Exploring Global Trends in Coral Reef Health with JMP Pro (2025-US-30MP-2478)

Hey, everybody. Thanks for tuning in. My name is Anderson Mayfield, and I'm a coral reef scientist working at Coral Reef Diagnostics. Over the next 30 or 40 minutes, I'm going to be talking to you about how I've been using JMP Pro to advance coral reef conservation and restoration, specifically by taking a more data friendly approach.

For those of you who've seen my talks in the past, you know a lot of it's driven by the plight of coral reefs. These are beautiful high biodiversity ecosystems. They're amazing in terms of their aesthetics, what you can see, the amount of species that live there, the ecosystem services they provide to us.

Unfortunately, corals do not like hot water, and because of climate change, it's now getting too hot for them to survive even in really remote areas. This is really scary because this is not something we think is going to happen in the future. It's already happening. Since the last discovery JMP discovery event, unfortunately, the planet experienced the fourth global bleaching event where virtually all reefs on the planet died.

But of course, like any ecosystem, you'll find areas that do better than others. You find reefs that for whatever reason don't bleach when they're supposed to, you find corals that are more robust or more resilient than others. Historically, we've kind of identified these reefs after the fact. But a lot of the work I've been doing the past few years is trying to use an analytics approach to identify these resilient reefs, resilient corals, and refugia, which is basically areas where corals for whatever reason are doing better than we would anticipate. That's actually what I'd be talking about today.

My background is actually in molecular biology and I do want to point you to forthcoming book chapter that I published on that I'm soon to be published on using JMP Pro to analyze large omics datasets.

Are things like genome sequencing, transcriptome profiling, things like that. I'm not going to talk much about molecular data today, but for those of you who are interested in this topic and to see how you could use JMP to make sense of these really complex data sets in a really easy way, reach out to me later, and I'll make sure you get a copy of the PDF once it comes out.

In past talks, I've been using a kind of data science approach to try to figure out where are the reefs we should care about. We still have not studied very many of the planet's coral reefs, most of them, you know, we only know by satellite imagery.

This is basically kind of an interpolation exercise. Other questions I've asked and presented on before are this idea where are these super corals, the corals that should be bleaching, but for whatever reason or not. We're going to touch on that today. Where are these refugia, so the entire reef systems that are doing better than we would expect.

Then what I'm really going to hopefully spend the last chunk of this talk on right now, for the next few minutes, is where should we plant corals? This has become a hot topic now trying to restore degraded habitats by planting new corals, but this is not a perfect science by any means. We don't have a lot of data for many coral species, but I have kind of a proof of concept I'm going to pitch using JMP that I think will really kind of expedite and optimize how we do our core restoration projects.

Then the in person Discovery Summit, I'm actually going to be talking about even further down the road. If this reef, given what we know about a particular coral reef, and we've got five or ten tools in our toolkit to maybe save it, which one actually gives us the most bang for our buck in terms of cost benefits?

If it hasn't gotten that bad, and there's a reef that we can actually protect before it is experiencing temperature stress or disease, can we climate-proof it? Can we make sure that the resident organisms are actually strong enough to withstand a future environmental catastrophe? How should we go about doing that? Those are the things I'm also going to talk about in person in Austin.

Today, I'm not going to go through this in too much detail. I just want to allude to some of the JMP tools that I'll be featuring, where I've gotten the data, much of the data are my own, such as the first data set I'm going to show you. But some of them are from public databases. I'm going to give you a really easy quick example of how I use JMP to characterize coral reefs. Then as we go down the list, kind of the analytical pipeline becomes a little bit more complex each step.

I'm going to be talking about identifying refugia from this mermaid dataset. Then I'm going to talk about this coral farming or coral planting endeavor using this ACDC dataset, which is based on a Caribbean coral. I had an amazing opportunity to go to the Sultanate Of Oman last year. It's one of these countries where people have been doing coral reef research there historically, but not really in the last 10 years or so. I didn't really know what I was going to see.

Spoiler alert, there's actually still really beautiful reefs there, some of which we didn't even know, you know, we didn't actually even know that there would be coral there. But I'm not going to dive into this dataset too much. This is really something that I made for coral biologists that I think will be wanting to tap into all these new artificial intelligence based image annotation tools.

