 Our next speaker is Amy Simstead. Amy is a research ecologist with the U.S. Geological Survey Northern Prairie Wildlife Research Center. And she's actually stationed in the Wind Cave National Park in Southwest, South Dakota. Her position was created specifically to work with National Park Service Units in the Northern Great Plains, which she has done for the past 20 years. Most of her work focuses on invasive plants and climate change adaptation in grasslands where fire, drought and grazing always play a role. Her title today is exotic perennial grasses are not easily moved, but prescribed fire provides some utility. Please welcome Amy. So by the title, you can tell that it's getting towards the end of the conference and unfortunately you've probably heard a lot of what I'm going to say today, but I add in the utility part because it's a little bit of a different twist. To begin, I want to acknowledge my co-authors, Heather Baldwin who did the lion's share of the work to create the model that I'll be presenting results from today. Max Post-Vanderberg who guided us through that process of building a model. And then of course, our National Park Service, I got this microphone, don't I? Our National Park Service partners, there are too many to name for individuals. There's a lot of programs and parks involved in this project. And it's been a pleasure to work with them to synthesize vegetation monitoring data from some of those programs to be used by all of those programs in making their decisions when it comes to managing their grasslands. So as we all know, Great Plains grasslands are highly adapted to frequent fire. And beginning back in the 1970s, I believe, some of the parks in the Northern Great Plains recognized that keeping that out of the system wasn't a good idea. So they started doing prescribed fires. And then about 20, 30 years later, they realized maybe they should be seeing if those prescribed fires were having the desired results. So they established a fire effects monitoring program. And that's been going on since 1998 in 10 parks in the region. And their monitoring system is a point intercept method where they're measuring the cover of all plants that they encounter. So to the species level, and they measure this one to two years before a prescribed fire and then one, two, five and 10 growing seasons after each prescribed fire. Then in 2011, the park service stood up nationwide a monitoring program called the Status and Trends Monitoring Program. And the idea of that was to have a better idea of what's going on in terms of the natural resources, park wide, not necessarily an effectiveness monitoring program. But in one rare flash of government programs getting along and doing the logical thing, these two programs decided to use the same protocol for collecting vegetation data in the same parks. But they do, so same type of data collected but on a slightly different schedule, but still very compatible data going into the exact same database, two programs working very closely together. After about five years of this park wide monitoring had been going on, we started to look at the data and we were particularly interested in the annual brome grasses because just by our experience in the field, it sure seemed like they were becoming a bit of a problem and the graph on the left confirmed that, we see that some of the parks, there's no trend line because they've always been pretty bad since the data started being collected. Others have seen an increase in annual bromes kind of going from bad to worse, but others started out with almost none and went from having almost none to some areas having more than 50% cover of those annual species. Of course, you're asking why is she talking about annuals? We're here for perennials. So I quick pulled some data from the last 11 years of monitoring to see what was going on with the perennials and saw a similar picture, perhaps even a little bit worse in terms of their relative cover and they're increasing in just the 10 years from this data in some of the parks. But looking at this data, so we realized the issue that these parks have invasive grass problems that weren't being treated. Another second issue was that they had all of this monitoring data, especially in the fire program that was basically only being used to evaluate the effectiveness of individual burns and not really being synthesized back into improving that fire management. So to take care of or at least address two of those issues and inspired by NPAM, in 2017, we set up the annual Brom Adaptive Management Project, which is a consortium of seven park service units in kind of the western part of the Northern Great Plains. So western, South Dakota, western Nebraska, eastern Wyoming, and southeastern Montana. And the parks are obviously a pretty small part of the landscape, but they're important to those of us in this project. So the goal of ABAM, just like for NPAM, is to maintain or attain high quality native vegetation while maximizing the cost efficiency of the management actions. But we call it ABAM for annual Broms, first because NPAM took the better name, but also because we decided at the outset to focus on the annual Broms because the managers had kind of the collective knowledge that the perennials are just really hard to deal with. They wouldn't have much luck in reducing those. However, as the goals here show, it's about native prairie and we wanted to take into account the fact that if you control one invasive species like the annual Broms that are up in the front here, you might actually make it good for something else to move in or you might just flourish, which is the smooth Brom in the mid here. So our adaptive management program wanted to take into account that possibility and avoid it if possible. So structured adaptive management is a cycle of assessment, prediction, management action and monitoring. We had the monitoring programs and we weren't gonna mess those up. So we built our framework around those existing monitoring programs and we had the management programs. What we needed was a way to, like I said, synthesize the information that's coming from those and we decided to do that like N-Pam by building a decision support tool to aid in decision making. And the ABAM decision support tool is considering not just the annual Broms but all