Episode 111      23 min 03 sec
A career in modelling: Assessing risk in natural resource management

Agricultural scientist Dr Andrew Hamilton explains how risk and uncertainty can be better modelled in both managing waster water and estimating species richness. With science host Dr Shane Huntington.

"What is the level of treatment that is actually required to get the water to a safe level, as opposed to always just going in and over-engineering a system?" -- Dr Andrew Hamilton





           



Dr Andrew Hamilton
Dr Andrew Hamilton

Dr Andrew Hamilton is Senior Lecturer in Sustainable Horticulture Management, in the Melbourne School of Land and Environment at the University of Melbourne. Andrew's resarch foucses on sampling protocols for horticultural pests, wastewater irrigation and waterbird conservation.

Dr Hamilton was a research scientist with the Victorian Department of Primary Industries (DPI) for around 11 years. Most of this research with the DPI was in the fields of entomology, especially sampling theory and practice, and wastewater reuse for horticultural irrigation. In 2005 Andrew and his colleagues were awarded the Daniel McAlpine Outstanding Achievement Award in recognition of their work on applied aspects of sampling, and in particular, the development of sampling plans for diamondback moth, a significant pest of brassicaceous crops.

Credits

Host: Dr Shane Huntington
Producers: Kelvin Param, Eric van Bemmel
Associate Producer: Dr Chrstine Bailey
Series Creators: Eric van Bemmel and Kelvin Param
Audio Engineer: Gavin Nebauer
Voiceover: Nerissa Hannink

View Tags  click a tag to find other episodes associated with it.

 Download mp3 (22.2 MB)

A career in modelling: Assessing risk in natural resource management


VOICEOVER
Welcome to Up Close, the research, opinion and analysis podcast from the University of Melbourne, Australia.

SHANE HUNTINGTON
I’m Shane Huntington.  Thanks for joining us.  In the management of natural resources, as in other complex and rapidly changing areas, there is a pressing need to quantify and analyse ever increasing amounts of information in order to arrive at decisions and make policy.  But, in many cases, we find predictions are made and actions taken based on what are little more than rough estimates, potentially leading to disastrous results.  So how might we mitigate these risks and do a better job collecting reliable information, particularly in natural resource management where the stakes are high and systems complex?  To discuss this issue we’re joined by Dr Andrew Hamilton, Senior Lecturer in Sustainable Horticulture Management in the Melbourne School of Land and Environment at the University of Melbourne, Australia.  Welcome, Andrew.

ANDREW HAMILTON
Good morning, Shane.  How are you going?

SHANE HUNTINGTON
Good.  Let’s start off with some of the statistical terms we’re using here and, for those of us who aren’t as literate in this area, what do the terms probabilistic risk and uncertainty analysis actually refer to?

ANDREW HAMILTON
When we’re making decisions in natural resource management, there is always uncertainty associated with them.  Often people try and separate the world into people who are modellers and people who are scientists, who - you are not into modelling, basically.  I would argue that everyone is a modeller and all that modelling is about is expressing and showing people how you have come to a certain conclusion.  What uncertainty analysis is about is not just showing people how you came to a certain conclusion, but also showing them all the steps in the process and including your uncertainty in that process.  It actually goes through to your prediction.  So you don’t just say I’m 90 per cent certain that this will happen; you actually say I may be 70 to 95 per cent certain that this will happen.  So you’re being honest about your understanding of the system.

SHANE HUNTINGTON
So you’re giving a range of confidence in a sense.

ANDREW HAMILTON
That’s right.

SHANE HUNTINGTON
So that people know the outlays and how far out you can be with your data.

ANDREW HAMILTON
That’s right.  Often you will get people in science, but also in politics and all different endeavours - they will say, I believe that this proposition is too risky; whether it’s drinking recycled water or whatever it is.  But to come to that conclusion they must have done some type of modelling in their head.  They’ve got some series of processes.  So, to me, mathematical modelling is nothing more than just laying your cards on the deck and saying to everyone, how did you get to that point?

