many phenomena, particularly bird migration. For
example, how many birds migrate on a given night?
How does the total density of migrants change over
time across a migration season or across years, or
even across the continent during the last two
decades? In what directions are birds moving and at
what speed? With what local and regional conditions
do large movements correlate?
In this article, we describe ongoing work to devel-
op an AI system that automatically interprets radar
data and extracts quantitative high-level information
about migration. A primary goal of the article is to
provide an accessible overview of a recent technical
paper that introduced a new probabilistic model and
approximate inference algorithm for reconstructing
the velocities of migrating birds from the partial
information collected by Doppler radar (Sheldon et
al. 2013). An accurate algorithm to estimate the
velocities of birds and other targets detected by radar
is critical to unlocking the potential of the data.
Velocity information is important for understanding
the biology of bird migration. In addition, there is
growing evidence that the structure of the velocity
field is a key to discriminating between birds, which
fly under their own power, and other targets such as
precipitation, insects, and dust, that are primarily
carried by the wind (Gauthreaux, Belser, and Van
Blaricom 2003; Dokter et al. 2011).
The article also highlights key themes that relate
to computational sustainability and AI. One theme is
the promise of data mining and machine-learning
techniques for interpreting large ecological data sets
Figure 1. Mosaic Image of the Continental United States Network of
WSR-88D Stations Depicting a Heavy Migration Event on the Night of 10 September 2010.
Precipitation appears as contiguous regions with blocky and irregular shapes. Birds appear as more uniform stippled patterns of primarily light and medium grays surrounding individual radar stations. Bird migration is evident across the entire continent, particularly across the central and eastern United States. Darker stippled areas represent bird densities approaching 2000 birds per cubic kilometer. On this night, many tens of millions of birds were aloft over the United States. Note, also, that areas with the uniform stippling
do not necessarily represent areas where no birds are present, rather these are places either without radar coverage or where the radar
beam is sufficiently far and high from the radar so as not to provide useful information.