[00:00:00] Bridget Scanlon: Welcome to the Water Resources Podcast. I am Bridget Scanlon. In this podcast, we discuss water challenges with leading experts, including topics on extreme climate events, over exploitation, and potential solutions towards more sustainable management. I would like to welcome Ruby Leung to the podcast.
Ruby is a Battelle Fellow at the Pacific Northwest National Lab, and her research covers modeling and analysis of climate and the water cycle, including land atmosphere interactions, climate extremes, floods and droughts, and land surface processes. Ruby is a member of the National Academy of Engineering and a fellow of the American Meteorological Society, AAAS, and the American Geophysical Union.
And she has received numerous awards, including the Hydrologic Sciences Medal from American Meteorological Society in 2022, and also a Department of Energy Office of Science Distinguished Scientist Fellow Award. Thank you so much, Ruby, for joining me today.
[00:01:09] Ruby Leung: No, thank you very much, Bridget, for having me.
[00:01:12] Bridget Scanlon: So Ruby has done lots of work globally and regionally, and today I think we would like to focus on the relative importance of natural climate variability versus climate change in the Western U. S., particularly in California, and how that affects drying conditions in recent decades and also talk about wildfires in these regions and how she has used machine learning to determine predictors of these wildfires.
And I also hope that we can talk about how she has involved stakeholders in her research in co-producing some of the work and trying to parlay the findings from this research to water management agencies. So first, Ruby, I guess maybe I would like to talk about internal climate atmospheric variability versus climate change.
Mostly in the news, when we hear about any climate event, we hear it's mostly attributed almost completely to climate change. And I really enjoyed your 2021 Nature communications paper looking at internal variability versus anthropogenic climate change on drying conditions in California. So I would love if you could describe how you quantify the relative contributions of these factors and the findings of that study.
[00:02:34] Ruby Leung: All right. Yeah. First of all, again, thank you for having me. And I would really like to talk about some of our studies, particularly as you mentioned, related to precipitation in California. So what motivated us to look at precipitation over California, both in the past and also projecting into the future, is that, like you said, there have been a lot of discussion about like, oh, wow, I mean, are we seeing climate change already?
I don't know. Especially in the last few decades, right? So California's precipitation has been reducing. So we, a lot of studies have been attributing that kind of decline in the precipitation to climate change. And then when we project into the future, unfortunately, there's a whole lot of uncertainty when we project into the future.
So California is one of those regions. Not all the regions are like that. But unfortunately, California is one of those regions where you, when you look at the future projection, the uncertainty is particularly large. So that motivated us to look at, like, first of all, asking the questions, whether the previous changes that we have seen in the past might actually be related to global warming, anthropogenic effects, or natural variability may play a role.
Thank you. And then also importantly to look at like how in the future, how much of the uncertainty in projecting the future changes might be related to our understanding or limitations in our climate models, or could it be also related to the natural variability, which we do not have a lot of control over, right?
So this is what we have been trying to do. So the tools that we have been using are essentially what we call initial condition large ensemble simulations. So most of us have heard about climate models. And so we use climate models to look at the past. We also use climate models to project the future.
And so the advantage of using climate model is that you can prescribe the forcing, right? For example, we know how carbon dioxide, the concentration has changed in the past, and we can also have different scenarios of how carbon dioxide concentration may change in the future. So we put these so called anthropogenic forcing into the climate model, and then we can look at the past and we can also look at the future changes.
So in the past, mostly, people would use a climate model and they run a single simulation or run two or three different simulations and then average out and then look at the future. But what we are finding now is that if the natural variability is very large, simply running a climate model, just running it one time, does not actually tell you the whole truth.
Because if you slightly perturb the initial condition of the climate model, even if you are prescribing the same forcing from the anthropogenic factors, you can actually get very different evolution of the climate, like in one of the simulations, you might be getting wetter, warmer and in another simulation, it might be completely opposite, even though over the long term, they might still give you a similar trend.
