 Our next speaker is Andrea Molod from NASA GMAO. Andrea is a research scientist at GMAO and also the lead of the S2S group. Andrea also gave a talk during the student colloquium. A couple of weeks back, thanks for that, Andrea. Looking forward to your talk. You're still muted, Andrea. I don't hear you, but I can see your screen. Sorry, I can't seem to unmute with the screen share. So let me go through it again. OK. Oh, and sorry, I missed your question to you, Andrea. Maybe you can come back to it. Yeah, that's OK. That's OK. So I think you guys should see my screen. We should be in full screen mode and you can hear me. Yeah. OK. So during the summer school, I spoke about some aspects of our geosubseasonal prediction data assimilation and model. And here I pointed out the issue, the use of the aerosol in there, but I'd like to focus a little bit today on what's going on with that. So these are our system characteristics for the new system that is running a new reanalysis now. And there are two aspects of this here that I'm going to focus this talk on today. One is the use of an interactive aerosol model and the aerosol cloud interaction that comes with the higher moment cloud microphysics. In the experiments that I'm going to talk to you about, they were not initialized like the lower one with the nudged optical depth. But that's to be looked at at another time. I'm going to focus on the use of the aerosol. So we haven't thoroughly analyzed how it behaves in our system. We're just starting to do that now. And I'm going to talk a little bit about some preliminary results. But why did we do this in the first place? There's a lot of evidence out there, first of all, that we can get useful subseasonal to seasonal prediction skill in the aerosol itself. Aerosol optical depth, PM, the particulate matter 2.5. That's relevant if air quality. There's also a lot of evidence out there that it including these kinds of effects increases the weather forecast skill, so-called forecasts of opportunity under certain conditions. That's been shown in the Geos system and in the European center by the forecast system as well. A recent paper shows that in the aftermath of a big volcanic event, and I'll show you another example of this later, that it can increase the seasonal skill, questions about increasing the subseasonal forecast skill itself. Benedetti and Vittart indicate that in a scorecard. Some local study also show that it seems to have enhanced the MJO activity, which for our model was a good thing. Decadal prediction, so there's a lot of reasons to think that you want the interactive aerosol model. And now why the second bit of it, which is the aerosol cloud interactions. And that has to do more with the physics of how the aerosols will affect the evolution of the clouds. One is the Tumi effect, which is basically you're going to increase the cloud albedo because you have more nuclei. The fan mechanism suppress collision and coalescence. And so you get longer cloud lifetime and higher frequency. The deep convection enhancement and more recently some issues related to ice clouds. In addition to that, there is a growing community activity in interest in at least the interactive aerosol model part, if not the aerosol cloud interaction part. And there's an intercomparison project going that are asking the models to run side by side suites of forecasts with and without the interactive aerosol and then presenting the presence of the indirect or the aerosol cloud interaction as an optional experiment to run. I'm going to show you some results from the GMAO's experiments with the so-called near real-time system that's running now. We had a couple of unique problems that had to be dealt with in order to do this. Because our default system has the interactive aerosol and aerosol cloud interaction, if you just replace one parameterization with the aerosol cloud with another, you're basically doing a different model and you can't look at the two of them side by side to attribute what is due to the presence of the aerosol cloud interaction. And so for a lot of different reasons, Dony von Barahona, who's our cloud two-moment microphysics expert at GMAO, decided that the best way to do this is this thing in red here. From our control runs, which have the aerosol cloud, he developed an climatology of the cloud drop numbers that the two-moment is predicting. And we use that to turn off the aerosol cloud interactions. And in the mean, it looks very similar to the control. And so that's going to be used for our aerosol climatology and for all of our no aerosol cloud interaction runs. And so the first bit of the result that I want to show you is the actual result of the AOD and of the cloud drop number. Because if we can't predict those, then looking at the impact of them on prediction skill for the temperatures and things like that is irrelevant. So here's just sort of a look back at global mean bias and RMS of the optical depth in two different seasons. I would argue based on this that at least it's credible for the aerosol optical depth, which is kind of the easiest thing to get. The bias in the aerosol optical depth, again, is it's there, but it's not substantial. On the right, we've got anomaly correlation for PM 2.5 over the US. This is one of the better regions in the world where we have this, but I don't have the scale here. This is anomaly correlations up in the 70s and 80s for one month, two month, and three month lead time. So I'm going to argue at least here that the aerosol prediction is credible. And now what about the cloud drop number? Here, upper left, we have some results from MODIS cloud drop number estimates. Upper right is the run with the aerosol cloud interaction. Lower left is no aerosol