A neighborhood statistics model for predicting stream pathogen indicator levels

Pramod Pandey, Gregory B. Pasternack, Mahbubul Majumder, Michelle L. Soupir, Mark S. Kaiser

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Because elevated levels of water-borne Escherichia coli in streams are a leading cause of water quality impairments in the U.S., water-quality managers need tools for predicting aqueous E. coli levels. Presently, E. coli levels may be predicted using complex mechanistic models that have a high degree of unchecked uncertainty or simpler statistical models. To assess spatio-temporal patterns of instream E. coli levels, herein we measured E. coli, a pathogen indicator, at 16 sites (at four different times) within the Squaw Creek watershed, Iowa, and subsequently, the Markov Random Field model was exploited to develop a neighborhood statistics model for predicting instream E. coli levels. Two observed covariates, local water temperature (degrees Celsius) and mean cross-sectional depth (meters), were used as inputs to the model. Predictions of E. coli levels in the water column were compared with independent observational data collected from 16 in-stream locations. The results revealed that spatio-temporal averages of predicted and observed E. coli levels were extremely close. Approximately 66 % of individual predicted E. coli concentrations were within a factor of 2 of the observed values. In only one event, the difference between prediction and observation was beyond one order of magnitude. The mean of all predicted values at 16 locations was approximately 1 % higher than the mean of the observed values. The approach presented here will be useful while assessing instream contaminations such as pathogen/pathogen indicator levels at the watershed scale.

Original languageEnglish (US)
JournalEnvironmental Monitoring and Assessment
Volume187
Issue number3
DOIs
StatePublished - Jan 1 2015

Fingerprint

Pathogens
Escherichia coli
pathogen
Statistics
watershed
water quality
Watersheds
prediction
Water quality
water temperature
water column
indicator
statistics
Water
Contamination
Managers
water

Keywords

  • E. coli
  • Markov random field model
  • Neighborhood structures
  • Stream water

ASJC Scopus subject areas

  • Environmental Science(all)
  • Pollution
  • Management, Monitoring, Policy and Law

Cite this

A neighborhood statistics model for predicting stream pathogen indicator levels. / Pandey, Pramod; Pasternack, Gregory B.; Majumder, Mahbubul; Soupir, Michelle L.; Kaiser, Mark S.

In: Environmental Monitoring and Assessment, Vol. 187, No. 3, 01.01.2015.

Research output: Contribution to journalArticle

Pandey, Pramod ; Pasternack, Gregory B. ; Majumder, Mahbubul ; Soupir, Michelle L. ; Kaiser, Mark S. / A neighborhood statistics model for predicting stream pathogen indicator levels. In: Environmental Monitoring and Assessment. 2015 ; Vol. 187, No. 3.
@article{c876712a729f48ceb99380a05442895b,
title = "A neighborhood statistics model for predicting stream pathogen indicator levels",
abstract = "Because elevated levels of water-borne Escherichia coli in streams are a leading cause of water quality impairments in the U.S., water-quality managers need tools for predicting aqueous E. coli levels. Presently, E. coli levels may be predicted using complex mechanistic models that have a high degree of unchecked uncertainty or simpler statistical models. To assess spatio-temporal patterns of instream E. coli levels, herein we measured E. coli, a pathogen indicator, at 16 sites (at four different times) within the Squaw Creek watershed, Iowa, and subsequently, the Markov Random Field model was exploited to develop a neighborhood statistics model for predicting instream E. coli levels. Two observed covariates, local water temperature (degrees Celsius) and mean cross-sectional depth (meters), were used as inputs to the model. Predictions of E. coli levels in the water column were compared with independent observational data collected from 16 in-stream locations. The results revealed that spatio-temporal averages of predicted and observed E. coli levels were extremely close. Approximately 66 {\%} of individual predicted E. coli concentrations were within a factor of 2 of the observed values. In only one event, the difference between prediction and observation was beyond one order of magnitude. The mean of all predicted values at 16 locations was approximately 1 {\%} higher than the mean of the observed values. The approach presented here will be useful while assessing instream contaminations such as pathogen/pathogen indicator levels at the watershed scale.",
keywords = "E. coli, Markov random field model, Neighborhood structures, Stream water",
author = "Pramod Pandey and Pasternack, {Gregory B.} and Mahbubul Majumder and Soupir, {Michelle L.} and Kaiser, {Mark S.}",
year = "2015",
month = "1",
day = "1",
doi = "10.1007/s10661-014-4228-1",
language = "English (US)",
volume = "187",
journal = "Environmental Monitoring and Assessment",
issn = "0167-6369",
publisher = "Springer Netherlands",
number = "3",

