If you’ve read through our “about section” or seen our blog in the past, you’ll know that kelp is threatened by climate change. This is perhaps the main reason why we are so interested in kelp in the first place, and so far Floating Forests has allowed us to build some of the most complete kelp forest datasets around. Soon we will be adding a new set of images to the project, this time examining a few key, high-risk locations with a fine-toothed comb. We’re interested in how coastal construction or development might impact kelp forests – it’s time to take our kelp research to the city! We will be taking a close look at kelp forests in California, Chile, Argentina, Australia, and New Zealand.
While global climate change is extremely important, there is great value in examining smaller scales as well. In fact, studies such as this one have shown that kelp forests are strongly affected by local conditions! As you might expect, changes to these local environmental conditions can have significant effects on the local wildlife.
We call changes that affect the environment “drivers”. For example, a construction project could “drive” environmental change by releasing sand and dirt (which are collectively referred to as “sediments”) into the water. When two different drivers interact and affect each other, we refer to them as “synergistic”. Occasionally they can create especially favorable conditions, but often one exacerbates the negative effects of the other, so we call these “synergistic stressors”.
One way to think about this is to say that one driver can amplify the effects of another. For example: a marine ecosystem might be able to handle a 2° Celsius increase in average temperature without collapsing. Separately, it also might be able to handle the effects of a large construction project. However, the effects of both at the same time could prove catastrophic: organisms already stressed by warming might not be able to cope with disturbances caused by the construction, and most importantly, things might get bad much more quickly than we might expect.
Because an emoji is worth a thousand words:
This example is one of the many that play out across an ever-developing world. One consistent trend over the last several decades has been a large increase in coastal population. To say this more plainly – more people than ever before live near the coast. As coastal cities grow, constant expansion is needed in order to keep up with the population. This conversion of land from natural to urban terrain is known as “urbanization” and brings with it a host of environmental impacts – some of which are potentially harmful for kelp forests.
By digging into the satellite record, we can get an idea of how kelp forests in urban areas have changed over the last few decades in response to this trend of development. What’s really exciting about this opportunity is that with the power of citizen science we can cover a lot of ground: we want to compare cities across the globe in order to build a better understanding of how kelp in different places is affected by human activities. We expect there to be many interesting differences between these locations – it would probably be more surprising to learn that kelp in Australia was acting exactly like kelp in Chile!
Stay tuned as we prep for the launch of this exciting phase of Floating Forests – details about our study sites and more about the connections between urbanization and kelp coming soon!
With the Falklands classifications wrapping up, it’s time to move on to our next phase of global kelp mapping – kelp on the edge! Giant kelp is a cold water species – warm nutrient poor water is a definite no. Their range extends towards the equator wherever they are found until they hit a wall of warm water. There they shall stop, and no further.
But what about when that wall – that range edge – starts heating up? That’s when you’ve got…
We’ve witnessed a variety of other kelps die back when things got to hot, and our own Jorge Assis has shown some projected major range shifts of kelps in the future.
With the data from Floating Forests, one question we want to ask is, how have kelps on the edge been faring? Over the past 35 years, have we seen kelps on warm water range edges dwindling? Have any of those populations blinked out? Have the ranges of giant kelp actually been on the move?
The great thing about this project is the simplicity of the question. Rather than circling kelp beds (although we’ll get there), we want to begin just by looking at range edges of giant kelp around the planet and the area up to 500km away and ask you just to note, do you see any kelp? Yes or no? That’s it!
The first set of images is up – from Baja California using Landsat 7 and 8. L4 and L5 imagery will come in the next few days, and more of Baja and then New Zealand next week.
So what are you waiting for! Pull out your phones and get swiping! (or click here.
Also… the music we can’t get out of our heads
There’s somethin’ wrong with the kelp today
I don’t know what it is
Something’s you’ll see with our eyes
We’re classifyin’ things in a different way
With swipe fast left or right
The data will surprise
Kelp’s livin’ on the edge!
Hey, all! You have asked us multiple times for a zoom interface. We have some reasons not to use a zoom tool – tradeoffs with speed, variability in accuracy between classifiers, etc. – but are intrigued and want to put our intuition to the test. If it’s better, we’ll switch to it full time, most likely! If the tradeoff isn’t worth it, we’ll know, and we’ll probably have an interesting paper come out of it.
