A geostatistical approach to optimize water quality monitoring networks in large lakes: Application to Lake Winnipeg

摘要

Water quality monitoring in lakes is often done by regularly sampling water quality variables of interest at fixed stations. Station location is a critical component of monitoring network design. Geostatistical methods were used to quantify redundancy in an intentionally dense network of lake stations on Lake Winnipeg, Canada. Water isotope samples (δ~2H, δ~(18)O) were collected approximately every 1-2 km during research cruises in September-October 2009, for a total of 240 lakewide stations. Two statistical approaches were used to assess redundancy: 1) kriging, a spatial interpolation technique, was performed in a reduced multivariate space and kriging variance was used to assess the suitability of the sampled network configuration; and 2) Local Moran's I values were calculated. Moran's 1 identified clusters of stations that were similar or different. Good kriging models were developed for both δ~2H and δ~(18)O. When assessed individually, a large number of stations were identified as redundant but redundancy was based on the premise that all other stations remained in the network. When the analysis was performed on clusters of stations, within each cluster, up to four stations could be removed without significant loss of information. Relationships of redundancy were confirmed by Local Moran's I values. In combination, these techniques identified stations that were statistically important or redundant. This study emphasizes the importance of completing the evaluation of information provided by an individual station with information from clusters of stations within a network.