← Back to all sparks
S

skytrackr

INFRA · APIS
Velocity0.0

R package skytrackr by bluegreen-labs — release notes from GitHub.

Light-based animal geolocation gets a second model that stops assuming the bird sat still.

geolocationanimal-trackingmodel-fittingoptimizationr-package
Current state
skytrackr estimates animal positions from logger light data by fitting a sky-illuminance model. Three releases span the history: a 2023 v0.9 the author called functional but unpolished, a CRAN-compliant v1.0 in October 2025 that added batch reading and twilight screening, and a v2.0 six weeks later that is the first structural expansion. v2.0 pulls the data-selection and calibration steps out of the main skytrackr() call and exposes them as stk_filter(), stk_calibrate() and stk_center().
Where it's heading
The arc is toward exposing internals the package used to hide. Each release takes a step that was buried inside the main fitting call and turns it into a function the researcher can inspect, plot and constrain, with the scale parameter now estimated across a whole dataset rather than guessed. The new individual light model is the sharper turn: it drops the stationarity assumption behind the original diurnal fit and solves each observation along a constant-bearing course from the previous position.
Prediction
The author describes the individual model's robustness as still being evaluated and its convergence as more fickle than the diurnal approach, so the next release most likely tunes convergence and documents when each model applies rather than adding a third.

Recent moves

  1. 8mo ago

    Per-observation 'individual' light model joins the diurnal fit

    ⚡ SPARK

    The release that turns skytrackr from a single-method tool into one with a modelling choice. Alongside the new individual model, the previously hidden filtering, calibration and centering steps become callable functions, continuing the pattern of exposing what the main fit used to do silently.

    View source ↗
  2. 10mo ago

    Batch reading and twilight screening arrive with CRAN compliance

    The release that made the package distributable and usable on more than one file at a time. Batch reading, batch twilight screening and parallel-processing vignettes set up the workflow that v2.0 would then decompose into separately callable steps.

    View source ↗
  3. 2y ago

    First working release of light-based geolocation fitting

    The starting point, described by the author as functional but needing polish. The template-matching and optimization core it introduced is what the two later releases have been progressively opening up and refining.

    View source ↗