Now you could actually do AI based image annotation and predictions in JMP using a plug in known as Torch deep learning. In this particular example, I used an open access Coral focused AI called CoralNet just because so many other people had trained it, I didn't need to reinvent the wheel. Maybe in the future, I'll consider using JMP for this.

But really what I wanted to just showcase here using this Oman dataset is how quickly I can go from taking this really ugly dataset from the AI. It's just giving me a big list of images and all you really need to know is this label code. This is what either myself or the AI thought was under each of these points. It basically puts random points on the image, and then you tell it, alright. Number 3 is water. Number 4 is water. Number 7 is a coral, and so forth, until the AI has enough confidence to start making the predictions itself.

The AI actually saves you a lot of time because it's going to take them weeks for me to actually annotate the hundreds of images I got. HC is really what I'm after here. That's the hard coral, the stony corals that build the reefs. That's going to be kind of our main ecological benchmark.

I've made this kind of bewildering looking workflow here because my memory is bad. This is a great way to not just remember what you did in your analytical pipeline. But one thing I really like, you've always been able to do this in scripting, you know, you could leave yourself little notes within the script that explain why you're doing what you're doing. But you could also do it as these little pop up windows in the workflow.

If I want to share this workflow with a student or somebody who has no idea, and I don't have time to babysit them or baby-step them through each step of the process, I can make all these little notes that say exactly what I'm doing at each step.

Right now I'm basically taking all those text data, converting them into series of zeros and ones, because it's basically presence absence data. Ultimately, I'm doing is way over here, I'm getting a count of how much of my image was actually the seafloor, that's the benthos, how much was actually coral. It's getting closer to what I actually want to know. Converting this to a percentage.

If you've got newer versions of JMP, you've got this great feature to look at the distributions, you know, without even calling the distribution menu. You've also got some summary statistics. Forty percent may not mean a lot to a non-coral person, but 40% coral cover, meaning 40% of the sea floor is covered by corals, actually really good. Doesn't mean this reef is safe from climate change, but it means that at the present time, there's still a lot of corals in Oman.

Then I keep going through this, I'm not going to explore this dataset in too much detail because it's not one of the most interesting, but I can do things like plot the data on a map, see how coral cover varies over space and time.

I can then plot these data on oops. I can then put them on my website by publishing to JMP public, which I'm not going to do now in the interest of time, but I've already actually got an example here. This is actually looking at either the coral cover or the algal cover. These are two of the most important things we measure as coral scientists.

I've got my sites here. You might notice one cool thing that is kind of a hack I got off the JMP community is if you want us to have your Tukey's post hoc groups plotted on the bars, there's not a native way to do it within graph builder. But within this workflow, there's actually a step by step protocol that shows you how to do this, which I know a lot of biologists will really enjoy.

Let's just see. We'll pretend like I published it to JMP public. I believe that's basically the last step of this workflow. Let's just see. Yeah, and then I basically, you know, I could do things like combine this with these data and make a nice dashboard to share with my colleagues if they want more detail than just this plot.

Simple data set, not really much to explore other than there seems to be a lot of corals in Oman, and we see some variation across the sites. But if anything, this is more of a plug for workflow builder and to show people that you can take these kind of dense, esoteric datasets you get from open source AIs and interpret it and hit the ground running really quickly using JMP.

I want to spend more time on this next demo that I've kind of alluded to. This is actually getting back to this idea of trying to find the refugia. The reefs that, you know, all the other ones around them are bleaching, but for whatever reason, they're not.

This is unfortunately a photo I took last year in Egypt, they would be the opposite example. This reef is so far gone, I don't even think it can recover. The black you see is actually a cyanobacteria that's overgrowing all the coral, which means that they basically have no chance to recover.

But on the other end of the spectrum, going back to Oman, people kept telling me, oh, all the coral there is dead, you know, you're not going to enjoy your dives, it's going to be depressing. But actually, you can still see really impressive coral in places where we didn't expect to find any.

Usually it's a doom and gloom situation. We're used to seeing this now as coral biologists, these kinds of images, but you know, there are these kinds of needle in a haystack reefs out there that are demonstrating resilience. The goal now is to see if we can figure out strategically where they might be, especially the ones that we may never have encountered before.