So, this is why now, a number of climate models have been used to run what they call initial condition, large ensemble simulation, meaning that they can slightly perturb the initial condition, and then they run a large number of ensemble members, something of the order of like 25 to even like 100 ensemble members to allow us to look at the natural variability.
So natural variability is particularly important if you are looking at a shorter time period, right? Because if, let's say, if you are looking at a period of 30 years, then depending on whether we are in the part of the cycle where it is in the wet part of the cycle or in the dry part of the cycle will give you a very different picture for the 30 year period.
So using this kind of large sample simulations from several different models, not just a single model. So we account for uncertainty in the climate models as well as uncertainty related to natural variability. Then we can identify how much of the uncertainty would be related to natural variability and how much of it may be related to like uncertainties because of the models, because we do know that different climate models, they have slightly different ways to represent the processes.
And so models themselves do not always agree with one another either. So using this type of analysis, what we find, quite surprisingly, is that natural variability, can account for 70 to 80 percent of the total uncertainty in looking at the past changes as well as looking at the future changes of precipitation in California.
[00:07:56] Bridget Scanlon: So that must have been surprising results to some people. So when those results were published, were many people surprised by the results? Because it suggests, when you talk about natural variability, you are talking about, I guess, like El Nino Southern Oscillation or Pacific Decadal Oscillation or these types of atmospheric variability.
So were people surprised by the results? Did you have a hard time getting it through review?
[00:08:28] Ruby Leung: Yes, I think people were surprised. I mean, I think, for California, number one, it's not surprising that natural variability plays a role because we do know like during El Niño year versus La Niña year, California experiences different amounts of precipitation and the temperature also affected as well, not just the precipitation.
So I think people in California are already quite used to like this kind of natural variability. But besides El Nino and the La Nina that we you just mentioned, which is really important for precipitation over California. El Nino condition they have a roughly a period between two to seven years but over the longer time scale they are also variability related to a natural part of the coupled atmosphere ocean system so this type of variability comes about not just because of the atmosphere but it's because of how the atmosphere interacts with the ocean and ocean has a much longer memory and therefore it can actually create variability.
Some natural variability that is much longer time scale, like on the order of 30 years or around that time frame. So when we were doing our study, we were particularly interested to look at the decadal trends, not just looking at, like, five years trend or 10 years. We were looking at the Cato trend.
So, as I mentioned earlier, the long-term trend in California in the last few decades has been a. decreasing trend is drying so an important question is this drying driven by climate change or not, right so what we found is essentially that a large part of this drying trend that we have been seeing in California is actually because of the natural variability so this particular type of natural variability is caught in the Interdecadal Pacific Oscillation or IPO.
It's quite similar to another term called Pacific Decadal Oscillation. So they have a periodicity of roughly between 25 to 30 years around that, and so about 30 years ago. So this oscillation kind of switched to a negative phase. And so during the negative phase, it would cause drying in California. And so what we find is that a large part of the drying that we have been seeing in California is actually because of this change of the phase of the IPO.
[00:10:59] Bridget Scanlon: And so that, that is really interesting because then that will, if you could predict that better then the IPO phase and when it might change and stuff like that, you might have a better chance of predicting rainfall in California. So in your analysis, then you have these large ensembles of models and then when you average out the large ensemble, because each ensemble might represent different phases of this natural variability. So averaging it out, then you would average out the natural variability. And so you'd be left with the anthropogenic climate change component. And so when you do this analysis, you have the total variability, and then you have the natural and climate change impact.
And it's amazing that like 70 to 80 percent of it is natural variability. And so I guess with that large natural variability then in the past when people said, Oh, there's discrepancy between your climate model and what we have observed. But what we have observed is just really one member of this ensemble.
And we think that's the only thing that, you know, and so we should match that. But with such large natural variability, that's not really the case then. And even going forward then to the end of the towards the end, towards 2100, then you're seeing still natural variability being very important related to these decadal cycles.
