Predicting the likelihood of water quality impaired stream segments using landscape-scale data and a hierarchical methodology

Abstract

The purpose of my dissertation research was to develop a methodology that can be used by states and tribes to comply with two requirements of the Clean Water Act (1972): 1) to obtain an estimate of regional water quality condition and 2) to identify water quality impaired stream segments. To meet this goal, I developed a predictive model that incorporated spatial autocorrelation and combined coarse-scale geographic information system (GIS) data, such as mean elevation, with data collected using field-survey methods. Geostatistical models are typically based on straight-line distance (SLD), which fails to represent the spatial configuration, connectivity, and directionality of sites in a stream network and may not be ecologically valid for studies in freshwater streams. Instead, hydrologic distances may represent the transfer of organisms, material, and energy through stream networks more accurately. I developed GIS tools that generate the spatial data necessary for geostatistical modeling in stream networks. I quantified patterns of spatial autocorrelation in pH, conductivity, nitrate, sulfate, acid neutralizing capacity, dissolved organic carbon (DOC), temperature, and dissolved oxygen using three distance measures: SLD, symmetric hydrologic distance, and weighted asymmetric hydrologic distance. My results indicated that spatial autocorrelation exists in stream chemistry data at a relatively coarse scale and that geostatistical models consistently improved the accuracy of model predictions. SLD appeared to be the most suitable distance measure for regional geostatistical modeling of water chemistry in Maryland due to the extensive pre-processing time required for hydrologic distance measures and the inability of the survey design to adequately represent hydrologic relationships in a stream network. I developed a geostatistical model and used it to predict DOC at 3083 unobserved stream segments in Maryland. DOC estimates were categorized using ecological thresholds and reported in kilometers. The predictions and prediction variances were displayed using a GIS, which provided a simple way to recognize and communicate regional patterns in DOC. This methodology has clear advantages related to regional water quality monitoring because additional field sampling is not necessary, inferences about regional stream condition are generated, and it can be used to locate potentially impaired segments in a rapid and cost-efficient manner.