A probability co-kriging model to account for reporting bias and recognize areas at high risk for Zebra Mussels and Eurasian Watermilfoil Invasions in Minnesota

Kaushi S.T. Kanankege, Moh A. Alkhamis, Nicholas Phelps, Andres M Perez

Research output: Contribution to journalArticle

Abstract

Zebra mussels (ZMs) (Dreissena polymorpha) and Eurasian watermilfoil (EWM) (Myriophyllum spicatum) are aggressive aquatic invasive species posing a conservation burden on Minnesota. Recognizing areas at high risk for invasion is a prerequisite for the implementation of risk-based prevention and mitigation management strategies. The early detection of invasion has been challenging, due in part to the imperfect observation process of invasions including the absence of a surveillance program, reliance on public reporting, and limited resource availability, which results in reporting bias. To predict the areas at high risk for invasions, while accounting for underreporting, we combined network analysis and probability co-kriging to estimate the risk of ZM and EWM invasions. We used network analysis to generate a waterbody-specific variable representing boater traffic, a known high risk activity for human-mediated transportation of invasive species. In addition, co-kriging was used to estimate the probability of species introduction, using waterbody-specific variables. A co-kriging model containing distance to the nearest ZM infested location, boater traffic, and road access was used to recognize the areas at high risk for ZM invasions (AUC = 0.78). The EWM co-kriging model included distance to the nearest EWM infested location, boater traffic, and connectivity to infested waterbodies (AUC = 0.76). Results suggested that, by 2015, nearly 20% of the waterbodies in Minnesota were at high risk of ZM (12.45%) or EWM (12.43%) invasions, whereas only 125/18,411 (0.67%) and 304/18,411 (1.65%) are currently infested, respectively. Prediction methods presented here can support decisions related to solving the problems of imperfect detection, which subsequently improve the early detection of biological invasions.

LanguageEnglish (US)
Article number231
JournalFrontiers in Veterinary Science
Volume4
Issue numberJAN
DOIs
StatePublished - Jan 4 2018

Fingerprint

Dreissena
Spatial Analysis
Dreissena polymorpha
kriging
traffic
Introduced Species
invasive species
Area Under Curve
Myriophyllum spicatum
risk estimate
roads
Human Activities
Observation
prediction
monitoring

Keywords

  • Early detection
  • Geostatistics
  • Observation bias
  • Reporting
  • Risk assessment
  • Spatial modeling
  • Surveillance

PubMed: MeSH publication types

  • Journal Article

Cite this

@article{5b09c375858e4aa890ae7fc30d7609c6,
title = "A probability co-kriging model to account for reporting bias and recognize areas at high risk for Zebra Mussels and Eurasian Watermilfoil Invasions in Minnesota",
abstract = "Zebra mussels (ZMs) (Dreissena polymorpha) and Eurasian watermilfoil (EWM) (Myriophyllum spicatum) are aggressive aquatic invasive species posing a conservation burden on Minnesota. Recognizing areas at high risk for invasion is a prerequisite for the implementation of risk-based prevention and mitigation management strategies. The early detection of invasion has been challenging, due in part to the imperfect observation process of invasions including the absence of a surveillance program, reliance on public reporting, and limited resource availability, which results in reporting bias. To predict the areas at high risk for invasions, while accounting for underreporting, we combined network analysis and probability co-kriging to estimate the risk of ZM and EWM invasions. We used network analysis to generate a waterbody-specific variable representing boater traffic, a known high risk activity for human-mediated transportation of invasive species. In addition, co-kriging was used to estimate the probability of species introduction, using waterbody-specific variables. A co-kriging model containing distance to the nearest ZM infested location, boater traffic, and road access was used to recognize the areas at high risk for ZM invasions (AUC = 0.78). The EWM co-kriging model included distance to the nearest EWM infested location, boater traffic, and connectivity to infested waterbodies (AUC = 0.76). Results suggested that, by 2015, nearly 20{\%} of the waterbodies in Minnesota were at high risk of ZM (12.45{\%}) or EWM (12.43{\%}) invasions, whereas only 125/18,411 (0.67{\%}) and 304/18,411 (1.65{\%}) are currently infested, respectively. Prediction methods presented here can support decisions related to solving the problems of imperfect detection, which subsequently improve the early detection of biological invasions.",
keywords = "Early detection, Geostatistics, Observation bias, Reporting, Risk assessment, Spatial modeling, Surveillance",
author = "Kanankege, {Kaushi S.T.} and Alkhamis, {Moh A.} and Nicholas Phelps and Perez, {Andres M}",
year = "2018",
month = "1",
day = "4",
doi = "10.3389/fvets.2017.00231",
language = "English (US)",
volume = "4",
journal = "Frontiers in Veterinary Science",
issn = "2297-1769",
publisher = "Frontiers Media S. A.",
number = "JAN",

}

TY - JOUR

T1 - A probability co-kriging model to account for reporting bias and recognize areas at high risk for Zebra Mussels and Eurasian Watermilfoil Invasions in Minnesota

AU - Kanankege, Kaushi S.T.