[00:12:24] Ruby Leung: Exactly, exactly. Yeah, so as you said, right, so oftentimes people look at climate models, and then they will see like, can the models actually reproduce the observation. So long time ago, most people when they look at climate simulations, they only look at the long term average, and then they compare the model simulated long term average, let's say, the spatial pattern of precipitation, and then they compare with the long term average spatial pattern of precipitation and say, Hey, do they match up or not?
Right? So this is one way to evaluate how well your models perform. But more recently, I think there is an increasing interest in seeing whether climate models can reproduce the observed trends, not just the long-term average spatial pattern or the long-term average quantity, right? So this becomes actually could become dangerous because as we said, right?
So the long-term trend that we have seen. Long term in this case, meaning like 30 to 40 years, not like 100 years. If we are looking at 20 years or 30 years or so, based on the past, and then if your models do not project the same track, did not simulate the same trend. So are you saying that the models are wrong or something?
And so, yeah, so this is a really important question. And as you said, what we can conclude is essentially that natural variability is a very big part of the climate in a particular region. We really need to consider the natural variability and take account of that when we compare the model simulated trends with the observed trends.
So in our study, we particularly looked into this. And so what we find is that if you do condition because now looking back in the last 30 years, we know that there was a shift In the IPO from a positive to a negative phase so when we look at the large ensemble sample simulations we can actually condition the large ensemble sample simulations to be in the negative phase and if we do that actually we find that the models were able to reproduce the observed drying trend.
If you condition it on the negative phase of the IPO so so this is really important and so when you then look at the future similarly you have to consider that what phase of the IPO might be in the future and so this is particularly important if you are looking at a shorter time period let's say what's the trend in the next 30 years then a lot of it would depend on whether we would be in a positive or in a negative phase of the IPO.
[00:15:06] Bridget Scanlon: Right. And since about the mid-seventies, I mean, we've gone from a positive to a negative phase. And now it's questionable whether we are transitioning back to a positive phase, but you can only determine that after the fact, after you have transitioned and you're pretty sure about it. But so are the droughts, I mean, we've, in California, you've had some long-term droughts, 80s, 1987 to 92, and more recently 2012 through 2016 and, and also after that. So if they're related to the negative phase of IPO, it means that maybe if the IPO changes phase, then you might get into a wetter cycle. And so it's not just a one-way street with climate change, it's up and down.
And so it seems a bit more hopeful than if it was a one-way street.
[00:15:57] Ruby Leung: Yeah, it is a bit hopeful, but it's also very challenging because this type of natural variability is pretty difficult to predict, like, let's say, if we wanted to predict 30 years from now, would we be in the positive or in the negative phase of the IPO?
This is still a very big challenge for our understanding as well as. Our models to be able to predict. And so at the end, I think it is important for us to recognize the long-term trends, very long-term trends that are due to anthropogenic forcing because they would really be driving the background, right?
So, so you might think of it like a trend that might be going up or going down depending on the region you're talking about for precipitation. But then on top of that, you have this fluctuation. Sometimes the fluctuation may be less with a cycle of 20 to 30 years. But if you are looking at an even shorter timescale, we even have to consider things like El Nino, La Nina, right?
So there are all kinds of variability adding on top of these longer term trends that are driven by external forcing.
[00:17:03] Bridget Scanlon: Right. And then another aspect of your work that you looked at, Ruby, was the sharpening of the seasonal cycle. I mean, California gets most of its rainfall in the winter period, but in recent times then you've been seeing that this has contracted and you're seeing most of the winter precip over a shorter time frame.
Can you describe that a little bit?
[00:17:25] Ruby Leung: Yeah, so again, this is a topic related to California's precipitation. Yeah, so we have been looking at the seasonal cycle, and that's important because in California, seasonal cycle is everything, right? So precipitation, we get the peak of the precipitation during wintertime, mostly from around November to March, but unfortunately, the time when we need water the most is in the summer for irrigation because California is after all a really, agricultural state with a lot of agricultural production, right?
