtered maps using 
a customised matched filter. This is the map that should used to detect point sources 
and measure their fluxes.

The 'fsigma' files are the corresponding instrumental noise maps. THESE MAPS DO NOT INCLUDE 
THE CONFUSION NOISE.

----------------
2.1. Astrometry
----------------

Astrometry of the maps of individual scans that form the mosaic has been performed by a 
stacking analysis on the SDSS. On the individual scan maps we found astrometric shifts of 
1-2 arcsec. In making the final mosaics, we have corrected for these shifts. 

-----------------------
2.2. Flux Measurements
-----------------------

The units of the map are Jy/beam. The calibration of the maps is SPIRE Calibration Tree 
Version 8 and is based on Neptune. The SPIRE data reduction pipeline is based on the 
assumption that the flux density of the sources depends on frequency^(-1). Users 
interested in accurately modelling SEDs should multiply the measured flux densities by the 
K_ColE parameter, which is given in Tables 5.6 and 5.7 in the SPIRE Handbook.

To convert the maps from Jy/beam to Jy/pixel, in order to carry out aperture photometry, 
the values in the maps should be divided by the ratio between the beam area and the pixel 
area in arcsec^2 (469/36, 831/64 and 1804/144 at 250, 350 and 500 microns respectively). 
The mean of the maps should also be subtracted. Users interested in the most accurate 
aperture photometry of extended sources should consider scaling the SPIRE fluxes using the 
K4 corrections for extended sources. They should also consider making an aperture 
correction.

There is a 5.5% calibration uncertainty in the fluxes. Some of this uncertainty is 
correlated between bands. For those wishing to take account of this effect, a useful 
prescription is to assume that there is a 4% calibration uncertainty which is perfectly 
correlated between the bands and a 1.5% calibration uncertainty that is uncorrelated. The
current recommendation (SPIRE Handbook) is that these factors should be added, obtaining a 
total calibration uncertainty of 5.5%.

-----------------------------------------------
2.3. Stacking Analyses - Matched-filtered maps
-----------------------------------------------

For obtaining the best S/N in the flux measurement of point sources, we have also provided 
maps that have been convolved with a matched filter derived from the point-spread 
function. The units of these maps are also Jy/beam. The mean of these maps is not zero: as
a result of the background  subtraction, it is the moda of these maps to be zero. 
THESE MAPS SHOULD BE USED IN STACKING ANALYSES ONLY AFTER THEIR MEAN HAS BEEN SET TO ZERO.
Estimates of the errors in stacking measurements should be obtained by Monte-Carlo 
simulations.

----------------
2.4. Noise maps
----------------

The 'sigma' noise maps contain estimates of the instrumental noise on each pixel in the 
data maps. They are created from the coverage maps (the extension called 'coverage' in the 
raw map fits file) by assuming that the noise decreases as sqrt(N_passes), and using the 
comparison between cross-scan measurements to determine the noise per pass. The noise per 
pass that we have assumed for each pixel for this data release is 30.1, 30.9 and 36.1 
mJy/beam at 250, 350 and 500 microns, respectively.  Pixels that have particularly high
values in the error map are flagged as 'hot pixels'. The thresholds for being flagged as 
a hot pixel depend on the field and are
G9: 0.1, 0.041, 0.03
G12: 0.14, 0.08, 0.065
G15: 0.27, 0.12, 0.086
Jy/beam at the three wavelengths.
These pixels are given the noise as estimated from the error map.

The 'fsigma' noise map is the noise appropriate to the noise-weighted filtered map. The 
instrumental noise is calculated as the square-root of the inverse variance-weighted 
filtered variance from above. Confusion noise is not included: it should e calculated using 
Eq. 13 in Valiante et al. 2016 and added in quadrature.

