Loading...
Loading...
RESEARCH METHOD / PROPOSAL
The academic method transforms latitude and longitude into suitable metric coordinates, then uses a continuous-time state-space model to distinguish GPS observations from the unobserved true trajectory.
Current status: This is a design proposal, not the method currently used for the rankings. Current rankings use observed intervals of at least 30 seconds and are not CTSD estimates.
01 — MODEL
CTSD does not treat each GPS position as the truth. It models the recorded position as a time-continuous latent trajectory plus observation error. Irregular sampling and gaps can therefore be handled on the same time axis.
Here, CTSD means Continuous-Time Speed and Distance estimation. Its foundation is a continuous-time state-space movement model such as CTCRW.
A Kalman filter or smoother is a computational method, not the movement model itself. For retrospective analysis of a complete track, smoothing uses observations before and after each time and is the default approach.
02 — PROCESS
Use full-resolution latitude, longitude and time before thinning, then check gaps, duplicate times and chronology.
Choose local metric coordinates, Earth-centred coordinates or geodesic calculations to suit the track. Do not force wide-ranging migrations into one UTM zone.
Provide instrument- and environment-specific position error using GPS accuracy, HDOP or stationary tests.
Fit candidates such as CTCRW and select a model suited to the animal and data using information criteria and diagnostics.
Solve the state-space model using Kalman smoothing or an equivalent method to estimate distributions of true position and velocity.
Use CTSD to calculate speed and total distance, storing uncertainty intervals as well as point estimates.
03 — INPUT
Latitude and longitude alone can be processed, but error and instrument metadata make it much easier to distinguish measurement jitter from real movement.
Without observation-error information, small GPS jitter cannot be fully separated from true movement. Error may be estimated from an empirical variogram or stationary tests of similar instruments, but the assumptions and resulting uncertainty must be reported.
04 — IMPLEMENTATION
Adding the explanation page is small; applying a reproducible academic analysis to every record is a medium-to-large project. It must cover large point sets, model diagnostics, retries and versioned results.
Select 20–50 tracks and compare speed, distance and runtime between the current minimum-30-second method and CTSD.
Ingest accuracy, HDOP and instrument-specific stationary tests, and define how missing values are handled.
Integrate R ctmm or a validated equivalent into batch processing, and store diagnostics and uncertainty intervals.
Calculate both during validation, check bias by species and instrument, then decide whether and where to change the display.
Fix the comparison metric—for example the 95th percentile of latent speed or expected total distance. Values from incompatible models or error-information quality should not be silently mixed in one ranking.
Display interval width and estimation quality with the estimate. When positions could plausibly switch, present them as similar instead of claiming a precise order.
05 — REFERENCES
This page is a design document for a proposed academic analysis. After implementation and validation, it will be updated with the model, settings, software version and scope actually used.
Back to the public preprocessing guide