citoMove

The goal of citoMove is to provide an user-friendly R package for analyzing movement data with DNNs.

Installation

The current version of citoMove depends on the development version of cito, which can be installed via:

# install.packages("devtools")
devtools::install_github('citoverse/cito')

You can install the development version of citoMove from GitHub via:

devtools::install_github("citoverse/citoMove")

Example - analysis of a deer movement track

set.seed(123)
library(amt)
#> 
#> Attaching package: 'amt'
#> The following object is masked from 'package:stats':
#> 
#>     filter
library(citoMove)
#> Loading required package: cito
library(terra)
#> terra 1.7.29

The following code creates a spatial covariate: the distance to the next forest.

forest <- amt::get_sh_forest()
forest <- terra::subst(forest, 0, NA)
forest_dist <- distance(forest)
names(forest_dist) <- "forest_dist"
plot(forest_dist)

Next, we load the tracking data of one red deer individual. From amt documentation: 826 GPS relocations of one red deer from northern Germany. The data is already resampled to a regular time interval of 6 hours and the coordinate reference system is transformed to epsg:3035.

The follwoing code orders the data, creates random absences, and extract the environmental predictor (distance to forest) create before for all true and random steps.

dat_ssf <- amt::deer |> 
  steps_by_burst() |> 
  random_steps() |> 
  extract_covariates(forest_dist) |> 
  time_of_day() |> 
  mutate(case_ = as.integer(case_))
#> Warning in random_steps.bursted_steps_xyt(steps_by_burst(amt::deer)): Some
#> bursts contain < 3 steps and will be removed

Now, we can fit the deep neural network (dnn) using citoMove:

model = dnn_ssf(case_ ~ forest_dist, data = dat_ssf, epoch = 30L, plot = T, verbose = FALSE)
#> Registered S3 methods overwritten by 'reformulas':
#>   method       from
#>   head.call    cito
#>   head.formula cito
#>   head.name    cito

The plot that is produced visualizes the improvement of the model during training. The default architecture of the neural networks entails two hidden layers. We can visualize the architecture via

plot(model)
#> Warning: Using the `size` aesthetic in this geom was deprecated in ggplot2 3.4.0.
#> ℹ Please use `linewidth` in the `default_aes` field and elsewhere instead.
#> ℹ The deprecated feature was likely used in the cito package.
#>   Please report the issue at <https://github.com/citoverse/cito/issues>.
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
#> generated.

A summary of the fitted model is provided via

summary(model)
#> Summary of Deep Neural Network Model
#> 
#> Feature Importance:
#>      variable importance_1
#> 1 forest_dist    0.1122311
#> 
#> Average Conditional Effects:
#>               Response_1
#> forest_dist 0.0004258009
#> 
#> Standard Deviation of Conditional Effects:
#>             Response_1
#> forest_dist  0.0081913

The cito package contains several options to create effect plots. Our current recommendation is to use accumulated local effect (ALE) plots to visualize the estimated selection effects

ALE(model)
#> Number of Neighborhoods reduced to 3