AU - Alkhamis, Moh A.

AU - Phelps, Nicholas

AU - Perez, Andres M

PY - 2018/1/4

Y1 - 2018/1/4

N2 - Zebra mussels (ZMs) (Dreissena polymorpha) and Eurasian watermilfoil (EWM) (Myriophyllum spicatum) are aggressive aquatic invasive species posing a conservation burden on Minnesota. Recognizing areas at high risk for invasion is a prerequisite for the implementation of risk-based prevention and mitigation management strategies. The early detection of invasion has been challenging, due in part to the imperfect observation process of invasions including the absence of a surveillance program, reliance on public reporting, and limited resource availability, which results in reporting bias. To predict the areas at high risk for invasions, while accounting for underreporting, we combined network analysis and probability co-kriging to estimate the risk of ZM and EWM invasions. We used network analysis to generate a waterbody-specific variable representing boater traffic, a known high risk activity for human-mediated transportation of invasive species. In addition, co-kriging was used to estimate the probability of species introduction, using waterbody-specific variables. A co-kriging model containing distance to the nearest ZM infested location, boater traffic, and road access was used to recognize the areas at high risk for ZM invasions (AUC = 0.78). The EWM co-kriging model included distance to the nearest EWM infested location, boater traffic, and connectivity to infested waterbodies (AUC = 0.76). Results suggested that, by 2015, nearly 20% of the waterbodies in Minnesota were at high risk of ZM (12.45%) or EWM (12.43%) invasions, whereas only 125/18,411 (0.67%) and 304/18,411 (1.65%) are currently infested, respectively. Prediction methods presented here can support decisions related to solving the problems of imperfect detection, which subsequently improve the early detection of biological invasions.

AB - Zebra mussels (ZMs) (Dreissena polymorpha) and Eurasian watermilfoil (EWM) (Myriophyllum spicatum) are aggressive aquatic invasive species posing a conservation burden on Minnesota. Recognizing areas at high risk for invasion is a prerequisite for the implementation of risk-based prevention and mitigation management strategies. The early detection of invasion has been challenging, due in part to the imperfect observation process of invasions including the absence of a surveillance program, reliance on public reporting, and limited resource availability, which results in reporting bias. To predict the areas at high risk for invasions, while accounting for underreporting, we combined network analysis and probability co-kriging to estimate the risk of ZM and EWM invasions. We used network analysis to generate a waterbody-specific variable representing boater traffic, a known high risk activity for human-mediated transportation of invasive species. In addition, co-kriging was used to estimate the probability of species introduction, using waterbody-specific variables. A co-kriging model containing distance to the nearest ZM infested location, boater traffic, and road access was used to recognize the areas at high risk for ZM invasions (AUC = 0.78). The EWM co-kriging model included distance to the nearest EWM infested location, boater traffic, and connectivity to infested waterbodies (AUC = 0.76). Results suggested that, by 2015, nearly 20% of the waterbodies in Minnesota were at high risk of ZM (12.45%) or EWM (12.43%) invasions, whereas only 125/18,411 (0.67%) and 304/18,411 (1.65%) are currently infested, respectively. Prediction methods presented here can support decisions related to solving the problems of imperfect detection, which subsequently improve the early detection of biological invasions.

KW - Early detection

KW - Geostatistics

KW - Observation bias

KW - Reporting

KW - Risk assessment

KW - Spatial modeling

KW - Surveillance

UR - http://www.scopus.com/inward/record.url?scp=85040466170&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85040466170&partnerID=8YFLogxK

U2 - 10.3389/fvets.2017.00231

DO - 10.3389/fvets.2017.00231

M3 - Article

VL - 4

JO - Frontiers in Veterinary Science

T2 - Frontiers in Veterinary Science

JF - Frontiers in Veterinary Science

SN - 2297-1769

IS - JAN

M1 - 231

ER -