Mapping Waters of the United States Using New Tools, with Simon Greenhill and Hannah Druckenmiller

Notable quotes

• “Waters of the United States” are open to interpretation: “One of the reasons that the Clean Water Act is so controversial, and you’re constantly seeing it in the news and the Supreme Court, is that [the waters of the United States] is actually not clearly defined in the original legislation. It’s actually up to the US Environmental Protection Agency and the US Army Corps of Engineers to interpret what [‘waters of the United States’] means.” —Hannah Druckenmiller (4:19)
• Predicting regulatory coverage: “A big-picture way of trying to think about what we’re doing is that we try to use machine learning to basically recreate the decision problem that the Army Corps faces. In particular, we take a bunch of data as inputs … and we feed those into a machine learning model … Out of that model, we get a decision which is equivalent to the decision that someone at the Army Corps might make—basically, a prediction: either regulated or not regulated.” —Simon Greenhill (12:26)
• Machine learning as a predictive tool: “Basically, this [tool] gives us visibility into what is regulated under different rules that have existed in the past … That can be useful to somebody who’s trying to understand whether the location that they want to develop might be regulated. It could also help … someone at the Army Corps make a decision and potentially help to provide decision support in that sense to the people who are actually implementing the law. It also allows us … to get these big-picture summary statistics that can be helpful simply for understanding what the Clean Water Act is actually doing.” —Simon Greenhill (15:00)
• Volume of regulated waterways can drastically change under different interpretations: “In total, we find that the rule change [from the Rapanos v. United States decision to the Navigable Waters Protection Rule] deregulates about 35 million acres of wetlands. How big is 35 million acres? That’s about the size of the state of Wisconsin or the state of Florida. This is a huge amount of area that’s losing regulation under this rule change.” —Hannah Druckenmiller (17:52)

Top of the Stack

• “Machine Learning Predicts Which Rivers, Streams, and Wetlands the Clean Water Act Regulates” by Simon Greenhill, Hannah Druckenmiller, Sherrie Wang, David A. Keiser, Manuela Girotto, Jason K. Moore, Nobuhiro Yamaguchi, Alberto Todeschini, and Joseph S. Shapiro
• Clean Water Act regulation map
• Clean Water Act regulation map explainer video by Simon Greenhill
• “Wetlands, Flooding, and the Clean Water Act” by Charles A. Taylor and Hannah Druckenmiller
• The Hungry Tide by Amitav Ghosh
• The High Sierra: A Love Story by Kim Stanley Robinson

The Full Transcript

Daniel Raimi: Hello, and welcome to Resources Radio, a weekly podcast from Resources for the Future. I'm your host, Daniel Raimi. Today, we talk with Simon Greenhill, a PhD candidate in agricultural and resource economics at the University of California, Berkeley, and Hannah Druckenmiller, an assistant professor of economics at the California Institute of Technology and a Resources for the Future (RFF) University Fellow. Along with several coauthors, Simon and Hannah recently published an innovative study that uses machine learning tools to try and determine which locations are subject to regulation under the Clean Water Act.

As we'll discuss, there are multiple ways to interpret what locations should or shouldn't be subject to regulation. But even once you decide on how to define “waters of the United States” and the locations that should be regulated, it can be really complicated to figure out exactly which specific locations are subject to the law. Simon, Hannah, and their coauthors have developed a really cool tool to help make this process easier and to better understand the environmental and economic consequences of Clean Water Act regulations. We'll talk about all of it in today's episode. Stay with us.

All right, Simon Greenhill, welcome to Resources Radio. Hannah Druckenmiller, welcome back to Resources Radio. It's great to have both of you with us.

Hannah Druckenmiller: Thanks. I'm really excited to be here.

Simon Greenhill: Yeah. Thanks, Daniel.

Daniel Raimi: As I mentioned, Hannah has been on the show before. But Simon, it's your first time—welcome. Thank you so much for joining us. I'd love to ask you the same question we ask all of our guests at the beginning of the show, which is, What inspired you to work on environmental issues?

Simon Greenhill: Thanks. It's great to be here. I grew up in a rural part of California and was just really fortunate to have great outdoor access there. As a kid I spent a lot of time outdoors. Then, as I got older, that developed into a love for hiking and then, later, rock climbing, backpacking, trail running—stuff like that. Then, as I was going through my education, I was really interested in human behavior and what determines human behavior. That started out as an interest in journalism. Then, I discovered economics and particularly fell in love with the approach of using data to try to learn things about the world and particularly to learn things about society.