The Wildlife Conservation Society has been not working on this topic per se, but they've been doing coral reef monitoring for a number of years in Tanzania, primarily in Fiji. What I've done is I've gone to this website, it's called Mermaid.

Unfortunately, of these 12,000 surveys, probably about 1,000 of them are actually open access. I don't know why you would go through the effort of putting your data up on this website and not making it publicly available, but I think some governments don't like to share data.

Anyway, so what I've done is I took about 900 coral bleaching surveys out of this 12,000. These are some of the ones that are actually open access. Then I said, Okay, what are some of the things we could measure that would actually maybe be a useful predictor of bleaching? This could be things like oceanographic parameters, temperature, salinity could be things like what the reef looks like, the properties of the reef, what's living there.

But in this exercise, we're basically looking at the percent of the reef that was bleached or dead. We want to minimize this. In some analyzes, actually, the main analyzes I'm going to show, I actually convert this into a binomial. Is the reef resilient and can withstand future heat waves? Or is it not resilient? Because most of the richest data set is from Kenya and Tanzania, I'm basically going to be focusing on those data.

Pull up the data table here. Again, there's going to be a lot of workflow steps that kind of go on behind the scenes in the interest of time. But in the workflow, it will tell you what's going on. This is really what I want to kind of belabor for a minute. This is actually kind of my hack or my tip of this entire module. The goal is to figure out where the refugia are. But in terms of the JMPsmanship, I think this is really cool because what I'm about to show you.

We do see some green up here. These are the reefs that are not bleaching as much. These are the refugia, and we've got some refugia down here. Maybe this isn't the best example because we have kind of a latitudinal variation. It's pretty obvious to see the winners and the losers.

But say you've got a bunch of green mixed in here, or maybe this is a good example, better example here, where we've got some green coral that's doing okay. Then there's some reefs here that are bleaching a little bit more. There's a few different ways you can do this. I can actually go in here with the lasso, or I could just simply click on the individual hexagons. I've actually already done this in the workflow. I'm going to deselect these in a second.

But the cool thing about this is that say I'm happy with this selection, and I want to call these reefs the resilient ones because remember, okay, I could just use the raw bleaching percentage, but I really want to know my definition of resilient are the reefs that are bleaching, that are not bleaching when others around them are.

I can go into this subsetted data table now that I've created, and it shows me these 34 rows of the resilient reefs I've selected. A cool tip is, while these are still selected, I can actually go into the row selection into this option name selection and column, and I can call this. The selected ones will be the resilient ones and the unselected ones will be no. Now I actually have a column property based on this, and this is actually what I'm going to be modeling in a second.

Just to do one more thing before I proceed with that, The Wildlife Conservation Society who curated these data and did these surveys, for some reason didn't measure temperature, which is insane. Thankfully, the temperature data was readily available on a public database managed by NOAA. All I want to do right now is just quickly kind of put these side by side, just to make sure that the corals in green were actually expected to be bleaching.

I think it's going to look better if I do side by side. Let me just drag, drag this over here, clean this up a little bit. In the temperature plots on the left, temperature is the left y-axis, that's these box plots. This red you'll notice is the DHW, which is not a term non-coral people normally use. It stands for degree heating weeks, and it's a measure of the thermal stress exposure.

These bleaching data I didn't mention are from 2024. You could see 2024 is when we had that global coral bleaching event. Fifteen degree heating weeks is really a dangerous number. We're usually starting to see bleaching above about five degree heating weeks. What that means is these corals, all of them were basically expected to bleach. That means the ones that showed low levels of bleaching are indeed that is indeed something interesting. They're tough, something about them is tough.

Let's go back into the workflow and try to dive into this resilience in a little bit more detail. Here what I've done in the predictor screen, I've just tried to see which of these potential predictors might be driving these differences in coral resilience as a binomial, either resilient or not. Latitude being number one is not that surprising given that we see this kind of, I won't say a gradient, but we do see distinct latitudinal differences here in coral resilience.

