Modeling wetland plant metrics to improve the performance of vegetation-based indices of biological integrity
The objective of this study was to determine if the accuracy and precision of wetland plant indices of biological integrity (IBIs) could be improved through the use of modeling techniques. To do this, we developed a modeled vegetation IBI (MVIBI) based on metrics previously used to develop vegetation indices of biological integrity (VIBIs) for Ohio wetlands (e.g. % invasive grass, % sensitive species, shrub richness). We selected 82 emergent, forested, and shrub-dominated reference sites distributed across the State of Ohio and built Random Forest models to predict plant metric scores at reference wetlands from naturally occurring environmental features related to climate, hydrology, geology, soils, and landscape position. The models explained between 14 and 52% of the variance in the scores of 21 metrics indicating that variation in wetland plant assemblages was significantly associated with naturally occurring environmental gradients. We used principal component analysis to identify ten groups of statistically independent metrics and selected one metric from each group that discriminated most strongly between reference and most degraded sites based on t-scores. Two axes did not contain discriminating metrics so we used eight metrics in the MVIBI. Analysis of variance of reference site MVIBI scores indicated that we could use one distribution of reference site scores to assess multiple wetland types, thus eliminating the need to separately designate wetland types. We used the MVIBI to assess 170 test sites and compared the accuracy, precision, responsiveness, and sensitivity of the MVIBI to those of the original VIBIs. The MVIBI was up to twice as accurate and precise as the original VIBIs, indicating that modeling can be used to improve the performance of vegetation-based IBIs. The use of model-based IBIs for wetland plants should reduce assessment errors associated with natural variation in plant metrics and should increase confidence in wetland assessments.