I found environmental economics as this marriage of the environment—something that I had been really passionate about throughout my life—and then these economic and social systems that I've always been interested in. That felt like a natural fit, and I’m excited to be doing that now.

Daniel Raimi: That's really cool. Where in California did you grow up?

Simon Greenhill: I grew up on the peninsula—the Palo Alto area, but more in the hills, as you go west, toward the coast. It's fun, because you're 30 minutes from San Francisco, you're close to Stanford's campus, but you also feel like you're up in the mountains.

Daniel Raimi: That's such a beautiful part of the world. I totally understand how inspiring it is.

Well, let's get into the substance of our conversation today, which is around a paper that the two of you recently published along with a team of coauthors. The paper's called “Machine Learning Predicts Which Rivers, Streams, and Wetlands the Clean Water Act Regulates.” It's in the journal Science. We'll have a link to it in the show notes at the bottom of the page. But one of the first things that's important to do in any conversation about the Clean Water Act is defining a couple terms, and the first term that we should define is “waters of the United States” (WOTUS). Can you define for us what the waters of the United States are, and why are they important?

Hannah Druckenmiller: Sure. I can do my best, but you'll see why this is difficult in just a moment. “Waters of the United States” describes a set of water resources that's regulated under the Clean Water Act. But one of the reasons that the Clean Water Act is so controversial, and you're constantly seeing it in the news and the Supreme Court, is that WOTUS is actually not clearly defined in the original legislation. It's actually up to the US Environmental Protection Agency (EPA) and the US Army Corps of Engineers to interpret what WOTUS means. This is repeatedly reinterpreted each time we see a new presidential administration, or each time the Supreme Court revisits what WOTUS was considered under the original writing.

Let me give a few examples of different water resources to help try and make this concrete. I think everyone agrees that major rivers are something that the EPA and the Army Corps called traditional navigable waterways. You can think of the Mississippi, for example. Everyone agrees that these water resources are regulated. There's also some things we all agree are not regulated. You can imagine a puddle in your backyard on a rainy day. No one is saying that's regulated by the Clean Water Act or that that's WOTUS, but there's a whole category of resources in between these two things where it's not actually clear whether they're a WOTUS or not.

To give a few examples, you can think of an ephemeral stream, which is a stream that only flows on rainy days, or an isolated wetland, which is a wetland that's not directly connected to a stream or a river. These are things that we're repeatedly debating whether they fall under the definition of WOTUS or not.

Daniel Raimi: That is great. Just one more clarification question before we talk a little bit about policy history, Hannah: What does it actually mean to be regulated? If I am a farmer, and there is an intermittent stream on my land, what does it mean for that stream to be regulated for me as a farmer?

Hannah Druckenmiller: Sure. Something being regulated mostly means that you have to get a permit in order to develop on it. I'm going to switch your example from a farmer to a developer, because farmers actually have exceptions under the Clean Water Act that make that example a little murky. But imagine you're a developer, and you want to buy a property and build a residential development there. If this land is not regulated, you can go ahead and build whatever you want at the market cost. If it is regulated under the Clean Water Act, you have to undergo an approval process in order to build there, which often requires costly mitigation measures for whatever you built.

To give an example of a wetland, if I wanted to build a residential development on a wetland, I might have to then invest in wetland mitigation banks, which is this system that's meant to restore the ecosystem services that you demolish when you pave over a wetland somewhere else in the watershed. You can see why this might make it prohibitively costly to develop in places that are regulated.

Daniel Raimi: That is a perfect example and also helps us understand why this is such a controversial and economically significant topic. I'm wondering if you can give us some policy background on how the Supreme Court and different administrations have interpreted WOTUS in recent years. You mentioned that these definitions swing back and forth with different administrations. Help us understand some of the environmental and economic implications of those different interpretations.

Hannah Druckenmiller: I'd say, loosely speaking, we see two big interpretations of WOTUS. One, which I'll call a broader interpretation, includes more of those ambiguous cases like ephemeral streams and isolated wetlands in the class of resources that is regulated in the definition of WOTUS. Then, there's a narrower interpretation which tends to exclude many of these water resources.

In our paper we look at two of these different rules. There have been several over the last few decades, but we focus on two in particular. One we call “Rapanos,” which is based on a 2006 Supreme Court case where the Supreme Court established a “significant nexus” criterion for WOTUS. That's a big fancy term, but what it meant was that any water resource that shared a physical, chemical, or biological connection to a traditional navigable waterway could be considered as part of WOTUS.