cloud interaction. And lower right is no aerosol cloud interaction, but no interactive aerosol model either. And the magnitude is off relative to MODIS, certainly. Again, patterns are credible. I can give you a better sense if you can take a regional look at this in these three different regions. We have numbers from Merit 2 for aerosol optical depth. I'll show you the cloud drop number in a moment. Merit 2 assimilates optical depth, and so we use that as our validation. We could just as well have used TAMS. The time series are credible. The red line is the ensemble mean, and the blue is Merit 2. So again, capturing trends is the easiest thing to do, but at least we can do that. Cloud drop number concentration also seasonally, this is the mean. Again, we argue here that it reasonably reproduces the mean cloud drop number concentration. The trend we're getting in the US, but East Asia trend we're not getting. And so it's a little bit more complex over there. So there's also one other issue to talk about here related to prediction of cloud drop number concentration, and that's so-called forecast of opportunity. And so the idea is that in the wake of an event like a big volcanic or a big kill way of degassing in this case, we expect the forecast skill to be better in the wake of an extreme event like this. And so the question is here are the time series, from MODIS and for the ensemble mean and the spread. And if you take a look in some detail on the left is the aerosoloptical depth anomaly itself. The middle is the number concentration on the right is the two meter temperature itself in that region. And so the idea is that there was this few year long kill away of degassing event in obviously not predictable in anybody's model. But we can see if you look at the dashed lines, you can see enhanced correlation with the observations in the wake of these degassing events. So once it happens, we are seeing increased prediction skill in the wake of this. And now the aerosol and the cloud drop number are within reason. And so now what is the impact on the meteorology, two meter temperature, for instance, here? And we're trying to also help sort out which of the bits are contributing to the prediction skill. So we're trying to separate out the aerosol-cloud interaction from the interactive aerosol, et cetera. So on the left is the two meter temperature difference from merit two. We could have used a better observational climatology for sure, but to point out the difference here. So the upper left is interactive aerosol and aerosol cloud. The middle one is interactive aerosol, but no aerosol-cloud interaction. And the bottom is the aerosol-climatology. And we see what's happening here. The control run is warm in the Northern Hemisphere and cold in the Southern Hemisphere. Removing the aerosol-cloud interaction fixes the warm bias in the Arctic and in the Northern Hemisphere, but makes it colder. It's made it colder everywhere. And if you look at the run with the aerosol-climatology, we see that that's very similar to the run without the aerosol with the interactive aerosol. So the point here is the aerosol-cloud interaction is what did the work. On the right, we just have some difference plots to help sort out what's going on on the left there. This is the no aerosol-cloud versus the aerosol-cloud. And this is the aerosol-climatology minus the aerosol-cloud. And again, aerosol-climatology is not what's making the difference here. It's the aerosol-cloud. And removing it basically just cools everywhere. Another little look at the two meter temperature basically is showing us the merit to at different lead days. This is North American average. And basically we're seeing that the track of it with the interactive and the direct and the indirect is closer to merit to than it would be without it for North America. Couple of other parameters going along with it to sort of help tell the story about what's going on here is what's going on with the cloud fraction. And so this cooling of the two meter temperature that we saw when we took away the aerosol-cloud interaction is associated in the Northern Hemisphere with an increase in low level cloud, but in the Southern Hemisphere it's a decrease. And so this confuses the issue a little bit. Also the interactive aerosol, the top and the bottom pictures are much more different from each other than what we saw for the two meter. In other words, the two meter temperature for the cloud fraction, the interactive aerosol is playing a role, but it's in the same sign. And so it increases the difference from the aerosol simulation. And now another player in this mix is the precip. So again, on the left is the difference due to the aerosol cloud. And on the bottom on the left is the difference due to both the interactive aerosol and the interactive cloud. So again, removing the aerosol cloud lowers the precipitation rates in the Northern Hemisphere in general, enhances the precipitation in the ITCZ. Neither one of those is a good thing. If you take a look at the lower right and the difference with GPCP in the control, we're seeing getting further. And so to sort of help tell the story in a little schematic with the aerosol-cloud interaction running over time, the depiction on the left and the one on the right are what happens over time. So you've got a lot of aerosol, it activates a lot of drops, makes a lot of drop numbers, rains more and scavenges more, so you lose the aerosol. So when you get later in time, there's less aerosol, less activation and less cloudiness. So over time, the scavenging and the inhibited activation that comes with it deplete both the aerosol and the cloud. If you turn off the aerosol-cloud interaction, the scavenging rates are never