Evaluating performance of macroinvertebrate-based adjusted and unadjusted multi-metric indices (MMI) using multi-season and multi-year samples

Multimetric indices (MMI) are extremely useful indicators of anthropogenic disturbance, but that usefulness is a function of how well they have been calibrated for natural spatial and temporal variability. Nonetheless, there have been very few studies regarding the effects of such natural variability on MMI scores and metric selection, and even fewer studies on the effects of minimizing the mean of metric correlations. To do so, we calibrated and validated several adjusted and unadjusted MMI for natural gradients with different metric compositions via datasets from 161 samples collected at 83 stream sites from 2008 to 2012 across the Li River Basin, China. We used random forests to model the variation of individual metrics with natural gradients, then used their residual distributions as the adjusted response range of each metric, independent of natural environmental influence. To select core metrics for MMI, the candidate adjusted and unadjusted metrics were all screened through reproducibility, sensitivity, and redundancy tests. The adjusted MMI were more precise, less biased, and more sensitive than the unadjusted MMI. Although the unadjusted MMI were as responsive as the adjusted MMI, the unadjusted MMI increased risks of inferring type I and type II errors. Minimizing the mean of metric correlations as well as avoiding biological information redundancy improved the performance of unadjusted and adjusted MMI, but only improved precision for the adjusted MMI. Using multi-season and multi-year samples proved to be an effective and efficient assessment method to diminish natural seasonal variation in adjusted MMI, but not unadjusted MMI. We found that the adjusted MMI with the lowest mean of metric correlations using multi-season and multi-year samples performed best and reduced seasonal variability.