Skip to contents

Evaluate Landscape Metrics

Source: R/metrics_evaluation.R
evaluate_landscape_metrics.Rd

Identifies the metrics most suitable for discriminating between different pattern types based on a specified selection method. The choice of method affects the ranking: parametric methods assume linear relationships and normally distributed residuals, while non-parametric methods are more robust to outliers and deviations from normality. This function is useful for selecting informative metrics to train the metric-based neural network.

Usage

evaluate_landscape_metrics(
  metrics,
  metrics_number = 10,
  method = "kruskal_effsize",
  exclude_NA_metrics = TRUE,
  exclude_metrics = NULL,
  correlation_threshold = 0.7,
  verbose = FALSE
)

Arguments

metrics

tibble. Metrics from calculate_landscape_metrics().

metrics_number

Integer. Number of top metrics to return (default: 10).

method

Character. Selection method to use (default: "kruskal_effsize"). See 'Ranking Methods' section below for details.

exclude_NA_metrics

Logical. Whether to exclude metrics with NA values (default: TRUE). This is recommended if data is later used for model training as this does not accept missing values.

exclude_metrics

Character vector. Metrics to exclude (default: NULL).

correlation_threshold

Numeric. Maximum allowed correlation between selected metrics (default: 0.7). If you do not want to filter based on correlation, set to 1.

verbose

Logical. Whether to print detailed messages on excluded metrics or just a summary (default: FALSE).

Value

Character vector. Names of metrics that best discriminate between pattern types.

Ranking Methods

coeffvar_all

Coefficient of Variation (CV = SD/mean). Ranks metrics by their relative variability across landscapes. Higher CV indicates greater spread. Best for identifying metrics with high variability regardless of pattern type.

lin_mod_r2

Linear Model R-squared. Fits value ~ pattern for each metric and ranks by R². Higher values indicate better ability to predict pattern types. Assumes linear relationships and normally distributed residuals.

mean_groups

Mean Differences. Calculates relative differences between pattern-specific means and overall mean, then sums across patterns. Higher scores indicate better discrimination between pattern types.

fisher_score

Fisher Score (ratio of between-group to within-group variance). Higher scores indicate better separation between pattern types. Assumes normally distributed data within groups.

kruskal_effsize

Kruskal-Wallis H test effect sizes. Non-parametric test for differences between groups. Higher effect sizes indicate better discrimination between pattern types.

See also

train_nn_metrics

Other metrics: calculate_landscape_metrics()

Examples

# Calculate most suitable metrics to discriminate between spots and random landscapes
landscapes <- create_landscapes(n = 50, patterns = c("spots","random"))
#> Warning: Regular spot placement requested 10 spots but only ~8 positions fit.
#> ℹ  Adjusting to maximum feasible spots. Consider decreasing `spot_radius`.
#> ✔ Successfully generated all 50 training landscapes
metrics <- calculate_landscape_metrics(
  landscapes,
  level = "landscape"
)
#>  ■■■■■■                            17% |  ETA: 13s
#>  ■■■■■■■■■■■                       32% |  ETA: 12s
#>  ■■■■■■■■■■■■■                     41% |  ETA: 13s
#>  ■■■■■■■■■■■■■■■■                  50% |  ETA: 13s
#>  ■■■■■■■■■■■■■■■■■                 53% |  ETA: 14s
#>  ■■■■■■■■■■■■■■■■■■■■■■            70% |  ETA:  8s
#>  ■■■■■■■■■■■■■■■■■■■■■■■■■■■■      89% |  ETA:  2s
metric_list <- evaluate_landscape_metrics(
  metrics = metrics,
  metrics_number = 5,
  method = "coeffvar_all"
)
#> Warning: Excluded 300 rows containing 6 metrics with NA values. Metrics removed:
#> "enn_cv", "enn_mn", "enn_sd", "iji", "pafrac", and "rpr" Use
#> `exclude_NA_metrics = FALSE` to retain (not recommended for model training)
#> Warning: Excluded 3 metrics with zero variance: "pr", "prd", and "ta"