So really trying to match up the precipitation, with the water demand, which peaking in the summertime is really what water managers have to do, which is why we needed to have reservoirs in order to manage the timing of that. Right. So, this really motivated us to look at weather and the climate change, whether the seasonal cycle could be changing, right? So currently we see a very big peak in precipitation over California, especially in December around December time frame, and then we still have quite a bit of precipitation from November to March. And what we're seeing in the climate model’s projection is that this seasonal cycle, although it still looks about the same shape, is the two major differences that we see. Number one, the beginning of the rainy season and the ending of the rainy season will be squeezed together so that the rainy season itself will become shorter. So, this is what we call a sharpening of the seasonal cycle, but at the same time, the peak.
We find them the peak would be increasing, although with a large uncertainty as well, partly because of the natural variability that we talked about. So the increase in the peak has large uncertainty. But interestingly, this shortening of the rainy season is robust across models. So we look at a large number of models by 36 or 37 models, most of them, 90 percent or more of the models, they all projected that under global warming, the seasonal cycle will be sharpened.
And interestingly, there has been also a study that tried to look for this in observation to see, can we already see this in the observation? And what they find is at least some indication, even based on observation, that the seasonal, the rainy season, the length has been decreasing already in California.
So this has become a really important question for us to answer, like, why? So why so many models project that, right? So, we published, several, two studies actually, try to look into the mechanisms of that. Maybe in a nutshell, I can just very briefly summarize it. So what we see is that the increase in the peak of the precipitation during the wintertime around December, that kind of timeframe is most, mostly because of two reasons.
Number one is because projecting into the future. We see the Pacific jet stream extending further to the east. So it is steering more storms towards California. So that is why the peak rainy season becomes rainier, although again with the Pacific. Quite a bit of uncertainty, and then there's also this low-pressure system sitting over North Pacific is called the Aleutian low is also strengthening.
And so that is also direct more moisture towards California, making the peak part of the precipitation seasonal cycle to be wetter.
But then the other part, why, why are we seeing reduced precipitation in the shoulder season, like November or March? And so that is very interesting. What we find is that it has, it has to do with the land sea warming contrast.
So, this is a very robust feature, that we understand is that under global warming, the land will warm more than the ocean. Which is quite easy to understand because land is the heat capacity is lower than the ocean, right? So this land sea warming contrast itself can actually turn into a difference in terms of how much of the moisture can be affected from the ocean towards the land.
And so this is a very robust feature, partly because it is not related to just changes in the winds, but rather it is related to the land sea temperature, the warming contrast, which is a very robust feature.
[00:22:03] Bridget Scanlon: That is very interesting. I mean, I know you mentioned this temporal disconnect between water supply in the winter from the winter precip and then the demand for irrigated agriculture in the summer, but I mean, I can remember being at various meetings and people would make a career out of saying whether they could predict precipitation and water availability in California. And I thought, well, if you do not have snow, you don't have water in the summer for irrigation. That seemed to be a lot better to me than say, for example, in North Texas or where the High Plains or whatever, where you are relying on summer convective storms. Yes. And so you do not know when you're planting in the winter, whether you will have any water or rain or not. So even though there is a temporal disconnect, and then you start planting. store it traditionally in snow reservoirs and then in actual reservoirs. So you would know early on whether you're going to have enough to support agriculture or not. So I think that was kind of an interesting aspect, but it's cool to, to try to understand the sharpening of the winter precip signal, why, and the reason behind that. That is amazing. And I guess with modeling, you can test. so many different things then, and you can amplify things. And so they are a great tool to evaluate the different causes or potential causes of different processes. So we have been talking a lot about drying conditions related to decadal climate cycles and the IPO and stuff.
And is that also related somewhat to blocking ridges that block rainfall from coming on land in California or atmospheric conditions that might reduce rainfall related to those decadal patterns.