SHANE HUNTINGTON
We often use the term mathematics as a catch-all cry for so many different areas of maths.  What particular sort of area of maths is used to do this sort of work?

ANDREW HAMILTON
I’m not actually a mathematician; I just draw on mathematics as a tool.  But most of this area is concerned with probability theory.  We construct models that basically link a series of processes together to describe our understanding of a system and at each step in the modelling we apply probability theory to say okay, what is the likelihood of this process happening?  So I would say that probability theory is the area of maths that we draw most upon.

SHANE HUNTINGTON
Andrew, when we talk about these areas and the uncertainty, presumably these is some issue of error and uncertainty in those measurements themselves.  How do you go about determining the range of uncertainty that you put forward, if it is plus or minus 10, or 10 per cent?  How do you know that that is a viable answer that you are giving at that point?

ANDREW HAMILTON
That is a very good question.  This is where, I suppose, philosophy comes into it as well.  We really have to think about what we are doing with modelling and what we’re doing is representing what we think is our best understanding of the situation, including our best understanding of the uncertainties.  Now, we could be hyper-conservative and basically sit back in our seats and say we just can’t model anything because we don’t know anything.  It comes down to a balance between our ignorance - basically our lack of knowledge of the system - and secondly, trying to represent natural variability.  So they are the two fundamental types of uncertainty.  The natural variability side of things is easier to quantify in a way because there is not really subjectivity left there.  But the ignorance, there is.  So it gets tricky.

SHANE HUNTINGTON
This is Up Close, coming to you from the University of Melbourne, Australia.  I’m Shane Huntington and our guest today is Dr Andrew Hamilton, who is speaking about risk modelling.  Andrew, let’s consider a specific example, that of determining species richness, which I know you have been working on.  How is this currently done?  I mean, before your modelling, how do we go about determining the number of species that are out there?

ANDREW HAMILTON
There have been a number of approaches to the problem.  One of the early methods was to look at just the simple relationship between body size and number of species.  If you plot just a relationship of elephants down through humans through - you get smaller and smaller organisms all the way to insects.  Basically you get more and more species of a smaller body size.  So if you plot that relationship, you could then extrapolate to the very small body sizes and say okay, therefore we think there are so many species because it’s obviously the small species that we don’t have the numbers for.  For two reasons: they are small, but also there is so many of them.  The problem with that method was though, that if you imagine a straight line going from whales through elephants and to smaller and smaller organisms, the relationship breaks down at the very small organisms.  We actually start to get less very small organisms. And because we don’t know why that is, that process - which Bob May put forward - that process just doesn’t work because we don’t know what’s happening.  So in the early ‘80s, an entomologist by the name of Terry Erwin from the Smithsonian Institute - he was a coleopterist, which is someone who studies beetles.  There are such people.  He went out into rainforest in Panama and basically fogged canopy with insecticide and all the insects would drop onto these sheets and they collect them.  And he found an enormous number of beetle species.  He just did a simple extrapolation and said okay, there is so many beetle species per tree, so many of those species will be specialised on that particular tree so they need that tree to live on and there is so many tropical tree species in the world.  He just multiplied up these simple steps and said therefore, I think there might be 30 million insect species, or arthropod species, in the world.  An arthropod is basically a group of invertebrates that includes the insects and most of them are insects.  So he did that and really shocked everyone by saying maybe 30 million, because before that people were thinking in the order of less than 10 million species.  Now, you will probably notice I switch almost interchangeably between number of species all up and number of arthropods.  That’s because most terrestrial species really will be arthropods, so it’s the arthropods that are the problematic group.  So, since Terry Erwin’s work, several biologists re-did the calculations using different values - and quite different values for the different steps in the model - and that produced an array of estimates.  But what I have done here is use, as we were talking about before, probability theory to represent the uncertainty for each one of those parameters.

SHANE HUNTINGTON
So when we look at some of those existing techniques what kind of numbers, in terms of total species for the planet, are we talking about?

ANDREW HAMILTON
As I said, Terry Erwin estimated about 30 million and this demonstrates the sensitivity of just using single values.  He then recalculated it and got 50 million and then even published a book chapter and said maybe 100 million; whereas, most people were more in the order of less than 10 million.

