A Scientist’s Take on Climate Models and Risk Management Applications

A Scientist’s Take on Climate Models and Risk Management Applications

– My name is Cynthia Vitters, and I lead the Enterprise Risk Management Practice at Deloitte for GPS, which is the government
public services sector, which includes the
federal government state, local government and higher ed. So, you know, before we get started
and I kick things over to start the webinar, I just wanted to take a few
minutes to talk to everyone about the Maryland Smith Risk Academy. And this series of webinars
is sort of in a sneak peek preview to what is to come
with the risk academy, which we're about to kick off. And, you know, just a
little bit about what it is. It's a collaboration between
the Maryland Smith school and Deloitte to strengthen and sharpen mid-career professional skills and applied risk management topics. And when we kick this off, we
will have a total of seven, two hour modules that will
be presented via zoom. And, you know, just to give you a preview of what those topics will include, the modules will include
building a sustainable risk management program, ERM, and credit risk, cyber risk and mitigation for regulators, risk modeling and governance, liquidity, and capital risk,
operational and regulatory risk management, and finally climate change
and financial management risk.

So each module will be led
by professor Cliff Rossi. Who's also here with us today, Deloitte professional, as
well as irrelevant module luminary who will provide
practical real-world examples in the space of the module
that we're conducting so that the program is set to launch on January 11th, 2022. So right around the corner. And it will conclude February 1st, 2022, the Smith school is hosting
informational sessions to provide greater context on the program, which are free and available
to the public for signup via their website. So again, this is a great
opportunity for each of us to learn and grow with
like-minded individuals and the broader financial services and government risk management space. So again, today's webinar
is just a sneak preview of what's to come early next year with the Maryland Smith
School Federal Risk Academy. So again, thank you all for joining and I'm gonna hand it over to Cliff Rossi, who will introduce the topic
and our guests for today's session on a scientists'
take on climate modules and risk management applications over to you Cliff. – Well, thank you Cynthia, for your support for our risk academy and this webinar series.

So today we are actually
deviating a little bit from our typical guests. That's usually either a senior
risk leader in government or industry, and we're
going to feature professor, Tim Canty, a real climate scientist to help us better understand what's behind the climate models, underlying much of the climate
risks scenario analysis that folks in the
financial services industry and their regulators are
dealing with these days, such as those that are
produced by such organizations as the network of central
banks and supervisors for greening the financial system or NGFS as you may have heard. So Tim is a faculty colleague of mine at the University of Maryland, and he's working with a number
of us from across the campus on an interdisciplinary climate
finance and risk program.

Tim is also an associate
professor in the department of atmospheric and oceanic sciences, where he serves as the
co-director of their undergraduate and professional master's
programs, which he helped design. He's also the director
of the Marine mastering and environmental sciences program, which is a statewide
graduate program across four universities and
university of Maryland center for environmental science, before coming to Maryland in 2007, he was a Caltech postdoctoral scholar at NASA's jet propulsion lab and a lecture at the
department of atmospheric and oceanic sciences at UCLA. He studied stratosphere. It goes on loss, air quality
policy and climate change. So, so welcome Tim. And thank you for joining
us for this webinar series.

Before we jump into the
discussion for everybody out there in the audience, I'd like to set the
stage just a little bit. The focus on climate
change has accelerated as you well know, it
seems within the last year with several notable highlights, the UN's climate change conference COP 26 just concluded last week, the IPCC released as much
anticipated AR six climate report and various financial regulatory
and oversight authorities in the U S are ramping up their activities around climate change,
including the federal reserve, the OCC, the OCC the federal
housing finance agency, New York state department
of financial services among others.

And central to those efforts our climate models used to
form the basis of scenario and stress testing and
risk assessments for things like climate disclosures for the TCFD task force on climate related
financial disclosures, for example, and the complexity of these
models tends to relegate a deep understanding of them to
the scientific community. And yet the companies that
we're talking about here and their regulators are
going to need to figure out new ways to adapt their
output to conducting these financial risk assessments. So Tim, with that as a backdrop, would you just mind maybe
kicking us off today by giving us brief overview of exactly what we're talking about
with regard to climate models that we read about in the popular press and what are some of the
key inputs and outputs? – Sure, thanks Cliff. And thanks for having me this morning. First off, I wanna share
everyone there won't be a quiz at the end of this hour, and also I'm going to
apologize in advance.

I think we can all agree
that within our communities, we are very comfortable
with our own acronyms and definitions. And I try very hard not
to get into that jargon, but if there's something I
say that doesn't make sense, please stop or put something in the chat. And I'll be sure to try to clarify that. And just a case in point, I was doing outreach awhile
back talking about satellite instruments, and this was to
a group of middle schoolers.

And the one student was
wondering why there's trombones on the space station like, oh yeah, we use the word instruments
very differently. (laughing) So with that in mind, so, you know what goes into climate models, what doesn't go into climate
models and how they've developed over the years, fundamentally climate models
are basically a mathematical representation of what I'm
gonna call the earth system, the atmosphere, the ocean,
the land, the biosphere, the cryo sphere ice, and how
all of these things interact.

And that's, it's a tall order to try to
represent all of these systems in basically lines of computer code. And back in the sixties and seventies, the computer models were
very simplistic compared to today's standards. Basically you had a surface
and you had an atmosphere and maybe you tried to
get precipitation correct in temperature, but as computers
have gotten more powerful, it's allowed the scientific
community to increase the complexity of the
computer models to represent interactions on finer scales. And fundamentally, it's basically just applying
a whole bunch of equations to understand the energy balance within the earth system and how that impacts things
like drought and rain and where clouds are forming
and what temperatures are.

