How Sequeira 2021, MegaMove L2, and IMOS clean tracking data
Sequeira et al. (2021) is a product-level standard, not a numbered filter recipe. MegaMove L2 is a keep/drop cleaner of observed Argos points. IMOS Animal Tracking Facility estimates a new track with a state-space model.
Paper: Sequeira et al. 2021, Methods in Ecology and Evolution DOI 10.1111/2041-210X.13593 (Sakaira / MegaMove data standards). Companion example: OTN biologging_standardization blue-shark workflow. IMOS: Jonsen, McMahon & Harcourt 2024 Best Practice Manual + Jonsen et al. 2020 + ArgosQC.
Where each pipeline lives
IMOS and MegaMove are not competing at the same level. IMOS’s published locations are a Level III-style estimate. MegaMove L2 is meant to stay at Level II: original points, errors removed.
Three philosophies
Keep observed points, then later interpolate
L2 must remain a clean version of the original decoded locations, “without manipulation or processing.” SSM belongs at L3 so the raw curated track is still available.
Delete bad observed fixes
Coastal GAT sharks: time window, drop Z, dups, land, optional distance, then speed. Output is a subset of Argos points.
Estimate the true path
Truncate to at-sea window, ignore extreme outliers in an SSM, write new lat/lon + uncertainty onto every dive/CTD record. Land is rerouted, not deleted from observations.
Actual step order
Blue-shark example does not run a land filter (pelagic). Paper text still lists land as a marine L2 example. Speed in the example: 4 m/s via trip::speedfilter.
Same skeleton as the paper example, plus land (needed for Shark Bay) and optional deploy radius. Speed method currently differs from the paper’s trip::speedfilter.
Speed is conservative and only to ignore outliers in the model (pinnipeds/turtles ~3 m/s). First/last points are still protected in Freitas/McConnell. Ghosts are usually absorbed by the SSM, not deleted.
Checklist from the paper
Items below are what Sequeira et al. actually specify for Level II (plus the companion L1→L2 example). The paper does not prescribe filter order beyond “use deployment metadata, then automate land/speed.” Freitas 2008 and Hoenner 2018 are cited as examples of impossible-location QC, not as a required algorithm.
| Sequeira 2021 / OTN example | MegaMove L2 | IMOS satellite QC |
|---|---|---|
| Keep L0/L1 originals; L2 is a clean observed subset, not interpolated | Yes L0, L1, L2 files retained |
L3-style Writes estimated coords |
| Deployment metadata: track start vs deploy date; end / detachment | Yes trackStart / deploy; end if present |
Yes NRT: first–last CTD; DM: manual at-sea |
| Remove points before deploy / after detachment (drifting tag) | Yes | Yes |
| Researcher flags at L1; repository clears those flags at L2 | Partial Metadata-driven, not a L1 flag column |
Partial DM can manually drop <1% |
| Land filter for exclusively marine animals | Yes Deletes on-land points |
Different Optional reroute of SSM path |
| Speed filter (paper example: trip::speedfilter 4 m/s) | Partial We speed-filter, but currently consecutive hops, not trip |
Pre-filter only Freitas/McConnell then SSM |
| Drop Argos location class Z (OTN example) | Yes | Yes |
| Drop duplicate timestamps (OTN example) | Yes | Yes In prep / SSM |
| Document a log of every curation step | Yes logger + QC plots + interim CSVs |
Yes ArgosQC diagnostics + metadata |
| Companion example: if filters miss a ghost, remove it by hand and note in metadata | Optional radius max_dist_km proposed instead of ad-hoc deletes |
SSM Smooths through extreme Argos (their Fig. 3a) |
| Level III SSM (Jonsen / foieGras / aniMotum) | Not in GAT L2 | This is IMOS QC |
| Error ellipses retained for later processing | Columns kept Not used in the speed rule |
Used in SSM |
| Hoenner et al. 2018 acoustic flags (DistanceRelease, Velocity, …) | N/A Satellite tracks, not detections |
Other IMOS stream Acoustic ATF, not ArgosQC |
What the NZ ghost would hit
Under the paper’s own L2 example
Temporal window makes the Tasman point the first retained fix. trip::speedfilter protects first/last points. The OTN example then says: if a spurious location survives, remove it manually and record that in deployment metadata. Distance-from-deploy is not in the paper; it is a coded stand-in for that leftover-ghost step.
Under IMOS
Same endpoint protection in the Freitas pre-filter. The SSM is then allowed to ignore the wild observation and estimate a location consistent with the rest of the track and the error ellipse. That is Level III behaviour. It is the paper’s next product, not a replacement L2 sequence.
Practical reading
MegaMove L2 follows the Sequeira Level II idea: time window from metadata, drop Z and duplicates, land for marine animals, speed, keep originals, log steps.
We diverge where the paper’s example uses trip::speedfilter and we currently use consecutive hops; and where we added optional max distance (the paper’s “manual leftover” slot).
IMOS implements the paper’s Level III (SSM), not Level II. Comparing IMOS filter order to our L2 order is a category error. If we wanted IMOS-like ghost handling without a distance cap, the paper’s next step is SSM, not reshuffling land vs speed.