SHANE HUNTINGTON
When you start applying your modelling techniques, presumably you need to base it on some of these other techniques for counting.

ANDREW HAMILTON
Yes.

SHANE HUNTINGTON
Then you bring in the uncertainty element.  How does that all work?  How does that fit together?

ANDREW HAMILTON
Yes, I got the information from all the published studies and instead of having what we call a point estimate, which is just a single number, I used that information to construct probability distributions.  But two of the important parameters in the model were the number of beetle species that use a particular tree species for feeding on and also the number of beetle species just found in a tropical tree species.  We didn’t have much information on that.  Terry Erwin’s model was based on just one tree species.  Subsequent studies were five, maybe up to ten, species.  We collaborated with Vojtech Novotny.  Vojtech is from the Czech Academy of Sciences.  He’s running a big program in New Guinea, where they have sampled about 56 tree species in Papua New Guinea.  Just by doing that, we’re getting a much better understanding of the variability; so natural variability, natural uncertainly.  New Guinea hosts a large proportion of the world’s tropical tree species, so that was a really good place to start and get more information.

SHANE HUNTINGTON
Why have they chosen the beetle and not the ant or something else that’s prolific in terms of insects and so forth that we find?

ANDREW HAMILTON
The simple reason is that there is a lot of beetles out there, a lot of beetle species.  Of described species, beetles account for about 40 per cent of all described arthropod species and arthropods account for 80 per cent of all species described.  So just in those numbers you can see how significant they are so it’s a very good place to base your model from.  You want to extrapolate from a group that is representing a large portion of species all up.  You’re still extrapolating from that group so there is obviously issues there, but at least you’re not extrapolating from something like primates, where you’re just starting with a few species.

SHANE HUNTINGTON:
Andrew, if I was to ask you to then put a number on the species and an uncertainty and then give us an indication of how this information will be used, what would you give us?

ANDREW HAMILTON
Basically for tropical arthropods, in the paper that we published it said 3.7 million tropical arthropods, with what we call a 90 per cent confidence interval - so basically 90 per cent that it’s between two and seven point four.  Now, I have re-run the model since then with yet more conservative assumptions.  I actually just re-ran that half an hour before I came in here.  But that gets us up to almost 7.1 million species, with a 90 per cent confidence interval about three to 14.  There is obviously uncertainty.  That’s the point of these models.  It’s not so much saying I think there are exactly this many.  But you can see that it’s nowhere near the 30 million.  I calculated the probability of that and that still holds with the refined model of less than .001 per cent probability of having that many.

SHANE HUNTINGTON:
Right.  So that number is essentially just wrong.

ANDREW HAMILTON
That is basically what I’m arguing, yes.  I’ve even constructed what we call probability bounds, which is the original model I’m talking about.  It is actually a probability distribution.  But then I’ve calculated the space that that probability distribution itself could fit in; so, basically, a distribution around the distribution, for want of a better description.  Even with that, you still don’t encompass the 30 million.  You go nowhere near.  As we get more data, the median values that I gave you then - the 7 million and the 3.7 initial one - they will move around a bit of course.

SHANE HUNTINGTON:
But never be 30.

ANDREW HAMILTON
That’s right.  That’s the crux of it.

SHANE HUNTINGTON:
Yes.  Now, how do we go about using this information now that we have a better idea?  So now that we know that we know that it’s between 2 million and 7 million, as opposed to 30 million, how is that going to lead to the way we deal with species in the future and how we care for them?

ANDREW HAMILTON
That’s a good question.  I suppose it gives us a target and something that we can really set our minds to.  So basically it says around 70 per cent of species still await description.  Now that’s a lot of species but it’s not necessarily insurmountable.  Obviously we need to know what is out there to know how much we’re losing.  Now there’s a lot of work done on extinction and it’s not actually my area of expertise but, just to cut a long story short, there is a lot of debate about extinction rates.  You know, are they 100 times background level or 1000 times or 10,000 times?  There’s a lot of discussion about that.  But at the end of the day, what we also need to know is how many species in absolute terms; so not just the rate at which we’re losing species, but how many in absolute terms we are losing.  So to know that, you actually have to know how many we have to start with of course.  That will help not just with conservation efforts, but also with justifications and a rationale for conservation efforts and for investment in conservation.  If we can say, “OK,  this is how many species we’re losing.” - in absolute terms, not just in relative terms - if we lose this much rainforest, then this amount of investment can save so many species for instance.

