Improvements in multi-metric index development using a whole-index approach

Multi-metric indices (MMIs) are the most common tool used to assess ecological integrity. Traditionally, only the metrics with the highest correlation to disturbance are incorporated. The rationale being that combining the metrics with the strongest individual relationships to disturbance should yield an MMI with the strongest possible relationship to disturbance. An alternative to this traditional metric-based method was proposed, which we term the index-based approach. Under this method, thousands of MMIs are generated through the random selection of candidate metrics and then a model competition framework is adopted to select the best performing MMI. An index-based approach may prove more sensitive to disturbance because even metrics weakly related to disturbance on their own may nonetheless contribute unique and valuable information to the whole MMI. Our goal was to compare the traditional metric-based method with the novel index-based approach to MMI development and validation to ascertain which generates a superior MMI for vegetation-based biomonitoring in prairie pothole wetlands. We sampled vegetation from 72 marshes from across prairie Alberta between 2014 and 2015. These ranged from temporarily- to permanently-ponded marsh classes and spanned a gradient in anthropogenic disturbance. Vegetation surveys yielded 732 initial metrics associated with abundance, richness and traits, 88 of which were reasonably sensitive to disturbance ( p < 0.1). We developed a four-metric MMI using the metric-based method and successfully validated it on an independent set of wetlands ( R 2 = 0.2518, F 1,22 = 7.404, p = 0.012). We also generated 50,000 four-, six-, and eight-metric MMIs using the index-based approach and contrasted these in a model competition framework. The best-fitting four-metric MMI developed using the index-based approach was also successfully validated ( R 2 = 0.2706, F 1,22 = 8.163, p = 0.009). It provided the best fit (AICc = 754.93) and was superior to the metric-based MMI (AICc weight = 0.97). We conclude that the index-based approach can yield a superior biomonitoring tool and recommend that it be adopted as the standard method for MMI creation.