Variable importance for sustaining macrophyte presence via random forests: data imputation and model settings.

Area of Science:

• Ecology
• Data Mining
• Environmental Modeling

Background:

• Missing data and model parameters commonly affect data mining and ecological models.
• Habitat suitability models are crucial for understanding plant distribution.

Purpose of the Study:

• To assess the impact of data preprocessing techniques on random forest habitat suitability models.
• To investigate the influence of random forest parameter tuning on model performance for macrophytes.

Main Methods:

• Utilized random forests for habitat suitability modeling of three macrophyte species.
• Evaluated effects of data reduction, outlier removal, correlated variable removal, and median imputation.
• Tuned random forest parameters, specifically the number of trees (ntree).

Main Results:

• Higher model performance (Cohen's kappa ~0.2-0.3) was achieved with significant data reduction and median imputation.
• Outlier or correlated variable removal did not improve model performance.
• Temperature, pH, and conductivity were key predictors for macrophyte presence.

Conclusions:

• Data reduction and median imputation are effective strategies for improving macrophyte habitat suitability models.
• Optimal macrophyte presence is associated with high temperatures, intermediate conductivity, and neutral pH.
• A multivariate management approach is recommended for supporting macrophyte presence, pending further validation.