of the components of the vegetation and it combines those components into what we call vegetation conditions. And the tool itself is a Bayesian network that quantifies basically the probability of transitioning from a current vegetation condition to a new vegetation condition in various environmental contexts, both static like soil conditions and also dynamic like weather under various management actions after different periods of time. And I'll just note that we use the term vegetation conditions that of state because state has a very specific meaning especially in the range land context and ours doesn't meet that definition. So the probability of being in a vegetation condition and these conditions aren't, you're in one or the other, you have a probability of being in a variety of them and that's because their definitions overlap somewhat and those definitions are based on how much of each of these components up in the column names here are and then also we include the diversity of native species. So for example, when it comes to condition the desired condition of course is high quality prairie which is defined as having less than 5% total exotic cover, a low or a high native species richness and a good balance between native forbs and grasses. A simplified grassland has low exotic cover but it's not very diverse. Exotic annual grassland, exotic perennial grassland and weedy forb mess, you kind of get the idea there. They're defined by their offensive groups. And then low quality prairie is kind of whatever's left. It's not quite, and it's not as bad as the weedy forb mess say but it's definitely not as good as high quality prairie. And I forget what else I was gonna say for this slide because PDF made it all gibberish in my notes. So I hope I don't miss anything important. Okay, so to build our model, we started out by consulting the experts in terms of the, to make a conceptual model of how the system works. And the main idea behind this was at first we thought we were gonna do something like Kami explained and have them put numbers on transition probabilities but we realized, well, wait a minute, we have all that monitoring data, let's use that. But we did want their input on what variables we should consider to include in the model. And this is the outcome of that. I won't have you look at all of that but it includes those dynamic and static variables that I mentioned earlier. At this point, you can see that we are lumping all exotic perennial grasses together. And we realize that that's, we all know that Kentucky bluegrass and smooth bromac differently but because our emphasis was on the annual bromes, we just lump them together, but we separated out the two annual brome species, Japanese brome and cheatgrass. Okay, so we have, we had all that data. So we use that model to plug into an R package called BN learn and use the hill climbing algorithm and that to figure out how many of those arrows that we had in boxes that we had in the conceptual model are actually really supported by the data. And this was the outcome here. And I won't, again, I'm not gonna go into the details of those, but one aspect of this type of model is that it works on categories for each of these variables as opposed to continuous variables. So we had to translate those into categories and we relied on another algorithm to objectively determine where the break points and say high, medium and low cover of each of our vegetation components should be. And interestingly enough, this algorithm which is designed specifically to reduce the amount or to minimize the amount of information that you lose by going from a continuous variable to a categorical seems to pick up that systems with no cover of these either the exotic annuals or the exotic perennials function a little bit differently than those with some of them. And you can see, and I wanna emphasize that again that these aren't necessarily ecological thresholds that would just define a state condition. It's just what was picked up by the algorithm. And you can see they're kind of low compared to some of the thresholds that we might see in ecological site model, for example. Okay, so we have the structure of the environmental variables and the vegetation components. Now we gotta bring in the management. And that of course came down to speaking to the managers and what they thought belonged in their decision making. We decided very early on that the only feasible tools for managing the vegetation in the parks were prescribed fire and herbicides. And of course, the herbicides that we were working with were targeted towards the annual bromes. So not really all that relevant to the perennials because they're targeting either emerging or very young plants. And let's see, something is supposed to appear here. Oh, okay, that doesn't, we had a lot of data for the fire and that's what's relevant to the perennial grasses. So that's what I'll be talking about. But I'm showing all of these other management options because it explains why we basically lumped all fall burns and all spring burns together instead of trying to separate those out more phenologically. Basically, we couldn't have too many actions in the model. Otherwise we'd never get enough information to fill in the blanks. And another reason that it's just lumped into fall or spring burn is that our fire managers were adamant that they didn't want to be told that they should be burning at a certain leaf stage of a given plant. So I think we all know why they were complaining about that. So, all right, so then we used the data that we had from the prescribed fire program to parameterize the model in terms of what management did. And when we used that same algorithm, we found that the only arrows that showed up that the direct links from management to any of our vegetation components were to those exotic components, the exotic grasses, even not the forbs. So that kind of shows in that the native species, fire, that's what they grew up with, doesn't change them much. Okay, so where am I? Okay, so what does our model say about fires effects on perennial grasses in these parks? To answer that question, we basically just summarize the output of the model for three different starting conditions. The high quality prairie with low exotic perennial grass abundance. The low quality prairie with medium