Using GPS data to evaluate the accuracy of state-space methods for correction of Argos satellite telemetry error

Recent Studies have applied state-space models to satellite telemetry data in order to remove noise from raw location estimates and infer the true tracks of animals. However, while the resulting tracks May appear plausible, it is difficult to determine the accuracy of the estimated positions, especially for position estimates interpolated to times between satellite locations. In this study, we use data from two gray seals (Halichoerus grypus) carrying, tags that transmitted Fastloc GPS positions via Argos satellites. This combination of Service Argos data and highly accurate GPS data allowed examination of the accuracy of state-space position estimates and their uncertainty derived from satellite telemetry data After applying a speed filter to remove aberrant satellite telemetry locations. we fit a continuous-time Kalman filter to estimate the parameters of I random walk, used Kalman smoothing to infer positions at the times of the GPS Measurements. and then compared the filtered telemetry estimates with the actual GPS Measurements. We investigated the effect of varying maximum speed thresholds in the speed-filtering algorithm oil the root mean-square error (RMSE) estimates and used minimum RMSE as a criterion to guide the final choice of speed threshold. The optimal speed thresholds differed between the two animals (1.1 m/s and 2.5 m/s) and retained 50% and 65% of the data for each seal. However, using a speed filter of 1.1 m/s resulted in very similar RMSE for both animals. For the two seals. the RMSE of the Kalman-filtered estimates of location were 5.9 and 12.76 km, respectively, and 75% of the modeled positions had errors less than 6.25 km and 11.7 km for each seal. Confidence interval coverage was close to correct at typical levels (80-95%), although it tended to be overly generous at smaller sizes. The reliability Of uncertainty estimates was also affected by the chosen sliced threshold. The combination of sliced and Kalman filtering allows for effective calculation of location and also indicates the limits of accuracy when correcting service Argos locations and linking satellite telemetry data to spatial covariate and habitat data.