Mapping seafloor habitats

Air versus water

On land we are used to observing the landscape and the animals living there in high definition from even great distances. On a good day we can observe objects several kilometres away and from space the entire planet Earth is observed daily! As soon as we put our head under water that changes. A diver can, under exceptional circumstances, see about 100m through the water (Silfra, Iceland), but in the coastal waters of the Baltic sea visibility will often vary between 0.5 – 10 meters. That means that satellites have a very limited view of down into the water and that we can only observe small patches of the seafloor when we send down a diver or a camera. Water absorbs light very quickly and even more so when it contains high levels of plankton and suspended particles. The longer wavelengths (starting with red light in the visible spectrum) are easiest absorbed, while shorter wavelengths (blue light in the visible spectrum) penetrate further. Therefore, everything turns blue or green as you get deeper down. In the coastal areas of the Bothnian Bay high levels of humus substances often tint the water a brownish green, absorbing light even faster. So, if we can only observe very shallow habitats from the air, how do we go about mapping anything deeper down? Fortunately, water has got one advantage to air, the speed of sound. Sound travels more than 4 times faster in water than it does in air – around 1500 m/s (varies depending on temperature and salinity). In the same way as whales, we can send out sound at different frequencies and record the echo as it comes back to us and turn it into an image. This allow us to capture detailed information of both the water, the seabed, the sediments, and the bedrock underneath.

SGU's ship sailing on the sea, on top of topography modelled seabed.
Habitat mapping with SGUs S/V Ocean Surveyor

Using sonars to map the ocean – from the water surface into Earth’s crust

A ship like SGUs S/V Ocean Surveyor is packed with different acoustic instruments; sonars using high frequency can map at high resolution but do not reach as far, while low frequency sound are used to penetrate down in the seabed. Heavy seismic instrument can penetrate several kilometers into the seabed (for example, for fossil fuel exploration). However, at SGU we normally only use lighter types of seismic instruments (penetrating a few hundred meters into the seabed) and sediment profilers (~1m -100m penetration) to map the sediment and bedrock underneath the seafloor surface. Our workhorse for seabed scanning is a 200-400 kHz multibeam sonar which can hardly penetrate the seafloor at all, but can scan a swath of the seafloor covering about 5-10 times the water depth at ~0.1 – 1m resolution. When analyzing the signal that comes back from the sonar we can determine the depth (i.e. just like any other echosounder), but we can also determine how much sound that the seafloor reflected and other objects that reflects sound in the water column. So, in addition to depth, we can also get information about seabed substrates (and sometimes organic material that cover the seafloor, like a seagrass meadow). In the water column there are fish and some escaping natural gas which both create air bubbles that can reflect sound very well and can be captured by the signal. In summary, we move like a giant lawnmower and scan the seafloor back and forth running lines about 100m apart at 6-9 knots speed (about 10-15 km/h), scanning both the seafloor surface and the seabed below. It takes a lot of work and dedication to make a detailed map!

10 pictures of habitat mapping with SGU's S/V Ocean Surveyor. Splitbeam sonar, multibeam watercolumn data, moving vessel profiles, infauna study, gilnet fish data, multibeam backscatter and bathymetry, high resolution photomosaics and video, sub-surface sediment samples, surface sediment samples.
Many different instruments and methods work together to collect oceanographical, biological, and geophysical information in the survey area when you conduct habitat mapping (the images are from SGUs survey of Hoburgs bank in the Baltic Sea 2016-2017). Sometimes we even work around the clock, surveying at night and taking samples and photos in the day

Direct observations from the seafloor – Ground truthing 

In addition to sonar and the oceanographic instruments that we need to calibrate the sonar (measuring salinity, temperature, and sound velocity), we also take samples and images from the seafloor at selected location. This is a key part of the map making process. By using the high resolution images from the seafloor we can locate not only the most common habitats/seafloor types, but also small reef structures and other seafloor types that can cover very small areas but are important nevertheless. We use a combination of stratified random sampling design and expert interpretation to sample these locations (ensuring we do not have a bias in our selection, but still sample the full variety of seafloor types in the survey area). From samples we can get detailed grain size information about infauna (SGU are not experts on infauna or fish but we collaborate with those who are, such as SLU Aqua), and environmental condition of the sediments. From picture and video, we can capture the patchy nature of the seabed substrates, geological processes (such as sand transport), and the flora and fauna that covers the different substrate types at different locations. Together the combination of geology and biology at the seafloor make up what we typically refer to as “benthic habitats” when we have an ecological perspective of the seafloor.