This allowed for the inclusion of ephemeral streams and isolated wetlands under that definition. The second rule that we look at is a narrower interpretation of WOTUS. This was the Navigable Waters Protection Rule, or NWPR for short, which is a Trump-era rule. This rule basically required that a water resource needed to share a continuous surface-water connection with a traditional navigable waterway in order to be regulated. That actually excluded some of these resources like ephemeral streams and isolated wetlands from the definition of WOTUS. You can see it gets quite technical. You can think of this as being a broader interpretation of what can be regulated under Rapanos and a narrower interpretation under NWPR.

Daniel Raimi: Perfect. All right, we're in a great starting point. Let's talk about the technical difficulty once you get past these legal questions about how to actually map the scope of water resources in the United States. In your paper, you cite EPA and the Army Corps as stating, in all caps (I don't know if the all caps is significant or not; you can tell me if they are), "EXISTING TOOLS CANNOT ACCURATELY MAP THE SCOPE OF CLEAN WATER ACT JURISDICTION." Why is that? Why is it so hard to map these waters?

Simon Greenhill: First, I think that all caps is significant, and in fact, it was even bold in the initial document. It’s a strong statement from the EPA and the Army Corps that, until now, there haven't been tools that allow us to map Clean Water Act jurisdiction. The reason this is difficult is that the main way that Clean Water Act jurisdiction is determined is through a process called the approved jurisdictional determination process, or AJD process. That's a fundamentally case-by-case process. The way it works is (picking up Hannah's example from earlier), if I'm a developer, and I think there might be a water of the United States on the property that I'm planning to develop, I go to the Army Corps, and I say, "Hey, could you please come and review this location and let me know if there are waters of the United States here or not?"

Someone from the Army Corps will either look at data from their desk and, in some cases, make a field visit and decide under the currently operative rule, whether that's Rapanos or NWPR or something else, if there is water of the United States here or there is not. Anytime the rule changes, it's not like the Army Corps does or even could ask its staff to make thousands or millions of fake determinations in order to be able to—I shouldn't say fake determinations, but pseudo-determinations—in order to be able to then create a map and say, "Here's regulated and what's not."

Instead, it's more like, as the rule is actually enforced and is put into law, we learn as we go what is regulated. What we do in our paper is create an automated way, based on the past decisions that the Army Corps has made, of determining jurisdiction by actually using the decisions that they made through that AJD process.

Daniel Raimi: That's great, and I'd love for you to take us a cut deeper on that and, particularly, what are some of the methods you use to try to map what is WOTUS and what is not WOTUS under these different broader or more narrow interpretations.

Simon Greenhill: A big-picture way of trying to think about what we're doing is that we try to use machine learning to basically recreate the decision problem that the Army Corps faces. In particular, we take a bunch of data as inputs. These are aerial imagery of the location that's being evaluated, maps of the wetlands and streams, the soil types, and the local climate, and we feed those into a machine learning model. We specifically use a convolutional neural network, which we don't need to get into the details of right now, but it’s a type of model that's particularly good at seeing patterns in spatial data. In that sense, it is well-suited to this task. Out of that model, we get a decision which is equivalent to the decision that someone at the Army Corps might make—basically, a prediction: either regulated or not regulated.

The sense in which this is recreating the decision problem that someone at the Army Corps would face is that we use the same inputs that they use. We know that because, in their decisions, they often write, "I looked at imagery from this location, and then I consulted the National Hydrography Dataset and came to this decision." We selected our inputs based on that, and then we're, as an output, predicting these actual legally binding regulatory decisions.

I think that makes our approach really powerful, because we're predicting actual decisions that the Army Corps made as opposed to some proxy for what might be regulated. Using that same data, once we have that, we can predict as many places as we want at very low cost, and that's what allows us to create these more comprehensive maps of what is and isn't regulated under different rules.

Daniel Raimi: That's super interesting. I'm wondering if you can say a word or two about the application of the information that you're developing with these tools. Are you envisioning that the Army Corps would use the model that you've built to help them make decisions, or is it intended for broader public consumption? Who's the audience, or who are the audiences that you think could be most benefited by using this tool?

Simon Greenhill: I think there's many potential applications of this tool. I think that, basically, this gives us visibility into what is regulated under different rules that have existed in the past. Of course, that can resemble rules that might exist presently or in the future. That can be useful to somebody who's trying to understand whether the location that they want to develop might be regulated. It could also help, as you mentioned, someone at the Army Corps make a decision and potentially help to provide decision support in that sense to the people who are actually implementing the law. It also allows us to think about the implications more broadly, to get these aggregate statistics of how many stream miles were regulated or unregulated under different rules in the past, or how many wetland acres are regulated or unregulated. Those allow us to get these big-picture summary statistics that can be helpful simply for understanding what the Clean Water Act is actually doing.

