How can software suppliers support best practice for uncertainty in engineering geology

Managing uncertainty in engineering geology is nothing new, but it is becoming increasingly important as 3D geological modelling becomes standard practice in civil infrastructure, and design and construction processes become increasingly optimised.

The panel examines ways that software suppliers can support best practise for uncertainty in engineering geology. Is there a silver bullet solution?

Overview

Video Transcript

[00:00:04.170]
<v ->Well, hello everybody.</v>

[00:00:05.400]
And welcome to this LYCEUM 2020 panel discussion

[00:00:09.510]
on uncertainty.

[00:00:10.410]
We framed up this discussion

[00:00:11.940]
to try to propose an answer to the question,

[00:00:15.460]
how can software suppliers support best practice

[00:00:18.610]
for uncertainty in engineering geology?

[00:00:21.720]
I’m Pat McLarin, I’m the segment director for Seequent

[00:00:25.400]
civil business.

[00:00:27.020]
And fortunately, with this very challenging topic,

[00:00:30.610]
I’ve got some big guns joining me today on my panel.

[00:00:34.130]
So welcome to Graeme Jardine from Jacobs.

[00:00:36.950]
Darren Paul from Golder Associates.

[00:00:39.220]
Philip Kirk from Aurecon.

[00:00:40.980]
And Ashton Krajnovich from Colorado School of Mines.

[00:00:44.300]
It’s great to see you guys.

[00:00:45.860]
I really appreciate you guys taking the time,

[00:00:48.630]
and joining us.

[00:00:50.750]
Welcome.

[00:00:51.810]
Are you all buckled in really for a good discussion?

[00:00:56.321]
<v ->Yeah, go for it mate.</v>

[00:00:57.425]
<v ->Let’s do it.</v>

[00:00:58.449]
<v ->Yep.</v>

[00:00:59.282]
<v ->Good one, thanks.</v>

[00:01:00.850]
So, I just got a confession to make first though,

[00:01:03.410]
for our audience.

[00:01:04.330]
Of course, we’re never going to tackle the topic of uncertainty

[00:01:07.377]
and do it any kind of justice in 20 minutes.

[00:01:10.390]
So, we all sat down together pre-LYCEUM

[00:01:15.607]
and had a bit of a workshop,

[00:01:17.800]
presented the different viewpoints,

[00:01:20.450]
and we’ve been collaborating since.

[00:01:23.300]
Perhaps Graeme will start with you.

[00:01:25.680]
<v ->Thanks Pat.</v>

[00:01:26.513]
Yeah, look in the recent years within certainly

[00:01:29.900]
the infrastructure industry, which most of the guys

[00:01:32.910]
on this call are involved in both in A and Z and globally,

[00:01:36.190]
there’s a significant shift towards

[00:01:38.750]
delivery of digital designs and a move away from

[00:01:41.580]
the traditional hard copy 2D production of sections.

[00:01:47.750]
This digital delivery, is now being used via 3D federated

[00:01:51.820]
model system typically within a BIM framework,

[00:01:56.640]
and it encompasses major structural components, utilities,

[00:02:00.720]
and our 3D geological models.

[00:02:03.920]
Obviously this brings different levels of risk

[00:02:05.780]
and uncertainties, not just for the us the practitioners,

[00:02:09.090]
but also for clients and stakeholders alike.

[00:02:12.600]
I think it’s a case of different scales of accuracy,

[00:02:15.920]
and with it differing levels of uncertainty,

[00:02:18.730]
and therefore risk.

[00:02:21.010]
Unfortunately A and Z market is one of the most litigious,

[00:02:24.660]
globally.

[00:02:25.580]
And it falls on ground engineers,

[00:02:28.210]
try and minimize the uncertainty as far as possible.

[00:02:31.540]
But also to communicate the uncertainty in a meaningful way

[00:02:35.210]
to stakeholders so that they understand the ground risk

[00:02:39.050]
and its implications.

[00:02:41.280]
<v ->Yeah, I agree with everything Graeme said there.</v>

[00:02:44.820]
I’ve noticed this phenomenon where the better

[00:02:47.900]
the model looks, just gets seen as having more certainty.

