You undertook your first degree in civil engineering, before moving on to a Masters in hydrology. What influenced your decision to focus your work in the water sector?
Actually, the very first degree I pursued was in Applied Physics. I’ve always been fascinated by how everything works in nature and I already knew for years prior that I wanted to study physics to get a deeper understanding of the natural world. But the program turned out to be way more theoretical and fundamental than I expected. So I dropped out after half a year, having no idea what to do next. I took on some student jobs and used that time to make a new plan. By then I had already built up a life at university, so I decided to stay and find the most interesting program there, which was indeed Civil Engineering. Water management was my favourite subject but once graduated I still had a craving for deeper understanding so a Masters in hydrology was a logical next step.
Water is also never far away in the Netherlands. We're a country shaped by water; naturally, historically and economically. Which also means there is a lot of research and economic activity related to water, so there are plenty of interesting career opportunities.
Your most recent project focused on river ice monitoring and early warning. Can you expand on the application of space technologies in this project and what you envisage the output and potential benefits of this project to be?
I’ve only just joined this project. My colleagues shown at the link further down are the ones who started it and know much more about it. But it is a great showcase of how earth observation can be used for early warning purposes. River ice jams can lead to catastrophic flooding, which for example occurred in 2020 around Fort McMurray, Canada. The general idea is that satellite data is used to classify various stages of river ice, which could then be used to identify potential areas where ice break ups and downstream ice jams could occur, which in turn could trigger early warning protocols. We’ve shown that the classification from satellite data works and are now working on coupling it with numerical physics-based models as well as integrating it into operational early warning systems. That work is expected to be completed before the end of the year.
Deltares develops software related to water for which it heavily relies on remote sensing data, can you share with us how knowledge of subject matter experts is shared with software engineers to be integrated into the software? How does mutual learning and interdisciplinary collaboration look like? What are the challenges?
Our software relies heavily on many different types of data as well as expert knowledge. The way to get this into the software is to work closely together. My own department, which focuses on operational water management and early warning (e.g. flood forecasting and water information systems), has a dedicated team of software engineers, who, while theoretically employed by a different department, are located at our office so they feel like part of the team and communication lines are very short. There’s also a community strategy board, which involves both software engineers and subject matter experts, specifically those from outside Deltares. This board takes decisions on new functionalities that should be added to the software.
The challenge is to get as much new knowledge into the software as possible, while also having computational speed and robustness as priorities. I think this is handled well with a modular approach, where the core backend can easily be supplemented with the necessary modules.
What excites you most about the work that you do in hydrological forecasting?
That’s a tough question! But I have to say the single most exciting thing is the fact that this work can really help people and contribute to society directly. The earth observation analysis and operational systems we develop are used every day all across the globe to help with water management, flood forecasting and disaster response.
Can you expand a bit on the different types of models, physical and such based on EO data?
Most of the models at Deltares are numerical physics-based models, as that allows us to really put our expert knowledge into practise as well as understand why a model gives a certain result. Of course, we’re also working with AI-based and more recently AI-physics hybrid models. We don’t have many models based purely on EO data, but many models do contain EO data up to a certain degree. Hydrological models need to be parameterized and/or calibrated (and validated), which is often done (partially) with EO data. Especially models that cover large domains (e.g. continental or even global) benefit from EO data, as it's unlikely that there are enough field measurements to cover the entire model area.
What kind of water-related problems should be addressed by means of supervised or unsupervised machine learning, and what differences do you see to traditional models on remote sensing data?
I don’t think the type of problem should be the main driver to decide what type of model to use. Both, AI-based and traditional models have their merits and can be used for all kinds of water-related problems. I’d say it’s more about what data is available and who would be using the model. I’m not sure there is a generic answer to this question, but I would like to stress that it’s not always about obtaining the most accurate or computationally efficient result; transferability, explainability, and with that, the available technological capacity of end-users, should also be considered.
