Experiments

Remarks

Because the fundamental importance of area and, closely related, climate niche fraction, shall be tested, species tested under criterion B were dropped

The indicators evaluated in each experiment are:

• Model performance TSS, Sensitivity + Specificity
• Variable importance scores for sets of selected variables with low corellation
• Partial dependence 1D, 2D
• Raw data distribution of low/high class for all variables, 2D plots
• For categorical destinctions, possibly train a submodel

Importance of climate compared to other variables

To test the relative importance of climate change variables to others (e.g. landuse) and umong each other (e.g. precipitation vs. temperature) I evaluate the abovementioned indicators following indicators in different experimental settings

Experiments

1. Select the following variables:
2. Run 100 supsampled (or upsampled?) RF models, report statistics from above.
3. Model without/with threats, test dependence on Realm, Habitat and Biological caracteristica (part. dep or supmodels)

Are climate niches a good paradigm for assessing species extinction risk?

1. Train one model with climate niche only and one with climate change only - use several competing models.
2. Look into the relationship between climate niche and area (make more concrete)