And the models are basically
they're designed to break up the entire planet into
a series of grid boxes as a function of latitude,
longitude, and altitude. So you take the atmosphere,
the land, the ocean, and just divide everything up into boxes. And the initial models were two boxes. You had land and you had the atmosphere and over time we've
gotten more and more boxes and they're much smaller. I'm gonna say finer resolution, but these comment of computational costs, which requires more computer code. And I think the modern models. Now, if you actually printed
out the computer code, some estimates I've read
up to about 20,000 pages of printed computer code
that go into these models to represent what's happened in the past and also use to make predictions of what happens in the future. And it's not just representing
the planet on smaller spatial scales, but also
time or temporal scales.

You know, maybe in the past
you could run a computer model for a given year and you
just do year by year by year, but then it's down to months and then days and then hours and then minutes. And so you're trying to represent
all these very complicated processes on finer and finer temporal time and spatial resolution to try to develop the best representation of our understanding of this earth system requires a lot of time, hundreds of scientists and engineers, just to develop the
inputs and the basic code, and then testing these
things are battleships or aircraft carriers. You can't just hit the
download wizard on windows and put it on your laptop. They require massive
technological investments and advancements and computer technology, which makes it difficult to develop, to maintain and only some countries.

And now some companies and
universities are able to develop and maintain these models. Fundamentally I'm a modeler. The old adage is all models
are bad, some are useful. And as a modeler though I'm
not beholden to the models, they're part of a toolkit to understand and probe our understanding of what's going on outside. And if the model doesn't
agree with the atmosphere or the ocean, I'm not gonna
to say that mother nature is wrong. Obviously the model is always
right, because I'm a modeler. No, it forces us as a scientific community to go back and say, okay, what's going on? What have we missed? And you know, so there's a lot of demands on the climate models to tell us exactly what's gonna
happen in the next 10, 20, a hundred years.

And also where exactly these
things are gonna happen. And in terms of, you
know, inputs and outputs, some of the inputs make
sense, greenhouse gases. If we, you know, agree
on the science here, that greenhouse gases lead to
kind of an increase of warming of the lower atmosphere. So how do greenhouse gas
concentrations change and Cliff and I are joking earlier, I've got my own CO2 monitor in my office, you know, globally, we're at
about, I think, four 20 parts, PPM parts per million, my office at 116 yet I'm
not living in a tropical paradise in my office.

So, you know, how do you
represent these small scale features and large scale features? So greenhouse gases are inputs. Things people may not
consider those important to population. You know, if there aren't any people, we don't care about climate,
cause we're not here, but fundamentally it comes down to energy use whether it's for growing food or to keep the lights
on and keep the heat on and how many people are
using what energy sources and how does that impact
greenhouse gas concentrations, land use, sit what I'll
call aerosols, you know, just dust and particulate
matter what we call in the atmosphere where
that's coming from, where that's going, how those things affect clouds.

So all of these inputs
have to go in the model to give us the outputs of interest. Temperature's one of the main ones. Precipitation is also another
one that a lot of people, are interested in because of
flooding in more recent years, drought, wildfires,
hurricanes, severe weather. This is what we're trying to pull out of, these climate models to help people understand their
risks moving forward. So, you know, that brief summary doesn't
do justice, I think, but hopefully kind of lets gives us some foundation of what
we're talking about. – Yeah, that was it was a tall order to ask
you to kind of condense all that down. And again, when we talk
about climate models, we're talking, there's not
just one climate model, there are many different
climate models, regional models, global models that people use.

And the like, and so it does get a little unwieldy very quickly. – Yeah, and on that note,
currently the AR six, the sixth assessment report
that was just published, you know, what models are
used and there's different flavors of models too. Currently there's about a
hundred different models that are used in these climate simulations coming from about 50
different entities globally. So this is kind of what
we're talking about. And as I said, different types of models, these integrated assessment models, these, global climate models, which are also called atmospheric
ocean general circulation models, that couple,
the atmosphere and ocean Cliff as you said, regional climate models that
can focus in higher detail on specific parts of the world.

This is the, these are the tool boxes
that are currently available to the community. – Yeah, and that's what
makes this a particularly, you know, those of us that are risk practitioners are in the
financial services industry or regulatory community. It's almost mind boggling the amount of, you know, modeling that goes on and how are we, you know, as I refer to it, the proverbial square peg
and the round hole syndrome, how are we to kind of take those outputs and then try and kind of
weave them into our financial and risk analysis? I'll save that for another moment here, but let me ask you as a follow-up to that.

So if you were put on the spot, how would you characterize
the accuracy of these models? Realizing, again, we're
talking about many models and some are more accurate than others, but how would you generally assess that? And how would you compare
since we now have AR six? How would you compare improvement
over time in that regard? – Yeah, and that is good, that is a tough question
and really gets to the heart of the matter is how
do you define accuracy? You know, what are you looking at to say, oh, our model's really good. And we have a lot of
confidence in our model enough that we're gonna
project into the future. And, you know, thinking back
to our undergraduate years, it's like having the answer
in the back of the book and trying to reverse
engineer that answer. Sometimes we run the global, we as a community, the climate models are run for
what's happened in the past to see if those global
climate models can reproduced what we already know happened. And if we do a really good
job of what's happened in the past, then we say we have good
confidence to move into the future.

