Australian Centre for Environmetrics. Developing Risk-based guidelines for Water Quality Monitoring and Evaluation Prof. David Fox CSIRO Land and Water

Australian Centre for Environmetrics


Developing Risk-based guidelines for Water Quality Monitoring and Evaluation

Prof. David FoxCSIRO Land and WaterUniversity of Melbourne

Melbourne University Private


http://www.deh.gov.au/water/quality/nwqms/


Chapter 1: Introduction• Rational for revision• Philosophical basis

Chapter 2: Framework• Key steps • Important issues

Chapter 3: Aquatic Ecosystems• Types & levels of protection• Default & site-specific guidelines• Use of biological indicators

Chapter 4: Primary Industries• Irrigation• Livestock• aquaculture

Chapter 5: Recreational WQ & aesthetics

• Swimming, boating, etc.

Chapter 6: Drinking Water• Safety & aesthetics

Chapter 7: Monitoring & Assessment

• Data collection & analysis




Environmental monitoring

Aim is to design and conduct scientifically credible programs of environmental surveillance

Aim is discover specific violations and force corrective action

information

data

Compliance monitoring


Risk-based Approaches

Evolution of conventions has a lasting effect on how risk analyses are conducted:

• USEPA has set mostly conservative defaults.

• US Nuclear Regulatory Commission generally avoids conservative assumptions, recommending that modelers use default values that are close to the central tendency of parameter estimates (Bier 2003).

• The Bayesian perspective is that there is a random variable, and the job of the analyst is to characterize how variable it may be.

The approaches share a common belief in the epistemic nature of risk: there is a state of nature and the job of the risk analyst is to define it.


Low High

Level of environmental protection

Protector Risk

Polluter Risk

'Acceptable' region of protection

Max. polluter risk

Max. protector risk

Risks and Trade-offs• Protector risk = prob. ecologically important impact goes undetected

• Polluter risk = prob. unimportant impact triggers further action


Trigger-values


Trigger-values for physico-chemical stressors


3

0

y

1.1110 x

mortality

100%

concentration

Distribtion of NOECS

Setting Risk-based trigger values : Aldenburg & Slob (1993)

Assumed log-logistic

Dose-response curves for selected species


3

0

y

1.1110 x

3

0

y

1.1110 x

0.95

Trigger value

Distribution of NOECs for all species

Setting Risk-based trigger values : Aldenburg & Slob (1993)


Species Test endpoint NOEC* (µg L-1)

Chlorella sp. Cell division

rate 129

Moinodaphnia macleayi

Reproduction 18

Hydra viridissima Population

growth 150

Mogurnda mogurnda Mortality 400

Melanotaenia splendida inornata

Mortality 810

* NOEC: no-observed-effect concentration

Example – Modelling Uranium NOECs

Chronic

Acute



Raw Data: x = {129, 18, 150, 400, 810 }

Trigger value = 0.49 g/L


1 2

1 1

( ; ) ( ; )j

n ny

X i Yi j

L f x f


Chronic data: denote by X with pdf ( ; )Xf x

Acute data:

• denote by Y

• distribution of Y/ assumed to be same as distribution of X where is acute to chronic ratio.

Given sample of n1 X observations and n2 Y observations, the maximum

likelihood estimator (mle) for is that value which maximises the likelihood

function:


0 5 10 15 20 25 30 35 40 45 50

1.861 1012

0

exp ll 1 2 t k

500.03 t k


Data: x = {129, 18, 150 } and y = {400, 810}

Likelihood function

Mle = 7.451



Modified Data: x = {129, 18, 150 } and y = {400 / 7.451, 810 / 7.451}

Revised trigger value = 5.34 g/L

cf 0.49 g/L (raw data)

5.8 g/L (DEH value)

3.11 g/L (using default = 10)


“Introducing Bayes Theorem, or any similar method, into a

criminal trial plunges the jury into inappropriate and unnecessary

realms of complexity, deflecting them from their proper task.”

Bayesian Methods – A Credible Alternative?

Bayesian approach:

Has advantage of introducing subjective assessment / expert opinion

But

May be perceived as being difficult to interpret & lacking objectivity.

London Court of Appeal:

The Times, November 3 1997



A Bayesian Approach

for(j IN 1 : 2)

for(i IN 1 : 3)

Y[j]

X[i]

mup

taup

lamdatau

mu

http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml


Data

Densi

ty

136.5117.097.578.058.539.019.50.0

0.25

0.20

0.15

0.10

0.05

0.00

Variablelamda_priorlamda_post

Prior & posterior distributions

node mean stdev P2.5 median P97.5

Lamda_prior 19.951 14.134 2.4 16.669 82.0

node mean stdev P2.5 median P97.5

Lamda_post 7.324 3.036 4.075 6.624 14.92

A Bayesian Approach




Modified Data: x = {129, 18, 150 } and y = {400 / 6.624, 810 / 6.624}

Revised trigger value = 6.64 g/L

cf 0.49 g/L (raw data)

5.8 g/L (DEH value)

3.11 g/L (using default = 10)

5.34 g/L (using mle = 7.451)

A Bayesian Approach


Reference site – Test site comparisons

3

0

y

1.1110 x

3

0

y

1.1110 x

Reference SiteReference Site

Test SiteTest Site

De facto ‘standard’

Test site median

Ref site 80th.percentile

Note:

• Normal distributions not a prerequisite

• Common distribution not a prerequisite

• 80th. Percentile at reference site must be based on minimum of 24 data values (2 years monthly data)


121110987654321

3

2

1

0

Month

conc

entr

atio

n un

its

Test site median

Reference site P80

next level investigation triggered

no action required

w arning - investigation may be necessary

Reference site – Test site comparisons


• Despite early attempts, development and adoption of a ‘standard’ risk metric seems a long way off (never?);

• Bayesian methods are becoming increasingly popular, although acceptance may be hampered by biases and lack of understanding;

• More attention needs to be given to appropriate statistical modelling. In particular:

- model choice- Parameter estimation- Distributional assumptions- ‘Outlier’ detection and treatment- robust alternatives (GLMs, GAMs, smoothers etc).

Observations & Challenges

Documents

Australian Centre for Environmetrics. Developing Risk-based guidelines for Water Quality Monitoring and Evaluation Prof. David Fox CSIRO Land and Water