[00:02:53.070]
And like it concerned by say structural engineers

[00:02:56.940]
who will take sections out of the model

[00:02:58.610]
and they’ll run analysis to predict displacement

[00:03:01.010]
to millimeter accuracy on a model that has uncertainly

[00:03:04.840]
too many more meters of accuracy.

[00:03:07.630]
Or another phenomenon I’ve seen is,

[00:03:11.410]
the people who have their fingers on the purse strings,

[00:03:14.410]
the project managers, who are the ones who decide

[00:03:16.710]
if it’s worth investing in say more investigation,

[00:03:19.670]
if it’s worth investing in spending money

[00:03:22.160]
to reduce uncertainty.

[00:03:23.710]
They’re the people we have to convince.

[00:03:25.850]
And same phenomenon, and they see a model

[00:03:28.930]
that looks fantastic, we know everything about the ground.

[00:03:33.760]
<v ->So, but it does boil back at the end of the day to,</v>

[00:03:38.060]
there’s a lot of onus on the modeler themselves,

[00:03:40.950]
and the biases that they bring,

[00:03:42.770]
and the interpretation they imply in the model, right?

[00:03:48.370]
In the workshop we talked a lot about something called

[00:03:52.500]
epistemic uncertainty.

[00:03:55.770]
So that’s a fancy word, perhaps someone, maybe Darren,

[00:04:04.090]
could you tell us a little bit about epistemic uncertainty,

[00:04:07.917]
and what it is and why that’s so important

[00:04:10.580]
in engineering geology?

[00:04:13.405]
<v ->Okay, so this chart here was our attempt to identify</v>

[00:04:17.460]
what it is that gives a ground model

[00:04:20.070]
certainty or uncertainty.

[00:04:22.300]
And it’s split into two main sort of sides on the left,

[00:04:24.950]
and the right here.

[00:04:26.520]
The left side is all about what observations we have

[00:04:29.860]
about the ground, which in our sphere is boreholes,

[00:04:33.000]
map of geophysics, observations, and measurements.

[00:04:37.210]
Now observations and measurements can have error

[00:04:40.280]
and uncertainty associated with them,

[00:04:42.170]
and that’s what we would call aleatory uncertainty.

[00:04:45.730]
You can reduce aleatory uncertainty

[00:04:47.930]
by taking more measurements,

[00:04:49.570]
or taking better quality measurements.

[00:04:52.260]
On the other side here, I’ve got what are called

[00:04:54.170]
ground complexity.

[00:04:55.970]
And this is all about how we interpret the data

[00:04:58.880]
that we’ve got, and it relies on the model of having a good

[00:05:01.830]
understanding of geological processes, geological history.

[00:05:05.650]
It’s a function of how predictable the ground is.

[00:05:09.170]
Now, understanding this geological history,

[00:05:11.070]
understanding if we can predict it.

[00:05:13.720]
And what we ultimately want to get is a situation

[00:05:16.490]
where our observations, and our concept,

[00:05:21.260]
or our understanding of the geology,

[00:05:23.280]
all marry together in a consistent,

[00:05:25.840]
this is what we call epistemic uncertainty

[00:05:28.080]
on the right hand side here,

[00:05:29.610]
because it relies on the knowledge,

[00:05:31.770]
the experience of the modeler (indistinct).

[00:05:34.120]
<v ->So certainly in Australia sense with,</v>

[00:05:37.690]
we losing as Darren says that that knowledge,

[00:05:40.880]
that the fundamentals of geological processes.

[00:05:45.233]
Which then is perhaps transmitted through a modeler’s bias

[00:05:50.790]
into what I would like to call dot-to-dot geology

[00:05:55.590]
in terms of modeling, where they’re actually joining

[00:05:58.910]
the lines of the top and bottom of each of the Stratus,

[00:06:03.710]
sometimes as much interpretation as you get now,

[00:06:07.210]
obviously that’s a bad scenario to be in.

[00:06:10.750]
<v ->Well yeah, there are similarities and differences.</v>

[00:06:13.970]
So I think what Graeme and Darren,

[00:06:17.160]
were talking about early on about the industry,

[00:06:21.619]
and how uncertainty translates into ground risk,

[00:06:24.620]
that’s exactly the same in New Zealand.