[00:23:48] Ruby Leung: Yeah. So in a sense, yeah, indeed, I mean, this type of inter annual variability or decadal variability are associated with large scale circulation pattern, which include like, for example, I mentioned about the Aleutian low, which is a low-pressure system sitting over North Pacific Ocean, but over some years, this can be, which meaning it's not like it becomes a high-pressure system, but it, but just that the strength of the low pressure system itself can also have very large variability from year to year or from decadal period to another decadal period. Yeah, so, so this type of large-scale circulation changes are very important for us to understand and predict the precipitation as well as the temperature as well.
Yes. Right. So because in California, we worry about not only the precipitation, but how warm the precipitation is, because that would affect how much of the precipitation accumulates a snowpack. As you said, right, if you have snowpack in the winter, then you worry less about the amount of water that you have in the summer.
[00:24:56] Bridget Scanlon: Right, right. And so with these drying conditions then in the last few decades, there has been increasing wildfires. And we have heard about insurance agencies pulling out of California and things like that. So a big impact. And I really liked your study of wildfires 2100 future, where you looked at potential predictors of these wildfires and there can be many factors that contribute to them.
And so using machine learning was a nice way to include all of those possible predictors and then try to rank them or figure out the relative importance of those. Maybe you can describe that a little bit, Ruby.
[00:25:37] Ruby Leung: Yes. So that's another line of research that we started a while ago. We, I have been talking to you about like climate models, right?
So we look at climate models to look at natural variability and we use climate models to look at how anthropogenic forcing, like increasing greenhouse gases may be affecting climate. So these are what we call physically based models, right? So we have equations, we write down the equations for the motion of the air, the thermodynamics of the air and all of these combined together. And so while we find these models to be very useful, for example, looking at precipitation changes and warming and things like that, wildfires are rather different.
Wildfires are quite complicated, partly because wildfires are not only related to natural processes Like, so we know some natural processes such as how humid the air is, how warm it is, as well as drought condition can contribute to wildfires. So there are lots of these factors that are natural, but wildfires are also very much affected by human factors too. So the population, if you have an area that is close to a populated area or during summertime when people go out to camp, so, so there are lots of human activities can also affect wildfires.
So for that, we were thinking, hey, maybe we can, give it a try using AI, machine learning type of model. because significant advances have been made about these types of techniques in the last 10 years. So. This type of method is nice about it is that the more data you feed into it, the better it can do, right?
It's basically what we call a data driven approach compared to the climate models, which would be what we call more like a physically based, approach so so then using machine AI machine learning we develop a model first of all to try to see whether we can actually predict wildfire so the first study that we did we try to see ok probably it would be very difficult to predict wildfire on a daily basis because whether you have a wildfire today or tomorrow depends a lot on whether there is like, like a lightning or some kind of ignition and that kind of activities would be very difficult to predict, but maybe on average, if I wanted to only predict what about the monthly over the month, can I predict whether this month is going to have more wildfires or fewer wildfires? So our first study, we have a more manageable goal.
We try not to predict wildfires on a daily basis, but we try to predict it on a monthly average to see whether we can actually predict the size, the area covered by the wildfires. And we use almost like 24 predictors in this AI machine learning model. So these predictors, as I said, include a lot of natural processes like temperature, precipitation, humidity, fuel, moisture, a lot of these natural factors, but it also includes but some of the predictors are also human factors like population and things like that.
So feeding all of this information into the machine learning model, we find that we have to divide it into what we call training period versus the validation period. So we first take half of the data, train our machine learning model, and see whether we can predict the other half of the data that the machine learning model has not seen before, right?
So, we were finding that the machine learning model was quite skillful in predicting the wildfires, the size of the wildfires on a monthly basis and also at about a quarter degree. So we divided the United States into quarter degree grid boxes. And so after we, learn that the machine learning model can do a good job in predicting the wildfires size.
