Silke Migdall in Spotlight for STELAR: Remote Sensing and Earth Observation
In episode four of the STELAR project’s Data Stories 360° podcast, we had the pleasure of speaking with Silke Migdall, the Managing Director at VISTA Remote Sensing in Geosciences. With over 15 years of expertise in advanced Earth Observation analysis, Silke specialises in optical and hyperspectral remote sensing, leveraging her skills for vegetation monitoring and agricultural applications.
Sandra Kolaric, Dissemination and Communication Manager from Foodscale Hub, one of the partners on the STELAR project, interviewed Silke to explore the advancements and real-world applications of technologies that enable early crop growth predictions.
Join us as we investigate her work, which not only shapes the future of satellite technologies but also ensures innovative ideas translate into impactful solutions.
Silke's Journey into Satellite Technologies
Can you tell us about your journey into the field of Earth Observation (EO) and remote sensing? What inspired you to pursue this path?
My journey actually started quite early in school. We had satellite reception, including Meteosat and NOAA reception, which are two weather satellites. We had an extra class where we covered everything from the technology to bringing the satellite dish up to the roof of the school and installing it ourselves.
This was in the 90s, so the tools were not quite as advanced as today, and we had to schedule every image and do the analysis with just our school knowledge. Although I lacked the in-depth skills and tools that I possess today, it was my initial introduction to a field that truly intrigued me.
With this class, we went to the Australian International Space School, a summer program that teaches everything about space technologies. While space technologies can either look away from Earth or at Earth, I found everything that looked down towards Earth much more fascinating. Our planet has so many great structures and so much to be figured out, as we have not figured out everything yet. So, I studied geography and Earth observation at college and have stayed in this field. Twenty years later, I still find it very fascinating.
"Our planet has so many great structures and so much to be figured out, as we have not figured out everything yet."
With over 15 years of experience, what have been some of the most significant changes or advancements you have witnessed in the field of optical and hyperspectral remote sensing?
In the beginning, there were very few satellites, and the technology was not as good as it is today. One major advancement has been the transition from very few satellites and scenes per year to a dense timeline.
In agriculture, where fields are dynamic and crops progress through stages quickly, having satellite imagery every three to five days provides a much more detailed view of crop development compared to just three or four scenes per year, though this increase in data frequency has also led to an exponential rise in data volume.
The Copernicus program and the Sentinel satellites have been revolutionary. We mainly use Sentinel-2, a multispectral sensor, and Sentinel-1, a radar sensor. Currently, our archive is processing nearly every Sentinel-2 scene across Europe that is not entirely covered by clouds, resulting in our dataset at Vista approaching two petabytes, illustrating the immense volume of data we are handling.
The next generation of satellites will be hyperspectral, with several hundred bands compared to the multispectral’s 7 to 12 bands, providing many more details but also much more data. This will be the next revolution – to have this data available operationally, and we are already working on how to handle these big data sets, store them, and make them accessible for agriculture.
VISTA is leading STELAR’s innovative Pilot B, which focuses on early crop growth prediction. What are the main challenges that farmers face with existing crop growth and yield prediction systems, and how does the "Early Crop Growth Predictions" pilot aim to address these issues?
For farmers, it is crucial to have the right crop variety, irrigation, and fertilisation to ensure good yield quantity and quality. These measures are mainly taken during the vegetative phase of the crop, which encompasses its growth period; however, in the ripening process, there is limited action you can take.
For example, you stop the last fertilisation measure for weed control about 8 to 10 weeks before harvest because, after that, it is just the ripening phase. This timing is crucial, as it significantly impacts the quality of the crop.
To achieve optimal crop management, it is crucial to obtain a yield prediction early enough, ideally 8 to 10 weeks before harvest. This allows for timely application of the last fertilisation measure, which not only stabilises yield quantity but also enhances yield quality. For instance, this timing is critical for ensuring the wheat’s quality, as it can make the difference between wheat suitable for baking and that which is only fit for animal feed. Currently, our systems provide forecasts four to six weeks in advance, which supports harvest logistics but does not address the needs for these crucial fertilisation decisions.
What makes the challenge of predicting crop development much earlier in the season particularly innovative and difficult?
Several challenges contribute to this complexity, including the difficulty of long-term weather forecasts, which play a significant role in crop development and ripening processes. Adverse weather conditions can delay harvesting if the fields are too wet for machinery. To tackle this, we are working on improving long-term forecasts using seasonal forecasts with multiple scenarios.