Daniel Raimi: Well, let's go to that now. I'm wondering if one of you or both of you can share some of those summary statistics, and what does the model tell us about how different interpretations of WOTUS—the broader Rapanos or the more narrow NWPR interpretations—affect the regulations of streams and wetlands and other parts of the United States?

Hannah Druckenmiller: I'll just pick up where Simon left off, and we can talk first about streams and rivers and wetlands. We find that Rapanos, which is that broader interpretation of WOTUS, regulates about two-thirds of streams and rivers in the United States. Then, when we move to NWPR under the Trump-era rule, we see that less than half of streams and rivers are regulated. This isn't equal across all types of streams and rivers. Consistent with the language of the rural change, we find that intermittent streams are more likely to be deregulated than perennial streams, for example.

Let's just provide you with a sense of the scale of those numbers. NWPR deregulates about 680,000 stream miles. That's the length of every river and stream in California, Florida, Illinois, New York, Ohio, Pennsylvania, and Texas combined. This is just a massive amount of water resources.

When we look at wetlands, we find a similar pattern. About half of wetlands are regulated under Rapanos—that broader interpretation—and that shrinks to about one-quarter of all wetlands under NWPR. We see that a quarter of all wetlands in the United States are deregulated with this rule change, or half of previously regulated wetlands. Consistent with the language of the rule change, we find that deregulation is more dramatic among isolated wetlands than wetlands that are directly connected to a stream or a river. In total, we find that the rule change deregulates about 35 million acres of wetlands. How big is 35 million acres? That's about the size of the state of Wisconsin or the state of Florida. This is a huge amount of area that's losing regulation under this rule change. In addition to those high-level statistics, we could also tell you about some case studies that we look at in the paper that reveal interesting patterns.

First, we find that both rules regulate all major rivers, which is consistent with how we think that the Clean Water Act operates in practice. But we do see that under the narrower interpretation of WOTUS under NWPR, there's a narrowing of water resources that are regulated in the areas surrounding major rivers like the Mississippi. Another place we take a close look at is the Mountain West. In the Mountain West, a large volume of water actually flows through ephemeral streams—those streams that only flow after rainy days. We find, again, a large deregulation of these resources when we move from Rapanos to NWPR.

Then, a last interesting case study is in the Upper Midwest, in a region that's characterized by prairie potholes, which are these freshwater marshes that tend to lack a connection to a surface water like a stream or a river. We see massive deregulation of these isolated wetlands in that area. This gives us a picture that's consistent with how we expected this rule change to affect different types of water resources, but for the first time, we're actually able to quantify the scale of these changes and think about just how many streams and rivers are affected so that we can start thinking about possible consequences for people and the environment.

Daniel Raimi: The images that are in the paper are really cool. They give you a great sense of that macro and micro picture that you just sketched out, Hannah—figures 3 and 4 of the paper in particular. I hope people are following along as they listen—not in your car, of course, but sitting at your desk listening and checking out these pictures.

I'd love to ask you to talk more about what you just said, Hannah. What are some of the consequences of these different levels of regulation? Obviously we're talking about huge amounts of land, huge amounts of water—both at the surface and potentially underground. What are some of the consequences for people and the environment?

Hannah Druckenmiller: This is something that we don't fully explore in the paper, but that many members of our team are really interested in and already working on in follow-up papers. But we do provide a few back-of-the-envelope calculations to put these numbers in context. To just give you an example, we find that there's this massive deregulation of streams and rivers. It turns out that about 30 percent of streams and wetlands that are in subwatersheds that provide drinking water are deregulated. So, this might have real consequences for the quality of water that is coming to our taps.

We also look at the value of wetlands that are deregulated. We found that about 25 percent of all wetlands in the United States were deregulated under NWPR. We calculate that that's somewhere in the range of $300 billion of land value in terms of the market value of that land. Another way we could look at this is the environmental services that that land provides. We look, in particular, at the flood-mitigation services that wetlands provide, because that's something that members of our team have looked at in prior work, and we estimate that that 35 million acres of deregulated wetlands provides somewhere in the range of $12 billion to $20 billion annually in flood-mitigation services.

Daniel Raimi: That's really interesting. I'm wondering, Hannah—is that drawing on your previous work estimating some of the benefits to society of wetlands?

