Oral argument — Alaska Wildlife Alliance v. USFWS

Good morning. May it please the Court, Bridget Pissarianos for Appellant's Alaska Wildlife Alliance. I'd like to reserve five minutes for rebuttal. The Marine Mammal Protection Act doesn't just prohibit killing, but harming and even harassing marine mammals, unless several narrow exceptions are met. Congress's intent was to keep populations at healthy and sustainable levels. The southern Beaufort Sea population of polar bears isn't there. It may be wiped out in the next 25 years. Because this population is on the brink, the service's scientists have acknowledged that the death of even one bear or cub from oil and gas activities may not be allowable under the MMPA. It's against this backdrop that we have this ITR. The service's model showed that the oil industry's activities here would cause a significant amount of take, and the Alliance doesn't challenge those models or the service's technical findings. But the problem is that the agency didn't grapple with its own findings. It subdivided them in ways that violate the statute and aren't entitled to deference. And any of the violations in the Alliance's brief is an independent basis for finding the ITR model. First, the service found a 75% chance of level A take happening each year. And it didn't dispute this in the ITR. But the service said the chance of level A take happening was too low to consider. So it didn't analyze the impact of that serious take on the population or consider ways to mitigate it. It avoided the question entirely by splitting this into a 29% chance of non-serious level A take and a 46% chance of serious level A take. And the service has never once used these categories before, even in its decades of administering the MMPA for polar bears and issuing ITRs. But what prevented it from doing that? Well, this violated the statute because Congress already defined take in the MMPA. And it established two categories of harassment, level A, which is the potential to injure, and level B, which is the potential to disturb. And Section 1371 allows the service to permit take if certain conditions are met. But the plain language of the statute doesn't let the service split injuries that caught or split potential injuries into different level A categories in a way that I don't understand why it doesn't allow it to do it. But I understood your point to be a little bit different. I'm going to refer to what they did with the serious and non-serious as sort of A1 and A2. They didn't call it that, but I'm going to call it that. Congress called A and B. And then service split level A into two more subcategories. And it seems to me that your argument is that level A is both 1 plus 2, and you have to add the 46 plus the 29, and that gives you 75 percent. Okay. That seems to me to be a very reasonable argument, and I'm going to have lots of questions for Fish and Wildlife when we get there. Now, what I want to know is, is there any overlap between those two categories, or is A2, the non-serious, subsumed into A1, in which case adding them would not be correct? The service specifically said that there is no overlap there, and that's on page 124 of volume 2 of the excerpts of record. You can find it in the first column. They explicitly explain that once a den suffered a level A take, that den was removed from the model such that it couldn't be exposed to another level A take again. So those categories are wholly independent of each other. The service didn't dispute that in the ITR or below, but it was raised on appeal for the first time. So what's the remedy if we've got this A1 and A2? It seems to me intuitively that the remedy isn't in the court of appeals to figure that out, but maybe you can tell me what do you think is the remedy if we determine that the splitting of it could potentially, you know, get you to too big of a number if you add them up? Sure. The vacanter is the default remedy under the APA. We do recognize here that there are some questions about ongoing oil and gas activities and the need to protect polar bears during their remand, such that it might be appropriate to remand this to district court to consider different kinds of remedies consistent with this court's opinion and to further develop the facts of that. I just want to go back to why what they did is wrong. It seems that you have two points here. One is that it violates the statute, and one is that maybe they didn't consider the implications of this as much as they should. So starting with the first one, why do you think it violates the statute? It violates the statute because the MMPA generally prohibits the service from authorizing take unless these certain conditions are met that Congress set out, and that's the overarching moratorium at Section 1371. And the service can't rationally make the findings that are required under Section 1371 unless it assesses how much and what type of take may occur. And so for purposes of assessing that, Congress specifically defined harassment and take for purposes of that exception at Section 1362. Right, but I don't think they're saying this is something other than level A take. What they're saying is it's important to look within that and to characterize the different types of risks within that. I guess I don't see why that's not a more precise way of doing it as opposed to just monolithically calling something all level A on the assumption that it might all be the same. Well, two points there, Your Honor. The first is that by doing it this way, the service basically entirely dodged the question of whether injuries were expected to occur, and it didn't consider how to mitigate impacts from injuries to polar bears because whether it's a serious injury or a non-serious injury, it's still an injury as Congress defined it, and the service needs to assess it on that basis. I'm not sure they disagree that it's an injury in that sense. It just seems