But I do have a slide I do wanna bring up, I think is represents this pretty well. If people can, can you see this? – Yeah. – Okay, so this is the temperature record. This is the global
surface temperature record going back to 1850 and the black line are the
observations for the global temperature record. The gray shaded region is the
representation of this global temperature record from the
models used in the AR six. And so this uncertainty, you know, we don't have just one line
superimposed on another line. This range is gray. Shaded region shows us the
range of the climate models. – We may have lost him. We'll give him a moment to. – That difference based
on different scenarios. did I freeze? – Just a little bit? – Okay.

Did he you anything important? – Probably, but maybe not.
– I don't know. So, yeah, the gray shaded
region is the ability of the models to reproduce the past. These colored lines are these scenarios, these socioeconomic
pathways moving forward that represent what if, what if we do this? What if this happens? So as I said, you can't, this isn't just one
perfect line superimposed on another line. This is considered good
enough for the climate models. And also considering the
further back in time you go, you know, you're tying the data
that is also more uncertain where we have fewer observations
the further back in time.

So you're trying to tie a model to a time period where we just don't have as much information. And part of the global climate models, what they have to do is represent physical processes that
would be happening. Even if people weren't on the
planet, people will say, oh, climate would change. Even if there wasn't
anybody on the planet. Yes that is quite true. And so how do you
represent those processes? Like the El Nino Southern
oscillations as a term, probably many people have heard, which has kind of helps us
understand El Nino, LA Nina, kind of the sloshing back
and forth of warm water in the tropical Pacific ocean.

You know, this process would happen, whether there are people here or not. And how do you represent
that in the models when we don't have really good
observations back in time, and we don't have a
satellite record from 1850. So you use proxies to
represent these impacts, you incorporate them in the model. And this is kind of the
state of the art state of the science of where we are right now. This is considered pretty
accurate from climate standards. Is this accurate enough for risk folks? And I think that cuts to really the, where we're at right now, the climate models are designed
to represent climate based on our best understanding of the science. I'm not quite sure these
are the right tools. They're not designed for risk managers. And I think the key is how you translate this into something that is usable by finance sector, governments,
that sort of thing. – Yeah, that is the
$64,000 question, right? Because as we've talked, if you look at the traditional, let's say stress test that
is imposed on a large bank or the GSA is, you know, they may only go out nine quarters and what you just showed
us there, you know, depending on these SSPs
could go all the way out to the year 2100 or, you know, you go out to even 30 years
and you have to ask yourself that cone of uncertainty is so wide.

How do you really, how do you feel reliable when
banks are making strategic decisions every day and
even on a strategic plan, how do they get comfortable
that implementing this is going to lead them down
a path that will be helpful and useful for them in
making those decisions. – Right, and also understanding
that those projections in the past those shaded
regions that I showed are based on an ensemble of models. This isn't just one singular model. It's all of the models that are available and they're all coded
slightly differently. And you may ask, why do we need all of
these different models? Well, as scientists, we
need access to these tools. And if there's only one model out there, we're not gonna get a lot of work done because everyone has different
areas of the climate system that they're focusing on.

And I need to do a simulation
that better represents El Nino. I need someone else
needs to do a simulation that better impacts or better understands the impact of Saharan dust blowing across the Atlantic
ocean that could impact air quality and cloud
coverage, over Florida. You know, so everyone
develops their tools. They all try to, they are based on the
fundamental same equations, but how they're incorporated
and understanding too. And we're getting in the
weeds a little bit here. This whole system that we're
talking about is not linear. You know, I can't just say, oh, I'm gonna increase carbon dioxide and everything else is
all these other things are gonna increase. You change the balance
of energy in the system. It takes a while for the
system to equilibrate, how it moves that energy
throughout the system. So I may increase temperatures, but that increase in temperature,
goes into sea ice melts.

Well, the sea ice melting
makes the surface darker, darker surfaces absorb more heat. That leads to a further
increase in temperature. These what are called feedback mechanisms are where there's a lot of
uncertainty in the climate models and how they are incorporated. How do you resolve where a cloud is? Weather models are designed
to do these things, climate models, aren't
necessarily gonna tell you that there's gonna be
a small cloud right there, outside my window here at, you know, two miles above the surface
because the spatial resolution is such that it's really
difficult to do that.

And so we're asking
the models to do a lot, but I think, you know,
one model can't do it. We use an ensemble. And from that ensemble of models, you can come up with these
probability distributions of what's gonna happen in the future. They give us, you give you a most likely scenario, but I think even then we
need to delve in further, why are the models saying that? And are these averages in the future, based on some outliers, are there a cluster of
models that are too high or too low compared to the past? What happens if you remove
those from these probability distribution functions, you can make all sorts of pretty pictures of what's gonna happen in the future.

But I know I'm gonna, take it on the chin for
the science community. I'm not sure that the
science community, you know, we try, but we're scientists
and we're, hyper-focused, on solving this problem at hand. And I think, a lot of
scientists realize this, and we're trying to do a
better job of communicating this to what we call
stakeholders or end users. I do a lot of this from
the air quality policy side of things, but also the
climate side of things. And I teach an introductory
class on climate to flute majors you know,
and basketball majors. And so I gotta make sure
that people can understand this without differential
equations and all of that. And it's a challenge,
it's a real challenge, but it's a chance we have to face.