----------------------------------------------------------------
----------------------------------------------------------------
3. NOTES ON PACS MAPS (see Paper I for further details)
----------------------------------------------------------------

The 'backsub' maps are Level 1 maps from HIPE, processed by JScanamorphos and filtered
using the nebuliser function from the CASU package to remove any residual large-scale 
artefact (see http://casu.ast.cam.ac.uk/surveys-projects/software-release/background-filtering
for further details). The units are Jy/pixel. These maps might introduce a bias if used to 
measure fluxes of extended sources with an aperture radius larger than 4 arcmin (see below). 

The 'nscan' maps are produced by JScanamorphos and include the number of scans in each 
pixel. It is used to derive the noise for each source (see below).

----------------
3.1. Astrometry
----------------

Astrometry of PACS maps are not exactly the same as for the SPIRE maps, because the PACS
images have been created using a different pointing model.
To calculate the offset between PACS and SPIRE maps, we have calculated the mean difference 
between the positions of the brightest PACS sources and their SPIRE counterpart.
We found an astrometric shifts of ~1 arcsec and we have corrected the final mosaic for 
this quantity. 

-----------------------
3.2. Stacking Analyses
-----------------------

The mean of the PACS maps is not zero, because of the background subtraction filter. THESE 
MAPS SHOULD BE USED IN STACKING ANALYSES ONLY AFTER THEIR MEAN HAS BEEN SET TO ZERO. 
Estimates of the errors in stacking measurements should be obtained by Monte-Carlo 
simulations.

----------------------------------------------------------------
----------------------------------------------------------------
4. NOTES ON FURTHER FILES (see Paper I for further details)
----------------------------------------------------------------

----------------
4.1. SPIRE PSFs
----------------

SPIRE PSFs have been derived by Neptune images (Griffin et al. 2013). They have been 
azimuthally-averaged and normalised. FWHM are 17.8, 24.0 and 35.2" at 250, 350 and 500um
respectively.

---------------
4.3. SPIRE MFs
---------------

SPIRE matched filters have been derived from the SPIRE PSFs and following the recipe in 
Chapin at al. 2009. The instrumental and confusion noises used in the calculations are
derived from the 'backsub' maps.

---------------
4.3. PACS EEFs
---------------

PACS Encircled Energy Functions have been derived for our maps by stacking at the position 
of optical counterparts with z<1 and optical size <5". FWHM of the beams resulting from the 
stacking are 11.4 and 13.7"" at 100 and 160um respectively. 
The EEFs are derived up at 30". At larger scales, we have adopted the EEFs provided by the 
PACS team (Lutz 2015).

----------------------------------------------------------------
----------------------------------------------------------------
5. NOTES ON CATALOGUES (see Paper I and II for further details)
----------------------------------------------------------------

Columns description is in the file HATLAS_DR1 CATALOGUE.COLUMNS

----------------------------
5.1. SPIRE Catalog Creation
----------------------------

This catalogue contains all SPIRE sources which are >4 sigma significance (including 
confusion noise) in any of the three SPIRE bands (250,350,500um).

The source catalogue is based on finding peaks in the noise-weighted matched filter 
filtered the maps using the MADX algorithm.

Maps are background subtracted before source extraction but be wary of the results where 
cirrus is strong, e.g. 9hr field. We recommend masking out regions of high cirrus before 
using the cats for studies of large scale structure.

FLUXES - The catalogue is produced by using the 250 micron map to perform the source 
detection. The flux in each band is then estimated as the flux from the noise-weighted 
mf filtered map at the same position in each band. Aperture fluxes are included where the 
identification in the optical suggested the source would be resolved by Herschel (see 
Extended source description).
'F<BAND>' columns are the point source fluxes from MADX, 'F<BAND>BEST' flux columns use the 
largest of the aperture or point source fluxes. To assess which one of the 2 values is the 
largest, we take into account the noise. The aperture flux is preferred over the point 
source flux if:
(F_AP-F_PS)>sqrt(sigma_AP^2-sigma_PS^2) 
Some sources have been assigned a point sources flux as best flux after visual inspection.
F<BAND>BEST values are what we consider the most reliable estimate of the flux for a source.