that they're discounting it as not as bad of an injury. They are, but that is still not permissible, and I think what Your Honor is referring to is some of the negligible impact findings, and we don't dispute that they could consider different types and severities of injury for purposes of assessing whether those injuries would pose a negligible impact on the population or something more, but it doesn't change the fact that at the end of the day, Congress said this is how we're defining level A, this is how you have to consider it for purposes of seeing if these exceptions are met. And I can turn next to the negligible impact piece of this, but even assuming that the service can split up level A for its negligible impact findings, the agency still found a 46% chance of that serious level A lethal take happening every year, and it didn't consider the total probability over five years. And this is the part that the service wants you to think is a complex math problem, but it's actually not. This is just a legal question of what reasonably likely and total of such taking mean. And to touch on the first problem, the service unlawfully raised the evidentiary bar for its negligible impact finding because under the service's regulations, an impact is negligible if, quote, it is not reasonably likely to adversely affect annual rates of recruitment or survival. And the reasonably likely standard doesn't mean more likely than not. We cite some cases on page 36 of our opening briefs, and the defendants don't argue otherwise here, that it doesn't require a preponderance of evidence standard. Your view is that more likely than not is the preponderance of the evidence standard, and it's generally 50% plus. Yes. And that reasonably likely is something less than 50%. Yes. Greater than 10%, less than 50%. Yes. The Supreme Court has said in considering that standard in other contexts that it means more than remote, outlandish, or simply hypothetical. And the problem here is that the service did apply this greater than 50% standard because it's undisputed that it found there was a 46% chance of a death or deadly injury to a cub every year that this five-year regulation was in place. But they've also said we're focusing more on the median because we think that's a more accurate figure. And that seems like a pretty technical judgment. So how do you respond to that? That's actually not a technical judgment. That's another way of restating that they didn't have to consider something that was less than 50%. And here's why. The median is just the middle number in a set of outputs when you arrange them in order of size. And so the service ran its model 10,000 times, and 54 out of every 100 times the model spit out a zero. But then the other 46 out of 100 times the model found there would be one or potentially more cub deaths. So the median is zero, and zero was the most likely outcome. But the problem here is that this median doesn't actually change the fact that the service was actually just saying we don't have to consider anything unless the probability is 50% or greater. Yeah, I mean, there's also other things they considered beyond the model, though, right? The models were computer-generated models, but there's lots of other parts of the record beyond just this. There are, but the problem with the service pointing to those here is that the service made it clear in the record that it was discounting the 46% of model runs based on the not more likely than not standard, and it also explicitly said that it was basing its take estimates on that model. And it says that at page 107, where it says its take estimates are being made on the model. And I'd also point the court to comment responses 30 and 33 in the ITR, and those are on page 140 of volume 2 of our excerpts of record, where the service, again, is tying its negligible impact determination to the fact that 46% isn't reasonably likely, and the median is zero, and zero is the most likely outcome. So it was actually using these quantitative numbers for its decision. And then the second point, too, is that the model actually incorporated the things that the service is pointing to as somehow determinative of the outcome. For instance, the service explained that its model took into account the footprint of the oil industry's activities. It simulated dens on the landscape. The model accounted for there being three aerial surveys for detected dens and assumed that detected dens would have a mile buffer. So the model actually incorporated these very things that the service points to as part of its justifications. Those were already baked in at the outset, and they don't really explain how those qualitative things that they point to here would have changed that percentage in any way. And the second problem with the service's negligible impact determination is that the service can only allow take where, quote, the total of such taking would have a negligible impact on the population. But here again, the service split up its findings, and it claimed that it could lawfully just look at annual take in isolation, not the total. And that is, again, at comment response 30 on page 140 of volume 2 of the excerpt of record, where the service says that the negligible impacts are best considered in annual terms. If the service is allowed to give an ITR of five years, but it doesn't have to, it could do it on a yearly basis, if they were to have broken up the ITR into one-year segments, then their estimates are still good, right, if they are only looking at one-year segments. Yes, if this had only been a one-year ITR. But this was a five-year activity. That's right. And that's what changes the calculus. Exactly, yes. Because what you wanted to, as I understand, what your expert did was said, okay, it's 46, we'll take your number on the serious, A1, 46 percent, but if it's 46 percent every year for five years, what is the chance of some take