– Let me ask you a question about this. You know, we're talking about, the complexity right? Of these models. And it reminds me you know back in the
day, having, you know, run groups where periodically
we'd be subject to, you know, our regulatory model risk
management oversight. And there are very, you know, robust requirements or guidance around how one goes about validating, let's say a financial
model, pricing model, or a loss forecasting model. And I'm thinking to myself, the complexity of these climate
models just goes way beyond. I think some of the
complexity that we have on the financial and risk side of things in banking. And so how do you as, a
climate or a science community validate your models? And I know we talked about
back testing and whatnot, but at the end of the day, I just have to kind of ask the question, how do you get comfortable
that these models are right? – Yeah well, you compare to
what's happened in the past. You know, we know what the world did, does the model represent what happened in maybe a particular region? And I think this is where
having a variety of models.

We have the large global climate models, these beasts of computer code. And I think this is where
regional climate models can be really helpful in terms of focusing on a specific area. And can you take, what's coming out of these
larger climate models, which again, aren't necessarily designed to do what the finance sector, perhaps the banking
sector, whoever asking, but to translate use what's
coming out of the climate models as inputs into a
specialized series of tools that can focus on supply
chain, flood risk, and flood seems to be, you know, the lower
hanging floor, you know, where is the water going? We know the model says, this has how much water should
be coming out of the sky. We can compare that to data in the past, whether it's streaming and inundation observations or data from
meteorological stations that measure rainfall rates
and the amount of rain, which model gets closest. Okay, why does it get closest to reality? And then, okay, the model says, this is what's falling out of the sky. You translate that into another model that will like an inundation
model that will tell you, okay, this is what's hitting the land.

We've characterized the
surface where the concrete is the slope of the land, where the water is getting funneled into to help basically translate
this global climate model output into something that is more usable for a specific stakeholders. And so, and this is, you
know, the federal government, NOAH is a, the national geographic
atmospheric administration is really focused on this
juncture of here are the weather forecast models, which
really get us about 10 days in the future. And many of us have, you know,
weather apps on our phone.

And then there's the climate
models that go out to 2100. So we've got 10 days and
2100 at the extremes. It's the seasonal, the sub
seasonal in the middle in there. That is a real focus of
the scientific community, recognizing that the global
climate models, you know, mortgage is 30 years in the future. You don't care what's happening at 2100. Well, what about Ms. 30 year time period? And how well can we represent
the seasonal changes over time in seasonal changes
under a climate scenario. And so the scientific
community, is working on this, recognizing this, but I
think new tools are needed. And I think tools designed
for the risk community. And I think this is where
we need a closer interaction and you know, this effort that Cliff
and I are working on here and, you know, over the past year, just recognizing that we
use the English language in a different way and
figuring out a new language to translate, to understand what the heck we're talking about and saying, for me as a scientist, like,
oh, this is what you're asking.

Okay, this is, I think the,
that the tool you need. And to me, I guess as
someone of my opinion, that's where I think we're
at where we are right now. – Yeah. One of the things that I think companies and the regulatory community at large may struggle with a bit is okay, these are not models
that you're typically going to be building in house, right? This is not a loss forecasting model where you're sitting on top
of your own level information, and you can import all the
macro economic factors. You need to kind of run
something, you know, this you're reliant on, let's say maybe in this case
more reliant on vendors, supplying climate models
to an organization. And of course, when you're talking about vendor models and I've vetted those in the past, the places that I've worked at, and they typically are black
box and I get it right, they're proprietary, all those things.

So, you know, we probably have a number of people that are here in the audience today. They're struggling with that
and so I'd asked you, you know, what should companies
be looking at or asking of their climate model
vendors about their models and what can they do for them
in terms of their climate risk assessments and
disclosures in general. – Sure, and just real quick,
you brought up a point, you know, that the models, the global climate models historically have an included kind of economic factors and the kind of feedback
of usage of energy, that sort of thing. The new, you know, SSPs are
called socioeconomic pathways, do try to represent
different scenarios in terms of the energy intensity, energy usage, how countries are willing to
share some of their cleaner technology versus keeping
it very much in-house.

And so these SSP scenarios moving forward, do try to take these different
types of future worlds into account. Maybe not exactly what these
different stakeholders need, but we're getting there. Now in terms of, I think what
vendors may want to be asking or what, you know, people here may want to be
asking of their vendors, who are translating a
lot of this information in the global climate model information, the outputs are freely available. If you know where to look or
who to ask, it's all out there. Now that you're talking
massive amounts of information in file formats that most
people aren't familiar with. And how do you take that information, understand that information
and make sense of it. And so this is, I think, where a lot of vendors
do is take on that role of translator.

But I think it also requires
some depth of knowledge, not just saying here's what
the climate models are saying. Here's your answer again, asking why are the models
giving us this answer, I think is a key here. You know, what assumptions
are the models making? How are the models? You know, if we're
looking into the future, handling some of these
uncertainties in terms of is what I said earlier, I kind of alluded to
these feedback mechanisms, greenhouse gases increase.