UNCERTAINTIES - The uncertainties maps are calculated from the coverage maps and assuming 
instrumental noise values of 30.1, 30.9, 36.1 mJy per pixel per pass per beam. The maps 
are deep enough that confusion noise is an important factor that must be included in any 
error estimates. We have estimated the minimum confusion noise in the mf filtered maps to be 
2.04, 3.06 and 3.91 mJy/beam. We have also derived a different estimated of the confusion
noise from our simulations, which depends on the flux of the source.
The noise on each flux in the catalogue is the sum in quadrature of the instrumental noise 
and confusion noise.

******************************************************************* 
A 5.5% calibration uncertainty should be added in quadrature to all
SPIRE fluxes in this catalogue.
*******************************************************************

POSITIONS - For S/N = 4 at 250um, the one sigma positional accuracy was found to be 2.4 
arcsec, reducing to 1.2 arcsec at 10 sigma (Paper II). We have set a floor of 1 arcsec for
the positional accuracy of the highest S/N sources as we expect this is the limit of 
systematics in the maps. No significant differences were found in DR1 compared to SDP.
NB EDIT: The results are consistent with Smith et al. 2011 using the same F250 cut, but 
results at S/N=5 for blue SPIRE colours are now consistent with the theoretical positional 
accuracy, whereas previously the positional errors seemed worse - this is due to the 
improved definition of the noise level.

COLOUR CORRECTION - The SPIRE data reduction pipeline is based on the assumption that the 
flux density of the sources depends on frequency^(-1). Users interested in accurately 
modelling SEDs should multiply the measured flux densities by the K_ColE parameter, which 
is given in Tables 5.6 and 5.7 in the SPIRE Handbook.

ISSUES - 
Flux boosting from Malmquist and Eddington bias and source confusion is present for 
sources near the detection limits. Please see Paper I for details and averaged correction 
values. This catalogue is NOT CORRECTED for flux boosting.

--------------------------------------------
5.2. Extended SPIRE source flux description
--------------------------------------------

The fluxes derived by aperture photometry are reported in the 'F<BAND>BEST' 
flux columns when they are larger then the point source fluxes, thus if 
(F_AP-F_PS)>sqrt(sigma_AP^2-sigma_PS^2). In these cases, with few exceptions, they are 
considered the most reliable estimate of the flux for a source. Exceptions are sources 
whose aperture would include a neighbour source and for which the customised aperture 
would have been smaller than the PSF: after visual inspections, these sources have been 
assigned a point source flux as best flux.

APERTURES - Aperture fluxes have been measured at 250, 350 and 500 microns for those 
sources which had RELIABILITY > 0.8, an optical diameter ISOA_R_ARCSEC>10*(FWHM/18.) arcsec 
and GSQ_FLAG!=1. Aperture fluxes were measured from the background-subtracted maps made by 
MADX, after setting the mean to zero. Pixels have been split into 16 subpixels to increase 
photometric accuracy. The aperture radius, r_ap, used for each source was: 
r_ap = sqrt(FWHM^2.0 + ISOA_R_ARCSEC^2.0) 
where ISOA_R_ARCSEC is the quantity ISOA_R, taken from the SDSS catalogue, converted in 
arcsec (all apertures were circular). Sources with MODELMAG_R>19.0 or 
ISOA_R(in pixels)>-35.7*ISOA_R+728 have r_ap calculated with ISOA_R_ARCSEC set to 0. The 
apertures were placed at the optical positions.  
In the table, the aperture used to perform the photometry of a SPIRE source is reported 
only in the row of the optical ID which was used to define the aperture and the position. 

Conversion from Jy/beam to Jy/pixel was done by dividing the maps by 469/36, 831/64 and 
1804/144 at 250, 350 or 500 microns respectively (SPIRE Observers' Manual).

Aperture photometry has been performed on the backsub maps made by MADX, after setting the 
mean of the map to 0. 
Fluxes might be slightly biased when AP<BAND> >1.5 arcmin, because of the background 
subtraction.

UNCERTAINTIES - 1 sigma errors have been derived performing aperture photometry with the
same apertures at random positions around each source. The Monte Carlo results are well
described by the relation provided in Appendix A1 of Valiante et al. 2016. 