It is also important to consider recent or future changes to the system. For example, large dams can completely alter the flow regime of a river, so how useful is data from before the construction of such a dam? Also, when the effects of potential flood mitigation measures are to be analysed (such as levees or retention basins), these could have a similar effect. This should of course be considered for all types of models, AI-based, as well as traditional. The suitability of a model to deal with this really depends on the assumptions behind it. Although probably traditional physics-based models have a slight edge over AI-based models in this regard (at least for now).
Specifically regarding (un)supervised learning on remote sensing data, these are of course very suitable for classification-type problems. For example, we’re using these techniques for river ice jams (see question on this topic) and floodplain vegetation monitoring. To successfully apply algorithms of this kind you generally need good training data. We are lucky that we get to team up with local parties that do extensive field mapping. Together with potential limitations on transferability to other regions, I’d say this is one of the biggest differences to traditional models. But as I already said earlier, there are a lot of other factors at play.
What drives innovation and development in the remote sensing community?
Oh, lots of things! I hope it’s mainly the desire to do good, but I suppose money also plays a big role. That’s not necessarily a bad thing as long as the innovations are put to good use and accessible to those that need it.
NASA/USGS opening up their Landsat data was very big for innovation and development, as was ESA's launch of the Sentinel missions. So open data plays a big role. Another big step was making data not just open but also easily available with large scale computational support. Think of Google Earth Engine (GEE) or Sentinel-Hub. Suddenly you could easily crunch satellite imagery from a small laptop, even on a global scale. Nearly every day there are new and exciting things coming out of the remote sensing community utilising this combination of unmatched data and computational access. Recently, we've also seen the rise of commercial satellite data providers, sometimes dubbed “new space” or "space 2.0" (or 4.0 for those in Europe). They are capable of providing data at unrivaled spatial and temporal resolution, which could be driving a new wave of innovations, but their data usually comes at a cost. And while it can be worth it for those with deep enough pockets, it is simply too expensive for many across the globe. Some of these companies, like Planet and ICEYE, are already starting to open up their data for research or humanitarian purposes. But it would be great if this could be increased and allow applications in developing countries. I don't think this would hurt their business much as these are not their usual clients anyway. In fact, the increased uptake and use of their data, with some good publicity thrown in, might even benefit them. But I’m not a business person, so perhaps I’m completely off the mark here.
Finally, we’re also seeing a push towards more Analysis Ready Data (ARD), which would enable non-EO-experts to also use satellite data. This could definitely be driving more innovations, but since this is only just starting I think we still have to wait a bit to see its effect.
Water is essential for life. The openness of states to share information on water varies. What are the benefits and dangers of sharing water data openly?
The benefits of open data, or at least the will and option to share with others, are huge. There are various studies highlighting this for open data in general, so it’s a pity this is not yet a common practise globally. Sometimes we even see that different branches of government within a country have to pay for each other’s data, even though it’s all funded with tax-payer money. While this is absurd to me, and sometimes holding back much-needed developments, data is still seen as power sometimes and that can be hard to let go.
But I believe water (data) can bring people together. Especially in the case of transboundary rivers, of which there are many across the globe, up- and downstream states can benefit from sharing their data. For example, reservoirs on either side of the border can greatly impact river flow and knowing how these are expected to be managed, is critical information for dealing with floods, droughts and other water management purposes. But this is by no means limited to rivers with reservoirs.
A good showcase in this regard is the Sava Flood Forecasting and Warning System (FFWS), where data and models of all relevant five riparian countries (Slovenia, Croatia, Bosnia and Herzegovina, Montenegro and Serbia) are unlocked in one shared system. This was the first time these countries were sharing data like this, and it is even more impressive considering the fact that some were at war with each other relatively recently. It’s a great example of what can be achieved with transboundary cooperation.
Increased EO capabilities utilized by open platforms (such as the upcoming Global Water Watch platform) might somewhat reduce the need for open water data from states, but it will not completely replace field measurements, at least not in the near future. Concerning potential dangers of sharing water data openly: I can’t think of any myself, but perhaps I’m too optimistic.
What do you see as the major challenges facing the remote sensing community in terms of flood forecasting, mapping, and early warning?