The amount of increase
increase will determine how temperatures change. So why is your model saying
the temperatures are changing, taking into account this
redistribution of energy across the planet. And is the model saying
this part of the planet is getting warmer and perhaps this part could be getting colder. And so are your, you know, if you're focused on, let's say a whole bunch of
mortgages along the Eastern seaboard is one particular, are some of the models
getting these differences in temperature? Are they contradicting
each other moving forward? So I would say a deeper dive
without going into equations and all of that. If you're like me, that's a
good way to put me to sleep, but really to say, okay, this is what the models are telling us. And just embracing, look, this is the best we can do right now. And being very clear and transparent. This is what the models are doing.

This is the best we can do right now. We recognize that the models
are the scientific community is still trying to nail
down the uncertainty of the impacts of aerosols
particulate matter dust suit in the atmosphere
and how that affects clouds and how clouds then will impact the climate system moving forward. Look, just to be very clear, this is the best we can do right now.

This is what we're working on. And I think also being able, I think what is important is to provide some honest feedback and say, okay, this is the question you
asked, and we did that for you. But actually this is the
question you should have asked now based on our analysis, I think that would be valuable to people working with vendors. Maybe some of the vendors
will disagree with me. I don't know, but I think
that's, and as a scientist, that's what I think is important. Kind of a, way to understand why maybe it's not the best available, or maybe it's not the best, but it's the best available and the best we can do right now.

– I guess I come back to
and the practical reality is how do you even tell, how do you differentiate once you go, you have three vendors
come in and, you know, they give you, you know, you run your loans through it and you, you look at them and you go, how do you select, you know, how do you know who's right
at the end of the day? – Yeah, well, and I think he used an
ensemble of the ensembles. I guess there's one thing
you can do and say, okay, this is the average of the vendors. But again, just as I tell my students
justify your answer, justify your results. Why did your model, why
is your model saying this? Did you apply some statistical down scaling, if the global climate model you're using has an a hundred kilometer resolution, and you're telling me what's
happening on a hundred meter resolution, how did you get there? If you're doing some statistical
analysis analysis to say, look, the model tells us on average, what's going on in this grid box and you claim to tell me
whats going on right there.

How did you do that? What did you do to constrain
this coarser output onto a finer resolution? If you can answer that
question with some certainty. Okay, great. And I think that can give a stakeholder better understanding at least better comfort
in saying, okay, you know, this was very complicated, what they did. It's kind of, state-of-the-art understanding the
limitation of the models, and this is what we're
gonna move forward with. You know, having, I'm not complaining about the models.

I think they're great for what
they do, the climate models. I don't think they're designed
to do what people need, but let's understand
that the climate models in their predictions have
been pretty consistent since, you know, the 80s and 90s have all kind of said, you know, this is what we expect to happen. If you take a model from 10, 20 years ago, they're not that far off
from where we are now and from what is actually happened. So the climate models are telling and have been telling a
relatively consistent story. Yes, there's a lot of
uncertainties involved, but I've been telling a
relatively consistent story. And the new AR six reports
has further emphasized on past reports and say, we really, really with high confidence
where in the past, they'd say with some
confidence or medium confidence with high confidence, we say that these things are happening. And I think what differentiates
this recent AR six report is that they're saying now, which they haven't said in the past, is that a lot of extreme
events, not a lot, but extreme weather events
now can be attributed to a changing system where
in the past they would say, it seems like this would be, you know, this exceptional rain fall events we think may be influenced
by a warming climate.

Now the AR six models saying, yes, we based on our analysis
is most likely or very high confidence that these extreme
events are being caused by a warming climate. – Interesting. I mean, at the end of the day, what do you think are implications then from this AR six report
for financial services and companies in general? – Well, I, you know, I think it's obvious that
risk is gonna increase and it's not just, so if you project on what's
based on what's happened in the past, here's, what's
happened in the past, we're gonna project a straight line. You're gonna be off because
these changes are happening. Non-linearly they're happening faster. And so if the past world
is a different planet, it's happened, it's changed now and is
changing in the future. And so we're trying to predict a world that hasn't really
existed yet in recent time. And so making these predictions, this is why you have all these
SSP scenarios to say, okay, we're not just gonna base on, you know, straight line extrapolation of
what's happened in the past, based on our best on our
understanding of the physics and the science and the
chemistry and the biology.

We think this is what the
system is gonna look like if we stay on this path
or this path or this path. And there was a lot of talk
recently about, you know, carbon dioxide is always the focus, but methane recently, and the AR six and a lot of discussion in cop 26 is let's try to regulate
methane on a per molecule basis. Methane is a much more
potent greenhouse gas on a per molecule basis,
then carbon dioxide, and about 25 to 30 times more powerful, but methane also has a shorter lifetime. And so the thinking is, well, we can affect some change now
by limiting methane emissions.

And that will have a more immediate impact while we're trying to solve
the carbon dioxide issue. There are other greenhouse gases out there nitrous oxide, CFCs, which were banned under
the Montreal protocol because they lead to the
destruction of the ozone layer, also have an impact. And, you know, again, without getting into
the mathematics of it, if I increase carbon dioxide, it's gonna have a different
impact on global temperatures. And if I increased methane
and trying to understand these varied impacts as
part of the complexity of the climate models and
then kind of the downstream, so to speak impacts, okay, we've incurred, we've changed the greenhouse
gases concentrations. This is how everything is going to adjust. And you want us to tell you if this city is gonna be flooded out
by this particular date, and that's where the difficulty lies, but the models are all saying
these things are happening. And now the latest climate
monitor results are saying more severe weather, more extreme events, and the data have shown
more extreme events.

