Total phosphorus inference models and indices for coastal plain streams based on benthic diatom assemblages from artificial substrates

We investigated the potential for using diatoms to monitor and assess nutrient enrichment in coastal plain streams using weighted-averaging inference models and diatom trophic indices. Samples were collected from low-gradient, clay- to sand-bottom streams in New Jersey (NJ), USA, using artificial substrates (diatometers). Multivariate analysis showed that conductivity was overall the most important explanatory variable. Total phosphorus (TP) explained a significant proportion of the variation in diatom species composition. There was statistical justification for development of inference models for TP but not for total nitrogen (TN). We developed and tested models for inferring TP using weighted-averaging (WA) and weighted-averaging partial least squares (WA-PLS) regression and calibration techniques. We also created a diatom TP index by rescaling the inferred TP values. WA-PLS provided the best model (n = 38), which showed moderate predictive ability (r boot 2 = 0.43; RMSEPboot = 0.30 log10 μg l−1 TP); it performed best at lower TP concentrations and tended to underestimate values above 100 μg l−1. The TP index performed well; it assigned the majority of the index scores to the correct nutrient category. TP models and indices developed for the Coastal Plain had lower predictive ability than those developed for northern NJ and streams in other comparable geographic regions of the US. This lower performance can be attributed primarily to a data gap in the TP gradient in the calibration dataset (lack of sites with TP concentrations between 240 and 560 μg l−1), and a smaller number of samples. We conclude that diatom-based TP inference models and artificial substrate sampling are useful for assessing and monitoring nutrient enrichment in coastal plain streams. Given the worldwide distribution of streams similar to those in this study, these tools should be widely applicable.