}

TY - JOUR

T1 - A neighborhood statistics model for predicting stream pathogen indicator levels

AU - Pandey, Pramod

AU - Pasternack, Gregory B.

AU - Majumder, Mahbubul

AU - Soupir, Michelle L.

AU - Kaiser, Mark S.

PY - 2015/1/1

Y1 - 2015/1/1

N2 - Because elevated levels of water-borne Escherichia coli in streams are a leading cause of water quality impairments in the U.S., water-quality managers need tools for predicting aqueous E. coli levels. Presently, E. coli levels may be predicted using complex mechanistic models that have a high degree of unchecked uncertainty or simpler statistical models. To assess spatio-temporal patterns of instream E. coli levels, herein we measured E. coli, a pathogen indicator, at 16 sites (at four different times) within the Squaw Creek watershed, Iowa, and subsequently, the Markov Random Field model was exploited to develop a neighborhood statistics model for predicting instream E. coli levels. Two observed covariates, local water temperature (degrees Celsius) and mean cross-sectional depth (meters), were used as inputs to the model. Predictions of E. coli levels in the water column were compared with independent observational data collected from 16 in-stream locations. The results revealed that spatio-temporal averages of predicted and observed E. coli levels were extremely close. Approximately 66 % of individual predicted E. coli concentrations were within a factor of 2 of the observed values. In only one event, the difference between prediction and observation was beyond one order of magnitude. The mean of all predicted values at 16 locations was approximately 1 % higher than the mean of the observed values. The approach presented here will be useful while assessing instream contaminations such as pathogen/pathogen indicator levels at the watershed scale.

AB - Because elevated levels of water-borne Escherichia coli in streams are a leading cause of water quality impairments in the U.S., water-quality managers need tools for predicting aqueous E. coli levels. Presently, E. coli levels may be predicted using complex mechanistic models that have a high degree of unchecked uncertainty or simpler statistical models. To assess spatio-temporal patterns of instream E. coli levels, herein we measured E. coli, a pathogen indicator, at 16 sites (at four different times) within the Squaw Creek watershed, Iowa, and subsequently, the Markov Random Field model was exploited to develop a neighborhood statistics model for predicting instream E. coli levels. Two observed covariates, local water temperature (degrees Celsius) and mean cross-sectional depth (meters), were used as inputs to the model. Predictions of E. coli levels in the water column were compared with independent observational data collected from 16 in-stream locations. The results revealed that spatio-temporal averages of predicted and observed E. coli levels were extremely close. Approximately 66 % of individual predicted E. coli concentrations were within a factor of 2 of the observed values. In only one event, the difference between prediction and observation was beyond one order of magnitude. The mean of all predicted values at 16 locations was approximately 1 % higher than the mean of the observed values. The approach presented here will be useful while assessing instream contaminations such as pathogen/pathogen indicator levels at the watershed scale.

KW - E. coli

KW - Markov random field model

KW - Neighborhood structures

KW - Stream water

UR - http://www.scopus.com/inward/record.url?scp=84928242578&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84928242578&partnerID=8YFLogxK

U2 - 10.1007/s10661-014-4228-1

DO - 10.1007/s10661-014-4228-1

M3 - Article

VL - 187

JO - Environmental Monitoring and Assessment

JF - Environmental Monitoring and Assessment

SN - 0167-6369

IS - 3

ER -