Qualifying Predictor Variable Data

All predictor variables should be researched to obtain as much information as possible on how they were collected and their uncertainty. For some predictors, such as climate layers, this may include working back through a number of processing steps to determine the original collection method.

Detecting Uncertainty

Processing Artifacts

Some data have artifacts introduced by processing. High frequency patterns can be determined by applying a high-pass filter (hill shading is one approach).

Correlation Between Predictors

Correlation should be checked between each combination of predictors. Predictors that have strong correlation (such as min/max temperature) should be eliminated.

• Pearson's for data that may correlate linearly
• Spearman's for data that may correlate non-linearly