You look at the rainfall
in New York city from Ida, you know, what four inches
of rain in three hours or in one hour rather, I
mean, that set a record. And that happened in a
way that no one predicted it could happen. So you're trying to get even these weather forecasting models to predict things that have never really happened before. and that's a big challenge. – Yeah, and it is the big
challenge and the big, you know, at the end
of the day, I know you, and I've talked about
this at to some degree that, you know, we've got
all these various, you know, cop 26 and a Paris core and
all those kinds of stuff, climate accord, and the like, and you mentioned something
that I thought this group might be interested in hearing about, which is, you know, there is hope, right.

There is a precedent, if you will, for how policy makers, industry and academics
scientists can come together to solve a huge environmental problem. And that was around ozone is that, there was, there's quite a
bit of discussion about that. You know, years ago, and I don't know that
it's fully been resolved, but it's a lot better off isn't it than. – Yeah, no, this is a, this is a very positive
example of some of the best environmental policy that
the world has ever enacted. And that was the banning
of chloral floral carbon CFCs for use in propellants
and refrigerants the pushback at the time as, oh, you know, the scientific community
is trying to try to get rid of our air conditioning in our hairspray. But as anyone upset, I
asked my students, you know, they're only born this century.

They're young. Like how many of you are upset
about the banning of CFCs? And they're looking at him like,
what are you talking about? And so the global community got together to solve the problem. And everybody benefited, the
ozone layer is recovering. It's getting better and
this is demonstrable. The data is showing that
the ozone hole is shrinking. Now you get year to year
variability, of course, but overall, it's getting better because everyone got together to agree. Okay, how, because no one wins. If you get rid of the ozone layer, you sterilize the surface of the planet and we can only live underwater. And most of us are ready to do that, but businesses weren't destroyed.

We still have air conditioner. We still have hairspray and propellants, you know, people adjusted, but DuPont was front and
center in this developing the replacements to CFCs. And also by banning CFCs,
there's this whole world avoided scenario because CFCs are
some of the most potent greenhouse gases ever created. And they're completely manufactured. Unlike carbon dioxide and methane, which have all sorts of natural sources, CFCs are completely manufactured. And they are a wonder of technology because they don't interact with anything. They're not dangerous. You can inhale and exhale
and they won't kill you. But because of that, they have very long lifetimes and they make it in the
stratosphere where the chlorine is stripped off, and that can lead to the destruction. The ozone layer, the global
community got together, banned them has been monitoring
CFC mounts in the atmosphere actually recently caught
a country cheating.

And the scientific community said, Hey, they CFCs has started to go up. And there is no way that should happen and figured out where that
was coming from contacted that country, that country
went to those companies and said, you cannot use these anymore. And because of this six
successful legislation, we are only talking about carbon dioxide, methane and nitrous oxide. Otherwise these CFCs would be
having almost the same impact that carbon dioxide would be
having with a thousand-year lifetime as opposed to
carbon dioxide, which is CO2.

So we can get together in
ways that everybody benefits. I mean, it's not easy, but how do we make changes
that allow people to benefit? And that really means we need
to take into the full costs of the taking carbon from
the ground and putting it in the atmosphere. What is the actual cost in terms of, and who's paying that cost
of flooding out all the coastlines, you know, the reinsurance companies who's gonna lose from these future scenarios
that we already see this happening. A couple of weeks ago, we had a lot of flooding here in Maryland from a low pressure system
moving through, you know, this was not a hurricane off the coast. It was just a storm. It was a mid-latitudes
cycle and that came through. And then they're talking about
another storm coming through and parts of the country
on the east coast, which I'm more familiar with.

You have sunny day flooding. You could go to downtown Annapolis here in Maryland. And on a sunny day on a king tide, you have streets, parts of the streets flooded
because sea levels rising. You know, Miami is
taking action against us, putting in pumping systems
to push the water back. That costs money, this adaptation, how much is it going
to cost in the future? How much are you willing to pay versus what if we solve the problem now? And I know people are more
reactive than proactive, but really fully accounting
for the costs and saying, okay, we want to avoid those
costs at all possible cases. How do we do that? How do we do that in ways
that people like me can invest their meager 401ks in, I personally, I think this
is the key to the future.

I think the business community
truly holds the key here. You can get all the scientists
together, seeing things, which many people ignore
half my students do because they're just trying
to get a gen ed credit. But I think the people
who are really control of a lot of the money, may talks, I'm going to be realistic here. And I think the finance sector
can really take the lead working in partnership
with the policy side and the scientific side. I, this is I guess my
opinion, but I really think the key moving forward is to
mimic the Montreal protocol. And subsequent amendments say, okay, how can we have a Montreal
protocol for climate? And I'm not the only
scientist saying this. There are a lot of
scientists saying this too. We have a roadmap. Let's try to use that
template to move forward. – Yeah, no, I think you're
spot on, on that one. And I guess I'll come
back a little bit more in a detailed way and just say, talked a lot about the climate
models in this session.

And what do you think has
to get better about those that help folks in industry
regulatory community, get their arms around this? I mean, at the end of the day,
they're gonna be, you know, asked or required to go down this path. What can a scientists
do to kind of help us and how do we do that? – Well, I think that's
what the scientists need is to hear more from stakeholders. You know, I'm hired to be a scientist I'm hired not necessarily
well because I do policy work on air quality to work with policy makers, but most scientists are
focused on solving the problem at hand. And this is a big task
just to get the funding, to do the job and focusing, hyper-focusing digging
into the minutia details to really understand what's going on. The scientists may be
largely unaware of the needs of different communities. And I think what we need
is a better opportunity to get everyone together
on a similar platform using a similar language, perhaps, or at least trained to
understand and to translate, okay, oh, I had no idea. This is what you really needed, okay.

