Predicting understory vegetation structure in selected western forests of the United States using FIA inventory data

Abstract

Understory vegetation structure and its relationship with forest canopies and site conditions are important determinants of carbon stocks, wildlife habitat, and fuel loading for wildland fire assessments. Comprehensive studies are needed to assess these relationships through the use of consistently collected field-based data. One approach to achieve this is to make use of preexisting forest inventory data to estimate understory vegetation height and cover from site and overstory attributes. In this study, overstory, understory, and abiotic data describing site conditions were obtained from over 6700 Forest Inventory and Analysis (FIA) fixed radius plots collected between 2000 and 2012 to assess how understory vegetation cover and height vary with overstory attributes and site characteristics. The focus was restricted to four common forest types including lodgepole pine (Pinus contorta var. latifolia), Douglas-fir (Pseudotsuga menziesii), ponderosa pine (Pinus ponderosa), and grand fir (Abies grandis) found on approximately 43 million hectares in the western United States. Random Forest regression classification trees were developed for cover and height of shrub and herb understories as a function of field-measured predictor variables. Separate analyses were undertaken for the Pacific Northwest (PNW) and the Interior West (IW) Forest Inventory and Analysis (FIA) regions. Models developed from the IW data generally performed better and the OOB (out-of-bag) percent variance explained varied from 8.08% for forb height to 39.24% for shrub height. For the PNW data, percent variance explained ranged from 13.82% for forb height to 27.4% for shrub height. Percent variance explained values were higher in all corresponding models for the IW than PNW, except for forb and grass height. Differences in model performance were smallest in the case of forb cover (27.17% vs. 26.15%) and greatest in the case of percent shrub cover (30.92% vs. 15.53%) for IW and PNW models, respectively. Cover models within each dataset performed better, on average, than their associated height models. The most influential variables for predicting understory cover and height were ones representing overstory conditions and conform to ecological expectation corroborated by many studies examining the influence of forest overstories on understory vegetation dynamics. Several variables, including aspect, slope, and stand disturbance and treatment, were not important and contrary to expectation. Predicting understory vegetation attributes to aid assessments of carbon, fuel, and wildlife habitat may be more generalizable across forests of the western U.S. using standardized national inventory data in conjunction with improved measurements.