Data-Driven Analysis of Forest–Climate Interactions in the Conterminous United States

• Main Authors: Olga Rumyantseva, Nikolay Strigul
• Format: Article
• Language: English
• Published: MDPI AG 2021-06-01
• Series: Climate
• Subjects: climate–vegetation interactions; data-intensive modeling; dimensionality reduction; forest inventories; multivariate statistics; machine learning
• Online Access: https://www.mdpi.com/2225-1154/9/7/108

A predictive understanding of interactions between vegetation and climate has been a grand challenge in terrestrial ecology for over 200 years. Developed in recent decades, continental-scale monitoring of climate and forest dynamics enables quantitative examination of vegetation–climate relationships through a data-driven paradigm. Here, we apply a data-intensive approach to investigate forest–climate interactions across the conterminous USA. We apply multivariate statistical methods (stepwise regression, principal component analysis) including machine learning to infer significant climatic drivers of standing forest basal area. We focus our analysis on the ecoregional scale. For most ecoregions analyzed, both stepwise regression and random forests indicate that factors related to precipitation are the most significant predictors of forest basal area. In almost half of US ecoregions, precipitation of the coldest quarter is the single most important driver of basal area. The demonstrated data-driven approach may be used to inform forest-climate envelope modeling and the forecasting of large-scale forest dynamics under climate change scenarios. These results have important implications for climate, biodiversity, industrial forestry, and indigenous communities in a changing world.