[00:06:28.040]
And certainly I think that similar trends to in what we call

[00:06:31.670]
or used to call accelerated projects,

[00:06:33.820]
and then super accelerated projects.

[00:06:37.870]
They’ve certainly taken their toll on ground risk,

[00:06:42.350]
and pushed it further and further down the path towards

[00:06:47.160]
what used to be say in the planning and early design stages,

[00:06:52.210]
now into tender and even into construction phases.

[00:06:55.860]
And so in many projects now we’re still doing GI,

[00:06:59.020]
to understand ground conditions in construction,

[00:07:01.730]
which is a pretty big difference to what we had in the past.

[00:07:06.010]
<v ->What are the steps that Golder</v>

[00:07:07.230]
are taking to kind of attack this challenge,

[00:07:11.730]
of epistemic theology and uncertainty in your day-to-day

[00:07:16.410]
practice with your reliability model approach?

[00:07:20.539]
<v ->Yes, so this came up, oh, it would have been about</v>

[00:07:23.080]
five or six years ago we had a particular project.

[00:07:26.520]
And we had to convince a director

[00:07:30.070]
within the state government, that it was worthwhile to spend

[00:07:34.130]
some more money on doing investigation.

[00:07:36.330]
Went to the client with a plan, it was just a map saying,

[00:07:39.707]
“Here’s red, this is where we don’t know much,

[00:07:41.720]
and here’s green, here’s where we know a bit more,

[00:07:43.610]
and hey, your tunnel goes right through this red bit.

[00:07:46.260]
You need to know more.”

[00:07:47.093]
And it resonated.

[00:07:49.180]
<v ->Right.</v>

[00:07:50.013]
<v ->This director looked at that and said,</v>

[00:07:52.037]
“Okay, I don’t want red on my project,

[00:07:53.900]
I need to do something about that,

[00:07:55.830]
and I’ll invest the money in doing it.”

[00:07:57.330]
So that sort of flowed on.

[00:07:59.980]
So this is the sort of thing.

[00:08:00.900]
So in a model like this,

[00:08:02.550]
you can see this boreholes in the middle there,

[00:08:04.490]
it’s obviously more certain around the boreholes

[00:08:06.520]
you see green lesser,

[00:08:08.560]
as you get further away from the boreholes you see red.

[00:08:11.500]
And what can be done with this model is this can be imported

[00:08:14.160]
into your BIM.

[00:08:14.993]
It can be imported and looked at alongside the project,

[00:08:18.640]
to give the designers, I think a powerful visual indication

[00:08:22.990]
of where they’re lacking in information.

[00:08:25.010]
So if they’ve got a critical piece of design in a red area,

[00:08:27.811]
that communicates to them, “Hey, you’ve got issue here.”

[00:08:31.570]
And they also use these sort of models to communicate

[00:08:34.350]
to directors, and project managers,

[00:08:36.400]
and those that we have to convince it’s a good idea

[00:08:38.730]
to invest in reducing uncertainty.

[00:08:41.440]
<v ->Makes sense, thanks for sharing that.</v>

[00:08:44.320]
If I jump back to everyone.

[00:08:45.980]
Graeme how have Jacobs kind of tackled that similar problem,

[00:08:50.923]
classifications, and systems, and frameworks,

[00:08:55.180]
to deal with it?

[00:08:56.570]
<v ->Yeah Pat, if you could just put up that slide,</v>

[00:08:59.320]
the next slide, please.

[00:09:01.729]
<v ->Sure.</v>

[00:09:02.562]
<v ->So we’ve looked at it in more of a qualitative way</v>

[00:09:06.110]
in trying to quantify uncertainty,

[00:09:08.360]
and it’s developed from some work done by Hails in 2004.

[00:09:13.800]
This was primarily based on more complex,

[00:09:18.910]
or more sensitive structures, such as,

[00:09:20.890]
this is an example from a dam.

[00:09:23.260]
Where we’re looking at different data sets,

[00:09:25.770]
and using descriptors to quantify the uncertainty

[00:09:29.540]
from a sliding scale.