EO-derived flood mapping has been making huge leaps in the last few years, owing to the availability of more sensors and cloud-based computations (see question on drivers for innovation and development in the RS community). Today, there are a lot of players offering flood mapping services for both disaster response as well as post-event analyses. Today, there are a lot of players offering flood mapping services for both disaster response as well as post-event analyses. There is the International Charter Space and Major Disasters, Sentinel-Asia, the Copernicus Emergency Management Service (CEMS) and of course your own UN colleagues at UN-SPIDER and UNOSAT, which are all activation-based. There are also new commercial players, such as Cloud to Street and ICEYE, but these can be expensive. Recent initiatives include continuous monitoring services, such as the newly-to-be-created CEMS component and SERVIR’s regional flood mapping service (see next question).
Regarding flood forecasting and early warning, there is a smaller role for remote sensing as this usually requires a forecasting element, while satellite observations are always in the past. Of course EO-derived data can be used to make forecasts for downstream areas or they can be combined with field measurements to improve model-based forecasts. In this realm you have the Global Flood Awareness System (GloFAS), the GEOGloWS Streamflow Service and our own inland as well as coastal global forecasting systems, hosted under the new BlueEarth Data platform. Also here there are other non-traditional parties stepping in, such as Google. Within the Earth Observation for Anticipatory Action (EO4AA) working group of the Anticipation Hub we are also looking at how EO can be used to support anticipatory action within the humanitarian sector.
One of the overarching challenges is exactly the fact that there are so many different players offering similar, but slightly different, solutions. As an end-user, it is difficult to know which product to use in which situation, especially if one does not have the technical capability to assess all aspects of these products (see question on knowledge transfer further down). Perhaps we could introduce a comprehensive benchmark, but I realize that would be a daunting task and also not without flaws.
Another challenge is to provide truly actionable information. Simply providing a flood map or a hydrological forecast is often not enough. It needs context (“how does this compare to previous years?”) and impact (“how many people will be affected?”), so that decisions can be made based on this information. Early warning is already a step forward from a forecast and can create actionable information, but it needs to be set up properly and in line with emergency management protocols. Sadly, this was highlighted by the recent deadly floods in Europe.
Of course there are also more technical challenges, related to spatial and temporal resolution, as well as sensor limitations, so that floods can be accurately mapped and forecasted also at very local scales. Some of this is already being tackled with new sensors, although most of these are not freely available (see question on drivers for innovation and development in the RS community). So this is something that still needs work from the larger remote sensing community.
Can you tell us about your experience in SERVIR-Mekong?
I could probably fill up an entire interview just about SERVIR. But I'll try my best and be brief. For reference, SERVIR is an amazing initiative by USAID and NASA, and I have the honour of being part of the Mekong hub. It was one of the projects I started on in my first year at Deltares and is now by far my longest running project. It has shaped my professional development, in large part due to the people I have had the pleasure to work with, both inside and outside of Deltares. My project leaders at Deltares gave me a large degree of freedom and responsibility as a junior, allowing me to learn quickly. Now that I'm project leader myself I try to put that into practice. Within the larger SERVIR family (and it really does feel a bit like family sometimes) there are also great people whom I’m very happy to call my colleagues. You really get the feeling that people are trying their very best to help things succeed and that rubs off on others. Perhaps it’s the way SERVIR is structured, with each hub always being led by a regional organization, supported by experts from NASA and other institutions, and its focus on impact-driven research directly benefiting local stakeholders. I think this works well.
My work within SERVIR currently focuses on near real-time flood mapping using EO and on hydrological forecasting. It is all a strong collaborative effort, backed up by recent scientific advances. We’re publishing our own research in open access journals and conferences, and the code, scripts and tools behind it all are released as open source. This should allow others to benefit from it too. We are continuously improving it and are also looking at implementing it for other regions than Southeast Asia. There are so many interesting developments within SERVIR, that it is hard to keep track of them all. It will be a sad day when the project finally comes to an end, but I hope we have managed to do a lot of good work with real impact before that time comes.
Some of the world’s worst flooding events occur in countries with lower technical availability and capability. How can the availability and knowledge of space technologies be better transferred and used across the world to help those regions most vulnerable to flooding events?