enhanced. And this may be part of why we're seeing an increased AOD and cloud fraction and reduced and more cooling and less precipitation when the aerosol-cloud interaction is turned off. So just in summary here, the first thing is dynamical predictions of the AOD, the PM 2.5 and the cloud drop number can be skillful. And the inclusion of the aerosol-cloud interaction had an effect on the surface temperature, the cloud fraction, the precipitation. We told a little story. The prescribed aerosol itself versus the interactive aerosol, the aerosol radiative interaction alone did not seem to have as big an impact. And we're getting a hint that the indirect seems to be a little too strong in geos leading to increased biases. This is all ongoing. We have a very nice suite of experiments to be looking into this. Donny von Barahona is leading the effort in GMAO to examine the impact of both the interactive aerosol and the aerosol cloud. Thank you very much for your attention. Thank you, Andrea. It was really interesting talk. We haven't had enough discussion about aerosols in this workshop. So thank you for this. It was really great. Any questions for Andrea? So I had one, Andrea. I think one of the regions that were highlighted with the interactive aerosol cloud effect was the Southern Ocean, right? And in the Southern Ocean, there's also like the mixed-face clouds that play a big role in like cloud radiative feedbacks where you have super cool liquid in the clouds. I was wondering if there were biases in both the representation of mixed-face clouds in the geos system and how does the ACI impact that? Yeah, so of course the answer is yes, like every other model out there. Mix-face clouds are really hard. Southern Ocean is hard enough. There may be a missing source of some kind of biogenic aerosol or something like that there, but the place where we're horrible at that is in the Arctic. The mixed-face clouds are, you have the ice over the liquid and things like that. And so none of that is captured well in these models. And so how much influence would they have on like the sub-seasonal timescale? Would they have a big impact on the sub-seasonal timescale because they're rapid or they would more come up on the longer-term seasonal and climate timescale in terms of scale? We've been looking a month out, okay? And that's sub-seasonal. And we're seeing the impact related to the lifetime of the clouds. Okay. Yeah. So it's a challenge for our S2S forecast systems, to represent these. Yeah. But I guess our indication at least, I haven't shown a lot of the other results. The indication at least is that it's worse without it. You know, the gap is still huge and we incrementally nudge our way closer to where we need to be by including the aerosol cloud interaction. Great. Any other questions for Andrea? So I don't, oh, Jana, yeah. Go ahead. Thank you for the great talk. So I have question on the dust to aerosol particularly. I assume that in GeoS S3S initialization for aerosols comes from Meridu. If I'm not correct, please correct me. So in Meridu, data assimilation for aerosol is only done based on AOD, aerosol optical depth. Correct. Yeah, not including like PM2.5 or PM10. So for the dust aerosol, does that undergo chemistry also? Like is the dust chemically active? I'm not sure. No, what do you mean chemically active? So that means like when dust is transported by winds, it undergoes many processes. Like is there heterogeneous chemistry, like chemically acting with other constituents? So we don't have an interactive chemistry model running here. It's just the aerosol. So yeah, that means like it's just like a tracer. Like, yeah, it only goes physical processes like deposition, wash out these things, but not chemistry. Okay, so the second question is, is vegetation interactive in GeoS S2S model? No, no. It's only prescribed, right? With some annual cycle? It's prescribed, yeah, with a monthly mean, the, you know, the leaf area index, things like that is all prescribed. And the vegetation fraction and things like that. Randy Koster, there is an interactive vegetation model that we have as an option. We've not run the seasonal forecast with it. There is a little nose of a hint that we get some increased predictability and prediction skill when we use it, but it wasn't enough to justify the computational cost, which is substantial. Just to be clear, like even for the hand gas, like there is set up hand gas, for example, for NMME, that GMAO provides to NMME. Correct. Even for those hand gas, so vegetation is not interactive, right? Correct. Thank you, thank you very much. Thanks, Jenna, thanks Andrea. Shui, you have a question. Hi Andrea, Shui Chen here, can you hear me? Yeah. Okay, I have a question about the sea salt aerosol. So that has been in GeoS and there seem to be a lot of time it gets transported really high. And I wonder if that's still an issue because we'll see hurricanes, you'll see the sea salt gets all the way to a stratosphere. So that's sort of one is a generation issue and the other is a transport issue. So has that been addressed or is there any change recently? Yeah, Shui, that's not happening anymore. Okay? Okay. Anton Darman fixed that for us by dealing with the settling and dealing with a few things. So this model doesn't have that in it anymore. This is a post-merit 2 model that doesn't have this issue anymore. Yeah, it had to get dealt with before we go running aerosol cloud interactions for sure, especially with sea salt. Yep. Great, thank you. Great. Thanks Shui, thanks Andrea. Thank you for your question. Thank you again. Very interesting talking. Also I see that you have answered your question on the chat. So yeah, thanks again to all the speakers from this morning session.