The longitudinal ones are less evident here to me, but this is still corroborating what we see on the map. Let's take this one more step of detail. I'm actually going to run a model in the background while I'm talking because I think it actually takes some time. But let's look at some of these univariate impacts while this model is running. Let me just move these out of the way. Estimated compliance, what does that mean? That means, was the reef in a marine protected area or a park, and was it being managed well? Low means basically, it wasn't being managed well, somewhat in the middle, full. Yes. Although this was not statistically significant… I actually think I've pulled it up here. Apologies for that.

Even though this effective compliance is not statistically significant, we do see about 55% bleaching in the low compliance areas and 45-ish percent in the ones that have better compliance. This is actually interesting because 55% versus 45% bleaching is a meaningful difference to a coral scientist like me. What's driving much more of the difference is the type of reef. Reefs in the lagoon, which tend to be more sheltered, they're bleaching, four times less.

Right there, this is probably going to be our number one predictor. This is interesting. Project includes GFCR. The Global Fund for Coral Reefs was a UN initiative where they're basically raising fundraising. They asked scientists to pick 50 resilient reefs.

I suspected that scientists just picked the reefs that they wanted to continue studying because they knew they could get money to continue studying these reefs. I thought it'd be funny to actually test this as a predictor. Are the "resilient" reefs that scientists nominated more resilient than an unpopular reef that nobody nominated. Sadly, you could see twofold higher bleaching in the resilient reefs that scientists chose. This is a shame on you, but I can't say if I'd been in that position, maybe I would have been tempted as well.

Silver lining though, is this means that for whatever reason, the reefs that scientists are studying are actually doing about two-fold worse than "unpopular" reef. Maybe that means, although we know things are definitely dire, maybe we're getting a little bit more of a pessimistic view of things just by virtue of the fact that we're studying reefs that tend to be more compromised. Probably because these are reefs that are easier to access, closer to population centers, which probably have more stressors. Anyway, that was a cool little side thing, a finding that came out of this analysis. This is interesting.

Wildlife Conservation Society and their surveys estimated how many stress-tolerant corals were on each reef. You would predict that reefs with less bleaching or more resilience would have a higher percentage of stress-tolerant corals. That is indeed what you see. It is a statistically significant relationship here with stress-tolerant corals on the x-axis, but it only explains about 6% of the variation. Many other things are accounting for this difference in reef resilience. It would make it a lot easier if it was all just simply a matter of how many stress-tolerant corals were on that reef. But unfortunately, it's not the case. There's a lot more going on in the background. Let's try to figure that out in a little bit more detail.

Let's go to the subsetted table where I've got mostly, I think it's almost exclusively Tanzania data here, a little bit of Kenya data. Because I'm interested in doing a binomial, so I could look at percent bleaching, but I really want to know is that reef resilient or not? Binomial, I'm going to put in these predictors. I'm leaving out a few because there's too much missing data.

I'm also going to include the different types of corals living on that reef. I'm going to include some other ecological benchmarks, put in my validation column. Then I'm actually going to switch this to generalized regression because generalized regression is going to allow me to fit a binomial distribution where I've defined my target level here. If it runs similarly to what I've seen before, it's actually a pretty good model. We can dive into it a little bit more using the profiler.

What I always like to do before I even start exploring, this was a tip some of you may not know, is, I want to see which variables had the most weight in the model. I'm going to do this independent resampled inputs. What that's going to do, it's actually going to reorganize my predictors in the order of impact. You can see here, this order matches this left to right order. Right now, you'll see the probability of resilience is showing me the conditions for basically zero, which I want the opposite.

I'm going to go in and set my desirability to maximize the probability of a reef being resilient. Then I'm going to maximize it here. Actually, another thing we need to do… This is always interesting. Sometimes I'll turn the extrapolation warning on, because it otherwise might be giving me conditions that would never actually exist either in this dataset or potentially even in nature in general. What I can do here is just to make sure, turn on the extrapolation control, re-maximize the desirability. It didn't change my probability, but you will notice that some of these conditions shifted. Some of them make sense based on what we explored earlier.

The lagoonal reefs were bleaching significantly less, those where there was good MPA compliance were tending to do a little bit better. The exposure is co-varying with the lagoon, so we don't read too much into that. But some of these things will be interesting for planning, basically.

