Modeling regional forest site productivity accounting spatial structure in climatic and edaphic variables

With increasing interest in sustaining productivity amid changing climate, disturbance regimes, and management practices, an accurate forest productivity estimate is important to develop sustainable management regimes. Our goal was to estimate and map the potential productivity of co-occurring tree species. We used forest inventory and analysis (FIA) data and climatic and edaphic variables to model the composite site index (CSI) as a proxy of potential forest productivity. Initially, we identified the site index model for selected species: slash pine (Pinus elliottii), longleaf pine (Pinus palustris), loblolly pine (Pinus taeda), and yellow poplar (Liriodendron tulipifera). We then standardized species-specific site index (SI) values to generate composite site index (CSI) values. Finally, we used a random forest (RF) machine learning algorithm (ML) to predict CSI values based on climatic and edaphic factors while addressing spatial dependencies in the data set. The RF model explained 81 % of the variation (R²_adj = 0.81), with a mean bias of 0.11 m and a mean absolute error (MAE) of 3.37 m. The accuracy of modeling and mapping forest productivity using CSI depends on the quality and spatial distribution of national forest inventory data at the species level and climatic information. We recommend modeling forest productivity that accounts for spatial structure in the data to reduce overinflation of overall accuracy.