'CUSTOMISED APERTURE' PHOTOMETRY - Some sources, upon visual inspection, needed a 
larger/smaller aperture. Typically these sources have either a strong ellipticity or have 
a neighbour source which would fall into the r_ap calculated as above. The apertures have 
been defined by hand and have a major semi-axis, a minor semi-axis and positional angle 
(columns AP*, AP_RMIN, AP_PA in the catalogues). For each source, the same aperture has 
been used for all wavebands. Errors were measured as above, using the custom-sized 
apertures. All sources requiring a customised aperture in SPIRE bands, also got a 
customised aperture in PACS bands.

APERTURE CORRECTIONS - The SPIRE aperture fluxes have been scaled to correct for the 
proportion of emission that is expected to fall outside the aperture. The correction has 
been calculated using the profile of Neptune (see SPIRE Handbook for more details).

-----------------
5.3. PACS FLUXES
-----------------

PACS flux densities are measured using circular apertures placed at the optical positions
when there is an optical ID with RELIABILITY > 0.8. Otherwise, the SPIRE positions are 
used. The PACS maps, similar to the SPIRE maps, have been background-subtracted, so we 
assumed that there is effectively zero background after we set the mean of the map to 0. 
Fluxes might be slightly biased when AP<BAND> >2 arcmin, because of the background 
subtraction. SPIRE sources falling outside PACS maps have PACS fluxes = -1.

APERTURES - Two sets of aperture radii are used: firstly a `point source' flux density is 
found using 11.4 and 13.7 arcsec radii apertures at 100 and 160 microns respectively
(the assumed FWHM of the beams). 
Next, additional aperture flux densities are measured for positions where a PACS source 
has a reliable optical ID (RELIABILITY > 0.8) using an aperture radius
r_ap = sqrt(FWHM^2 + ISOA_R_ARCMIN^2)
where ISOA_R_ARCSEC is the quantity ISOA_R, taken from the SDSS catalogue, converted in 
arcsec (all apertures were circular). Sources with MODELMAG_R>19.0 or 
ISOA_R(in pixels)>-35.7*ISOA_R+728 have r_ap calculated with ISOA_R_ARCSEC set to 0. The 
apertures were placed at the optical positions when a reliable optical ID has been found.
In the table, the aperture used to perform the photometry of a SPIRE source is reported 
only in the row of the optical ID which is used to define the aperture and the position.  
If no optical ID is found, then the aperture photometry is performed at the position of 
the SPIRE source.

APERTURE CORRECTIONS - Aperture corrections have been made calculating the EEF of the PSF 
derived from stacking analysis in our maps (see Paper I). The EEFs files are part of this
data release. 
Fluxes measured at SPIRE positions have been increased by 5 and 10% at 100 and 160um 
respectively to take into account the SPIRE positional uncertainty.

UNCERTAINTIES - The 1 sigma error for each source will depend on its aperture and on the 
number of scans in the pixels inside the aperture. They have been calculated by placing 
apertures of different sizes at random positions on the images.

**********************************************************************
A calibration error of 7% should be added in quadrature to all errors
quoted in this catalogue.
********************************************************************

-------------------------------
5.4. Optical IDs from SDSS DR7
-------------------------------
IDs are sought via a Likelihood-Ratio analysis of optical candidates within 10 arcsec of 
all SPIRE sources with S/N>=4 at 250mum (defined as F250/E250 in the matched-filter 
catalogues). The data release main catalogue (HATLAS_DR1_CATALOGUE.FITS) contains only the 
'best' candidate ID to each SPIRE source (where available). Most users will find in this 
catalogue everything they will need for their science purposes. A second catalogue is also
available (HATLAS_DR1_CATALOGUE_ALLIDS.FITS), which contains all possible counterparts 
within the search radius of each SPIRE source, and provides the full LR 
statistics so that these may be independently analysed as the user wishes. To select only 
sources which have reliable optical IDs, we recommend applying a cut of Reliability>=0.8, 
although other cuts on Reliability or LR may be suitable for different purposes as 
discussed in Bourne et al. (2016). The flags 'R' and 'M' in the ID_FLAG column indicate 
that the reliability has been modified for sources which are resolved in SPIRE ('R') or 
which have a merger ID ('M') which caused the reliability of a genuine counterpart to be 
underestimated by the formula. Note that in the catalogue of all potential counterparts
(HATLAS_DR1_CATALOGUE_ALLIDS.FITS), in a small number of cases it is possible for a source 
to have two reliable IDs if it is a merger ('M' flag).

