Turbid oceans better understood with satellites

Within Edito Model Lab, partners work on tools and data on ocean knowledge with the aim of making them available and accessible to a wider audience. Felix Dols, researcher at Deltares, is working with a team on this digital platform. By combining tools and digital technology, insights into turbid ocean water can be obtained faster and more accurately.

Learning from the Wadden

Felix: "Besides applications of existing tools and models from Deltares, in Edito Model Lab we are also working on a development for using Artificial Intelligence (AI) in estimating the turbidity of the Wadden Sea. This AI algorithm has already been applied for conditions of the ocean. We chose to look at the Wadden Sea, because of the high degree of difficulty in estimating and predicting turbidity. Parts of the Wadden Sea alternately flood and then fall dry again due to the tidal cycle. Consequently, there are more complex dynamics than further out into the ocean. We trained a mathematical AI algorithm, 4DVarNet from our French partner IMT Atlantique, using data generated by Delft3D FM, an existing numerical model from Deltares. With the trained model, we filled gaps in satellite data."

The problem of turbidity

"Turbidity is the opposite of light transmission," Felix explains. "Light is also important in water for photosynthesis. The less light, the less algae grow and that affects the food supply for other organisms. Therefor light is essential for all life underwater. At Deltares, we have models that can simulate growth of organisms such as shellfish, algae and seagrass based on water conditions. The amount of light is an important indicator of how well an organism will be able to survive and reproduce.

Until now, in our ecological modelling studies, we used a simplified turbidity data - seasonal averages, based on what we already know. We used very limited measured data from incidental measurements, meaning measured at a single location, at a single time. As a result, we know that we are making (too) rough assumptions about the change in turbidity over time. It also does not yet tell us much about the whole sea.

A lot changes over the seasons and in the day and night cycles, so we wanted to have a continuous data stream. In recent years, colleagues did extensive research with numerical models on turbidity in the Wadden Sea for the years 2016 and 2017. Years of collecting measurements and modelling, yielded a lot of useful data.

Faster insight with satellites

"For faster, and in some cases more accurate, predictions of turbidity fields, we saw an opportunity to derive accurate data from satellites " says Felix. "The challenge now is that the satellite enters orbit and takes a photo capture at most once a day. There are also gaps in satellite data due to cloud cover. So with that, we don't yet get continuity in the data. Something we want precisely in order to also understand the influence of tidal currents and the day and night cycle."

"AI can be used to fill these gaps in the measurement series. This goes much better if the output data from a simulation model such as Delft3D FM is used." Explains Felix.

"The AI model is then trained on both the satellite data and the numerical simulation model. So we trained the AI model with correlations with which we can fill the missing data and thus fill a turbidity field. We can now create a turbidity map for each year in which the right satellite data is available, so we now have a time series of the past decades, instead of just 2016 and 2017. This data can be used by biologists and ecologists!"

Project leader Edito Model Lab at Deltares Lőrinc Mészáros adds a surprising learning point: "Incorporating the physical knowledge from our numerical model, Delft3D FM, helped us achieve better machine learning results. This is a very important point because it shows complementarity, rather than competition between numerical and data-driven tools. It also positions Deltares as a key player in the future, even for companies who are trying to do solve the same problems but with machine learning."

With this model and data, we can establish relationships between human activities, such as dredging, and the effects on turbidity. It is useful to monitor turbidity, and artificial intelligence helps us do that more accurately and completely.

Felix Dols

Lessons from algorithms and rivers

"By applying it, we learned how this algorithm works and how satellites can extract useful information for us. From this, we have derived how accurate our model is. With this, we now have a new reference that we can use to improve our numerical models. With that, we deliver new and improved products. For example, the previous simple models also did not contain additional information from storms.

Furthermore, we have seen that around the mouth of the Ems River, that discharges into the Wadden Sea, we had difficulty making good predictions because we did not have the river data as training data fed into the algorithm. A next step is that for the Rhine delta - where we have more information on turbidity in the river and mixing of river discharge and sea - we use this data additionally to the turbidity data in the coastal zone. This is a big improvement!"

"So, there is always the next step to start investigating. Our French partner used the models mainly for parameters in the ocean. In coastal areas, there is often more at play at the same time, there are more dynamics, so you need to know a lot about the system. Like where how the rivers mix in the coastal zone and how exactly does water retreat during falling tide. We first want to better understand and include the influence of rivers," says Felix.

Our collaboration with Deltares has enriched our understanding of marine and coastal systems, particularly in the Wadden Sea. At IMT Atlantique, we aim to use our machine learning skills to enhance the awareness of ecosystem dynamics. With the collaboration, we address complex environmental challenges more effectively.

Nga Nguyen, Researcher Artificial Intelligence, IMT Atlantique

Benefits for the food chain

"The results are qualitatively better than what we have had so far, which is important for modelers who want to say something about the food chain that starts in the water at the so-called "primary production" or algae. Algae growth is highly dependent on turbidity and affects the amount of food for fish, mammals and birds.

At a detailed level, Deltares can now complete simulations for 20-30 years. Here, the numerical model helped well in making the connections. "We noticed that AI is a reinforcement of the numerical models, and the application is in line with each other. Now we work with satellites, but we build on the numerical models. Expert modelers such as Thijs van Kessel, among others, have helped make this work possible."

We increasingly understand the impact our activities have on the sea and on the food chain and nature. This helps us to be able to make responsible choices in which activities to undertake and which not to. We can thus protect and conserve nature more

Felix Dols

Hackaton

In this impact year, large-scale sharing of EDITO Model Lab results is on the agenda. In June, there is the United Nation Ocean Conference in Nice. Then the platform with these and many other tools will be presented to a larger audience. Expect live demonstrations in a control room, training courses for policymakers and the launch of a hackathon for technical users.

What the future holds

This model can be used wherever decent satellite data is available. "I expect barriers to be removed in international and interdisciplinary cooperation because data and tools with knowledge about oceans will be pooled and made available in one place. A great basis to be able to solve problems together. Taking this course, could also have a big impact for Deltares too. Dare to share not only through our own media, but also through the Edito platform!" adds Felix.

Artificial intelligence has emerged as a pivotal catalyst for transformative advancements in tackling environmental challenges. The Deltares approach, combining advanced machine learning, numerical modelling, and remote sensing represents a significant advance in turbidity assessment of complex coastal areas, enabling a better understanding of coastal surface turbidity dynamics.

François Courteille, Principle Solutions Architect at NVIDEA, Advisory Board of EDITO

Role of Deltares

Deltares is a major contributor to the development of a turbidity field of the Wadden Sea. "We have developed a number of software packages with manuals that are offered on that platform. There are open-source tools available such as SFINCS, D-EcoImpact and dfm-modelbuilder, which can already be found online.

The simulation software - Delft3D FM - is also free but Deltares can provide training to make it easier to use. This is because a lot of technical knowledge is needed for advanced applications of the software and Deltares mission is to make this more accessible." explains Felix.

Dream!

"That everyone who wants to can contribute to a sustainable ocean themselves! I hope that the threshold will be lowered when the tools and data become available. Furthermore, I am very enthusiastic about European collaborations to get a picture of how other countries deal with similar problems. It is great to see how great the will to share (for free) is. This motivates and creates a positive working atmosphere!"

Collaborating with the right parties to learn from each other and complement each other is key! In this case, Deltares brings in the system knowledge and numerical tools, while IMT Atlantique provides deep learning knowledge and reliable, robust algorithms.

Lőrinc Mészáros