during that five-year period? And that's what he calculated at about 94 percent. That's correct, yes. The service didn't argue that that 94 percent was wrong. They simply said they didn't have to look at it, because they could look at the 46.2 percent chance each year over five years. But, again, that's not the total. That's each year in isolation, and no one disputes that the oil industry applied for a five-year take permit, and that that is what Fish and Wildlife Service granted. So that's what it had to look at. And the service can't point to this mention of annual rates of recruitment and survival and its regulations as a basis for ignoring the MMPA's clear requirement that it's the total of such taking that has to be considered for the negligible impacts. So you interpret total to mean you have to add it up over the five years? Yes, or at least do something. They couldn't just look at each annual year and call that the total, because that was just one year when the total was five. So, yes, here we think they should have, because they had these numbers, look at that total probability. They weren't free to disregard it. Has any court addressed this issue? Not that I'm aware of. This is new. The service just came out with this model in 2020. It's in our record. So this probability model hasn't really come forth yet, but the idea that courts need to look at the total take is not new. That hasn't come up specifically in a numerical context, especially in these ITRs for polar bears, because historically they've been a lot of qualitative assessments. But the service does recognize that the statute requires it look at the total take. But here, when you look at their comment responses, including on page 125 of Volume 2 of our excerpt of record, where it's the sum of harassment of all sources, the service says that the sum is this 46.2 number, which Table 8 on page 124 makes clear that number is actually the annual number. Could you also comment on this issue of the same bear versus different bears? Yes. The service's model specifically assumed that each take, for purposes of its small numbers finding, would be to a different bear. And just to be clear, the small numbers argument is tied to Level B harassment, not this A2 or serious Level A take. So that's Level B harassment. And the problem there is that the ITR authorizes take of 443 individual bears by Level B harassment, and that is nearly half the population of 907 bears. And no one argues that take of half the population is small. And this court explained in Center for Biological Diversity v. Salazar that that small numbers requirement is a distinct check on the service's ability to authorize take under Section 1371. So splitting up the total into smaller chunks and ultimately allowing take of half the population would just render the small numbers check meaningless, and that's contrary to what Congress intended in the MMPA. Unless the court has further questions, I'll reserve the remainder of my time. Okay. Thank you. May it please the Court. Rebecca Jaffe, appearing on behalf of the Fish and Wildlife Service. I'm going to take 15 minutes today, and the Intervenor Alaska Oil and Gas Association is going to take five. For decades, the service has issued incidental take regulations for polar bears in the southern Beaufort Sea area for oil and gas exploration, development, and production activities. This rule covers similar activities that past rules covered, albeit with requirements for far more extensive mitigation measures. And like past rules, this rule complies with the MMPA. The MMPA mandates that the service shall allow incidental taking of polar bears so long as certain requirements are met. I'll focus here today on the small numbers requirement and the negligible impact requirement. Starting with the small numbers requirement, the service reasonably assessed small numbers on an annual basis. Plaintiffs are adding up the small numbers over all five years of the rule, but the service reasonably looked at it annually. And there's a couple reasons why that's reasonable. The first is that the phrase, total of such taking in the statute, is in the negligible impact phrase, not the small numbers part of the statute. Congress said small numbers in the statute, but it did not say over what time period. It did not define the term small numbers at all. And in the legislative history, it even recognized that small numbers is an imprecise phrase. The annual approach evaluating small numbers on an annual basis is consistent with the Fish and Wildlife Service's past practice and current practice, and it's also consistent with the practice of the National Marine Fisheries Service. And the annual approach makes sense because there's some changes in the population every year. Overall, the population has been roughly consistent, but some bears are born and some bears die every year, because there's also some variability in industry activities every year. The service referenced these annual metrics in 2ER138, and it referenced the variability in industry activities in several places, but 2ER78 is one example of that. As to Judge McKeon's question, the model is a new model, and it does have some limitations, which the service acknowledged. One of those limitations is that the service could not reliably calculate how many takes would accrue to the same animals. So it conservatively assumed each take would be of a different bear. And the model indicated there would be a maximum of 92 takes per year, which is 10.14% of the population. And the service reasonably concluded that's a small number. That's what the service authorized, but it also did a qualitative analysis and recognized it had to look at the history of take under, for the past 30 years, there have been nine previous incidental take regulations. It also looked at the geographic scope of industry activities compared to the range of the polar bears. And the model