`corr_check()` is an optional function for the `envar` package workflow. It allows you to check for variable correlation and multi-collinearity among the variables over the study area. It can be used before or after `extr_check()`.
Arguments
- x
A `SpatRaster`, `data.frame`, or a list containing `data` or `extracted_df` (e.g., output from `extr_check()`).
- pearson
Numeric or `NULL`. Threshold for the absolute Pearson correlation coefficient above which variables are flagged with a warning. By default (`NULL`) a default threshold of `0.7` is used; supply a value (e.g. `0.6`) to be warned about variable pairs whose absolute correlation exceeds it instead.
- vif
Numeric or `NULL`. Threshold for the Variance Inflation Factor above which variables are flagged with a warning. By default (`NULL`) a default threshold of `3` is used; supply a value (e.g. `5`) to be warned about variables whose VIF exceeds it instead.
Value
A `list` object containing:
`data`: The input environmental data used.
`correlation_matrix`: Pearson correlation matrix.
`vif`: Variance Inflation Factor data frame.
`summary`: Character vector highlighting high correlation or VIF, based on the supplied thresholds (or the defaults of `0.7` and `3`).
`plot_path`: Path to the saved correlation plot.
`vif_path`: Path to the saved VIF table.
Any additional elements from input list (e.g., `extrapolation` from `extr_check()`).
Details
Citation:
Wei T, Simko V (2021). "R package 'corrplot': Visualization of a Correlation Matrix." GitHub.
https://github.com/taiyun/corrplot
Naimi B, Hamm NA, Groen TA, Skidmore AK, Toxopeus AG (2014). "Where is positional uncertainty a problem for species distribution modelling?" Ecography 37, 191-203. https://doi.org/10.1111/j.1600-0587.2013.00205.x
Regardless of whether `pearson`/`vif` thresholds are set, the function writes two files: the correlation plot (`Corr_plot.png`) and a table of VIF values (`VIF_table.csv`). In an interactive R session it asks, at the console, for the directory in which to store them every time it is called. In non-interactive sessions (e.g. scripts or `R CMD check`) a temporary directory is used and no prompt is shown. Their paths are returned as `plot_path` and `vif_path`.
Examples
# corr_check() runs offline on the small example raster bundled with the
# package (a real WorldClim extract for Switzerland), so it needs no download:
switzerland <- terra::rast(
system.file("extdata", "switzerland.tif", package = "envar")
)
cc <- corr_check(switzerland)
cc$summary
# It also accepts a plain data.frame of predictor values:
set.seed(1)
example_df <- data.frame(bio1 = rnorm(40), bio12 = rnorm(40))
example_df$bio11 <- example_df$bio1 + rnorm(40, sd = 0.2) # correlated with bio1
corr_check(example_df)$summary
# \donttest{
# The typical workflow chains corr_check() after downloading variables.
# Example 1: Basic usage after environmental variable extraction
processed_bilayer_corr_check <- par_set(country = "Italy", crs=3035, buffer = 10) %>%
melc(vars=c("ice")) %>%
chelsa(vars=c("pr"), months= 12, year=2015) %>%
corr_check()
# Example 2: Chain with extr_check() (corr_check before extr_check)
result <- par_set(country = "Italy") %>%
chelsa(vars = c("bio1", "bio12")) %>%
corr_check() %>%
extr_check(calib_points = my_points)
# Example 3: Chain with extr_check() (extr_check before corr_check)
result <- par_set(country = "Italy") %>%
chelsa(vars = c("bio1", "bio12")) %>%
extr_check(calib_points = my_points) %>%
corr_check()
# Example 4: Custom thresholds for high correlation (>0.7) and VIF (>5)
result <- par_set(country = "Italy") %>%
chelsa(vars = c("bio1", "bio12")) %>%
corr_check(pearson = 0.7, vif = 5)
# }