[00:09:30.730]
So, you can see at the top of the table,

[00:09:33.020]
we’ve got implied, and then you go down to actual verified.

[00:09:37.560]
And say for instance, for implied again,

[00:09:42.080]
this probabilistic epistemic way of looking it

[00:09:45.550]
in terms of there’s no specific data available.

[00:09:49.830]
So you are using your knowledge and experience.

[00:09:53.170]
And then that carries obviously with a low level reliability

[00:09:59.050]
of that model, in that data set.

[00:10:01.970]
And then as you go down, and then you obviously,

[00:10:04.760]
you get to the end when you’ve actually got verified data,

[00:10:09.530]
where you’ve got a lot of local data available,

[00:10:13.500]
the rock mass, you are pretty happy

[00:10:17.270]
with the geological boundaries,

[00:10:18.940]
that have also been mapped in the field,

[00:10:20.680]
and you’ve been able to project them into the model.

[00:10:24.900]
And, you’ve got a high level of reliability,

[00:10:30.220]
the highest level you can of the model itself.

[00:10:33.670]
So that’s the way we’ve looked at it.

[00:10:36.500]
<v ->Yep, now that’s great.</v>

[00:10:38.640]
I mean, I think these kinds of frameworks to be systematic,

[00:10:44.210]
you need these kind of practices in there,

[00:10:47.130]
so I guess we’re also thinking about,

[00:10:50.540]
how can we support people with the implementation

[00:10:55.380]
of these frameworks in the model.

[00:10:57.420]
But Phil, if you take what your approaches at Aurecon,

[00:11:03.120]
and you look at towards the future,

[00:11:06.500]
what kinds of things are you aiming to do?

[00:11:12.060]
<v ->Thanks, yeah (indistinct),</v>

[00:11:13.614]
if you can just bring the slide up.

[00:11:15.660]
And certainly I support what Darren and Graeme

[00:11:19.300]
are talking about, those look great.

[00:11:21.930]
We are thinking about uncertainty

[00:11:24.040]
in a slightly different way.

[00:11:27.840]
And one note I’d make is perhaps the geology

[00:11:30.730]
we are working on in New Zealand tends to be

[00:11:33.120]
a bit more complicated,

[00:11:34.370]
and we’ve got multiple processes going on.

[00:11:36.380]
So for example in Auckland, where I do most of my projects,

[00:11:39.550]
we’ve got Quaternary volcanism, plasticine, glacial,

[00:11:44.645]
sea level changes, complicated mine seed rocks.

[00:11:48.450]
And there’s multiple processes

[00:11:50.440]
in each of those group of rocks that can and do cause

[00:11:53.830]
trouble to project.

[00:11:55.840]
So what we are excited about

[00:11:57.480]
in terms of representing uncertainty is doing it

[00:12:01.420]
quantitatively using techniques such as entropy,

[00:12:05.600]
Shannon entropy, which has been popularized recently

[00:12:08.590]
by a group of workers.

[00:12:09.780]
And Ashton of course is talking about as well.

[00:12:12.250]
And also Monte Carlo, which is the multiple interpretations

[00:12:17.130]
with perturbations of the input data,

[00:12:20.780]
according to understanding of the uncertainty

[00:12:25.890]
of each of those data points.

[00:12:28.310]
Now these are not complicated concepts,

[00:12:33.570]
but they’re a little bit hard to implement.

[00:12:36.770]
And I think that’s where Seequent can step in

[00:12:40.170]
by making it easier for us to adopt these.

[00:12:44.710]
<v ->Yeah, I mean, actually, that’s probably a good Seequent</v>

[00:12:47.650]
to some extent to the work that Ashton has done,

[00:12:50.060]
because he’s been also looking at that at that exact kind of

[00:12:56.049]
problem space and approach with his work.

[00:12:59.092]
Ashton, could you tell us a little bit about

[00:13:04.300]
how you’ve approached your work ,

[00:13:06.260]
and this notion of perturbing different inputs,

[00:13:09.220]
but not randomly with a plan?

[00:13:17.850]
<v ->Yeah, so I think it all starts from,</v>

[00:13:19.550]
so as the other panelists have shown it in great detail,

[00:13:22.800]
we have this multitude of techniques that we can use

[00:13:25.980]
to identify and characterize these different

[00:13:28.150]
sources of geologic uncertainty that could be affecting

[00:13:30.930]
our geologic models.