exotic component and then the exotic perennial grassland with high, according to those levels that some of you might be able to see down at the bottom. Okay, so here's the results. Again, lots of numbers that are, I'll try to walk you through in interpreting. So the number it's in all the tables that you'll see in the next few slides represent the probability of the exotic perennial grass cover being in the, I haven't been able to get the pointer thing to work. So being in the after column at a certain time step, one, two to three or four to five years growing seasons since the management actions. So the fire either in fall, spring, or not occurring. So this slide focuses purely on maintenance, which means that the before and after levels of those exotic perennial grasses is the same. And the big take home message and the reason that I changed my title to these things are stubborn is because the highest probability for any management action almost in any time step is that you're gonna stay in the same level that you started out with. But you can see that these numbers are not all 0.99 or one. There is, it isn't always gonna stay like that. So let's see if there's some hope of changing, moving things a little and which direction they go. And I'll start with the most invaded areas. So that starting out in the high come condition. Again, the maintenance levels that were in the previous slide are all shown here. But what I wanna look at with now is, is there any chance of getting to that lower levels with prescribed fire and one year or one growing season out after a fall, spring fire, or sorry, a fall prescribed fire, we do see some chances of improvement. And it's actually, you got a 50, 50 chance or better than a 50 chance of improving rather than just staying in that high level. However, that's just one year after the fire. That's the wrong arrow. Two years or two to three years out, the spring and the fall fire are looking about the same in terms of your chance of improvement. But the chances is lower than it was in that one year after. And then four to five years out, the spring fire is looking better. What happens if you're in moderately invaded areas? So starting out in the medium level. Well, the fall fire there is actually being fairly consistent in that you have a higher probability of improving than you do of getting worse, getting to that high level about twice as high, usually throughout all those time steps. So fall fire is looking pretty good for those moderately invaded areas. But in the longest term, spring fire is showing the best improvement. And notably, however, no action is showing the greatest chance of you getting in a worse condition of going up, which we've all seen before. In the least invaded condition, again, maintenance is the most likely thing. Whether or not you burn or not. But a spring fire actually has a fairly high chance of worsening your condition, or it has the highest chance of worsening your condition compared to the other doing the nothing or the fall fire. And fall fire offers the highest probability of maintaining the low invasion in all, but the shortest term. So I guess to sum it all up, in most of ecology, what does fall, or what does fire do to exotic perennial grasses in these systems? It depends. And that's pretty much the same story for the annual grasses, just in case you're wondering. The results are also fairly messy. However, generally improvement is more likely than maintaining that your state or worsening. But of course, it depends on your starting point. And in fact, burning in uninvaded areas seems to have a risk of opening up the area to invasion by the annual species. And it depends on species. So just as, you know, smooth brome and Kentucky bluegrass differ, it's interesting to see that, at least what we're initially seeing is that Japanese brome and cheatgrass are susceptible in different seasons of the year, burn seasons. So given all these complexities and contingencies, the conundrum that we heard in our last talk about helping your native species with a spring fire, or herding native species with a spring fire, but also herding your Kentucky bluegrass, what's a manager to do? Well, this is where the utility comes in part. Our decision support tool doesn't just say what's gonna happen to the vegetation. It quantifies how happy the managers will be with the change in vegetation by multiplying basically their preference for each of these different conditions by the probability of being in that condition. And when you do that, you get a single number that describes your current utility for a given management unit. What will happen, what the model projects will happen if you don't do anything. It identifies the optimal action given your conditions and what your utility of that will be. Or, and then it also compares the action or the utility of doing nothing, or that action or our fall or spring burn actions, because that's the easiest thing to think of on a management wide scale. It compares that to doing nothing. Basically, you don't have to look at these numbers to you don't have to, but the managers do look at these and they can evaluate, will we get the most utility? Will we be happiest by doing something in this management unit versus that management unit? And it's a lot easier way of making, or it makes it possible to make a decision in the face of all of this complexity. So to wrap up, as we all know, perennial exotic grasses are stubborn, but prescribed fire offers some promise of reducing them. And decision support tools can use data or a good way of using the data that we collect to help managers deal with that or complex ecological behavior. We've only actually been using the model for two years, but we have all of that back years data from the monitoring programs. And based on that, we know what we would like to do to improve the model already at this stage. First of all, of course, splitting out Kentucky bluegrass and smooth brome would be good. And then we'd also like to be getting into more detailed analyses of the fire effects. We do have severity data for most of the prescribed fires. We can look at weather before and the weather after the fires. And so just dig into it a little bit more, see if we can hone in or to tailor the prescribed fire actions to get the best results even more. And I'm done.