Predicting biological impairment from habitat assessments.

The goal of biological monitoring programs is to determine impairment classification and identify local stressors. Biological monitoring performs well at detecting impairment but when used alone falls short of determining the cause of the impairment. Following detection a more thorough survey is often conducted using extensive biological, chemical, and physical analysis coupled with exhaustive statistical treatments. These methods can be prohibitive for small programs that are limited by time and budget. The objective of this study was to develop a simple and useful model to predict the probability of biological impairment based on routinely collected habitat assessments. Biological communities were assessed with the Index of Biotic Integrity (IBI), and habitat was assessed with the Qualitative Habitat Evaluation Index. Two models were constructed from a validation dataset. The first predicted a binary outcome of impaired (IBI < 35) or non-impaired (IBI ≥ 35) and the second predicted a categorical gradient of impairment. Categories include very poor, poor, fair, good, and excellent. The models were then validated with an independently collected dataset. Both models successfully predicted biological integrity of the validation dataset with an accuracy of 0.84 (binary) and 0.75 (categorical). Based on the binary outcome model, 22 sites were observed to be impaired while the model predicted them to not be impaired. The categorical model misclassified 47 samples while only seven of those were misclassified by two or more categories. The impairment source was subsequently identified by known stressors. The models developed here can be easily applied to other datasets from the Eastern Corn Belt Plain to aid in stressor identification by predicting the probability of observing an impaired fish community based on habitat. Predicted probabilities from the models can also be used to support conclusions that have already been determined.