[00:13:32.450]
And I think really the tool that Seequent helped develop

[00:13:37.010]
for us was having this interface between

[00:13:40.680]
the geologic uncertainty characterization

[00:13:42.560]
that I’m performing outside, on the data set itself,

[00:13:45.540]
and with the actual process of geologic model creation,

[00:13:49.540]
opens the door to having this probabilistic approach

[00:13:52.290]
to geologic modeling.

[00:13:54.040]
So take, for example what is showing right now,

[00:13:57.190]
where we’re trying to assess the uncertainty

[00:13:58.620]
of a fault zone intersecting a tunnel alignment

[00:14:00.980]
from part of my research.

[00:14:02.810]
So we use a schematic diagram like this to identify

[00:14:05.810]
and conceptualize what are the key sources of uncertainty

[00:14:08.790]
that we need to consider?

[00:14:10.040]
And we use that to inform a probabilistic model,

[00:14:12.870]
to characterize these geologic uncertainties

[00:14:14.820]
using various probability distributions

[00:14:18.160]
which ties into this Monte Carlo simulation

[00:14:20.500]
that Phil just brought up.

[00:14:22.500]
Then with that bridge into leapfrog,

[00:14:24.570]
which supported automated perturbation

[00:14:26.660]
of an initial geologic model of fault zones,

[00:14:29.200]
we’re then able to look at the geologic model,

[00:14:31.700]
as a range of possible models, each of which conforms

[00:14:34.720]
to initial structures and uncertainty envelopes

[00:14:37.030]
defined by the modeler.

[00:14:38.590]
I think now is a good time to switch to the next visual.

[00:14:43.450]
And, with this new view of this subsurface geology,

[00:14:47.130]
this probabilistic geo model I’d like to emphasize,

[00:14:49.710]
it’s not a replacement for the deterministic model,

[00:14:52.830]
but it’s just an additional way to view and analyze

[00:14:55.150]
the conventional geologic model

[00:14:56.790]
and uncertainties associated with it.

[00:14:59.230]
And from what I’ve seen today, this cutting edge format

[00:15:02.300]
for subsurface characterization

[00:15:03.980]
using this probabilistic geologic model,

[00:15:06.570]
is really pioneering novel visualization

[00:15:08.830]
and analysis techniques.

[00:15:10.830]
For example, visualizing the maximal zones of uncertainty

[00:15:14.780]
in your geologic model using this information entropy,

[00:15:17.480]
which is shown in the contours in the block model shown.

[00:15:21.060]
Or by using thresholds on the likelihood

[00:15:23.700]
of fault zone occurrence,

[00:15:24.880]
to give a straightforward communication of uncertainty

[00:15:27.170]
to model users, so it’s not just, you have a deterministic,

[00:15:29.950]
you’re in the fault zone or you’re not,

[00:15:31.260]
you now have a 75% chance,

[00:15:33.530]
85% chance to be in a fault zone here.

[00:15:36.240]
And I’d like to add one more thing on that,

[00:15:38.780]
is that as we talk about these probabilistic methods,

[00:15:41.490]
I want to emphasize that just as the initial geologic model

[00:15:44.350]
is defined by the modeler themselves,

[00:15:46.380]
and subjected to their own biases,

[00:15:48.740]
so is the probabilistic model that you parameterize

[00:15:51.720]
for uncertainty characterization.

[00:15:53.250]
So there’s that necessity to have a careful

[00:15:56.030]
consideration of these epistemic sources of uncertainty

[00:15:58.670]
when creating the geologic model and the probabilistic model

[00:16:01.820]
to go along with it.

[00:16:03.940]
<v ->Yeah, you can’t escape the epistemic</v>

[00:16:05.650]
at the end of the day.

[00:16:07.900]
I don’t think.

[00:16:10.340]
So, it’s great to see these realizations,

[00:16:14.640]
I think they’re very powerful.

[00:16:16.520]
What do we need to keep our eyes on when we are talking

[00:16:19.800]
about handing off models to other users perhaps down stream

[00:16:24.350]
in these beam processes?