Positions:
----------
The optical catalogue is based primarily on SDSS-II DR7 primary object IDs with 
MODELMAG_R<22.4, obtained from CASJOBS. These have been supplemented with by filling in 
objects from SDSS-III DR9 that were missing from DR7. This only affects small regions 
close to two bright stars in the GAMA9 and GAMA12 fields, where the masking was especially 
aggressive in DR7, and so several SDSS-III objects were added from DR9. 
SDSS positions and OBJIDs from either DR7 (SDSS-II) or DR9 (SDSS-III) are included in the 
table to facilitate matching back to the CAS databases. Note that the OBJIDs of a given 
object are not the same in SDSS-II and SDSS-III. Due to the length of OBJIDs, these must 
be stored in a 64-bit integer. 
SDSS_OBJIDs OPTICALRA and OPTICALDSEC in the catalogue are generally from SDSS-II unless 
the object was missing from DR7, in which case they are from SDSS-III.  

Positions with SDSS deblend flags have been eyeballed, leading to the removal of a number 
of erroneously deblended SDSS galaxies, which had been shredded into multiple sources by 
the SDSS source extraction.

Redshifts:
----------
Spectroscopic redshifts are obtained from SDSS DR7, DR10 (which contains significant 
additional redshifts for LRG and QSO samples), the 2SLAQ LRG and QSO samples, 2dF, 6dF and 
GAMA-II (SpecCat/SpecAll v27). The GAMA catalogue combines redshifts from AAO and LT as 
well as many historical surveys (see Bourne et al. 2016 and Liske et al. 2015 for full 
details and references). GAMA redshifts are used in preference to others, where the 
quality is nQ>=3, although note that many additional redshifts are sourced from SDSS DR10 
and WiggleZ in particular for the fainter extragalactic targets. 
The source of each redshift in the catalogue is given by Z_SOURCE. Codes are as follows:
1    SDSS DR7
2    6dFGS
4    2SLAQ-QSO
8    2SLAQ-LRG
16   GAMA HATLAS filler targets
32   GAMA Main Survey
64   2dFGRS
128  SDSS DR10
256  Wigglez
512  GAMA (but object not in current TilingCat of GAMA-II)

The quality flag is Z_QUAL. For GAMA redshifts, this is equal to nQ. SDSS redshifts with 
spectroscopic flags (ZWARNING) have been flagged as follows:
Z_QUAL=2:	Z_FITLIMIT (chi-squared minimum at edge of the redshift fitting range(Z_ERR 
			set to -1))
Z_QUAL=3:	MANY_OUTLIERS (fraction of points more than 5 sigma away from best model is 
			too large (> 0.05))
          	or NEGATIVE_EMISSION (a QSO line exhibits negative emission, triggered only in 
          	QSO spectra, if C_IV, C_III, Mg_II, H_beta, or H_alpha has 
          	LINEAREA + 3 * LINEAREA_ERR < 0)
Z_QUAL=0:	Any other spectra flag.
Z_QUAL=5:	No spectra flags.
For more details of the SDSS spectroscopic flags see 
http://www.sdss3.org/dr10/algorithms/bitmask_zwarning.php

Only spectroscopic redshifts with Z_QUAL>=3 are recommended to be reliable. Photometric 
redshifts have been obtained using ANNZ for all objects in the equatorial fields with 
photometry from SDSS and UKIDSS LAS (see Smith et al. 2011 for full details). In Bourne 
et al (2016) these were compared with the SDSS photometric redshifts and found to have 
both less scatter and smaller bias.

