Of climate models and trust

00:00:02: A hoi and welcome to the Gen Z podcast, a podcast by early career marine scientists.

00:00:10: Hi I'm

00:00:11: Leonard And i am currently writing my master thesis To finish up my Master in Environmental Modelling at The University of Oldenburg In Germany!

00:00:20: And hi ,I Am Shalotte !

00:00:22: I have recently graduated from the Masters in Marine & Coastal Systems At The University Of Algarve in Portugal And together we are going to record some episodes for this podcast over the next few weeks.

00:00:34: So, Leonard!

00:00:35: Let's dive right in!

00:00:39: Good morning, Leonard.

00:00:40: Morning, Charlotte.

00:00:43: We talked about sea ice models... ...we talked about supermodels and

00:00:49: today

00:00:49: we're gonna talk about climate models.

00:00:52: Yes

00:00:52: I mean to round it up you have to talk about a general model

00:00:56: exactly.

00:00:58: so everyone who is still with us super excited that you made it to this episode with us and we're happy to feel or see other people also think models are cool.

00:01:17: Leonard, can maybe introduce a little bit the history of climate modeling?

00:01:23: It seems like the idea has been around for quite some time already.

00:01:29: Yes funny enough the idea of weather forecasting obviously is around for way, way longer than this.

00:01:37: I mean you have like sayings or farmers for example when birds fly low and will have rain so on.

00:01:45: it's kind off already whether forecasting if your think about since its using observations to then extrapolate a future draw conclusions out And then later on we got a bit more, well let's say advanced in our things.

00:02:09: How do you think about this?

00:02:10: So there is this Norwegian meteorologist who was big part of it as he called William Björknas and so... He has been around for quite awhile that did some interesting stuff!

00:02:24: ...and wanted to describe the atmosphere with equations with those partial differential equations that we use for our models.

00:02:33: He was one of the meteorologists who then tried to use this for actually forecasting.

00:02:42: and there were some people around beginning of the twentieth century, Louis Frye Richardson is also quite famous there who tried to develop a forecast but he did it still by

00:02:57: hand.

00:02:58: So he imagined weather forecasting centers as super human computers.

00:03:07: You would have a big hall with people and everybody gets like sheet of paper, then you calculate for let's say this grid cell the pressure or whatever.

00:03:22: And this idea was always taken by people like Jules Charnet, I think.

00:03:30: I'm not sure if he's French or English who based his ideas on Richardson and then developed from these models in the nineteen fifties actually got first forecasts which were usable.

00:03:49: We have been always increasing and I mean we are, like those climate models in the nineteen eighties were already predicting basically climate change.

00:03:58: Since then actually also in a quite correct way on good amplitude for example.

00:04:06: So it's

00:04:07: very interesting to see how fast the development also was.

00:04:10: I mean, yes from Louis Richardson in the nineteen twenties till now that is a hundred years and somebody seeing people sitting at home calculating by hand too having super computers and super models know they're very fast development in such a short time

00:04:28: indeed.

00:04:42: So from calculating everything by hand to actually having these very sophisticated computer models, can you go a little bit more into detail how the sophisticated computer model do work?

00:04:57: Well so I mean what we call the dynamic core.

00:05:03: That's those partial differential equations or mathematical equations mainly like the Navier-Stokes equation for example of fluids.

00:05:12: So you can have them for the atmosphere and also for the ocean.

00:05:15: And here then, yeah... You calculate time step by time step how physics evolve.

00:05:31: closely related to dynamic core is the numerical algorithm.

00:05:36: so that's a choice of how do we solve these navier stokes or equations in a numerical way?

00:05:43: There's just different mathematical methods to use them.

00:05:53: They're like simple, well pretty simple time-stepping or Eulerian method Or there are more advanced where you used several times steps and so on.

00:06:06: So that is depending upon the grid And it depends what kind of algorithm we use.

00:06:12: There are slight differences if you calculate same thing for example And then also related to this, again is the discretization.

00:06:21: So on which grid do you actually calculate all of these stuff?

00:06:26: This has impacts because for example it's a witch processes that can resolve.

00:06:35: so if we have big grids and cannot solve stuff like certain ocean eddies they're too small scale to one hundred and fifty kilometers.

00:06:50: Also, it depends where you place variables for example on the grid.

00:06:58: so you can have like regular grids or just rectangles Or yeah kind of irregular grids which then would be more dry angles And then you need to alter your numerical algorithm To fit these grids.

00:07:17: So those are some of the important decisions you make when you're developing a model by choosing specific grid size and choosing your time frame, and your timesteps in way that you can represent what actually want to research.

00:07:38: Your timesteps aren't too long or your grid is not wide.

