Measuring the biotic integrity of stream ecosystems with restoration

Robin M. Henderson

One goal of my research was to determine whether the traditional River InVertebrate Prediction and Classification System (RIVPACS) framework for developing Observed/Expected (O/E) indices from Discriminant Function Analysis (DFA) could be improved upon by using newer statistical methods like random forests (RF) with more ecologically-relevant predictor variables. I used benthic macroinvertebrate data to develop classification models for 105 reference sites and 414 test sites and compared overall model performance (i.e., accuracy, precision, bias, responsiveness, sensitivity to stressors) to determine which would accurately portray biotic integrity. I determined that RF models have greater crossvalidation accuracy compared to DFA models (82% to 75%; 62% to 53%, respectively) and greater precision (RF r 2 = 0.77, 0.65; DFA r 2 = 0.52, 0.59). Index precision varied (SD = 0.13-0.16), however the RF 4 group model produced SD = 0.13. Additionally, responsiveness varied (t = 7.0-10.9) as did model sensitivity (31% - 50% in most disturbed condition), with DFA models typically outperforming the RF models. The final model selected, based on the above metrics, was the 4 group RF model. The coefficient of variation (CV) was then calculated for the final O/E index and 9 other biotic indices (richness measures, Hilsenhoff's biotic index, and Simpson's diversity index [SDI]) for 14 reference sites to determine which index provides the least variable method of predicting biotic v integrity. The SDI and the O/E index had the lowest CV; however, the SDI had a greater range of CV values indicating that the O/E index provides a less variable method of predicting biotic integrity for the region. To test if the Middle Fork John Day River (MFJDR) sites have improved biotic integrity scores following restoration, t-tests and box and whisker plots were utilized to detect differences in the O/E scores. Results indicate that there is not yet evidence of improvement in biotic integrity scores for restored MFJDR sites. Since predictor variables used to develop the O/E index provide insight into the niche axes that partition taxa among sites, measures of ecosystem integrity, metabolism, and functional processes should be incorporated in models designed to monitor the effectiveness of restoring biotic integrity

Measuring the biotic integrity of stream ecosystems with restoration

Files and links (1)

Abstract

Metrics

Details