Compliance assessment oriented microcystin prediction: A Bayesian adaptive LASSO Tobit quantile regression approach

Abstract

Microcystin has been one of major contaminants impacting health of aquatic ecosystems and threatening human health. The development of drivers-microcystin relationship is of vital importance to microcystin management. However, current practices often focused on the mean response of microcystin concentration and cannot meet the requirement of percentile-based compliance assessment. Despite of many informative studies on the development of drivers–microcystin relationship, there remains a gap between the relationship development and the percentile-based compliance assessment of microcystin concentration. In this study, Bayesian adaptive LASSO Tobit quantile regression (BALTQR) model was introduced to environmental and ecological studies for the first time. The model is specially designed for the prediction of left-censored response variable. We applied the BALTQR model to develop the drivers–microcystin relationship of lakes across the US continent. Based on the results of parameters estimation, Chlorophyll a (CHL), pH, and water temperature (WT) were identified as key drivers to the microcystin concentration. We found that CHL was approximate the same important as pH and both of them had positive effects on the microcystin concentration at all the five regression quantiles. WT was relatively less important and had a surprisingly negative effect at the 0.7, 0.8, and 0.9 regression quantiles. We demonstrated that the BALTQR model successfully established the linkage between the development of drivers–microcystin relationship and the compliance assessment of microcystin concentration. We further revealed important implications of these findings to microcystin management. We believed that the BALTQR model has great potential of generalization to model other left-censored response variable in environmental and ecological studies.