Then the question becomes like, why the machine learning model can do that? It must be something that the machine learning model has figured out, right? So, so, so we wanted to tease apart. What has the machine learning model learned? Which variables, which predictors? What is actually used most importantly by the machine learning model to be able to be so skillful.
So we use a type of technique which is called explainable AI, which essentially tease apart to try to see which predictors were actually used by the AI machine learning model in order to be better. able to predict. So, so, so by doing that, then we can learn which predictors are actually more important and rank them and see, does it actually make sense?
I mean, are these really the more important predictors or not? And so I think I find that this technique is really quite useful. We learn a few things. Number one, mostly when we think about predicting wildfires, because it's an extreme event. So you would imagine, well, I can give the machine learning model predictors local predictors. Like if I want to predict wildfires here in this quarter degree by quarter degree grid box, I can give it the temperature at this location, the precipitation, etc. I don't need to give it any other information from far away. It doesn't help here. But what we find is that actually some large-scale pattern is very important. Like for example, in order to predict wildfires in California, we find that a large-scale pattern, particularly related to a high pressure system, how persistent that pattern is actually becomes a very important factor. Like when we look at the daily variability of this high-pressure system, if it persists over over let's say a week or more than the chance of having a wildfires become much larger so it makes sense because this kind of large scale high pressure system induce downward motion and cause clear sky so it warms up the air and so it can increase the chance of wildfire so so this is number one that we learn in only in two regions in the united states no three regions in the united states but we find that this kind of large scale patterns are important for predicting wildfire. The first one is California. The second one is over the Rocky Mountain region. We also find that some high-pressure system sitting around is also important. But also over the southeastern United States, we find that some kind of large-scale pattern also important, but it is the large-scale pattern two months before the wildfire that gives you the predictability of the wildfire. Because apparently some wildfire some large scale patterns cause more stormy weather and the more stormy weather actually gives you more precipitation two months before wildfires. It gives you more vegetation. So, so, so that it actually contributes to the wildfires two months later. So this is one thing that we learned that we find quite interesting and then the second thing that we learned is that this kind of technique can be used to look at individual year like not only like over the last 20 years which predictors are more important we can actually look at it individually like for example in California2020 was a really big wildfire season right so was 2017.
Also a lot of wildfires and we wanted to see are they because of the same reason or because of different reason and so this machine learning model we tease it apart and see like which predictor actually tells the machine learning model that 2020 is a big fire season and we find that actually it was because of different things. We find that in 2020, the biggest factor for why the wildfires were so big in California was because number one, there was this high-pressure system sitting there for pretty long persistently. And number two, it was. In there was a drought ongoing drought condition, but in the 2017 year, when there were also pretty big wildfires, we find that it wasn't because of that, not much drought condition, no high-pressure system sitting there. It was simply because that year, the relative humidity was kind of on in general, lower than average. So I think this is also telling us that it is pretty useful tool for us to learn about work. physical world.
[00:34:14] Bridget Scanlon: Yeah, I think with so much data available now, more and more agencies and groups are using this data driven approach.
I mean, I think if the U. S. Geological Survey and we did analysis of water quality in the U. S. using machine learning, you can throw in the kitchen sink. Yes. And so it does not matter if those, those, parameters are correlated or not. And so you do not have all of the limitations of traditional statistical approaches.
And so it's a really kind of a robust way of looking at these things. And so it seems like some of your findings are a little bit counterintuitive then. So if you have increased rainfall in the Southeast two, two months ahead of time, and then you have increased fuel basically, fuel load, I guess, makes it more susceptible to fire.
But when you mentioned 2017 in California, I mean, that's when the drought ended from that, when extended from 2012 through 2016, and you had all those atmospheric rivers, you think there was more vegetation or did that contribute at all, you think, or that was a wet year.
[00:35:21] Ruby Leung: Right, that is a very interesting point.
