Smooth and thin data

This vignette shows how to smooth and thin WATLAS data.

# Packages
library(tools4watlas)
library(ggplot2)

# Path to csv with filtered data
data_path <- system.file(
  "extdata", "watlas_data_filtered.csv",
  package = "tools4watlas"
)

# Load data
data <- fread(data_path, yaml = TRUE)

Median smooth data

To reduce error in the position data, a basic smoother such as a median filter can be applied. the function atl_median_smooth() calculates the median coordinates within a window of positions set by moving window.

# Smooth the data
data <- atl_median_smooth(data, moving_window = 5)

The resulting table overwrites the smoothed coordinates in the columns x and y and keeps the original ones in the columns x_raw and y_raw.

Calculate speed and turning angle

After median filtering the data, the speeds need to be recalculated. We will also calculate turning angles.

Note: the distance between median smoothed positions can be 0 and therefore will produce NAs and a warning

# Recalculate speed
data <- atl_get_speed(data, type = c("in", "out"))

Look at the data

This plot shows one example of a raw (black) and median smoothed (red) track.

# subset first tag
data_subset <- data[tag == data[1]$tag]

# subset some data to look at
from <- min(data_subset[, datetime]) + 1 * 3600
to <- min(data_subset[, datetime]) + 12 * 3600
data_subset <- data_subset[datetime %between% c(from, to)]

# Create basemap
bm <- atl_create_bm(data_subset)

# Plot
bm +
  geom_path(
    data = data_subset, aes(x_raw, y_raw),
    color = "firebrick3", linewidth = 0.5
  ) +
  geom_path(
    data = data_subset, aes(x, y),
    color = "black", linewidth = 0.5
  ) +
  geom_point(
    data = data_subset, aes(x_raw, y_raw),
    color = "firebrick3", size = 1.2
  ) +
  geom_point(
    data = data_subset, aes(x, y),
    color = "black", size = 1
  )

Smoothed track (black) on top of raw track (red)

Save data for the next steps

# Save data
fwrite(
  data,
  file = "../inst/extdata/watlas_data_smoothed.csv", yaml = TRUE
)

Thin data

Depending on the desired analysis, it might make sense to thin data, either by aggregation or by sub-sampling. Both methods return fixed time steps (depending on the interval).

By aggregation

Returns the mean of all columns for each time step. The additional column n_aggregated shows how many positions were aggregated for this position. Time and datetime are returned rounded down to the desired interval (in seconds).

# Thin the data by aggregation with a 60-second interval
thinned_aggregated <- atl_thin_data(
  data = data,
  interval = 60,
  id_columns = c("tag", "species"),
  method = "aggregate"
)

# Show head of selected data
head(thinned_aggregated[, .(tag, time, datetime, x, y, n_aggregated)]) |>
  knitr::kable(digits = 2)

tag	time	datetime	x	y	n_aggregated
3027	1695438780	2023-09-23 03:13:00	650705.6	5902556	3
3027	1695439140	2023-09-23 03:19:00	650722.1	5902562	4
3027	1695439200	2023-09-23 03:20:00	650712.0	5902563	10
3027	1695439260	2023-09-23 03:21:00	650702.9	5902562	1
3027	1695439440	2023-09-23 03:24:00	650705.2	5902576	6
3027	1695439500	2023-09-23 03:25:00	650700.1	5902562	17

By subsampling

Returns the first position for each time step. The column n_subsampled shows from how many positions this position was sampled.

# Thin the data by subsampling with a 60-second interval
thinned_subsampled <- atl_thin_data(
  data = data,
  interval = 60,
  id_columns = c("tag", "species"),
  method = "subsample"
)

# Show head of selected data
head(thinned_subsampled[, .(tag, time, datetime, x, y, n_subsampled)]) |>
  knitr::kable(digits = 2)

tag	time	datetime	x	y	n_subsampled
3027	1695438802	2023-09-23 03:13:22	650705.6	5902556	3
3027	1695439189	2023-09-23 03:19:49	650721.0	5902559	4
3027	1695439201	2023-09-23 03:20:01	650723.1	5902564	10
3027	1695439261	2023-09-23 03:21:01	650702.9	5902562	1
3027	1695439477	2023-09-23 03:24:37	650702.8	5902562	6
3027	1695439501	2023-09-23 03:25:01	650709.9	5902598	17