[00:16:28.300]
Phil, I think you’ve had some experiences in this area.

[00:16:31.790]
<v ->Yes, we have had trouble, and I think we all have</v>

[00:16:35.840]
the same experience, Darren and I,

[00:16:38.150]
in a modern infrastructure project,

[00:16:40.210]
there’s hundreds of people, and multiple disciplines,

[00:16:43.940]
all working very capably within their own

[00:16:47.420]
software preferences.

[00:16:49.030]
Then the models get extracted into revert

[00:16:53.550]
or multiple other softwares to suit

[00:16:57.610]
the needs of other disciplines,

[00:17:00.170]
and that’s where we’ve had some trouble.

[00:17:03.610]
Because the the data or the surfaces that people want

[00:17:08.050]
to design to, get separated from

[00:17:11.840]
the representation of uncertainty,

[00:17:14.340]
get separated from the supporting data.

[00:17:18.148]
And, the understanding of these features may vary in space

[00:17:25.280]
quite considerably is not clearly understood.

[00:17:29.030]
That’s where I would like some sort of uncertainty actually

[00:17:35.478]
in the surface, so within the data structure

[00:17:38.200]
that supports the functional surface,

[00:17:40.560]
if we could have some indication,

[00:17:42.510]
or some representation of uncertainty,

[00:17:45.940]
that would be exceptionally good.

[00:17:48.096]
Of course, it’d have to be developed by Seequent,

[00:17:51.170]
and then taken up by other software vendors,

[00:17:53.970]
but that’s the sort of thing that I would like to see.

[00:17:57.050]
<v ->Really appreciate everyone’s time, it’s been fantastic.</v>

[00:18:00.590]
If I could just summarize what we Seequent has taken away,

[00:18:04.530]
hopefully, this represents the big takeaways

[00:18:07.416]
from our group here as well.

[00:18:09.650]
I’ll flip back to the presentation.

[00:18:11.790]
So, the big takeaways for us, I think,

[00:18:14.880]
was that both the deterministic and probabilistic

[00:18:18.040]
methodologies have a role to play,

[00:18:19.960]
as we try to deal with these different

[00:18:22.860]
sources of uncertainty.

[00:18:25.170]
And the fact they’re intrinsically interrelated.

[00:18:28.480]
We need to support the entry of these quality indicators.

[00:18:32.590]
Basically, try to provide hooks in our models that

[00:18:35.905]
enable the frameworks that have been built,

[00:18:38.570]
the rubrics that are developed within companies like

[00:18:41.910]
Jacobs, Golder, and Aurecon, to enable them to just easily

[00:18:46.880]
express their own approach to certainty and uncertainty.

[00:18:51.690]
And, potentially then use those things in downstream

[00:18:54.780]
calculations and workflows.

[00:18:57.520]
It did strike me that if we can support better

[00:19:00.860]
review processes, and create this collaborative environment,

[00:19:04.010]
connecting those with the experience to those that are new

[00:19:06.840]
to modeling, that we were going to be able to help reduce

[00:19:10.360]
some of that epistemic uncertainty,

[00:19:12.870]
that’s dependent on the modeler,

[00:19:14.190]
if you get more eyes on the model.

[00:19:16.160]
And then, in this handoff sense,

[00:19:18.320]
the reliability information needs to flow with the model,

[00:19:22.490]
and it’s key outputs into other software environments.

[00:19:24.860]
And that’s a challenge for not only Seequent,

[00:19:27.110]
but the rest of the software industry that supports

[00:19:31.150]
civil infrastructure projects.

[00:19:32.880]
And, with that, I’d like to say,

[00:19:35.430]
thank you very much gentlemen for this very engaging panel,

[00:19:40.360]
it’s a massive topic.

[00:19:41.770]
I think we scratched the surface,

[00:19:43.340]
but I’d certainly got a lot to add to it,

[00:19:45.437]
and I hope you have too.

[00:19:46.830]
<v ->Thanks Pat.</v>

[00:19:47.690]
<v ->Thanks very much</v>

[00:19:48.900]
<v ->Thank you all.</v>

[00:19:49.733]
<v ->(indistinct).</v>