SHANE HUNTINGTON:
I’m Shane Huntington and my guest today is Dr Andrew Hamilton.  We’re talking about risk modelling here on Up Close, coming to you from the University of Melbourne, Australia.  Andrew, the second big issue that you have been working on is waste water irrigation.  Tell us about the risks associated with waste water irrigation and broadly what these risks are?

ANDREW HAMILTON
I suppose there is two main areas of concern when it comes to risks with waste water irrigation.  There is microbial pathogenic organism risks but there is also chemical risks.  The chemical risks, in an irrigation sense, are probably much less of a concern but still possibly a concern, depending on the source of the waste water.  But they are also very difficult to model, which is not necessarily an excuse for not modelling them, but they are very difficult to model.  The reason for that - waste water is just a mix of hundreds of different organics and heavy metals and all sorts of things.  There is the possibility of synergistic interactions between chemicals and antagonistic interactions, so it’s very hard to say exactly how chemicals might be affecting us.  That being said, the reason we tend to be more concerned with pathogens is that there is a lot of waste water used in the developing world for irrigating crops.  There has been some very rough figures that say 10 per cent of the world’s population consumes food irrigated with waste water - with raw or partially diluted waste water I should say.  The pathogen load is what is of immediate concern because there is lots of children dying of rotaviruses.  There is other diseases associated with waste water.

SHANE HUNTINGTON:
Andrew, can these risks sort of not be just readily mitigated by careful planning and use of some of these new technologies around water testing and so forth?

ANDREW HAMILTON
Yes.  Well, microbiological risks can be dealt with quite effectively in developed countries.  But those systems are highly engineered and they reduce pathogens to a level where the risk of infection is incredibly low and basically not a concern.  That being said, I think it is still interesting to model risk, even for these situations, because there is the potential for a lot more waste water re-use, even in developed countries but maybe we don’t always need to be engineering to such a high degree.  So there is still potential to use more but have a slightly higher risk.

SHANE HUNTINGTON:
On that, I guess, one of the issues is just how high the risks are in a particular area.  How do you go about modelling that and determining, in a quantitative way that you can rely on before you start spending money, how high those risks are in some of those locations?

ANDREW HAMILTON
First of all we develop what we call an exposure model; basically a model that describes the series of events that determine the likelihood of a person consuming, say, vegetables from a particular scheme, taking in a pathogen.  That will include obvious things like the concentration of pathogens in the irrigation water and then the amount of water that ends up on the surface of the plants, the survival rate of the pathogen, the amount of produce actually consumed by the person - so a very kind of logical series of events.  So that’s telling us how many pathogens are likely to come in in a particular consumption event, for want of a better word.  Then we have a second part of the model, which is called a dose response model.  That tells us the relationship between the pathogen and the human and the likelihood of infection being established once you take in the pathogen.  Those dose response models are based on data sets where people have actually been challenged with pathogens and we look at the proportion of people that actually become ill.  Now, there’s a bit of a problem there because we don’t have a lot of good quality data for dose response models and a lot of that comes down to the ethics of conducting such trials.  We don’t normally like to challenge people with diseases.  A lot of the data are quite old.  They are from when they used to use soldiers, for instance, and they would challenge them with pathogens and look at the proportion - who would become ill.  That being said, there was a big study done in the US over the last five or six years on norovirus, which is a major cause of gastrointestinal illness in developed and developing worlds.  They were actually challenging people with the pathogens and obviously prepared to treat them when they became infected, et cetera.  So that has produced a really good data set.

SHANE HUNTINGTON:
Once you have sort of determined this risk, what sort of improvements do we see over the way we have been looking at these numbers in the past?