We did not see in 2017 a particularly large contribution coming from the fuel. We were seeing a larger contribution coming from the relative humidity. Yes, yeah, good point about this ending of the drought. Yes, indeed. So that is why in the model, in 2017, drought was not selected to be the predictor.
[00:35:43] Bridget Scanlon: you normally think of wildfires being associated with dry conditions and drought, but 2017 was not one.
And of course, I do not know if you have looked at other regions globally, but certainly Australia has been subjected to these wildfires and extensive wildfires. And trying to understand the causes of those will help you try to solve some of those problems and develop appropriate solutions. If you do not know what's causing them, you can't resolve the issue.
So one of the aspects of your work that I really like is that you work with stakeholders. And so you're not siloed and just doing your own thing. And then at the end of the day, then trying to communicate with stakeholders, but you actually work with them through the process. And I think you had a program called HyperFACETS or a DOE Department of Energy program, where you work continually with stakeholders.
I would love to hear you describe that and how successful you think it has been in parlaying your, knowledge that you gain from modeling and other things to water managers and resource managers.
[00:36:51] Ruby Leung: Yes, thank you for bringing this up. Yes, so this particular project, as mentioned, is called HyperFACETS. it's funded by the Department of Energy.
it is one of the few projects that actually has a pretty big stakeholder engagement component. it, the project itself involves a large number of scientists. I am only one of the scientists, and I have to give all the credits to several people on the project. They are particularly really leading the effort to do stakeholder engagement.
But it is interesting that across the project, including people like me. I do not really have much expertise in stakeholder engagement. But despite that, I would say that all of the scientists in the project are interested in contributing. So we have monthly meetings with stakeholders and we participate. And sometimes we give a short presentation. Short little presentations about the science that we are doing, and stakeholders also have chance to give presentations as well. So this is one way where we can really exchange information towards co production, meaning that the type of information that we provide about climate change would be, we hope would be useful for the stakeholders.
So, and that in through this kind of co-production, they better understand what are the limitations of the information that we provide and how we might improve on the information that could be more useful for the stakeholders. So I think this kind of two way dialogues have been really important. So as part of this stakeholder engagement, as I said, a few people on the project have particularly been very active in that they went out and recruited stakeholders, talking to people from related to the water resources side. And also, we have some people who are interested in wildfires and energy and these other many different types of topics. So yes, so through this type of dialogue, regular monthly meeting with them, we have been really able to better understand really what kind of information could be more useful.
[00:38:54] Bridget Scanlon: And, and I forgot to mention, I wanted to check with you briefly, Ruby about, atmospheric rivers, because I know we have been talking so much about drought, drying and wildfires, but also the other side of the coin with these atmospheric rivers. And I know you have had papers on atmospheric river tracking, model comparison projects and things like that.
How does that fit into your decadal variability in your cycles and your natural variability? and how do you think atmospheric Do your climate models suggest they might change in the future or because they seem to play an important role in water resources in California?
[00:39:32] Ruby Leung: Yes, so atmospheric rivers are definitely an important topic phenomenon that we care about if we look at California's precipitation.
So, just to mention that when I talk about the sharpening of the seasonal cycle of precipitation, indeed, as a supporting evidence, when we look at the climate model simulations, they also project a sharpening of the seasonal cycle of the frequency of atmospheric rivers making landfall. So, so in the atmospheric rivers, the changes in the atmospheric rivers contributed also to these sharpening of the seasonal cycle.
So climate models generally project that atmospheric rivers frequency will be increasing, partly because in a warmer climate condition, then the air become more moist and therefore you would have a more chance of these type of atmospheric rivers transporting enough moisture, towards the West Coast and therefore producing more precipitation.
I think this is a very important topic, not only that we need to better understand, like the changes in the intensity, like how moist the atmospheric rivers might be, but I think now there are also a lot of concerns about whether that the, how, these atmospheric rivers may come in sequence become also a big concern as well, because, like, over the last few years, sometimes we've seen atmospheric rivers one after another.