But – we want to hear from you! In addition to the zoom interface itself (basically, the classify interface + a pan and zoom tool), are there questions you think we should be asking you after you complete a zoom classification that could yield new and interesting insights? Let us know over here!
I’ve been playing around with some of the incoming classifications and thought I’d share this little map as a preview of what is possible with the data created by all of you! Click the map to launch an interactive map of the Falkland Islands! Use the menu in the top right to toggle each season on/off, and try zooming in to see individual patches!
The green patches represent the aggregates of everyone’s classifications – what you are looking at here is all of our data from the Falkland Islands. All years are lumped together for now as there are still some gaps in the time series (we’re almost there, just 16% to go!). In other words, these patches represent maximum patch size for each season. Another technical note: these polygons are based off a 6 user consensus, our current candidate for best accuracy.
Giant kelp growth patterns are often dominated by seasonal processes such as storms. This can lead to an annual ebb and flow in which kelp patches grow rapidly throughout the spring and summer before dying back (or getting wiped out) in the winter.
We can see this pattern in our data too! Note how the annual minimum kelp extent occurs in the austral winter (June, July, and August), with growth beginning in the austral spring (September, October, November) and increasing throughout the summer.
One great thing about Floating Forests is that I am constantly amazed with how much we can learn even with a quick look at the data. Check out the waypoint marked “Tyssen Patch”. This is actually an undersea shoal (note the color difference on the basemap), and notice how much kelp is there! Here is the google maps link if you want to prove it to yourself – definitely #sokelpy!
Once we get the time series data online I am very interested to see how persistent this patch has been over the years. Stay tuned!
Often when citizen scientists view an image, they want context. Where is this? Am I really seeing kelp, or is this sand or mudflats? Fortunately, we have you covered. In the video below, I show you how you can view the metadata about each individual image, including how to view the area pictures in Google Maps. Now, the map on Google Maps isn’t going to be from the same time as the Landsat image, so, there may or may not be kelp in the same places. But you can at least get a better high resolution view of the area to make decisions about your classifications if you want it.
One question that has come up a few times with our consensus classifications is, does the level really matter when it comes to looking at change in kelp forests over the long-term?
While our data isn’t quite up to looking at large-scale timeseries yet (we’re still digging through a few thorny methodological issues), I grabbed the complete dataset we have for the Landsat scene around Los Angeles from work Isaac is doing and decided to take a look. Note, this is totally raw and I haven’t done any quality control checking on it, but the lessons are so clear I wanted to share them with you, our citizen scientists.
After aggregating to quarterly data and then selecting the scene that had the highest kelp abundance for that quarter (i.e., probably the fewest clouds obscuring the view), we can see a few things. First, yeah, 1 classifier along is never good.
Note, I haven’t transformed the data to area covered, instead we’re just going with number of pixels (1 pixel = 30 x 30m). But, wow, we need consensus!
But what if we impose a more conservative filter? Say, a 4-10 user agreement threshold? What does that show us?
What I find remarkable about this image is that while we see the effect of decreases in detection when more and more citizen scientists have to agree on a pixel, the trends remain the same. This means that while we will try to chose the best threshold that will give us the closest true estimate of kelp area, that there will be multiple intermediate thresholds that give us the same qualitative results to any future attempts at asking questions of this data set.
This is a huge relief. It means that, as long as we stay within a window where we are comfortable with consensus values, this data set is going to be able to tell us a lot about kelp around the world. It means that citizen science with consensus classifications is robust to even some of the choices we’re going to have to make as we move forward with this data.
It means you all have done some amazing work here! And we can be incredibly confident in how it will help us learn more about the future of our world’s Floating Forests!
In working on a recent submission for a renewal grant to the NASA Citizen Science program, I whipped up a quick script that takes the data posted and overlays it with the actual image. I really like the results, so here’s one. Feel free to grab the script, data, and play along at home!
Some of you have noted in earlier posts of this preliminary dataset that some classifications show up on land – particularly at low thresholds. This is likely due to some images being served up that, shouldn’t have been (we’ve fixed this in the new pipeline), and the zeal of some classifiers. Regardless, we can crop out those areas, as we know that there’s no real kelp there. But do to it, we need some very very good maps of the coastline. Fortunately, there’s a solution!
