This thesis presents investigations into different methods of associating astronomical
sources detected at different wavelengths, and describes the development of a tool for
AstroGrid to enable users to associate sources in a fully automated manner.
At present when associating sources at different wavelengths it is common for astronomers to select IDs by eye or at least verify probabilistically -determined counterparts by eye. With the new trend for large surveys this is no longer practical as datasets
may contain millions of objects. Previous work on association algorithms has focussed
on case -specific techniques which typically only match a restricted number of objects
with counterparts, and often only those with small positional errors. This thesis addresses the issue that these methods are not adequate in the general case where datasets
may be enormous and source error ellipses large. In such situations matching based
purely on spatial proximity is deficient since there may be hundreds of candidate counterparts within a source error ellipse. We therefore investigate the likelihood ratio as an
association technique, as this allows incorporation of data such as object magnitudes
as well as positions, and prove its applicability in the (difficult association) case of the
FIRBACK survey. We also develop the application of a machine learning technique,
the EM algorithm, and test it against the likelihood ratio method. We determine that
it may be effectively applied to find IDs in surveys with a magnitude distribution with
unrestricted shape. These different association methods are successfully developed into
a tool for AstroGrid to enable users to associate sources in a fully automated manner.
We describe detailed analysis of the likelihood ratio method through the association
of a population of far -infrared sources from the FIRBACK survey with optical counterparts from the INT Wide Field Survey. This is a challenging association problem since
the far -infrared sources have a large positional error due to the poor resolution of the
instrument and the relatively long wavelength. We compare two different variants of
the likelihood ratio method in detail, and use the better one to derive optical counterparts for the far -infrared sources. This proves the applicability of the likelihood ratio
method in the case of large source error ellipses where there are numerous candidates
to choose between.
The scientific benefits of associating multiwavelength data are illustrated via deducing, for the first time, the nature of the FIRBACK sources. These are identified with
not only an optical counterpart but also with data at up to nine further wavelengths.
Their properties are examined through the comparison of their observed spectral energy
distributions with predictions from radiative transfer models which simulate the emission from both cirrus and starburst components. The far -infrared sources are found to
be 80 per cent star -bursting galaxies with their starburst component at a high optical
It is a common situation in astronomy to wish to investigate a source population for
which we have no prior knowledge about the properties of the source counterparts expected at another wavelength, for example through observations with a new instrument.
In such a case it is necessary to estimate the counterpart magnitude distribution to use
the likelihood ratio association method. Since little was known about the FIRBACK
sources, prior to our research, their optical magnitude distribution had to be estimated
in order to assign them optical IDs. To alleviate this problem we develop a new astronomical application of a machine learning technique known as the EM algorithm which
is used in the field of informatics. This is able to `learn' the source magnitude distribution iteratively. The algorithm is tested on the FIRBACK sources and also radio
sources from the HI Parkes All -Sky Survey (HIPASS) catalogue and is found to be a
very effective association method in the HIPASS case where the background magnitude
distribution is of unrestricted shape.
We use the FIRBACK survey far -infrared sources as a test -bed for several different
association methods. The value of bringing together multiwavelength observations is
illustrated through the insights that are gained into the nature of the sources. This
work culminates in the development of an association tool for AstroGrid, the UK Virtual
Observatory project, offering three different association methods: the Poisson method,
the likelihood ratio method and the EM algorithm. This tool is able to return a user
specified number of possible counterparts along with a figure of merit for their match
with a source. We also implement the AstroDAS system to store resulting object pairs
in a database for future use. This prevents the same cross association tasks being
carried out numerous times by different users. The Virtual Observatory aims to link
diverse datasets from across the globe. The extra knowledge available from these may
only be extracted after establishing links between detections in these datasets. Our
AstroGrid association tool is therefore vital to the success of the Virtual Observatory.