ANDREW HAMILTON
In the past, often we just had a very simple threshold approach to dealing with waste water.  We said if the waste water contains more than X concentration of what we called an indicator organism - and it was normally something like E coli, so a bacterium - if it contained more than that, we considered the waste water was dangerous for a particular purpose.  That was basically arrived at very informally; often just by observations on disease rates associated with different irrigation schemes, et cetera.  But also, it was sometimes just some countries would say, okay, we’re going to go one or two orders of magnitude less than another country just to be conservative.  What we are doing now is we’re actually modelling the risk for a particular situation.  It’s a much more sophisticated approach than just having a simple threshold for source water.

SHANE HUNTINGTON:
How exactly does that sort of process work in terms of the modelling?  I understand you use something called quantitative microbial risk assessment?

ANDREW HAMILTON
That’s right.  That’s another bit of lingo.  What quantitative microbial risk assessment is, it’s that process that I briefly described before.  We identify a hazard - so it might be norovirus, for instance.  Then we construct an exposure model to determine the likelihood of someone taking in the pathogen.  Then we have this dose response model that tells us the likelihood of them becoming infected.  The last step is bringing it all together and that actually uses some of the computational techniques that I was talking about before.  We tend to use a process called Monte Carlo simulation but there are other methods.  Basically, the computer draws on all these probability distributions so they’re representing all the parameters and it gives you an output.  That output is not just a single estimate of risk.  It’s a probability of the risk again.  So just like the species richness model, we have confidence bounds around it.

SHANE HUNTINGTON:
Andrew, what is the next step in the modelling for you now, moving forward?

ANDREW HAMILTON
I suppose with the uncertainty in the species richness estimation, it’s to figure out how we can improve on what I call the ignorance.  So as part of that work, I did what we call a sensitivity analysis and I identified which parameters actually had the biggest influence on the uncertainty in the output.  So now we can focus on those parameters and say, okay, we need to improve our understanding of those parameters.  So you basically start narrowing down the uncertainty.  For the microbial risk assessment work and the waste water work more generally, what we want to do - and I hinted at it before - is look at it from another perspective and just say what is the level of treatment that is actually required to get the water to a safe level, as opposed to always just going in and over-engineering a system?  Which is fine - it’s obviously good to be conservative if you can afford to be conservative - but that doesn’t necessarily mean we’re going to make the best use of our water resources.  There’s going to be situations where we might be put off going ahead with the waste water re-use scheme because we can’t afford to be so conservative and have such fancy treatment.  Whereas, we could probably get away with a less intensive treatment system and still not be putting the public health at risk.

SHANE HUNTINGTON:
Do you also have a hope that the scientific and engineering and other communities involved in a lot of this work will start to take on the need to, I guess, give more appropriate parameters around the data that they are publishing?

ANDREW HAMILTON
Yes.  I think most - well, engineering and just natural resource management in general - are starting to go that way.  And we’re really starting to be more explicit with the uncertainty associated with our parameters and not just using these simple, what we call deterministic models.  We’re really starting to express uncertainty in models a lot better.

SHANE HUNTINGTON:
Dr Andrew Hamilton, Senior Lecturer in Sustainable Horticulture Management in the Melbourne School of Land and Environment at the University of Melbourne.  Thank you very much for being our guest on Up Close today.

ANDREW HAMILTON
Thank you, Shane.

SHANE HUNTINGTON:
Relevant links, a full transcript and more info on this episode can be found on our website at upclose.unimelb.edu.au.  Up Close is brought to you by Marketing and Communications of the University of Melbourne, Australia.  This episode was recorded on 27 August 2010 and our producers were Kelvin Param and Eric van Bemmel.  Audio engineering by Gavin Nebauer.  Up Close is created by Eric van Bemmel and Kelvin Param.  I’m Shane Huntington.  Until next time, good bye.

VOICEOVER
You've been listening to Up Close.  For more information visit http://upclose.unimelb.edu.au.  Copyright 2010, the University of Melbourne.


show transcript | print transcript | download pdf