The Global Self-consistent, Hierarchical, High-resolution Geography Database is an incredible resource, with some coastline data files that are remarkable in their detail. The data is also, of course, huge. So, for anyone playing along at home, we’ve subsetting it down to a few files for you delectation. These are all in the common ESRI Shapefile format, but if folks want them otherwise, we’re happy to provide. Here’s what we’ve created for you. Click on the names of the areas below to download the zip files.
And last, the absolutely stunning Falkland Islands
When we used them for coastal cropping, they worked great – we’ll show some timeseries with cropped data next week!
At Floating Forests, we get confronted with two issues a lot. First, how good is the data? A lot of scientists are still skeptical of citizen science (wrongly!) Second, many citizen scientists worry that they need to achieve a level of accuracy that would require near-pixel levels of zooming. We approach both of these issues with consensus classification – namely, that every image is seen by 15 people as soon as it’s noted that it has kelp in it. We can then build up a picture of kelp forests where, for each pixel, we note how many users select it as, indeed having kelp. You can read an initial entry about this here.
So, how does this pan out in the data I posted a few days back? Let’s explore, and I’ll link to code in our github repo for any who want to play along at home if you’re using R – I’d love to see things generated in other platforms!
I love this, as you can see both at the 1 threshold at least one person was super generous in selecting kelp. However, at the 10 threshold (unclear why we’re maxing at 10 here – likely nothing was higher!) super picky classifiers end up conflicting with each other so it looks like there’s no kelp here.
How does this play out over the entire coast? Let’s take a look with some animations. First, the big kahuna – the whole coast! Here are all of the classifications from 2008-2012 overlapping (I’ll cover timeseries another time).
I love this, because you can really see how crazy things are at a single user, but then they tamp down very fast. You can also see, given that we have the whole coastline, how, well, coastal kelp is! Relative to the entire state, the polygons are not very large. It’s a bit hard to see.
Let’s look at the coastline north of San Francisco Bay, from Tomales to roughly Point Arena.
Definitely clearer. You can also see we accidentally served up some lake photos, and folk probably circled plankton blooms. Oops! Now the question becomes, what is the right upper threshold? Time (and some ongoing analysis which is suggesting somewhere between 6-8) will tell!
If you have ideas for other visualizations you want to see, queries for the data that you want us to make, or more, let us know in the comments! If you want to see some other visualizations we’ve been whipping up, see here!
The day has come – we’re finalizing our data pipeline to return data to you, our citizen scientists! It’s been a twisty road, and we’re still tweaking, but we’ve begun to build some usable products for your delecation and exploration!
We want to know more from **you** about what you want and what is interesting for you to explore, so, today, I’m going to post some demo data for you to look at and give us feedback and comments on. This is a data file from our California project that consists of polygons for each kelp forest at different levels of user agreement on whether pixels are kelp or not. So, first, here’s the file in three formats (depending on what you want) (we can also add more if asked for)
You can do a lot with these in whatever GIS software is your preference, and if anyone has examples, we’d love to post them! For now, here’s a quick and dirty visualization of the whole shebang at the 6 users agreeing on a pixel per threshold (source.
Neat, huh? You can even see where something in one image was confusing (no kelp on land!) which now I’m *very* curious about.
So, what’s in this dataset? There’s a lot, but here are things most relevant to you
threshold – the number of users who agree on the pixels in a given polygon are kelp
zooniverse_id – the subject (i.e., tile) id of a given image, if you want to just look at a single image, subset to that id
scene – Individual Landsat “images” are called scenes. So, every subject that we served to users was carved out of a scene. You can look at a whole scene by subsetting on this column. For more about what a scene name means, see here
classification_count – number of users who looked at a given subject
image_url – to pull up the subject as seen on Floating Forests
scene_timestamp – when was an image taken by the satellite?
activated_at – when did we post this to Floating Forests?
There’s a lot of other info regarding subject corner geospatial locations. We might or might not trim this out in future versions, although for now it helps us locate missing data and see what has actually been sampled.
So, take a gander, enjoy, and if you have any comments, fire them off to us! This is just a sample, and there’s more to come!