Background is high resolution model from sea bottom, showing bottom shape. On top are 4 small photographs of different bottom substrates: sand ripples, rocks and boulders, soft clay bottom and compacted hard clay structure.
High resolution depth model from the Seamboth survey and images from our drop-camera that show what the seafloor look like at the different locations (these images are taken from similar locations and not exactly where the arrows point). The sonar map in the Seamboth survey cover about 200 km2, while the seafloor images we have from 200 location cover only about a total of 0.002 km2 (a staggering 100 000 times smaller area is covered by seafloor photos compared with the sonar map).

Making a map out of all of this…

So, how do we make a seafloor map? There are two main ways we go about it.

The manual way: We use expert knowledge to interpret what we see in the sonar data, sediment profilers, the samples, and the images and we draw a map by hand (though in a digital format on the computer). The advantage with this method is that the expert can use all the experience from previous work as well as the new survey to try and make the most accurate map of the geology in an area. It takes quite a lot of hard work, but in the end, you get a very useful geological map that’s been verified by expert knowledge, samples, and survey data. This is the way SGU used to always go about mapping. However, there are some drawbacks that makes us explore new options. A detailed map drawn by an expert is quite time consuming, but it is also limited by the themes that are feasible to draw this way (and we are mostly limited to a geological perspective). The maps are also subject to the knowledge of different map makers, if two equally knowledgeable experts draw the same map it will still look quite different even if the general trends are the same. Finally, if new information becomes available we need to redraw the whole map for the map to get an update.

The modelling way: By combining sonar data and information from the direct observations (i.e. camera, samples) we can also use computer algorithms to make a prediction for us. This is commonly referred to as machine learning or modelling. The underlying mechanism is the same for any kind of prediction, we have data that we use for training (i.e. our observations), and some data that we train on (i.e. sonar images or any other environmental data that capture the seafloor). “Computer vision” that can be used to identify your face in an image does a similar thing, as does the prediction for your next search in your preferred search engine. The main advantages with mapping this way is that we can focus on producing good training data and quality sonar data from our survey, then the computer does the map drawing for us. If we have poor data however, the model will also be poor. Once the modelling is set up with all ingredients we can model any themes we have data on with little extra effort, this includes geology, substrate grain sizes, environmental condition, biological cover, or any other observational data you have available on the seafloor and which relates to the environmental data you have available (i.e. depth , sonar mosaics, shape of the seafloor, oceanography…). Another advantage is that we can update our maps when new data becomes available with only moderate efforts (and we can repeat making the same map repeatedly, independent of which person is pushing the button).

Modelling seabed habitats using high resolution data. Chart picture.
Modelling the seafloor is very complicated but also quite simple at the same time. Once you have the right ingredients you can make the same soup over and over again! Just make sure you have a good master chef 😊

Sounds too good to be true? Well…. There are some things you need to be careful about when using computer algorithms to produce maps: the models only get’s as good as the data you have, and expert validation and training is very important part of the process. Also, colourful and highly detailed modelled maps can look great without being very good (this is true for all kinds of maps). Fortunately, we can avoid creating colourful but misleading maps by using test data not included in the models to get some statistics about how good our maps are and even display this in a spatial context. As a rule, when we have a very detailed survey available, our modelled maps can be produced with a high quality and resolution.

Putting the final map together in the Seamboth area

So, how do we create the maps for the Seamboth project? We combined the best of what we know about expert interpretation and modelling. This allows the experts to contribute to the modelling process and generate additional training data to improve the models in areas with lower quality data. 

3 maps. Geological interpretation shows bottom material. Surface substrate models show hard bottom coverage. Biological cover models show freshwater hydroids coverage.
Examples of the maps we are making in the Seamboth project using a combination of modelling and expert interpretation methods

We are also using expert knowledge to interpret the sediment profiles and to draw a geological map (mapping the upper 1m of the seabed substrate) using manual methods. The traditional approach allows us to make seamless maps over the border with Finland and Sweden and connect the data with our older maps made the same way. The geological map also helps the modelling of the even more detailed geological, substrate, and biological components. Due to the low visibility in Bothnian Bay, the areas mapped in the deeper areas do not have much in terms of flora and fauna attached to the seafloor, so the main product in the survey area are the detailed substrate and geological maps. We are still in the works of finalizing the process. One of the final steps is to get permission from the Swedish military to publish the full resolution of these maps, it is still pending due to the sensitive nature of underwater information.

Seamless geological maps GTK/SGU. Combined geological maps in the Haparanda/Tornio region covering 350 square kilometers.
SGU and GTK have worked together to create a unique and seamless geological map in the border region between Finland and Sweden. The Swedish side is blurred due to pending permission to publish the maps (seafloor maps are considered sensitive information for the military and needs permission to be released at higher resolution)

Written by Gustav Kågesten, the Geological Survey of Sweden

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s