Improved Recommendations Through Data Fusion (SEGES Case)
SEGES Innovation, a leader in digital agricultural tools, operates CropManager, among other tools, to help farmers optimize crop management with variable-rate application maps. SEGES has decades of experience developing agricultural recommendations for farmers and creates value for farmers by combining deep agricultural knowledge with new technologies. However, weather conditions often pose challenges, particularly cloud cover obstructing satellite images. When creating application maps based on satellite images, farmers rely on cloud-free images. But what can be done when there are no cloud-free images? ClearSKY uses AI to generate cloud-free satellite images, ensuring reliable data for agricultural decisions. Below, we present a case from November 2023, where ClearSKY supplied vegetation maps to SEGES to enhance their nitrogen recommendations for rapeseed fields in Denmark.
In the autumn of 2023, SEGES faced the challenge of sourcing scalable data due to continuous cloud cover obstructing most satellite images. This impacts the accuracy of nitrogen recommendations crucial for rapeseed fields in Denmark. Every January, SEGES releases nitrogen recommendations for 13,000–18,000 rapeseed fields to help farmers prepare for the coming season. Late autumn vegetation maps accurately reflect nitrogen deficiencies in rapeseed fields, forming the basis for spring recommendations. However, relying on satellite imagery in late autumn in northern Europe is problematic due to inconsistent data availability and unpredictable timing caused by persistent cloud cover. This issue is present all year round but is especially critical in spring and autumn for agricultural use cases.
Navigating the Cloud Cover Challenges
As is typical in autumn, while Sentinel-2 revisits Denmark two to three times a week, the available, useful optical data is sparse and predominantly obscured by cloud cover. In precision agriculture, timing is everything. For crops like rapeseed, which are sensitive to nitrogen levels, receiving up-to-date information on vegetation health is crucial. Delays or inaccuracies in data can lead to suboptimal nitrogen application, affecting yields, cost-efficiency, and environmental sustainability. ClearSKY’s partnership with SEGES aims to cut through the uncertainty caused by cloud coverage. By providing reliable, cloud-free satellite imagery, SEGES can stop relying on uncertain data availability and enhance their nitrogen distribution recommendations with unparalleled precision. This ensures that each farmer’s decision is data-driven, timely, and yields the best possible outcome for their crop, regardless of weather conditions.
“With the wetter weather we are facing, situations without satellite images will occur more often. It emphasizes the importance of having access to AI-generated satellite images.” — Rita Hørfarter, SEGES Innovation
Assessing Crop Health with NDRE
To avoid oversaturation, SEGES employs the Normalized Difference Red-Edge (NDRE) index. The NDRE measurement assesses crop health by analyzing light reflectance, which is linked to chlorophyll and, consequently, to nitrogen content. Accurate NDRE readings are essential for SEGES’s targeted fertilizer recommendations, ensuring efficient use of resources and maintaining environmental balance. Let’s examine the NDRE values across multiple rapeseed crop fields during the autumn months of 2023 to illustrate why timely decision-making is a cornerstone of modern sustainable agriculture.
The first thing that comes to mind is that not all crop fields are created equal, as NDRE-values for these fields are not uniform from field to field. This is why an average growth curve modeled through historical data is not a precise way of estimating future needs for individual crop fields. Another insight to share is how most of these fields had their last cloud-free image in mid-October, while a few crop fields had usable imagery in November. Others had no useful original Sentinel-2 data since mid-September (based on a 4% cloud tolerance with Sen2Cor Scene Classification). The average change between the last original cloud-free Sentinel-2 image and ClearSKY’s top NDRE prediction in November is 0.061 NDRE-values — or a relative NDRE change of 28%. This indicates that we believe there have been significant changes during that period, which would not have been visible without newly generated AI satellite images during November.
Advantages of ClearSKY
The primary advantage of adopting ClearSKY over traditional Sentinel-2 is its reliability, ensuring that satellite-based services can operate effectively and deliver substantial benefits to farmers even under challenging conditions. The actual effect on a field-level is difficult to measure in a live scenario, but based on historical equivalents, we know the accuracy in various situations. If the use-case only needs average NDRE values for a crop field, SEGES has documented an r2-score of 0.98 when using ClearSKY on various crop types — with cases up to 32 days of cloud cover. If the use-case on the other hand needs in-field variance data, which is often true for variable-rate fertilizer applications, we see a lower r2-score as it’s significantly more difficult to estimate accurate pixel-level information. However, even with intervals ranging from 2 to 37 days since the last cloud-free image, SEGES’s analysis revealed a robust r2-score of 0.89, demonstrating significant reliability in measuring in-field variance.
“The test showed that a better description of a field’s variation is achieved by using images from ClearSKY than by using the old images — which would be the alternative.” — Rita Hørfarter, SEGES Innovation
Assessing the Impact on Nitrogen Recommendations
If we continue with the absolute NDRE change found by analyzing the sampled rapeseed fields, while keeping all other parameters from the SEGES redistribution calculations equal, we can calculate the difference made by our cloud-free images. In practice, working with thousands of crop fields like SEGES does, some fields had already peaked in October, showing no meaningful NDRE change between the latest original Sentinel-2 and ClearSKY recommendations. However, looking back with the benefit of hindsight, most fields had little useful imagery after their growth curve peaked in October or November — making it difficult to assess in real-time.
The average change of approximately 0.061 NDRE, identified between ClearSKY and Sentinel-2 earlier, would have resulted in a recommendation difference of -24 kg N per hectare on average (with a difference range between -177 kg N per hectare and +18 kg N per hectare). The histogram below visualizes this difference on the one hundred sample fields:
With more than 125,000 hectares of rapeseed fields managed through SEGES, this would amount to a total of 3,698 tons of nitrogen. In practice, it is not so straightforward, as SEGES would not utilize a severely outdated image and could instead use historical averages for fields with no cloud-free images. This approach might be somewhat cumbersome to organize but is certainly feasible for an organisation such as SEGES.
In a field as dependent on timely, accurate data as agriculture, ClearSKY’s AI-powered satellite imagery can be a game-changer. By providing reliable, cloud-free imagery, SEGES Innovation can continue to empower farmers with precise nitrogen recommendations, enhancing both crop yields and sustainability.
ClearSKY satellite data is available as a beta feature for all users of CropManager. This uniquely includes true-color images as well as vegetation maps, provided in near real-time. This steady stream of data is accessible to all SEGES farmers, enabling informed and timely decisions for optimal crop management. Try it today at cropmanager.dk or learn more at clearsky.vision.
