You are here:
Spatial Statistical Network Models for Stream and River Temperature in the Chesapeake Bay Watershed, USA
Citation:
Detenbeck, N. AND A. Morrison. Spatial Statistical Network Models for Stream and River Temperature in the Chesapeake Bay Watershed, USA. Northeast Arc User Group (NEARC), Amherst, MA, May 09, 2016.
Impact/Purpose:
Our goal for this study is to develop predictive models describing the thermal regime of streams across the Chesapeake Bay Watershed using landscape attributes as well as meteorological variables.
Description:
Regional temperature models are needed for characterizing and mapping stream thermal regimes, establishing reference conditions, predicting future impacts and identifying critical thermal refugia. Spatial statistical models have been developed to improve regression modeling techniques by taking into account spatial covariance structures inherent in stream networks. Unlike earlier approaches, spatial statistical models describe spatial autocorrelation based on distance along flow networks as well as standard Euclidean distances between observation points. Our goal for this study is to develop predictive models describing the thermal regime of streams across the Chesapeake Bay Watershed using landscape attributes as well as meteorological variables. We developed a database of Chesapeake Bay monitoring stations using existing stream temperature time series data obtained from state, federal, and nongovernmental sources. After applying quality control criteria, we described thermal regimes of Chesapeake Bay rivers and streams based on a set of reduced metrics chosen through principal component analysis. We used a variety of GIS tools to develop parameters used as independent variables in our spatial statistical models. In addition, we utilized GIS-based tools to facilitate watershed processing. Landscape networks were developed using the NHDPlus hydrographic dataset and the STARS (Spatial Tools for the Analysis of Rivers) geoprocessing toolset.We predicted monthly median July stream temperatures as a function of air temperature, drainage density, land cover, main channel slope, watershed storage, percent coarse-grained surficial deposits, stream flow and velocity and watershed area, with an overall root-mean-square prediction error of 1.5○ C. Predictive models for the remaining variables are being developed.