How To Utilize Relevance Vectors To Collect Required Data For Modeling Water Quality Constitu-ents, And Fine Sediment In Natural Systems? Case Study: Mud Lake, Idaho

Document Type


Journal/Book Title/Conference

Journal of Environmental Engineering






American Society of Civil Engineers

Publication Date



The development of monitoring programs for water quality and habitat assessment in surface waters is an ongoing challenge because of inherent difficulties in determining the effective spatial and temporal distribution of sites and trips. Recent advances in statistical learning theory, in which system characteristics are learned from data, point to the possibility of using the information content of data to shed light on monitoring results that provide sensitive and independent results. One of those techniques, multivariate relevance vector machines (MVRVM), creates as part of its algorithm subsets of a data set, called relevant vectors (RVs), that are most relevant for building an empirical model of a process, thereby revealing the information-rich portions of the data and potentially providing unambiguous information for management decision making. In this paper, we explore this concept in the context of a shallow lake, Mud Lake in SE Idaho, used as a sediment trap and as a wildlife refuge that is suffering from excessive sediment deposition and lack of data for water quality constituents that might affect the habitat for resident species. Data collected extensively in the lake over a 2-year period were used to train the MVRVM model and revealed that more than 40% of the observations contained little or no model-relevant or, by extension, management-relevant information. It is further demonstrated how RVs can help decision makers understand the practical problems of how much data are sufficient to support this class of model in order to tailor monitoring programs to efficiently maximize information.

This document is currently not available here.