What I can now do, which I'm not going to do in the interest of time, although I might actually have it in the workflow is… Have I done it? I've actually got another data table that's much bigger, like this one here. It's instead of 100 rows, this has 55,000 rows from these countries. I can go in there, use that model, or I could even use the data filter simply. I could be like, look, show me the reefs that match the conditions shown in my desirability analysis.

I think I actually pulled up a subset for this demo. It looks like maybe not. But anyway, I was able to go in, use this model, and then go into this larger dataset. I found some reefs in Indonesia that actually had a lot of the same properties, which would be the ones I would want to go have my colleagues test and see if they're indeed REEfoluti like the ones that we were able to find in Tanzania.

My last demo that I've already mentioned earlier is this idea of planting corals, putting on the farmer's hat. I love this idea of the seed packet where it's telling you the conditions in which to plant your corals. I actually have a demo on that. It's called ACDC best conditions. It's named after Acropora cervicornis data coordination hub. This is a NOAA program in which they basically coalesced all the data for this one species, which has become the main target species for Caribbean coral reef restoration.

It's probably the only species where you could do the analysis I'm about to show you. The rest of the species, there's just too little data out there. But there's another example you could work through with the workflow called best conditions. That's the simulation where I'm saying, I have this particular genotype, where should I plant it? That's a cool exercise. I'm actually going to show you the opposite side of the coin.

I'm working in this nursery, which corals should I plant in my nursery? Think about it as you're essentially a farmer going shopping for corals. There's about 200 different genotypes of Acropora cervicornis. You obviously want to do this in a strategic way. These corals are endangered. You don't want to go and plant corals that have no chance of surviving. That's going to be irresponsible. Let me first close out some of these windows. I'm just going to roll back. Here is my little simulation I alluded to earlier, where I'm trying to find REEfoluti in Indonesia. It was there. Closing the mermaid dataset.

Now we're going to open this ACDC dataset. 28,000 rows, so it's fairly large. These are the genotypes, so there's over 230. These are the different nurseries. Not a lot of data here, a lot of missing data for temperature pH. The planting month, that's going to be something we want to look at. All I want to show you right now is RY, so what we're actually trying to optimize. It's TLE which means total linear extension. These corals are basically growing in a three-dimensional way. That's why we basically have this unique metric for growth.

Let's just have a quick look at the distribution, highly skewed. This is a relatively fast-growing coral. But you could see the overall mean is 30 cm. It's not super important, just to give you some context for what we're going to be showing in a second. Let me just pull up this workflow. In this simulation, we're pretending that my farm is called the South Nursery. Pretty sure it's in Florida.

Other than that, I don't really know much about it. Before we do anything, we can go in here and basically… Let's just subset. Just show some basic properties of the South Nursery. I chose it because it's the one with the most data. What I can do now, just going to subset here. Right now there's about 23 different genotypes in that nursery. It has a static depth of 6 m, that's going to be important to know. The temperature ranges between 25 and 31. Corals have been planted most months there. Those are going to be our main predictors amongst maybe some others.

Let's kick off this workflow. It's actually going to do quite a few steps. Sorry for bouncing everywhere. But basically what we're going to do is we're going to make a model that will maximize coral growth irrespective of genotype and site. This might sound like a weird exercise. If I'm only interested in this South Nursery, why would I not first subset into only looking at the South Nursery and then running the model? It's because I want to see if there's genotypes that are not already in that nursery that would make sense to plant there. This model is based on the whole 28,000 row dataset. But that means there's going to be caveats. This is what I think is a really cool thing you can do. There's probably other ways to do this that are easier. But let's just have a whirl.

Right now, I'm going to see this amazing RSquared value. Not going to get too excited yet because remember, there's going to be some constraints here because these might correspond to conditions that are not in the South Nursery. I have a look. I have a validation column that I made behind the scenes. Still getting a pretty good fit here based on the RSquared, but really, we're not going to have anything useful until we get into this profiler. Again, I'm going to do the independent resampled inputs to assess the variable importance.

We've got about, let's see, five that we want to focus on maybe even four. This doesn't really matter. But sometimes I like to clean up my appearance here, where I can really zoom in on the most important variables. I haven't done any desirability analysis yet, but let's go through that. Should remember that I want to maximize total linear extension or growth. This number is about to JMP to something really high, 558, which you remember the average is about 30. This is really good. But we need to consider the fact that Mote Tree Nursery is not an option. A cool thing you can do, I don't actually even need to go back into the data table. I can find the South Nursery here, option click on a Mac, and lock this factor setting.