Star-galaxy Separation:
-----------------------
The SDSS positions and modelmags are matched to UKIDSS LAS (DR8) from which we use 
aperture magnitudes in J and K bands to assist with star-galaxy separation. Galaxies and 
stars were discriminated using a g-i/J-K colour-colour diagram in addition to r-band 
PSF/model magnitude criteria (see Bourne et al. 2016). To avoid contamination from quasars, 
anything classified as a star which had a spectroscopic redshift z>0.001 (with Z_QUAL>=3), 
was reclassified as a quasar. All matches with SDSS DR12 quasars (based on spectral 
classification: CLASS='QSO' and ZWARNING=0 in SPECOBJ) were also classified as quasars. In 
addition we identify photometric "quasar candidates" 
which have no spectroscopic redshift but lie outside of the stellar locus of the g-i/J-K 
colour diagram (delta_SGJK>0.4, as defined in Bourne et al. 2016 and Smith et al. 2011). 

The GSQ_FLAG column contains the star-galaxy separation results as follows:
0 = galaxy
1 = star
2 = quasar (point-like source with spec-z>0.001 or SDSS CLASS='QSO')
3 = candidate quasar (no spec-z)

For a clean sample of quasar IDs we suggest using SGSEP==2 only.
For a complete sample of quasar IDs we suggest using SGSEP==2||SGSEP==3.
For a complete sample of star IDs we suggest using SGSEP==0||SGSEP==3.
For the purposes of LR statistics, all extragalactic sources (SGSEP!=1) are treated 
together, while stars are treated separately as described in Bourne et al. 2016.

Likelihood ratio analysis:
--------------------------
Likelihood ratios are calculated for every optical object within 10 arcsec of each SPIRE 
source, following the procedure described in detail in Bourne et al. 2016. The procedure 
consists of three stages. 
i) The first quantity to define is Q0, which is the underlying true fraction of SPIRE 
sources which exist in the SDSS catalogue (ie which have r magnitude <22.4). Q0 is 
measured using the blanks-counting technique from Fleuren et al. 2012 for the 
galaxies+quasars sample, and for all candidates, thus providing an estimate for stars from 
the difference. This is the same as in previous H-ATLAS internal releases, but differs 
from the method in SDP, as discussed in Bourne et al. 2016. 
ii) The r-band magnitude dependence is calculated based on the normalised magnitude 
distributions of real counterparts [q(m)] and of background objects [n(m)]; this is 
estimated separately for stars and extragalactic sources respectively. At bright 
magnitudes (m<14 for galaxies; m<21.5 for stars), the Poisson noise is too large to 
estimate this reliably and the distribution q(m)/n(m) is assumed flat and is given by the 
average across all magnitudes 10<m<14. This assumption differs from that in previous 
H-ATLAS catalogues, in which the q(m)/n(m) for brighter galaxies was simply fixed at the 
brightest bin where it could be measured; and for stars a constant q(m) was assumed rather 
than constant q(m)/n(m). The reasons for these decisions and the effects on the results 
are discussed in Bourne et al. 2016.
iii) The positional offset distribution is used to describe the probability f(r) of a true 
counterpart having a given radial offset r. This is estimated from blue SPIRE sources only, 
as a function of 250mum SNR (F250/E250, based on the matched-filter catalogue). Red sources 
(F250/F350<2.4) are excluded from this analysis in order to avoid the bias due to lensing 
described in Bourne et al. 2014. The functional fit measured from blue SPIRE sources (and 
applied to all SPIRE/SDSS pairs, regardless of colour) is
sigma_pos/arcsec = 2.10*(SNR/5)^(-0.88)
This defines the width of the Gaussian f(r). This empirical fit gives results very close 
to the theoretical 0.6*FWHM/SNR (Ivison et al. 2007), although note the weaker SNR 
dependence which may be explained by the increased positional errors of bright sources 
which are more likely to be (marginally) resolved by SPIRE.
Modifications to sigma_pos are made for bright SDSS galaxies (r magnitude <20.5) to account 
for the increased error in the optical centroid:
sigma_mod^2 = sigma_pos^2 + (0.05*ISOA)^2
where ISOA is the r-band isophotal semimajor axis IN ARCSEC. 
Additionally, for all sources we impose a floor of 1 arcsec on sigma_pos since smaller 
positional errors are considered unrealistic.

