SQU's Satellite AI Spots Dhofar Frankincense With 93.7% Accuracy

A Sultan Qaboos University researcher has shown that satellites and deep learning can identify frankincense trees in Dhofar with 93.7 percent accuracy, a breakthrough that could reshape how Oman monitors one of its oldest and most culturally significant natural resources.

🌳 Key Takeaways

• SQU's Remote Sensing and GIS Research Centre developed a method combining field spectroscopy, WorldView-3 satellite imagery, and AI detection models to classify Boswellia sacra (frankincense) trees in Wadi Dawkah, Dhofar.
• The study, led by Associate Professor Yaseen Ahmed Al-Mulla, achieved 93.7 percent overall classification accuracy using a technique called Spectral Angle Mapper.
• Three deep learning models, SSD, YOLOv3, and RetinaNet, were tested for automated tree detection; the Single Shot Detector (SSD) performed best, with a precision of 0.83 and recall of 0.93 across 701 validated tree detections.
• The research was published in the peer-reviewed journal PLOS ONE on August 13, 2026, and reported locally by Oman Observer on August 31.
• Wadi Dawkah, the study site, sits inside the UNESCO-listed Land of Frankincense, giving the research direct relevance to heritage protection as well as agricultural and environmental planning.

🛰️ How the study worked

According to Oman Observer, the research team first built a spectral "fingerprint" for Boswellia sacra using field-based spectroscopy, then tested that signature against high-resolution WorldView-3 satellite imagery captured over Wadi Dawkah. The goal was not simply to spot vegetation from orbit, a relatively routine task in remote sensing, but to distinguish frankincense trees specifically from the surrounding arid scrub and shrubland that shares much of the same visual and spectral profile.

The team applied a classification technique called Spectral Angle Mapper (SAM) to compare the spectral signature of known frankincense trees against pixels across the satellite scene, and the method correctly classified frankincense cover with 93.7 percent overall accuracy, as confirmed by the peer-reviewed study in PLOS ONE, titled "Spectral signature mapping and deep learning detection of Frankincense (Boswellia Sacra) using worldview-3 imagery."

The researchers then went a step further, testing whether individual trees could be automatically detected rather than just classified as landscape cover. They compared three deep learning object-detection models, SSD (Single Shot Detector), YOLOv3, and RetinaNet, against the same imagery. SSD came out on top, achieving a precision of 0.83 and a recall of 0.93 across 701 validated tree detections, meaning the model correctly flagged the large majority of actual frankincense trees while keeping false positives relatively low.

🌍 Why satellites matter for frankincense monitoring

Frankincense resin, harvested from Boswellia sacra, has been central to Dhofar's economy and identity for centuries, and Wadi Dawkah forms part of the UNESCO-listed Land of Frankincense sites. Traditionally, tracking the health, spread, and density of frankincense stands has meant researchers physically walking or driving through difficult, often remote terrain to survey trees one by one.

As Oman Observer reported, satellite-based mapping could eventually complement that fieldwork by helping authorities identify changes in distribution, target field inspections more efficiently, and compare the same landscapes over time without repeated ground surveys. That kind of longitudinal monitoring is particularly valuable for a species facing pressure from overtapping, grazing, and climate stress in parts of the Arabian Peninsula.

🎓 Part of a broader pattern of SQU AI research

The frankincense study adds to a growing list of Sultan Qaboos University research projects applying artificial intelligence to problems with a distinctly Omani dimension, from telecom networks to environmental monitoring. It follows other recent moments of international recognition for SQU-affiliated AI work, including an Omani student's AI project ranking in the world's top 13 at a London forum earlier this month. Together, these developments point to a university research pipeline that is increasingly translating machine learning expertise into tools with practical, sector-specific applications inside Oman rather than purely theoretical output.

The Remote Sensing and GIS Research Centre, which Al-Mulla directs, has previously worked on projects spanning agriculture, water resources, and land-use mapping, making frankincense a natural extension of the centre's applied focus on Oman's natural landscape.

🔬 Why this matters for Oman

Beyond the specific case of frankincense, this research demonstrates that Omani universities are building homegrown capacity in satellite-AI pipelines, work that combines remote sensing infrastructure, spectral science, and deep learning models rather than relying entirely on imported analysis. For a country whose Vision 2040 strategy emphasizes both economic diversification and the preservation of cultural and natural heritage, tools that let a small team of researchers monitor a UNESCO-listed resource across large, difficult terrain using satellite data rather than exhaustive ground surveys have practical value for conservation authorities, tourism planners, and agricultural researchers alike. It is also a reminder that Oman's AI talent pipeline extends beyond fintech and telecom into environmental science, an area where the Sultanate has unique assets worth protecting with modern tools.