For temperature, it's going to be a little bit different, but I want to make sure that I'm not being given temperatures that are impossible for this wreath. 25, I know if you remember, it's between the range we see at the South Nursery, but just in case, I'm going to constrain this to the actual temperatures we observe. Not going to lock it, but I am just going to give it this range. Ideally, it's not going to give me values outside of this range when I rerun it. You can already see since I've made those two changes, the number dropped down to something a little bit more reasonable. I think it pretty much already maximized it. This is cool because this is actually an answer that makes sense.

Plant genotype ML5 in September. September is probably going to be warmer than this, but I forgot one thing. Remember that nursery kit is only at 6 m depth. I don't think this is going to affect it a lot because depth was not one of the more influential factors. Let's see if this drops much when we constrain the depth. It didn't really affect it too much. Basically, we want to plant genotype ML5 in September, and its growth will be predicted to be 111. If you remember… Actually, I'm just going to show you. Let's go back to the workflow. This is the average growth of the corals in the South Nursery.

It's only 20 or actually less than 20. Was the genotype it told me to plant already in the South Nursery? That would be interesting. It's actually not. It's saying that should introduce a genotype that's not every day there. Remember the one it wants me to plant is called ML5. It's not in this list. Let's see what I've done here. What I've shown here is the variation in growth rates across the genotypes. Remember, if we plant genotype ML5 in the South Nursery in September, we're predicting the total linear extension should be about 111. Is that good for ML5? We know that's good for the South Nursery, but let's actually look here. This has already been subset.

Let's go back into the main data table. Maybe it's here. ML5. Actually on average in other nurseries, that coral normally grows at a total linear extension of 200. That's almost two times higher. That's interesting because it basically, if it was introduced to the South Nursery, it would far and away be the best genotype compared to all the others. But compared to its performance elsewhere, it's not doing that well. What does this mean? It basically means there's something wrong with the south coral nursery.

I did this simulation a few different times, so you see slightly different results here. ML5 would be the best genotype to plant there. It would do better than the other corals at that nursery. But suffice to say, the corals in this nursery are not growing very well. Yes, you've optimized something, you've made a strategic choice, but maybe a better question you would have at this point is why are the corals doing so poorly in this nursery? That might be actually be what you want to do next.

Just to give you a recap, I know I've flown by a bunch of different things. All the slides, all the datasets, all the workflows have been posted on my JMP discovery page where you'll hopefully be viewing this video, so you should be able to download whatever you want, follow along. I've got the Oman example where I'm just showing you some very rudimentary re-characterization analyzes in JMP as well as a pipeline that takes you from an AI based image annotation server.

We have this REEfoluti demo we did using data primarily from Kenya and Tanzania, where we found some funny things like scientists were not good at nominating reefs based on their resilience. But then we've learned some things that were actually some good news, management's working to some extent, and sheltered reefs in the lagoons tend to be the ones that are doing better. A lot of useful data there, could get it all into a nice dashboard, share with your colleagues, put it on a website.

Then this final example using the Acropora cervicornis data coordination hub database, we're basically trying to use a predictive modeling framework with a desirability analysis to help us optimize where we would plant our corals and which genotypes to plant in our nursery. I think a lot of potential there for people working on coral restoration and probably even for agriculture. You know, I stole most of these ideas from people working in agriculture. I'm sure the agricultural industry is way ahead of us and using this data driven framework for optimizing crop yield that we should be adopting in the coral field.

Hopefully, if you guys are able to catch my in-person talk in October in Austin, I'm going to be talking about coral like I always do, but instead of looking at REEfoluti and where the strong corals are and where to plant corals, I'm actually going to be trying to optimize the methods we use to save a reef.

You've got this amount of money, this amount of time, and these parameters are in your reef looks like this. Should you create a marine park? Should you be restoring it? Should you be doing mitigation? What's going to give you the most conservation or restoration bang for your buck? This is something I've been working on for a while, and I'm excited to share that with you guys in Austin. Otherwise, I'll wrap this up now, and thank you so much for your attention.