A mixed‐model moving‐average approach to geostatistical modeling in stream networks

• DOI: 10.1890/08-1668.1
• Publication Date: 2010-04-09T20:47:22Z
• Authors: Erin E. Peterson; Jay M. Ver Hoef
• Keywords: Autocovariance; Mixed model; Geostatistics; Matérn covariance function

Spatial autocorrelation is an intrinsic characteristic in freshwater stream environments where nested watersheds and flow connectivity may produce patterns that are not captured by Euclidean distance. Yet, many common autocovariance functions used geostatistical models statistically invalid when distance replaced with hydrologic We use simple worked examples to illustrate a recently developed moving‐average approach construct two types of valid based on distances. These were designed represent the spatial configuration, longitudinal connectivity, discharge, direction network. They also exhibit different covariance structure than true difference way relationships represented. Nevertheless, multi‐scale complexities be fully using model one structure. advocate variance component approach, which allows mixture (Euclidean models) incorporated into single model. As example, we fit compare “mixed models,” multiple structures, for biological indicator. The mixed proves flexible because sources information can