had limitations, which the service had to acknowledge. Turning to the negligible impact finding, which is separate from the small numbers finding, I do want to clarify, to be very clear, the categories, as Judge Bybee called them, of A1 and A2, the service used those only for the negligible impact analysis and for no other purpose. The statute says that the service must ensure that the taking authorized under a rule will have a negligible impact. But it doesn't define negligible impact. The regulation defining negligible impact says that it defines negligible impact as whether any impacts are reasonably likely to affect the survival and recruitment of the population. Level A and B harassment are concepts in the statute, but the statute does not mandate that the service use those concepts specifically as part of the negligible impact inquiry, and neither does the regulation. Okay. I don't think there's any inherent problem in the service deciding to subdivide level A. My problem comes when you calculate A1 and A2 separately and then fail to add them together because A1 plus A2 is what A equals. Once you subdivide them, level A is still what Congress said requires certain kinds of actions or calculations by the service. So we can't – it seems to me that when you only look at the 46 percent of A1, you've completely ignored A2, but A2 is a part of A. So I don't understand what the service has done here. Your Honor, I beg to differ for two reasons. The first is that those are – that's just one piece. That's one data point that the model produced. It also produced a median output. It produced a median – Just keep me – just keep with me on my logic. Is there something wrong with my logic that A1 plus A2 equals A? Yes, Your Honor. Are those overlapping categories? So are we double counting? Is that the problem? There could be double counting. Plaintiffs note that the model assumed if there's one particular den that has take in the early denning period, for example, then it wouldn't have take later on. That is true, but the model was looking at take across the whole population. It wasn't – Right, but when you subdivide – I mean, Fish and Wildlife was not obligated to do this subdivision, but once it decided to do that and it gave us separate percentages for A1 and A2, and both of those are part of A, I don't understand why you don't have to add those two percentages together. Now, if you're going to tell me that there's some overlap there, then I need the service to explain to me how much overlap there is, because if A2 is completely subsumed in A1, then it's irrelevant. But if it's not, then it seems to me you've got to add them together. Your Honor, I do think the service was obligated to define them, because the question under the negligible impact inquiry is, will the take affect the population's survival and recruitment? And the science indicated that the A1 impacts, i.e., lethal take or injury that could cause mortality, the service said, those maybe could affect survival and recruitment of the population. A2 impacts were the uncertain fitness cost. For example, a cub may be a little bit less mobile. And the service said there's no data indicating that those A2 impacts can affect survival or recruitment. So it would have been arbitrary for it to discount those scientific differences. And it explained those scientific differences between the A1 and A2 categories, although it was using different names, at 2ER148 and 2ER146. I mean, it seems to me it's not that different than having said – I mean, this is a sophisticated model, right? So the model is now very detailed. But one could have just said we're estimating a 75% probability of Level A take. However, within that Level A take, there's considerable variation of what that means, and we need to probe deeper and explain why not all of that is the same in doing our negligible impact. Here you basically just up front broke it into two and analyzed it that way. But I don't see why you couldn't have just described it the way I'm describing and essentially the same thing. I suppose I do want to flag that the 75% figure is plaintiff's figure. I get that. But even on those terms, right, you could have done some addition. You didn't even have to have A1 and A2. You could have just said A and then had a further analysis that said, let's now explain what this A number is really composed of. And, in fact, in the service's view, it seems that there's some different things within A that are worth considering. Yes, Your Honor. Alternatively, the service could have said we're calling Category 1 happy and Category 2 sad. I mean, it could have used any names whatsoever because it doesn't have to use the Level A and Level B names. It has to question whether the particular take will impact the survival or recruitment of the species. It almost feels like you've taken that A2 category and sort of downgraded it to a Level B. It just feels like you felt like you're going to second-guess Congress on this one and that the things that the service has under its model still defined as Level A categories somehow should not count. Again, I beg to differ, Your Honor. First of all, the service did separately look at Level B harassment in its negligible impact inquiry. It said Level B would likely accrue to the mother bears or the female bears. But there are specific places where Congress said in the MMPA, you have to use Level A and Level B. For example, in the Authorizations for Bonafide Scientific Research, that's at 16 U.S.C. 1374 C.3.C. It said there's different requirements for bonafide scientific research depending on whether Level A or Level B is predicted. Also, in terms of permissions under the MMPA for photography, it said that could only be Level B. And that's at 16 U.S.C. 1374 C.6. So I don't think that the service was in any way— Okay, but—I'm sorry, go ahead. No, I mean, what's confusing here is that you did decide to divide