Very relevant question! It is no use to create the most sophisticated tool or dataset if there is no capacity to use, maintain or update it after the project ends.
One potential path is a global application which could be tuned for regional use. Something like Global Forest Watch but then for floods. There are already some relevant initiatives (see question on challenges facing the RS community). Regional applications have the advantage that they can focus on use instead of focussing on technical developments. The downside is that you’re not developing technical capacity in those regions and they are then stuck with a single provider. And of course there are challenges remaining to get local actionable information from such global applications (again, question on challenges facing the RS community), also related to the fact that local hydro-meteorological data is often not shared (see question on open data). National hydro-meteorological agencies usually have the mandate to issue forecasts and that also isn’t something where one global player can simply step in and take over, nor would you want that to happen. Finally, this overlooks the importance of local hydrological knowledge from experts with decades of local experience.
Therefore, while I do think such a global application would be very useful and a good step forwards, for truly meaningful flood forecasting and early warning you would need regional applications, which are developed together with hydro-meteorological agencies and local experts. This should be coupled with capacity building and training so that they can take ownership and are not completely dependent on whoever constructed an application.
But truly improving the knowledge and technical capabilities across the world has to start much earlier, in education. This takes time, so it is truly a long-term endeavour. But I think it is the only way to really ensure there is enough technical capacity, also in those regions most vulnerable to flooding events.
How can one best learn about remote sensing algorithms and applications, specifically those that require the detection of water from space? What is a good starting point? What is missing in freely available open online courses?
Personally, I find the best way to learn is to use it in practise. Of course, you do need to start somewhere and if that’s not coming from (higher) education and/or your (early) career, then online courses can indeed be a good option. It then boils down to which platform and/or programming language you prefer, as there are many options now to get started on various platforms. Both GEE and Sentinel-Hub have good documentation and include educational content (which in case of GEE also includes a training on detecting water from space written by a colleague of mine, although I have to stress it was created over 4 years ago so it can be a bit outdated).
When you start working on a project, how do you assess the data needs and research / identify the available data sources?
That depends a lot on project scope, budget and time. For large projects this can indeed be a separate step, while for smaller projects I usually keep it limited to what is easily available. For example, restrict it to whatever the client has available or global datasets that are already accessible online. I try to keep up to date on the latest releases of relevant datasets through various channels. But I'd like to emphasise that not every new dataset released, at ever higher spatial resolution, is necessarily better for all applications.
A good example are elevation datasets, which are an important factor in hydrological applications. There are now many of these with global coverage, but I’ve not found a single one to always be superior, depending on the region and specific application. Sometimes lower resolution is better. This requires careful analysis of the data, as well as a clear conversation with the client about what can be expected.
How can the space and water communities best encourage the young generation to engage with these communities and to consider them as viable options for future study and employment? Do you have any advice for young people who are already interested in these sectors?
It depends on what drives people. But it starts with them knowing what’s out there and becoming enthusiastic about it. I think both the space and the water sector are booming, which automatically attracts a new generation of students and young professionals. Space is getting plenty of attention with all the tech billionaires throwing their money at it. In the Netherlands the water sector is also relatively big, owing to our historic ties with it as both, a friend and foe.
Regardless, I think it is always good to make the younger generation enthusiastic about the wonders of our planet and the role that water plays in it. As for advice, I would say “stay true to your course and find your own niche”. Don’t simply follow the herd in what seems ‘hot’ right now. In the end, you’ll get more satisfaction and reward if you focus on something you truly find interesting and/or believe in. Of course, if you need to land a job you might need to compromise this a bit, but that does not mean you have to settle for it for the rest of your career.
What is your favourite aggregate state of water and why?
Liquid, definitely. I enjoy ice skating in winter and a nice hot steam sauna from time to time. But liquid water forms our rivers, lakes and oceans, creating some truly beautiful natural areas. The sound of crashing waves or rushing streams is very relaxing to me. And I have good childhood memories of building small stone dams in alpine streams or sandcastles to withstand the waves and incoming tide at the beach. Perhaps that’s actually what subconsciously led me into civil engineering, hydrology and the water sector...