Combining these elements, the Likelihood ratio of each SDSS counterpart is given by
LR = f(r) q(m,c)/n(m,c)
where c indicates stars/extragalactic objects in SDSS, whose magnitude dependence is 
treated independently. The reliability of each potential match then depends on the 
likelihoods of all possible matches to a given SPIRE source, in addition to the 
probability that the true counterpart is absent from SDSS (1-Q0):
R_i = LR_i / [Sum_j{LR_j} + (1-Q0)]
Reliabilities given by this formula range from 0 - 1; we consider R>0.8 to be reliable 
although this cut is somewhat arbitrary (see Bourne et al. 2016). Note that some 
reliabilities in the catalogue have been modified following by-eye validation, as 
described below (under Flags), and these will fall outside of this range. In these cases 
the reliability has been set to R+1, so that matches deemed reliable by eye will satisfy a 
cut of R>0.8. 

Flags:
------
The column ID_FLAG contains flags assigned during post-processing. The brightest 300 SPIRE 
sources in each of the three equatorial fields were examined by eye after the completion 
of the LR analysis in order to validate automated results in these special cases where 
they are most likely to fail. In addition, the positions of bright stars from the Tycho 
catalogue were eyeballed to ensure that their IDs were not missed and that they had not 
been misclassified. Also, for the subset of sources with no potential counterparts in SDSS 
DR7, we checked DR8 for objects that may have been missing from DR7.
As a result, the following flags were assigned in the ID_FLAG column:
R 	= Resolved SPIRE source where the counterpart was missed or appeared unreliable due to 
		a large optical-SPIRE separation. The reliability was fixed manually after visual 
		inspection (R_new=R_formula+1) to ensure the ID meets the reliability cut.
M	= SPIRE source with merger ID(s), confirmed by z_spec, which caused the reliability to 
		be underestimated. The reliability was fixed manually after visual inspection 
		(R_new=R_formula+1) thus ensuring the ID meets the reliability cut. Note that in a 
		small number of cases two 'reliable' IDs were given to a single SPIRE source where 
		it was obvious that the emission was roughly equally shared between the two 
		merging galaxies.
V	= Visible galaxy in optical image but missing from all cats searched (SDSS DR7-9, NED, 
		GAMA), for example because the galaxy falls under a bright star mask. No optical 
		ID given in catalogue, but the flag indicates that it is visible in the optical.

GAMA Panchromatic photometry:
-----------------------------
GAMA CATAIDs and matched-aperture photometry from GALEX, SDSS, UKIDSS, VIKING and WISE are 
provided for galaxies in the GAMA Main Survey (roughly defined as those with magnitude 
r<19.8). The photometry are from the GAMA DMU 21BandPhotomCatv03. CATAIDs are from 
InputCatv07.

----------------------------------------------------------------
----------------------------------------------------------------
6. SUMMARY OF CHANGES FROM THE V3 INTERNAL RELEASE
----------------------------------------------------------------

1. Completely new reduction of the SPIRE maps with background subtraction using the 
	Nebuliser method, revised instrumental and confusion noise estimates, and new 
	matched-filter source extraction with MADX.

2. New optical source lists, including few additional sources from SDSS DR9, many new 
	redshifts in SDSS DR10, WiggleZ & GAMA-II (especially for fainter/high-redshift 
	galaxies and QSOs).

3. Improved star-quasar separation using (i) additional quasar redshifts and (ii) 
	photometric selection via J-K and g-i colours where redshifts are not available

4. Overhaul of the LR calculations, measuring the positional errors on blue sources only 
	to avoid the bias (probably due to weak lensing) in positional errors measured for 
	redder sources.