up Level A. And you're now telling us, if I understand it, A1 is could affect. Is that right? And A2 is uncertain. But that A2 being uncertain doesn't discount any of the percentage. I mean, you can't just, like, wipe it out of the calculation, can you? No, Your Honor. I may have misspoken. The A2 category is uncertain fitness cost. There's no—to the Cubs— there's no data indicating that that would have any impact on survival or recruitment. The service wasn't saying it's uncertain whether this might affect survival or recruitment. It was saying there is some fitness cost. But there's a big difference between having a fitness—you know, impairing fitness in some way and killing a Cub. Right, but then how do you reconcile the two groups? That's the question that we're having. Once you lay them out there. Level A is injury. Level B is disturb. And both of those are harassment. Level A and Level B are both harassment. Level A is injury. You've described—the service subdivided those into serious injury and non-serious injury. But they're both categories of injury. And that's what Congress said was Level A. I don't see how you get to ignore non-serious injury as though it's not an injury. It may not be serious, but it's still an injury, and that's what Congress defied. Congress—the service wasn't ignoring it. It was saying, we have to do something specific for the negligible impact inquiry. We have to say, will these potential—will the taking affect the survival and recruitment of the population? That's a specific inquiry. And the service knows there are some injuries that may affect survival and recruitment, and there are some that won't. It would have been arbitrary to disregard that science indicating the differences between different types of injuries. When you're doing a negligible impact assessment, do you need to use—does the service need to use Level A and Level B? Not at all, Your Honor. It can use any categories—or it can use whatever the science takes it to, I think. Right. That's why I wonder whether some of the confusion here is just simply over labeling, because it seems that the service could have done exactly as you've described it and not called it Level A. You could have just said, well, we've done this analysis and modeling, and we're seeing some injuries that are more serious and some injuries that are more speculative, and discount the speculative ones, focus on the serious ones, and make a judgment as to whether that's negligible impact. That has nothing to do with whether you're calling it Level A or Level B or anything else. Yes, Your Honor, exactly. It could have called the A1 category potential survival impact, and it could have called the A2 category small injury, no potential survival impact. It could have used any names at all. Why did it use Level A here? What was the thinking behind that, if you know? I do not, Your Honor. So I want to follow up on Judge Bress's question, because I think it's a very important question. So what is Level A and Level B relevant to that? It's defined in 1362. Is it not operational in 1371? Congress did not require the service to use the Level A and Level B categories in the negligible impact inquiry. Did it use the term injuries or disturb elsewhere in 1371? They're defined in 1362. Are they used elsewhere in 1371? That's a pretty long, pretty complex statute, so I have a statute in front of me. Not to my knowledge, not elsewhere in 1371, but as I mentioned earlier, Congress did use those terms specifically for the requirements for issuing bona fide scientific research authorizations and photography authorizations. So Congress knows how to say, I want you to use Level A and Level B, and it did that with specific types of authorizations, but it did not with the authorizations under 1362. It has used Level A and Level B as two different forms of harassment, and the term harassment certainly appears in 1371. Yes, Your Honor. So it suggests that those are relevant to that, a taking of harassment or small numbers. Yes, Your Honor. And the service did do a Level A. I mean, it only authorized Level B harassments with the rule, and the small numbers are not. If Level A and Level B have sort of no correlative in 1371, why did the service use them at all? I don't know, Your Honor. But it was making different categories because that's where the science took it. It had to use the best available science. But it certainly seems confusing for us. We are non-experts. If you were to take something that's defined in 1362 and use it to help calculate something in 1371 but then tell us that it's ultimately irrelevant, they didn't have to do it. Yes, Your Honor. That makes the path of the agency's reasoning harder to follow. It doesn't make it arbitrary and capricious necessarily, but it does make it a whole lot harder to follow. Yes, Your Honor. I think there's a lot of things looking at. Okay. I would like to turn to the five-year question as to whether this is the question as to whether even taking your numbers of 46.2% on the A1, the serious injury, and whether you had to calculate that for a five-year term, which their expert says comes out to about 94%. So we do agree that the service has to look at the total of such taking over the five-year term. But we disagree that the service has to use the probability output specifically. Plaintiffs prefer that output, but the service had a variety of data points in front of it and it reasonably focused on the median output. And the total of such taking per the median output was zero over five years. It was zero each year and it was zero over five years. Well, yeah, but that's because you calculated it at less than 50% once you subdivided it. So you subdivided it, come up with 46%, said, okay, so the probability in any given year is likely to be zero because it's only 46% possibility. But if you have 46% each year over five years, it seems to me that during that five-year