00:07:41: if you go into more short-term fine scale processes for example such as eddies?

00:07:49: Yeah, it's depending on your research question.

00:07:51: What would you apply?

00:07:52: because we have this sites of computational calculation and grid resolution or accuracy.

00:08:01: so there You have to weigh them and say okay We can go down to this resolution.

00:08:05: that still fine.

00:08:06: We can still calculate.

00:08:07: This doesn't take a month to take two calculated one step

00:08:19: And also depending on your research.

00:08:20: question is then what parameters you are choosing to include into the model, I assume.

00:08:27: Yeah when do want?

00:08:29: because like simply said because this isn't a real podcast here if processes in the upper ocean, you will always have to include certain wind parameters such as direction and so on.

00:08:46: These parameters have a quite important influence or even are triggers for certain processes of the upper oceans.

00:08:53: Exactly!

00:08:53: And also if we think about carbon uptake which is what were going talk next episode For example, in a lot of models the uptake which is done by bacteria and so on.

00:09:12: It's basically represented maybe but very simple way because we just don't know or it's not included yet.

00:09:26: So that also needs to be considered.

00:09:30: If you want to know how much carbon the ocean will take up in two hundred years, one hundred years and fifty years.

00:09:38: there is some process that we need consider which are not necessary still already inside models.

00:09:50: And we talked about complex climate system because a lot of factors feeding into making our climates.

00:09:59: Yeah, so I think this is also an important aspect of climate models that you have to consider the so-called boundary conditions.

00:10:10: Indeed!

00:10:10: Of course especially boundary conditions would be more important if you have climate models because they are running in the future.

00:10:19: So it's more about a boundary That we impose.

00:10:23: for example For solar radiation there're these things called the Milankovic cycle you know, when the Earth orbit is changing a bit.

00:10:35: And that would imply some changes in solar radiation coming onto the earth.

00:10:41: So thats boundary condition and for example especially important to us it will be something like CO₂.

00:10:50: so these CO₀ emissions are they gonna stay like this?

00:10:55: Also everything around them For example land use or land change.

00:11:00: So those are like boundary conditions that we don't know them.

00:11:06: We don't how they will evolve, and thats why we have these different kind of scenarios where you say ok the emissions are going to stay like this or they're gonna flat line or decrease even so Those is what we call a boundary condition.

00:11:25: Well when your look at initial value, so the initial point where we start a model.

00:11:31: That's the initial condition and that is for example way more important than something like weather models because you need to capture state of climate or whether system at this moment should be able forecast it in next week

00:12:03: trying predict future.

00:12:05: But how predictable is the future, actually?

00:12:07: And I think this is one of key issues or like one of the key challenges in climate modeling as well.

00:12:14: Well it's once again you have this differentiation between weather and climate because if would say... If we look at the weather models You can see that forecast deteriorates pretty quickly.

00:12:30: The weather system too chaotic predict way longer.

00:12:35: So I mean there is predictions going up to several weeks or so on, also seasonal predictions but the pure weather forecast that you can see in your phone.

00:12:45: normally those are not gonna be very reliable for more than five days while for climate models they use different.

00:13:01: Well, they don't care that much about this weather noise but it's more of a long-term climate.

00:13:07: So the climate and weather differentiation is like this thirty year period where you average.

00:13:15: so in the climate models you would predict larger patterns.

00:13:20: basically That why for example in climate model to have those modes or variability which are more important than on these longer term time scales.

00:13:33: And yeah, so we can say there's like... We can predict the future or well you know project for climate scale and also weather scales just different limits.

00:13:54: I think this is an important distinction to make.

00:13:57: Also, when you talk with people because a lot of people are not working in this field.

00:14:03: They're using climate and weather interchangeably exactly.

00:14:08: I think that's very important point to make.

00:14:10: they're not interchangeable.

00:14:18: i Think one can point out however That the weather forecast on the climate models there kind Of little bit underneath one roof institutions that Are providing data or use models for?

00:14:31: our weather predictions or our weather forecasts.

00:14:36: They tend to also work on the more long-term climate models as well, such as European Centre for Medium Range Weather Forecasts.

00:14:45: they do not only provide this real time data of all member states but are also engaged with the European Space Agency and they are helping to build these climate models.

00:15:02: These weather forecast institutes, they're also very important because... They have huge archives of weather data observations And if you want to feed observations into your climate models I think those are the institutes that will go through to get observational data.

00:15:25: or when you're, for example running a model like a model hind cast over four.

00:15:32: Like period in the past and then try to compare it with outcomes of the model what actually happened in the passed.

00:15:42: that's something where weather forecasting and whether archives come into play with modeling.

