In ianjonsen/foieGras: Fit Continuous-Time State-Space and Latent Variable Models for Quality Control of Argos Satellite (and Other) Telemetry Data and for Estimating Changes in Animal Movement

knitr::opts_chunk$set(echo = TRUE)

install aniMotum dev branch

remotes::install_github("ianjonsen/aniMotum@dev", dependencies = TRUE)

require(tidyverse, quietly = TRUE)
require(aniMotum, quietly = TRUE) 

Fit crw SSM with 12 h time.step

fit <- fit_ssm(sese2, model = "crw", time.step = 12,
               control = ssm_control(verbose = 0))

Simulate 200 tracks from the 2nd fitted track, ensuring they return to the start & plot

trs <- sim_fit(fit[2,], what = "p", reps = 200, cpf = TRUE)

plot(trs)

Filter simulated tracks using similarity flag = 1, keep top 30 % & plot

trs.f1 <- sim_filter(trs, keep = 0.3, flag = 1)

plot(trs.f1)

Filter simulated tracks using mean lon-lat as similarity metrix, keep top 30 % & plot

trs.fll <- sim_filter(trs, keep = 0.3, var = c("lon","lat"), FUN = "mean")

plot(trs.fll)

Comparison of mean lon-lat simulated track locations to the fitted track appears to give a slightly better match than similarity flag 1.

Use path re-routing to move simulated tracks off of land

trs.fllr <- route_path(trs.fll, centroids = TRUE)

plot(trs.fllr)

Use Chl raster & extract values at track locations

## read SpatRaster
chl <- terra::rast("chl.grd")

terra::plot(chl)

trs.df <- unnest(trs.fllr, cols = c(sims))

## extract chl values at track locations
trs.df <- trs.df |> 
  mutate(terra::extract(chl, cbind(lon, lat))) |>
  rename(chl = chl_summer_climatology)

## convert back to nested tibble
trs.n <- nest(trs.df, sims = c(rep, date, lon, lat, x, y, chl))

## append new nested tibble with correct aniMotum classes so sim_filter works
class(trs.n) <- append(class(trs.fllr)[1:2], class(trs.n))

Run sim_filter to select top 15% of tracks with chl most similar to the SSM-estimated track

## try filtering on mean chl
trs.chl.mn <- sim_filter(trs.n, keep = 0.15, var = "chl", FUN = "mean", na.rm = TRUE)

Visualise

plot(trs.chl.mn)

## all tracks
tmp.a <- unnest(trs.n, cols = c(sims))

## filtered tracks
tmp.f <- unnest(trs.chl.mn, cols = c(sims))

## Kernel densities of all Chl values
d.all <- density(tmp.a$chl, na.rm = TRUE)

## densities of Chl values from estimated track
d.e <- density(subset(tmp.f, rep == 0)$chl, na.rm = TRUE)

## densities of Chl values from filtered random tracks
d.s <- density(subset(tmp.f, rep > 0)$chl, na.rm = TRUE)

## plot density of all chl values
plot(d.all, ylim = c(0, 1.75), lwd = 2)

## add density of Chl from estimated track
lines(d.e, col = "red", lwd = 2)

## add density of Chl from filtered random tracks
lines(d.s, col = "blue", lwd = 2)

ianjonsen/foieGras documentation built on Jan. 17, 2025, 11:15 p.m.