The Douglas Argos-Filter Algorithm

The Douglas Argos-Filter Algorithm ingests Argos satellite tracking data (DIAG format only) and flags improbable locations based on user-defined distance and velocity thresholds. The algorithm writes all results to ASCII comma-delimited files for spreadsheet or GIS applications, as well as KML 2.0 files compatible with Google™ Earth.   The entire unfiltered data set is also output, so in its minimum application, the algorithm is a useful tool for parsing Argos DIAG data into a more convenient digital format. 

Service Argos disseminates each location with a location class (LC) accuracy index that falls under two broad categories: standard and auxiliary.  Standard locations (LC 3, 2, 1) have an estimated 1-sigma error radius of 250, 500, and 1500 m, respectively, while accuracy of the auxiliary locations (LC 0, A, B, Z) is highly variable and undocumented by Service Argos. Auxiliary locations are largely unsuitable for interpreting animal movements without first addressing accuracy issues.

The Douglas Filter assesses plausibility of every Argos location using two different methodologies based on: 1) distances between consecutive locations; and 2) rates and bearings among consecutive movement vectors.  Both filters independently move chronologically through the raw tracking data of each individual animal, evaluating 3 locations at a time. Results of both filters are written to separate output files. The first filter searches for spatiotemporal redundancy, under the premise that significant locational errors rarely occur in the same geographic locale consecutively.  The user inputs a distance threshold that defines the spatial scale of “locale”.  The second filter constructs 2 adjoining vectors and evaluates their velocity, internal angle of intersection, and length, under the premise that significant locational errors are commonly associated with tracking signatures that create implausible movement rates or anomalously acute movement angles.  User-defined thresholds for implausible velocity and suspect angles allow the user to adjust the filtering to better conform to a species’ movement capability and behavior.  Depending on species, one of the two filtering methods typically possesses advantages over the other.  For avian species, the algorithm also produces a third output that optimally merges the two filters, since birds intermittently migrate long distances with high velocity and directionality, but during non-migratory periods they tend to occupy relatively localized nesting, molting, staging, or wintering areas.