findScale: Find an appropriate smoothing parameter

In track2KBA: Identifying Important Areas from Animal Tracking Data

Find an appropriate smoothing parameter

Description

findScale takes a tracking data set and outputs a series of candidate smoothing parameter values. Additionally, it compares the scale of movement resolved by the sampling resolution of the data set, to a grid of desired resolution.

Usage

findScale(
  tracks,
  scaleARS = TRUE,
  res = NULL,
  sumTrips = NULL,
  scalesFPT = NULL,
  peakWidth = 1,
  peakMethod = "first"
)

Arguments

tracks	SpatialPointsDataFrame. Must be projected into an equal-area coordinate system; if not, first run projectTracks.
scaleARS	logical scalar (TRUE/FALSE). Do you want to calculate the scale of area-restricted search using First Passage Time analysis? NOTE: does not allow for duplicate date-time stamps.
res	numeric. The desired grid cell resolution (square kilometers) for subsequent kernel analysis (NOT performed in this function). If this is not specified, the scale of movement is compared to a 500-cell grid, with spatial extent determined by the latitudinal and longitudinal extent of the data.
sumTrips	data.frame. Output of tripSummary function. If not specified, tripSummary will be called within the function.
scalesFPT	numeric vector. Set of spatial scales at which to calculate First Passage Time. If not specified, the distribution of between-point distances will be used to derive a set.
peakWidth	numeric. How many scale-steps either side of focal scale used to identify a peak. Default is 1, whereby a peak is defined as any scale at which the variance in log FPT increases from the previous scale, and decreases for the following one.
peakMethod	character. Which method should be used to select the focal peak for each ID. Options are "first", "max", and "steep". "steep" is a value of scalesFPT at which the variance in log FPT changes most rapidly compared to the surrounding scale(s).

Details

The purpose of this function is to provide guidance regarding the two most sensitive steps in the track2KBA analysis: specification of the (1) smoothing parameter and the (2) grid cell size for kernel density estimation (KDE). Specifically, the goal is to allow for exploration of the effect of these parameters and their inter-relatedness, so that an informed decision may be made regarding their specification in subsequent track2KBA steps.

Kernel density estimation has been identified as particularly sensitive to the specification of the smoothing parameter (AKA bandwidth, or 'H' value), that is, the parameter that defines the width of the normal distribution around each location. Many techniques for identifying 'optimal' smoothing parameters have been proposed (see Gitzen, Millspaugh, and Kernohan for a classic review; see Fleming and Calabreses 2017 for a later implementation) and many of these techniques have their merits; however, in the track2KBA implementation of KDE we have opted for simplicity.

In the context of the track2KBA analysis, the smoothing parameter ought to represent the relevant scale at which the animal interacts with the environment. Therefore, when selecting a Scale value for subsequent analysis, the user must take into account the movement ecology of the study species. For species which use Area-Restricted Search (ARS) behavior when foraging, First Passage Time analysis may be used to identify the scale of interaction with the environment (Fauchald and Tveraa 2003), however not all species use ARS when foraging and therefore different techniques must be used.

What minimum spatial scales are detectable by the data also depends on the sampling resolution. Therefore, when applying First Passage Time analysis, findScale sets the range of scales at which movements are analyzed based on the distribution of forward, between-point displacements in the data.

The grid cell size also affects the output of kernel density-based space use analyses. Therefore, by specifying the res parameter you can check whether your desired grid cell size is reasonable, given the scale of movement resolved by your data.

Value

This function returns a one-row dataframe with the foraging range in the first column (i.e. 'med_max_distance') calculated by tripSummary, and the median step length (i.e. between point distance) for the data set. The subsequent columns contain various candidate smoothing parameter ('h') values calculated in the following ways:

1. 'mag' - log of the foraging range (i.e. median maximum trip distance)

2. 'href' - reference bandwidth a simple, data-driven method which takes into account the number of points, and the variance in X and Y directions.

   sqrt((X + Y)* (n^(-1/6))); where X=Longitude/Easting, Y=Latitude/Northing, and n=number of relocations

3. 'scaleARS' - spatial scale of area-restricted Search behavior as estimated using First Passage Time analysis (see fpt)

If the scaleARS option is used, a diagnostic plot is shown which illustrates the change in variance of log-FPT values calculated at each FPT scale. Grey vertical lines indicate the peaks identified for each individual using peakMethod method chosen, and the red line is the median of these, and the resulting scaleARS in the output table.

All values are in kilometers.

Examples

## make some play data
dataGroup <- data.frame(Longitude = c(1, 1.01, 1.02, 1.04, 1.05, 1),
  Latitude =  c(1, 1.01, 1.02, 1.03, 1.021, 1), 
  ID = rep("A", 6),
  DateTime = format(
   lubridate::ymd_hms("2021-01-01 00:00:00") + 
   lubridate::hours(0:5)
   )
 )
 colony <- data.frame(
  Longitude = dataGroup$Longitude[1], Latitude = dataGroup$Latitude[1]
 )
 ## split data into trips
 trips <- tripSplit(dataGroup, colony=colony,
  innerBuff = 1, returnBuff = 1, duration = 0.5, 
  rmNonTrip = TRUE
 )  
 ## summarize trip characteristics
 sumTrips <- tripSummary(trips, colony)
 ## project tracks
 tracks_prj <- projectTracks(
   trips,
   projType = "azim",
   custom = "TRUE"
 )
 ## calculate candidate smoothing parameter values
 h_vals <- findScale(tracks_prj, sumTrips = sumTrips, scaleARS = FALSE)

track2KBA documentation built on Sept. 27, 2023, 5:08 p.m.