period, there's a pretty good chance that at least one year there's likely to be a taking. And that doesn't seem to me to be rocket science to calculate. Your Honor, per the probability output, I mean, I do want to note that the 94% calculation is plaintiff's calculation, not the service's. But in any event, the model produced different outputs. And aside from the model's outputs, the service also did a qualitative analysis. It has 30 years of history of oil and gas activities in this area. And it was looking at a host of information, not just the model. Within the model's outputs, it focused on the median, which, as I said, was zero. It was the median result. It was the most likely result. But that's only for one year. That was the median per year, right? Yes, Your Honor. And the model calculates each year independently because there's different industry activities each year. And this was all relevant to the question of negligible impact. Yes, Your Honor. Where the statute does use the word total. Yes, and we don't dispute that. But if your number is zero, the total every year, then the total over five years is zero. Right. It's a little bit of a fallacy of composition, though. If you keep subdividing things, then we may end up with, you know, what is the chance of taking a bear in the next second, you know, approaches zero. But if we continue to add those over the course of five years, it changes things. But if we look at any individual second during that five-year period, now there's no chance we're going to take a bear. Your Honor, I beg to differ. The service, I mean, looking at each year individually for the model made sense for a variety of reasons. First of all, there's some variability in industry activities each year, and it had to account for that. Second of all, denning season happens once a year. The service wasn't subdividing things by the day or the second. It was looking at each year, and the result each year was zero, and it focused on that. I see that I'm quite over my time. Let me see if there are further questions for you. I do have one last question. If we were, if, and I'll say it's a big if, if we were, if the panel were to find that we thought that there was something that was incomplete, maybe arbitrary and capricious, maybe not, but at least incomplete, is vacature the appropriate remedy, or is there something less than vacature required here? Something less than vacature is absolutely required here, Your Honor. And what would that be? Is that a remand from us to the district court to order something from Fish and Wildlife or to the district court to take additional evidence, or can we just send this straight back to Fish and Wildlife and say, run the numbers on these things? You can, you could remand to the Fish and Wildlife for further explanation. I think if the court were considering vacature, then we would respectfully request that instead it remand to the district court for further briefing and argument about remedy so that the parties can present evidence about the disruptive consequences of vacature. But if the court is disinclined to vacate and wants more explanation, for example, about the 46 percent issue, then you can remand for further explanation. It would be highly disruptive to industry to vacate, and the rule provides important conservation benefits. Thank you. Thank you. Okay, we'll hear from Mr. Steen now. Thank you, Your Honors, for still allowing me some time. I appreciate it. Well, the benefit of being the intervener and going last is I can see how the issues have been delineated, so I'm just going to jump right in unless you have any questions you want to start with. But I'm going to start with this Level A issue, which is distinct. It goes to negligible impact. It does not go to the small numbers issue, which we haven't really talked that much about. I'm happy to answer questions about small numbers, but we've addressed it in the briefing. On Level A, I think one of the first things to understand is that we're talking a lot about probabilities. We're focusing on a single metric of probabilities. And the plaintiffs have said, well, yeah, that probability makes sense because negligible impact refers to reasonably likely. The definition actually doesn't refer to take being reasonably likely. It says an impact resulting from the specified activity cannot be reasonably expected to and is not reasonably likely to adversely affect the species or stock through annual rates of recruitment or survival. So it's the impact that has to be reasonably likely. The service has to first determine what the impact is. This exercise of running the model and looking at all the various products of the model and looking at the qualitative information is so that the service can get a feel of what's the impact. And then it has to decide is that impact reasonably likely to affect annual rates of recruitment or survival. And so it's not that take has to be reasonably likely. It's that the service has to determine the impact. Here they said we're going to determine the impact as being these types of take. As Your Honor said, level A, A1, and A2, and then there's level B. They didn't have to do any of that for the negligible impact determination, but they did. But I think that's a really important point, that service determining the impact, the standard that applies to that is not is it reasonably likely. It's did they act rationally? Did they explain this? Were they transparent? Did they disclose it? Did they explain what they did? And I would submit they did. The plaintiffs disagree. They think they should have focused on one metric, and only one metric, and that metric said, oh, it's probable, and so you had to conclude there's level A take. It's not what the service did, though. It ran the model, but I think with the model, the first thing we have to start with is the idiom, all models are wrong, some are useful. This model was definitely