Yeah. I think I can go back and
either tweak my model to give us specific outputs, or we've
got a regional climate model that can take what's coming
from the global climate models to give you specifically what you need, which we were largely unaware for a specific sector. Maybe not a specific company, but maybe one company
their needs represent the needs of a larger portion
of the sector to say, oh, this is what you need. Cause I think that's where we're lacking. And honestly, a lot of
the scientific community, maybe skeptical of the private sector, because we've taken a beating
over the years from private, you know, 10, 20 years ago,
all the scientists, you know, they're just these elitists
using government funds just to maintain their jobs. So we really want to solve a problem. And I think we need
stronger, clear partnerships, both with the policymakers, but
with industry to find, okay, what do you, what do you need? Do we already have this? In-house maybe NASA NOAH
already have this information that can be translated for you. We just didn't know you needed it.

And I think that's at some level, and I think this is
happening here and there, but I think we need a
stronger push in there. – Yeah, there's, there's a whole segment of models that we've actually put on
the side purposefully today. And that is these
integrated assessment models that we, that you referred to earlier, the dice model and the like. And I think that intersection, you know, while people won Nobel prizes, for their innovation, I'll say, and kind of bringing all this together, those models themselves are fraught with all sorts of assumptions and, you know, things
where they're taking, you know, outputs from climate analysis
and trying to come up with damage multipliers and assessments.

And so when you talk
about that intersection of trying to take those, you know, outputs, the greenhouse gas up what's in the light, coming from the climate
models into a digestible form that can be, say converted into what typically a financial services company is most used to seeing, which are things like
changes in home prices and interest rates and right, and GDP and unemployment
and commodity prices.

I know it can do some of those things, but therein lies also some
of the struggle, right? You've got climate models over
here with their own issues. You got the integrated
assessment models over here with their own issues. And this is where I keep coming back. And I struggle with how do we kind of better build those kinds of tools that can
be leveraged by, you know, the financial community anyway, that's kind of what I say. – No, I agree. And this is how the models are developed. Initially, you make assumptions
because you don't know.

So we're gonna assume
this is what's happening. And then you realize, oh, that assumption is much more nuanced. And so you dig, take a deep dive in understanding the
impact of land use change. You know, what happens
when you chop the trees down and convert to farmland? this has a larger impact than
we thought based on, okay. We just had a damage multiplier in there to represent everything. You know, we have to
make this more complex.

And I think this is where we are now with the climate models. They are so complicated
that only a small number of people really can fully
understand all of the impacts here and how to translate it. But yeah. And so now you're
developing these risk models that people have won Nobel prizes for, but you'd start digging into that, what assumptions were made. (inaudible) – Correct. And this research. – Absolutely and I
think, you know, people, I call it shiny object bias all the time that we get these very complicated, elegant looking
mathematical representations that may not hold up very well. And yet we're being
asked to implement them, but I'm wanna be mindful, of the time here we have 10 minutes left. I want to
– Question just pop up.
– Yes – Awesome, perfect segue.

You're talking about these shiny objects. And so the question is, you know, artificial intelligence machine
learning neural networks are the big buzzwords right now. Let's train a computer to solve
all these problems for us, but you have to train a computer. You can't, this isn't star trek or star wars computer solve this problem for me
are holographic displays where you can move things. No it's grinding out code, you know, black and white text to
try to represent this. And so there are a lot of, you know, models that are popping up
now that use machine learning, artificial intelligence. And you know, the key
here is that the computer can be more sensitive to patterns that will in much more
quickly pull up patterns and eyes a scientist can do. There may be patterns going on in the data that I'm simply not seeing, but these tools now machine learning and neural networking and
artificial intelligence in the past 10 years, it used to be, you had to be very skilled in how to write these codes yourselves.

Now there are codes you can download tools you can use for machine learning and artificial intelligence,
neural networking. And now we have the
black box problem again. we ran our neural network. All right, this is the answer you got is that physically possible? Did you get negative carbon
dioxide out of your model, which is physically never going to happen? And, and this is, you know, how, how much are these fancy new shiny tools tied to actual science? Because I've seen talks people
and not for climb models, but for some other models, oh we ran our neural network
and this is a result.

But yeah, but if you think about it, that's not physically possible. Why would you even say that? Just because the model says it, you paid a lot of money
for this new tool great. It's nice and shiny with this
nice shiny user interface. And it makes beautiful pictures. It's fundamentally still wrong just because it looks nice. Doesn't mean it's correct. And so this is where you
need critical thinking. And this is where I'm not
the only one here saying that's okay, this is a great tool, but you really need some well versed scientists
who really understand what's going on and say, okay, this is what the model said, but hmm, I don't believe it. It doesn't pass the sniff test. And I'm worried about some of
these artificial intelligence machine learning models. I'm not saying they're not
valuable I think they are. And I think they will become
another tool in our toolbox. – Right. – But as a skeptical scientist, I always have like, do I
trust these people or not? It's the newest thing. I'm not a 100% I do think there will be
tremendous value because computer can do things that I can't, but you're training the computer on things that have happened in the past.

