Applying interpretable machine learning to assess intraspecific trait divergence under landscape-scale population...

Area of Science:

• Plant biology
• Ecology
• Machine learning

Background:

• Investigating intraspecific variation in functional traits is crucial for understanding plant adaptation.
• Sunflower (Helianthus annuus) populations exhibit significant divergence across contrasting ecoregions.

Purpose of the Study:

• To apply interpretable machine learning to identify functional traits predictive of ecoregion origin in Helianthus annuus.
• To understand the ecological strategies associated with trait divergence across different environments.

Main Methods:

• Utilized recursive feature elimination and the Boruta algorithm on functional trait data from the HeliantHOME database.
• Trained and validated Random Forest and Gradient Boosting Machine classifiers.
• Visualized results using accumulated local effects plots.

Main Results:

• Identified functional traits in leaf economics, plant architecture, reproductive phenology, and floral/seed morphology as most predictive of ecoregion.
• Desert genotypes showed shorter stature, fewer leaves, higher leaf nitrogen, and longer phyllaries compared to Great Plains genotypes.
• The machine learning approach successfully distinguished between sunflower populations from the Great Plains and North American Deserts.

Conclusions:

• Interpretable machine learning effectively identifies traits linked to contrasting ecological strategies within a species.
• This methodology can parse large plant trait datasets to explore adaptive divergence at intraspecific scales.