wrong, and the service acknowledged that in the sense that it dramatically overestimated the probabilities of take. So they explained that to ER-124. They say very transparently we evaluated the most impactful scenario, and then they gave many examples of how that would be. Like, for example, seismic survey, which under the model draws the greatest impact. They said we know they're not going to occur all five years, but we're going to assume they occur every five years for purposes of running the model. So boom, that gets plugged into the model. That affects the result. They said we have this really important road that a lot of trucks go on, and the way the model works is it just assumes if there's somebody in a facility, just somebody there, it's occupied, and a truck goes by the facility, and there could be polar bears in this area, just because it's occupied or just because a vehicle's there, it assumes a likelihood of take immediately. It doesn't assume what's going on, and so it said, okay, we have this road. We could have trucks going back and forth, could cause take, but there's another route that wouldn't cause take, but we're going to assume they use the road that would always cause take. Boom, that gets plugged into the model. And so at every point, they used the most conservative assumption in the model. They say they use the 99 percentile of probability of impact in the model. And so you have a model that's going to produce these likelihoods, okay, that's not reflective of what's really happening. And they acknowledge that, and that is one of the qualitative factors they considered. They said, we recognize that the model is the most impactful scenario, and that's one of the things we're considering when we're going to ultimately conclude what kind of impact is going to occur. So what is the relationship between the level A and B categorizations and the negligible impact determination? The relevance of the negligible impact is, one, they don't have to split them up, but it's really – you have to go back to the definition is, does the impact have an adverse effect on annual rates of recruitment or survival? So level A – This is out of the – now you're talking about the reg. Right. Well, that's how they've defined it in the regulation. But in terms of, in practice, how they're using it, the reason they look at level A and level B for purposes of the negligible impact determination is they have different effects on a bear. So, like, level A – level B harassment, for example, is behavioral harassment. The bear might run away from an activity. That's behavioral harassment. It's not removed from the population. That's it. Level A harassment, potential for injury, could be a spectrum of things, and that's why the service divided it up into serious and non-serious. You could have non-serious injury that does not remove a bear from the population. It doesn't create the risk of removing a bear from the population. And so the service categorized it that way, I assume, because that has a different effect on the annual rate of recruitment or survival. Serious level A is something that's more serious. It could result in the death. It could result in removal of a bear from the population. And so they considered that differently because that could have a different rate on annual rates of recruitment and survival. Okay, and I guess the way I was conceiving of this is that these – level A and level B are sort of categorizations that exist elsewhere that can be useful for examining negligible impact, except potentially when they're not useful, in which case you might make an adjustment. My understanding was that was what they had done here. It's not necessarily laid out in the most clear way, but that is how I was reading what the service was explaining in their decision. Yeah, I mean, it's one way that you could do it. It's the way the service chose here. Ultimately, they have to determine whether the total take has a negligible impact, and they could have done that by lumping all take together – level A, level B – saying, here's the total take. Here's why we think it has a negligible impact. And they could have explained that. They could have said different types of the take have different types. Is this A1, A2 methodology something they've done before, or is this a maiden voyage on this approach? In my experience, I have not seen this before. The question was asked earlier, is serious injury something that is in the statute? It's not anywhere else in Section 101A5. It is – it does appear elsewhere in the Marine Mammal Protection Act. So, for example, it applies to some of the provisions governing commercial fisheries. So looking at the types of impacts from fisheries, do you have something that's a serious injury, which the National Marine Fisheries Service has defined as something that's likely to result in the death. So it's not an uncommon term in the MMPA, but it was – I haven't seen it used in this context before. Can you address the five years versus one year before you sit down? On the negligible impact analysis? On the – not the small numbers, but on … Not the small numbers. So getting back to my point about the models, they use the model. They have to look at the results of the model. But the model produced a mean, it produced a median, it produced a probability, it produced a confidence interval. The plaintiffs hone in on one of those. The service said – and they explained it. This is rational basis review. Did they explain it? They did explain it. They said, we think when the data are significantly skewed, like the data we have, we, the service, think the median is the most informative. And that's all they're required to do. That's the agency's explanation. The plaintiffs disagree with that, but that's the agency's explanation. They also did consider qualitative factors, recognizing that we've always done this qualitatively, so we can't ignore