So these are the type of situations where we really worry about because the first atmospheric river can cause a lot of damage. Wetting and soaking up the soil. The second one. There is no way there's the soil moisture can go anywhere except to generate runoff and flooding. So I think understanding how this type of frequency of atmospheric rivers coming in sequence is a very important thing.
And as a matter of fact, you might have heard about this. USGS has developed scenarios for California. They call it the wetting
[00:41:36] Bridget Scanlon: Oh, the flooding, was it an ARkStorm or something?
[00:41:38] Ruby Leung: Yes, so essentially it was based on what was observed back in 1860s when there were several atmospheric rivers, one after another, almost over the whole month that flooded California with lots and lots of water.
So, I think this is an, a scenario that USGS developed, to, in order to see like how this type of sequence of atmospheric rivers may be changing in the future.
[00:42:07] Bridget Scanlon: Yes, I think people's psychology changes over time with these, I mean, I guess some people refer to them as families of atmospheric rivers.
And so initially there, they've been in a drought for years and, and they welcome them and it's a drought and then as it gets more and more intense and everything, and there's no room for the moisture in the soils and then it's flooding, it's becoming, Oh my gosh, how are we going to cope and stuff? And so I think that is one of the things in water resources is trying to manage these extremes and I think that the Army Corps of Engineers and others are looking forecast informed reservoir operations and things like that and subsurface storage in depleted aquifers and things.
So I think once we know what we have to deal with, we can usually develop some ways to adapt and manage these things. and I think, Ruby, you mentioned stakeholders and, but you've also been working a little bit or reviewing some of the documents that the California Department of Water Resources, their Water Plan.
And when you mentioned 20-to-30-year time cycles, that's usually sort of the planning horizon for these water managers. And it may not be that concerned about 2100, but maybe 2030 to 2050. And so it seems like your work is highly relevant to those timescales and I think, how do you think they pick up on your work and incorporate it into their management plans or is that kind of something that will happen more in the future or what are your thoughts?
[00:43:50] Ruby Leung: Yeah, I hope that will happen more in the future, but I have, as you mentioned, I have reviewed the plans for the Department of Water Resources in California. I think they have done a really wonderful job in making information available, right? So first of all, looking at the past 100 years, so fitting a trend to see that, oh, so in the past 100 years, 100 years.
It is already changed. It is not like stationary. So you do have to account for the changes that have already happened to account for the current situation, right? And then you take in information from climate models to project to the future. I think this is a wonderful job that they have done. Taking into the information about climate variability would be important.
But as I said, unfortunately currently the skill for predicting which phase we would be in whether it's a positive or negative phase in the next 30 years we have very little information so I think this type of uncertainty we just have to account for them and there's not too much that we can do to reduce the uncertainty in that regard but being just be aware that these are the uncertainty that we have to cope with, I think is an important factor already.
[00:45:05] Bridget Scanlon: But also I think it's kind of hopeful that, yeah, we're, California's been in a negative phase of the IPO and that's associated with drying and that's a 25 30 year cycle and so maybe come out of it then and maybe will be a wetter phase.
So it is not just all drying and scarcity. So I think that is somewhat promising. And then they developed their management plans, accordingly that would consider those sorts of cycles and stuff and not, just think of a one way street. So, well, I really appreciate your time. I know we haven't covered 10 percent of all the work you do but I think I am really impressed with the insights that you bring with the various testing using the ensemble member approaches to the modeling and trying to isolate natural variability versus climate change impacts, anthropogenic climate change, and highlighting that 70 to 80 percent of the variability in California is natural variability.
And then wildfires and the machine learning approaches, trying to understand the causes of those. and those large-scale patterns then that impact those. And most of all, it is really impressive that the Department of Energy is working with stakeholders closely to help translate the science that you are doing to water resource managers and others.
So thank you so much for your time, Ruby. I really appreciate it.
[00:46:28] Ruby Leung: No, thank you for the opportunity.