Hannah Druckenmiller: That number is built on an estimate that came out of a paper I have with Charles Taylor that looked at what happens when you lose a wetland to flood damages downstream. We combined those spatially resolved estimates of wetland value for flood mitigation with our spatially resolved predictions of where wetlands lost regulation to come up with that number.

Daniel Raimi: One question that you explore in the paper as a validation of the technique is looking at how accurately the model is able to match real-world determinations that the Army Corps has made over the years. Can you talk a little bit about the accuracy of the model and how well it lines up with what the Army Corps might've decided in a given case?

Simon Greenhill: Absolutely. This is, of course, crucial, right? We have this fancy machinery and these terabytes and terabytes of input data, but it's all for naught if we don't have a model that we actually believe. The first thing that I'll mention is that anytime we're talking about accuracy metrics, we're talking about the accuracy on a sort of held-out test set. This is the general best practice anytime you're dealing with a prediction problem. But it's always worth putting front and center, especially for audiences that maybe don't deal with this kind of thing.

Often, anytime you're trying to predict something, you break up your data into some data that you're going to use to actually fine tune and train your model, and then some data that is going to be out of sample and that's going to give you an approximation of how accurate your model is likely to be when you predict on some place where you don't know what the true regulatory status of that place is, or whatever the sort of true value you're trying to predict is. On that held-out-of-sample test set, our accuracy is about 80 percent overall, which we think is pretty good.

It's better than any kind of random guessing that you would get. We do a number of things to try to unpack, as you mentioned, the accuracy and understand where the model may be more or less accurate. One of the tests that we do, which Hannah already mentioned, is that we feed in points along navigable waterways into the model, and we find that 100 percent of those are predicted as regulated. This is comforting to us, I think, because navigable waterways actually don't appear in the AJDs, in the approved jurisdictional determinations, that the Army Corps is making very much, right? People don't often try to build something in the middle of Lake Michigan or the middle of the Mississippi River. But as we've discussed, those locations are most definitely regulated, and we definitely want a model that claims to predict regulation to get it right in those obvious cases.

The other thing that we find and that's useful about the model is we find it's well calibrated. What that means is that the model produces a score, a number between zero and one. When a model is well-calibrated, that means that the score it produces actually maps well to a probability. If the model gives me a score of 0.05, that means that there's approximately a 5 percent chance that that point is regulated under the rule that I'm trying to evaluate.

What that means is there's subsets of the data where the model is particularly accurate, and those are the cases where the model might be particularly useful, especially when we're talking about decision support. Looking at that, about 30 percent of all the AJDs that we held out of sample can be predicted with a 95 percent or higher accuracy, and about half of them can be predicted within 90 percent or higher accuracy.

There's these pockets of the data where the model has particularly high accuracy, and because of this calibration aspect, we can actually determine where those pockets are and maybe get more out of those predictions and have more confidence in some of those predictions. In a prediction where the model's saying, "Hey, this is a toss-up”—speaking of toss-ups, I'll just add one more thing, which is that one of our case studies is the location of the Sackett property, which was the subject of last year's Supreme Court case and, of course, is an area of a great policy interest and attention. Our model does tend to predict around a 50 percent chance of regulation under pretty much all the rules, if you're looking at exactly that location. Our model’s predictions are consistent with the chain of events that ended up with the Sacketts in the Supreme Court.

Daniel Raimi: That's really interesting.

One last question to you all before we go to our Top of the Stack segment, which is building off something you just mentioned, Simon. The model tends to be maybe more or less accurate under certain conditions or in certain locations. I'm wondering if you can give an example or help us understand what are the factors—geographical, topographical, maybe something having to do with the satellite imagery—that might make the model more or less able to give an accurate prediction in any given location?

Simon Greenhill: There's many, and we look into many of them, but I think a really crisp example is the example of a field visit. The one thing I've talked about is how we're replicating the Army Corps's decision problem and so on and so forth. One thing that our model most definitely cannot do is get up and go on a field visit of a location. This is something that the Army Corps does pretty frequently. Something like 45 percent of its decisions include a field visit, and on that field visit they might find something that … The case might be marginal, and they're not really sure, based on the imagery and the other datasets they look at, so they go to the field and discover something that's not well-reflected in the data, whether it's something that's going on in a forest under a bunch of trees, or maybe there's some sort of groundwater or other subterranean water flow going on.

We do find our model performs less well in the cases where there was a field visit. I think that's intuitive and speaks to … I think we've developed something really powerful and potentially useful here, but there is no world in which our machine learning model replaces the job that the Army Corps is doing here entirely.