When working on a large scale, such as for food security, obtaining information on crop types and development across entire countries or continents becomes challenging. Unlike individual farmers who know what they have planted, large-scale monitoring requires classifying crop types before flowering, which is typically the most accurate point for differentiation using Earth observation.
This task is difficult due to the limited availability of technologies for early in-season crop classification. At STELAR, we are addressing this by improving early crop classification methods.
The accuracy of crop predictions improves with the quality of the time series data on crop development. Historically, we have relied on Sentinel-2 due to its high-quality data and relatively dense time series, providing updates every three to five days. However, there are other satellites that can offer even denser time series data.
We aim to integrate these additional data sources through data fusion and address any gaps through data imputation. This effort is a key component of our STELAR work package, with the goal of enhancing crop growth predictions and yield forecasts.
"This will be the next revolution - to have this data available operationally, and we are already working on how to handle these big data sets, store them, and make them accessible for agriculture."
Can you provide some examples of real-world applications where early crop growth predictions have significantly impacted farm management decisions or regional food security planning?
For farmers, the timing of the last fertilisation measure is crucial. This measure, taken about 8 to 10 weeks before harvest, impacts the quality of the wheat and its protein content. If the wheat does not have enough protein, it cannot be used for baking, and will instead be used as animal feed, which is less desirable because the goal is to bake bread.
On a larger scale, early crop growth predictions are vital for food security. For example, in recent years, we have conducted yield forecasting for Ukraine, which is in a crisis situation due to war. This conflict has hindered the ability to assess agricultural conditions on the ground and has prevented accurate reporting on crop development.
Earth observation serves as a valuable tool for monitoring and evaluating the effects of the crisis, identifying where agriculture is still taking place and where it has ceased due to a lack of resources like fuel or seeds.
This information is crucial because the absence of Ukrainian wheat, which is typically sold to Northern African countries, can lead to food shortages in those regions. Earth observation helps in planning and addressing these broader food security issues. We have worked on these forecasts for the Ukrainian Ministry of Agriculture to assist in managing the situation.
In normal conditions, early crop growth predictions are also important for harvest logistics. Knowing how much harvest to expect and where to deploy machinery helps in planning, especially if using contractors. This prevents yield losses due to poor timing in harvesting and ensures that logistics are managed effectively.
VISTA is also leading tasks on spatio-temporal data alignment and labelling. Can you explain what this entails and why it is crucial for the project?
The title of our work package is “Space-Temporal Data Alignment and Labelling.” To break this down, “space-temporal” refers to the fact that our data consists of geographic information with a spatial component, as each image pixel corresponds to a location on Earth. Additionally, it includes a time component because vegetation is dynamic and grows over time, requiring multiple images to analyse its development.
Our task involves aligning different data sources to create a coherent time series. This means integrating data from various sources so that every pixel corresponds to the same location and the time series is consistent. This allows us to analyse the data as a unified whole rather than as separate individual sources.
Labelling involves classifying what is growing where. For instance, we need to identify specific points in the time series as wheat or maize. We also label the data based on the phenological development of the plants. This means knowing the developmental stage of plants, such as flowering, to predict future stages.
If we know that flowering of wheat occurred at a specific stage in one year, and the flowering stage is slightly delayed in the following year, we can use this information to adjust our predictions.
By understanding the phenological stage of the wheat, our models can align with the current developmental stage. This allows us to make accurate comparisons by referencing earlier or later developmental stages from previous years, thereby enhancing the modelling accuracy when the data is correctly labelled.
What do you envision as the long-term impact of the STELAR project on agriculture and food safety?
The long-term impact of the STELAR project on agriculture and food safety largely depends on the continued availability of the tools we develop. We are working to ensure that many of these tools will be open source, remain available, and be maintained for future development.
Currently, we are at the midpoint of the project. We have set up and demonstrated that our workflows and tools are functioning together effectively. The next steps involve calculating services for various test sites, both large-scale and individual fields, and evaluating their performance. This phase is particularly exciting, and we look forward to sharing the results once they are available.
In the future, STELAR tools are expected to support accurate yield estimates well before harvest, providing valuable information to farmers, regional planners, public institutions, and policymakers. This will help improve agricultural management in future.
Conclusion
This concludes our discussion with Silke Migdall, wrapping up another insightful episode of our podcast.
More STELAR episodes are coming soon, so be sure to follow our Youtube channel and our Blog. See you soon!
