2014 Stantec_SSRO Presentation - auprf

Alternative Tier 2 Process
Model Refinement, Implementation
and Validation
PTAC: March 20, 2014
Research Providers:
Dr. Melissa Whitfield Aslund
Drs. Eric Lamb & Steve Siciliano
Barry Zajdlik, PhD. Candidate
Robin Angell & Gladys Stephenson
Kathryn Bessie
Agenda
1
2
3
4
5 Vision for 2014-2015
Background
Review Progress (Year 2)
Aims & Objectives (Year 3)
1
Background
• Weathered PHCs in soil = recalcitrant residuals
less bioavailable and less toxic; Tier 1 SQGs
generally overly conservative
• Direct eco-contact pathway = F2/F3 driver
• Provision for Tier 2 Pass/Fail approach
• Tier 2 Pass leads to closure
• Tier 2 Fail leads to further remediation or
management measures; caveat on land title
Develop a Tier 2 process that would
allow the derivation of site-specific
remedial objectives or clean-up values
2
Review (FY2; 2012-2013)
Project Goals
1
2
3
• Develop models that could predict “effects” for soils with
known properties and contaminant profiles
• Use the distribution of predicted effects to guide
management decisions
• Establish a national database where ecotoxicological data
(response variables) are linked with soil and contaminant
characteristics
2
Review (FY2; 2012-2013)
Feasibility of Four Approaches
GeoMean Approach - Geometric Mean
of NOAEC and LOAECs
PLS Approach – Partial Least Squares
multiple regression analyses with leave
one class out cross validation
DRAMA Approach – Data exploration,
Reduction And Model Averaging
SEM Approach – Structural Equation
Modelling
2
Review (FY2; 2012-2013)
Conclusions (Year 2)
Final Report September 2013
• PLS and DRAMA showed that it was possible
to link multivariate soil properties to biological
responses
• These soil properties were as, or more,
influential in explaining the variability in the
response data than the contaminant
variables
• Critical variables include – texture, OM, TOC,
pH, Ca, Mg, fertility (P,K,N), EC, MC, SAR
2
Review (FY2; 2012-2013)
Conclusions (Year 2)
Final Report September 2013
• SEM demonstrated that confirmatory factor
analyses to aggregate multiple endpoints
into a single latent response variable could
then be incorporated into standard nonlinear modelling to estimate any ICp value
• Cross-site model predictions successfully
explained the aggregate species responses
(R2=0.70)
• SEM is “data hungry”; model adequacy
questionable
2
Review (FY2; 2012-2013)
Summary (Year 2)
Final Report September 2013
• Tier 2 Pass/Fail is of limited value when site
soils fail
• Failure to meet Tier 2 criteria is not
necessarily a failure because of
contaminant toxicity; substantial portion of
the observed significant biological
responses is attributable to noncontaminant variables
• Statistical tools can be applied to data to
clarify interpretation of toxicity
3
Aims & Objectives (Year 3)
Collate data and expand the dataset
Refine the models based on larger dataset
Secure a site that failed the Tier 2 Pass/Fail
Collect field soil samples and conduct an
ecotoxicity assessment
• Apply the refined models (PLS, DRAMA, SEM)
using the physical/chemical data for the site to
predict effects across the site (independent)
• Compare the model predictions to the
measured effects data – verification/validation
•
•
•
•
4
Status Update (Year 3)
• Landowner cooperation for use of a site was
secured
• Field soils and reference soils were collected
and characterized
• An ecotoxicity assessment comprising 3 plant
and 2 invertebrate species was conducted
• Tier 2 Pass/Fail Approach was applied
• Model refinement is still in progress
• Prediction of effects is pending
• Validation/verification is forthcoming
4
Status Update (Year 3)
Barley
E. andrei
Northern Wheatgrass
F. candida
Alfalfa
4
Status Update (Year 3)
Tier 2 Pass/Fail Approach
Site Soil
# Criteria Pass:Fail
Conclusion
S1-1
1:3
Fail
S1-2
2:2
Fail
S1-3
2:2
Fail
S2-1
4:0
Pass
S2-2
4:0
Pass
S2-3
4:0
Pass
Status Update (Year 3)
% Effect Inhibition Relative to Control
4
S1-1
S1-2
S1-3
S2-1
S2-2
S2-3
4
•
•
•
•
•
•
Status Update (Year 3)
Model refinement
Implementation - predictions
Comparison (prediction vs response)
Validation
Further model refinement
Reporting
5
Vision (FY4; 2014-2015)
• Implementation of models in site
management
• Geospatial modelling of effects data
• Incorporation of predictive modelling into
risk assessment
• Development of accessible database
• Publication of approach and models
• Additional applications
Questions?