Daniel Raimi: That's really interesting. Not a job-killing computer program.

Simon Greenhill: No, we strongly think it could be a job-augmenting program, maybe, right? Help people be more productive and focus on the most controversial or the most interesting or most edge cases. But no, certainly not a replacement for anybody's job.

Daniel Raimi: This has been a fascinating discussion, Simon and Hannah. These topics are so complex and rich and thorny in many ways, and you've explained them so clearly. I really appreciate that and would love to ask you now to recommend something that you think is great. It could be related to the environment or not. We're not particularly picky on that front—just anything that's at the top of your literal or metaphorical reading stack that you think our listeners would enjoy.

Hannah Druckenmiller: I have one that's not a new book, but it is one that I recently reread after about a decade, and I loved just as much the second time, which is called The Hungry Tide. It's a fiction novel by Amitov Ghosh, and it is about a marine biologist who goes to this rural part of India, the Sundarbans, to find a rare species of dolphin. On her adventure, she meets up with two locals, a translator and a fisherman, and they go together through this very dangerous forest with crocodiles and tigers and tides that rise on a whim and with devastating consequences. It's a really beautiful novel about people trying to live in an unstable environment, and there's political tensions and interpersonal drama, and it's a lovely story about sustainability and also collapse.

Daniel Raimi: That's so interesting. We actually just hosted Ben Cahill from the Center for Strategic and International Studies to talk about liquified natural gas on a recent podcast, and he recommended a different Amitav Ghosh book, so it's in the air.

Hannah Druckenmiller: His books are wonderful. I've read most of them.

Daniel Raimi: Fantastic. How about you, Simon? What's on the top of your stack?

Simon Greenhill: Mine is a book that I read over the Christmas and New Year holidays that were just a couple months ago. It's The High Sierra: A Love Story by Kim Stanley Robinson. Kim Stanley Robinson is this celebrated science-fiction author. I guess he's probably best known to the Resources Radio audience for being the author of The Ministry for the Future. This book is nonfiction, and I think it's his only nonfiction book so far.

It's a combination of an autobiographical account of his own life and how it's reflected through different trips to the mountains that he's taken over the course of 40 years. That really resonated with me. Even though I am much younger than Kim Stanley Robinson, I've definitely felt that outdoor trips have punctuated my life, but then it's also mixed in with this amazing natural and human history of the mountain range. Everything from geology to the prehistory or early human history in the range, and then to the arrival of Europeans—people like John Muir and their efforts to conserve the range in the modern era.

Then, the final part of the book, which I really love, is almost like a guidebook. It has recommendations for trips and very opinionated ideas of what kind of gear you should and shouldn't be taking into the mountains. There's beautiful pictures throughout. For me, it was definitely a really rejuvenating read over the holidays and reconnected me with the mountain range that I really love and inspired some trips that I'm hoping to take this summer. It was a beautiful package of a book. I don't think I've ever read anything quite like it.

Daniel Raimi: Excellent. That is a fascinating recommendation, and it's actually the same recommendation that we recently got from Lisa Rennels, who also lives in your neck of the woods in California and recommended that book, as well. Definitely something to check out.

Simon Greenhill: And it’s proof that I'm very cool, because Lisa is extremely cool.

Daniel Raimi: Lisa is extremely cool; I agree. Well, you guys are also extremely cool. Your paper is extremely cool. Simon Greenhill, Hannah Druckenmiller, thank you so much for joining us today on Resources Radio.

Hannah Druckenmiller: Thanks so much for having us. It was really fun.

Simon Greenhill: Thanks, Daniel. This is great.

Daniel Raimi: You've been listening to Resources Radio, a podcast from Resources for the Future, or RFF. If you have a minute, we'd really appreciate you leaving us a rating or a comment on your podcast platform of choice. Also, feel free to send us your suggestions for future episodes.

This podcast is made possible with the generous financial support of our listeners. You can help us continue producing these kinds of discussions on the topics that you care about by making a donation to Resources for the Future online at rff.org/donate.

RFF is an independent, nonprofit research institution in Washington, DC. Our mission is to improve environmental, energy, and natural resource decisions through impartial economic research and policy engagement. The views expressed on this podcast are solely those of the podcast guests and may differ from those of RFF experts, its officers, or its directors. RFF does not take positions on specific legislative proposals.

Resources Radio is produced by Elizabeth Wason with music by me, Daniel Raimi. Join us next week for another episode.