00:15:49: Yeah especially ECM WF is super important because A lot of the, as you said, observational data to compare models is for example produced by them.

00:16:03: And I just read yesterday next year they will release a new era six dataset so we'll have new data.

00:16:11: Wow!

00:16:14: That's cool.

00:16:27: We've talked about... Metrological archives we use as observations for observational data and that's a very nice way into our next topic.

00:16:39: So how do we actually evaluate model performance then?

00:16:43: Yeah, it is... As you said with Heinkast for example.

00:16:46: so they used models to re-forecast past times And then you evaluate them against observations.

00:16:55: So, of course the main part is you just compare it to observations.

00:17:00: So two weather stations satellites ships planes whatever we will find.

00:17:06: Also proxy observations for like a very long time ago For example You can have ice water isotopes or corals or tree lines?

00:17:14: For example if you want to Compare how your model actually did in the place-to-seen so on Exactly.

00:17:25: And then also what you can do is, um... You can study certain events for example like volcano eruptions because they have a huge impact.

00:17:37: They shield the atmosphere from solar irradiation so that it will drop in temperature.

00:17:43: So

00:18:01: we know now how to evaluate models.

00:18:05: But also the question always is how to improve models and how to account for the uncertainties we are dealing with, especially when we're trying to predict such long time frames.

00:18:16: So can you maybe go a little bit into detail on some of the possibilities?

00:18:21: How do I count these uncertainties?

00:18:23: well as we have seen in last episode already may normally use this ensemble model so used different models, but also you run them several times in a perturbed version.

00:18:36: So you would like alter the initial state of it or you can also alter some of the parameterizations so that they have slightly different physics and That would account for The uncertainty that we have.

00:18:51: and those estimates for the parametrizations are for observations And then we just run them and then you kind of average And if you average them, normally we'll get a better result.

00:19:03: So it's just like... We run them lot of times and then see the rough direction they go in.

00:19:14: That is probably going to evolve In IPCC.

00:19:24: there are those intercomparison projects projects where you have different setups of the models to then compare how this would affect.

00:19:40: So there's like atmospheric comparison, oceanic comparison and there is also geoengineering or paleoclimate climate.

00:19:50: so there are specific experiments set up for those specific cases to see how the models are able to represent us or how uncertain they are.

00:20:06: And another thing that's used a lot is, for example you can use limited area models and regional models.

00:20:15: That would give your higher resolution in certain places.

00:20:21: For example if you're looking at CIS You have like very high-resolution model and see how the sea ice affects.

00:20:34: um, ocean physics for example.

00:20:36: And then you can kind of scale up to the whole Arctic Ocean How the sea is affecting the ocean.

00:20:44: physics very simply put

00:20:46: Yeah exactly.

00:20:48: or if we could do this.

00:20:49: For example there's city models.

00:20:52: So yeah you can basically take temperature, humidity and all of this from a bigger model.

00:21:02: You put it into the smaller model then run on different scale.

00:21:07: So that's maybe important for something like the urban heat islands.

00:21:11: so heat stress onto places where most people actually are.

00:21:20: Coming back to the intercomparison projects because the geoengineering case called my attention.

00:21:29: Do you maybe have a short example for such an intercomparison in this context?

00:21:36: I think that's, for example something like ocean fertilization would be something that is looked at there.

00:21:44: probably actually...I'm not sure if i've looked into it but because we are Not representing all of the processes or not in a very high resolution.

00:22:00: For example, I think like phytoplankton and so on there are sometimes only two classes Or bacteria And so that's just a lot of Lot of process is missing.

00:22:14: or yeah, not as fine-scale resolved.

00:22:19: So when you would put in iron or something to increase bacteria growth, like phytoplankton growth.

00:22:27: To take up more carbon into the ocean and so on I guess then there you could compare for example models who have just as an example now i don't know actually um Have more classes for phytoplacton Or a different version of the carbon uptake.

00:22:46: Yeah,

00:22:49: different species composition exactly.

00:22:51: Exactly.

00:22:52: and then they would say okay maybe this actually Accounts for the fifty percent in value of carbon uptake.

00:23:02: so them we'd have this uncertainty somewhat And that we know.

00:23:07: ok because if we change this parameterization or this process a bit it already takes way less away more carbon.

00:23:15: So we are not exactly certain how much this is going to account for in the future.

00:23:21: Yeah, so you're again using um...the fact that different models are simulating different processes differently and then I suppose- You get a range of potential outcomes from including a process into a model?

00:23:38: That's how I understand it now!

00:23:39: Yeah

00:23:40: yeah that's a way..you can do.

00:23:42: I

00:23:55: think this is one of the advantages of models that you have for all these possibilities, changing parameters or changing boundary conditions and then just running a huge experiment on the whole globe to see what happens.