How do you predict something
that has never happened before? How do you tell the computer
to anticipate a record event, a record hurricane a
record rainfall event, a record snowfall event,
once it's in the records, then you can train it. – Yeah, yeah. Here's here's another one says
what's the current approach to get data, to conduct
a climate risk analysis for a portfolio of obligor to assess the financial implications of different types of climate events
impacting underlying exposures, unexpected loss as well.

I'm going take that one. So I just had an article
published in the journal of risk management and
financial institutions, where is an example of this, where I took the public
is publicly available, a large sample of the
Fannie and Freddie credit performance data and built, a default model with
the usual risk factors that we think about there, a borrower credit
characteristics, property loan, all the other usual things,
house prices, et cetera. And then I, looked at augmenting that
data with FEMA and NOAA data on hurricane disaster
declaration information that's available. That is both the rating of the hurricanes that came through from 2000 and from 1999 up to
the present, basically. And then also the number of
hurricanes that came across the areas that are in the sample. And it turned out when you
control for all these things. In this case, the default
risk was amplified.

It was statistically significant in terms of both hurricane intensity and frequency, kind of no surprise, but that's one way you
could think about building and some of these physical
risks, not the transition risk, but the physical risk directly into some, of that kind of analysis. So we got a lot of wood, the job I'm thinking along
the way on all of these. And that's just one
kind of a little example of money right? You know, you think about the, you know, the trading books that are out there and trying to kind of come
up with a market value estimates of those. There are, all sorts of software
that's out there that, that can help you do that sort of thing. But I think that's kind of
where the world is right now. And, we're gonna have to
accelerate a fair amount of that effort, I think. – Yeah, kind of along those lines, you mentioned hurricanes, you know, what are the climate models
predicting for hurricanes? The planet's warming does
that mean more hurricanes, but there are a lot of different
conflicting factors here.

And what's many of the models are saying is that you may have fewer hurricanes, but the hurricanes that do form are gonna be larger and stronger. – And how do you incorporate
that in your risk assessment? – Right, well, again, in
that simple example, I gave, you know, that analysis could inform, and I did a little bit
of sensitivity analysis , from those estimates to
say, okay, let's set, let's suppose the NAOH long range forecast over the next 30 years,
says that we're gonna see twice as many category three
to five rated hurricanes. Okay, let's pop that into the model now. And that's one of our new scenarios. We can drive that through
and we can see, okay, defaults are going to be, you know, 30% higher than what we
expected along these areas that are gonna be seeing
more of these hurricanes come through. That's the kind of stuff that I think, we could try and tease out, or whether it's, you know, broader base, you know, flood risk,
trying to get a handle on that sort of thing, I think is, you know, another way to go,
particularly as it relates to changes in property values.

That's another whole kettle
of fish when we're talking about the mortgage side and in particular. But I think generally speaking, and I know we're kind of
getting close to the top of the hour, you know, we've talked about, we've talked about a range of things here, and I guess the one of
the last things, you know, I would ask you Tim, to maybe comment on, we've talked about this
too in the past between us, you know, weather versus climate, we can't forecast the
weather more than 10 days in the future. You know, why should we
believe climate model result 20, 30, 50 years from now? What's the difference between
weather and the climate model.

Maybe we kind of tie that off. – Weather models are one
constrained to observations and they're much higher resolution. And they're designed to
represent smaller scale features. You know, our 10 day forecast now is where our five day forecast, you know, back in the 70s. And it's basically,
you're, fighting chaos. You know, the results
your output of your model is dependent upon how
accurate your inputs are. And the further out in time you go, the more that uncertainty
to kind of takes over the forecast community has, you know, given us more accurate
forecast out to 10 days, but these models are
run in rerun every hour, every three hours data
are fed into the model saying, here's the state of
the atmosphere right now. Okay, now let's go out another 10 days. Another hour, more data comes in they're rerun and they're run
operationally, continuously. All the weather forecast
that you're getting on your phone are based
on continuously you know, functioning models that
are digesting millions of data points from all
the weather stations and whatnot and satellites
all over the globe all the time, constantly at really high resolution climate models can't do that.

You can't just continuously
run a climate model. Currently the computers aren't there. We don't have big enough
computers to do this. So how do you predict in the future? Well, let's take a random year, 2050, very basic, which month
is going to be hotter June or January? Sorry, I said there
wasn't gonna be a quiz. So, but this is a basic one. What month will be hotter
in 2050 in Maryland, June or January? I think most people would
think June because it's summertime, you've just
made a prediction based on your understanding of the physics. And so this is what we're
doing with the climate models is taking this larger scale physics, understanding that the climate model will not tell you when your neighborhood will get a rainstorm, but we'll tell you the
probability in a sense of increased rainfall
amounts, increased droughts, that sort of thing.

But again, this is where the, you know, the scientific community
is trying to develop those mid range models. This is a big focus, these seasonal to sub seasonal models to bridge the gap between
the climate models and the forecast models. – Well, we've got to
leave it there, my friend, that was a lot in a short
amount of time to digest. And, again, many thanks for you joining us and kind of walking us, lay people through the intricacies of these climate models, many things. (laughing)
(inaudible) – Absolutely. And thanks everybody who was on today, the recording will be available later on and we will have another
one of these very soon. Thank you all..

As found on YouTube

Looking to see what kind of mortgage you can get? Click here to see

Leave a reply

Your email address will not be published. Required fields are marked *