Ecohydrological model parameter selection for stream health evaluation

Sean A Woznicki; A Pouyan Nejadhashemi; Dennis M Ross; Zhen Zhang; Lizhu Wang; Abdol-Hossein Esfahanian

doi:10.1016/j.scitotenv.2014.12.066

Ecohydrological model parameter selection for stream health evaluation

Sci Total Environ. 2015 Apr 1:511:341-53. doi: 10.1016/j.scitotenv.2014.12.066. Epub 2014 Dec 29.

Authors

Sean A Woznicki¹, A Pouyan Nejadhashemi², Dennis M Ross³, Zhen Zhang⁴, Lizhu Wang⁵, Abdol-Hossein Esfahanian³

Affiliations

¹ Department of Biosystems and Agricultural Engineering, 524 S. Shaw Lane, Room 216, Michigan State University, East Lansing, MI 48824, USA.
² Department of Biosystems and Agricultural Engineering, 524 S. Shaw Lane, Room 216, Michigan State University, East Lansing, MI 48824, USA. Electronic address: pouyan@msu.edu.
³ Department of Computer Science and Engineering, Michigan State University, East Lansing, MI 48824, USA.
⁴ Physical Sciences Division, Department of Statistics, University of Chicago, Chicago, IL 60637, USA.
⁵ International Joint Commission, Great Lakes Office, Windsor, ON N9A 6T3, Canada.

PMID: 25553548
DOI: 10.1016/j.scitotenv.2014.12.066

Abstract

Variable selection is a critical step in development of empirical stream health prediction models. This study develops a framework for selecting important in-stream variables to predict four measures of biological integrity: total number of Ephemeroptera, Plecoptera, and Trichoptera (EPT) taxa, family index of biotic integrity (FIBI), Hilsenhoff biotic integrity (HBI), and fish index of biotic integrity (IBI). Over 200 flow regime and water quality variables were calculated using the Hydrologic Index Tool (HIT) and Soil and Water Assessment Tool (SWAT). Streams of the River Raisin watershed in Michigan were grouped using the Strahler stream classification system (orders 1-3 and orders 4-6), k-means clustering technique (two clusters: C1 and C2), and all streams (one grouping). For each grouping, variable selection was performed using Bayesian variable selection, principal component analysis, and Spearman's rank correlation. Following selection of best variable sets, models were developed to predict the measures of biological integrity using adaptive-neuro fuzzy inference systems (ANFIS), a technique well-suited to complex, nonlinear ecological problems. Multiple unique variable sets were identified, all which differed by selection method and stream grouping. Final best models were mostly built using the Bayesian variable selection method. The most effective stream grouping method varied by health measure, although k-means clustering and grouping by stream order were always superior to models built without grouping. Commonly selected variables were related to streamflow magnitude, rate of change, and seasonal nitrate concentration. Each best model was effective in simulating stream health observations, with EPT taxa validation R2 ranging from 0.67 to 0.92, FIBI ranging from 0.49 to 0.85, HBI from 0.56 to 0.75, and fish IBI at 0.99 for all best models. The comprehensive variable selection and modeling process proposed here is a robust method that extends our understanding of watershed scale stream health beyond sparse monitoring points.

Keywords: Biological Integrity; Clustering; Fish; Fuzzy logic; Macroinvertebrate; SWAT.

Publication types

Research Support, U.S. Gov't, Non-P.H.S.

MeSH terms

Animals
Bayes Theorem
Biodiversity
Ecosystem
Environmental Monitoring / methods*
Fishes
Hydrology
Insecta
Michigan
Models, Theoretical
Rivers*
Water Pollution / analysis
Water Pollution / statistics & numerical data*