00:24:10: Or could happen potentially in future?

00:24:13: And yeah there's couple other advantages through models.

00:24:18: so they're very important complement to our sometimes rather limited observational data.

00:24:29: And this is, for me in my work one of the biggest advantages.

00:24:35: models.

00:24:36: if you're working in Arctic or also Antarctica just because winter climate and sea ice that limits observations so much because it takes time and money to go on research cruises up there due the harsh conditions.

00:24:58: Also, observation from space is rather limited with ice cover.

00:25:02: This is where models can really be a great addition in our understanding of what's happening Exactly.

00:25:13: And then another advantage of it is that you are able to, in quotation marks observe different systems over much longer periods of time and also trying to understand what may happen in the future.

00:25:29: Yeah especially observing longer times can be important for statistics basically.

00:25:36: So when you have events like for example ENSO again We only observed for a hundred years, bit more.

00:25:45: Um

00:25:46: we have good data for this time span.

00:25:49: so we need the models to run them for millions of years basically To then actually get some better statistics or some robust statistics

00:25:59: Yeah and also I think in this case specifically And the mechanisms behind these long-term processes as well, because it's very hard to observe this if we're talking about such a long time scales.

00:26:13: But there is also of course some shortcomings with models.

00:26:17: so I think one of the big ones that maybe gives people a bit of false feeling of certainty.

00:26:25: definitely yeah i think thats just somewhat slight difference between prediction and projections.

00:26:34: Because, I mean what we see is that we project like those scenarios.

00:26:39: We talked for the IPCC about Those are projections because they assume so much boundary conditions and so on.

00:26:49: It's just not a certainty That will end up with this in the future.

00:26:57: And thats also why you change from these absolute values.

00:27:01: You changed your probabilities Right, so Because then what we do in these or what?

00:27:09: We can see in those reports is that they normally would say okay It's more likely That this happens than that it not happens because more than fifty percent of the models.

00:27:21: Say This is gonna happen.

00:27:23: let

00:27:23: me have more less what I was saying.

00:27:26: Yes, that's a very important point when it comes to communicating the results off such studies too decision makers and the public and stakeholders that are involved.

00:27:37: It's very important to use the correct wording in order also manage expectations.

00:27:44: And I guess you can say there is a lot of research going on, as well as development in environmental modeling happening at the moment.

00:27:53: so they're improving and getting much better.

00:27:58: The more precise they become, the higher are usually also computational costs.

00:28:11: For example we talk about having a much finer grid to represent all of these small scale fine-scale processes better.

00:28:19: but that really increases the computational cost and then raises with question whether increasing costs is justifying outcomes?

00:28:32: Yeah, a little bit.

00:28:32: So I think it's also somehow societal question how much money should be invested in this?

00:28:38: Of course!

00:28:39: How Much Money and Also how much energy?

00:28:41: i mean if you're running those models they are also...I'm in there running huge data centers.

00:28:47: yeah

00:28:48: so They need water or power.

00:28:53: There is some costs connected to it And that's definitely A Question That Needs To Be Asked.

00:28:59: At Which Point Is It not beneficial enough anymore.

00:29:05: And another point of this then also would be the explainability, This is especially good things for models that use machine learning and or AI because Then there's a point.

00:29:18: okay do we really understand why something happening in the model?

00:29:25: That's just question it needs to ask

00:29:28: But nevertheless I think you can say Models are very important for climate science.

00:29:35: And there still needs to be done a lot of development and improving the quality of models.

00:29:41: at this point in time certainly, Still has Very high value and it's worth the investment.

00:29:50: I guess we agree on that.

00:29:52: so yes

00:30:06: So Leonard thank you very much for giving these insights and these additional explanations on how climate models work.

00:30:15: I think, again...I learned a lot today!

00:30:18: And it was very interesting to talk with you about this.

00:30:22: Likewise

00:30:23: Maybe just as like the final question of this episode... Leonard Do we trust models?

00:30:30: Yes We do

00:30:32: But..we

00:30:33: trust them but also know that they are not one hundred percent able to give us certainty.

00:30:41: So you just need to keep in mind the limits?

00:30:44: You have to keep your mind at the limit, this is why we still need observations and humans to work with them and interpret it.

00:30:52: I think that's a good closing for today.

00:30:55: It definitely is!

00:30:59: See ya next time!

00:31:06: Gen C is an initiative of The Blue Capacity Development Project of the ECOB Program hosting & editing Lennart Wontag & Charlotte Walter.

00:31:16: Music, Florentine Seifert.

00:31:19: Production and Graphic Design, Annabel von Jakowski Management, Eva Rolfe.