qualitative factors. The qualitative factor of the model being overly conservative, we have to take that into account. The qualitative factor of we didn't consider all mitigation measures in the model. Plaintiffs' counsel is correct. They did consider the beneficial effects of flying surveys to detect polar bear dens in the model. They didn't consider any other mitigation measures in the model, though. And they recognized at 2ER140, 2ER145, that there are beneficial effects of mitigation they didn't take into account. They also recognized that there's never been an observed level A harassment in the history of this program, decades of polar bears. And then that was something else that was relative. It was the service as the expert can take all of this information and make a conclusion. The plaintiffs might disagree. The court might disagree. But the question is, was that ultimately rational? And I would submit they explained it. I think we have your argument, so thank you very much for your presentation this morning. And we'll go ahead and hear rebuttal. Thank you. There are three points I'd like to hit on in rebuttal. The first is that the service's failure to consider level A take holistically is not harmless error, as the defendants seem to imply. The fundamental problem with the way the service broke out level A take in the way that it did is that, again, they didn't assess the impacts of level A take because they said it was so unlikely to occur. And they talk about this on page 124 of volume 2 of the excerpt of record. And as a result, they only look at the impacts of level B, which is this potential to disturb on the population. So while the federal defendants have said it served no other purpose but informing the negligible impact determination, the record's clear that it did, in fact, inform the service's determination that no level A take was expected either annually or over the entire course of the five years. And that played a role in the agency's ultimate findings. And the service explains several times in the ITR that it is not considering impacts of level A take on the population, specifically on volume 2 excerpt of record, pages 102, 125, and page 143, comment response 49. And the reason this matters is that even assuming this non-serious level A harassment doesn't impact rates of survival because the service defined it to be an injury that doesn't cause death, this court said in Natural Resources Defense Council v. Pritzker that even negligible impacts still have to be mitigated to the greatest extent possible. This is a fundamental precondition to the service allowing any take at all. So on remand, the service would have to consider whether it can issue an ITR if there is projected level A take from these activities. And the crux of that is it would force the service to assess how to mitigate those impacts and ensure the least practical adverse impact. I would also note that, yes, Congress did use the word serious injury elsewhere in the MMPA, but as the alliance explained in its brief, when Congress uses words in one section of a statute but omits them in others, it shows that Congress knew how to make that specification, and including it one place and excluding it elsewhere should be given meaning by the courts. I'd also just briefly touch on the fact of the five years being zero. Again, a roughly coin-toss chance of there being a death of a cub every year over five years increases the probability greatly, and this is not technical, and I think Judge Bybee was asking some questions to get to that, but I think a reasonable person realizing that there's a 46% chance of death or deadly injury from an activity knows that that probability increases substantially when that activity is repeated five times in a row. For instance, if I had a 46% chance of dying on my way to work every day, I would be a lot more nervous going to work five days a week than I would one. And to touch on the point about the model being overly conservative, the service did note that the model was conservative, as Mr. Steen pointed out, and that's on page 124. But the service also repeatedly said, well, it also explained that the reason the model was conservative was because it was using the information that industry provided about the activities it was going to do. So conservative or not, it's based on the reality of the information that AOGA provided as to what it was going to be doing over five years. And the service also repeatedly stated in comment responses that it was not overestimating the potential for take, and that's in volume two, excerpt of record pages 141 to 142 in response to comment 41, pages 148 to 49 in response to comment 83, and page 154 in response to comment 129. And regarding the comment that these oil and gas activities have gone on for decades without any problems for polar bears, that ignores that the service developed this quantitative model because it recognized that its prior findings and past reported take had actually underestimated take of denning cubs, and that new information showed impacts to denning bears were more significant than the service previously thought. And that's explained at volume two of our excerpt of record, page 139, pages 141 through 142, page 251, and page 257. And regarding level A take, as we pointed the court to earlier, at page 261 a study by the Service of Scientists does question whether any level A take of this population could be permitted, consistent with the statute. So this is a really significant question that the service would need to consider on remand as to whether it could allow even level B harassment for these activities where level A take is essentially certain to occur. And unless the court has any further questions, I'll sit down. Okay. Well, thank you to all counsel for the briefing and argument in this case. This matter is submitted, and we'll stand in recess until tomorrow morning. All rise.