At the 2025 ArcheoScienceFest in Castellum, Utrecht, we took part in a lively science dialogue session titled Airborne Multispectral Imaging, led by Matthias Lang and Wouter Emaus. The event combined Roman ruins with remote sensing — and offered a glimpse into how drones, wavelengths, and algorithms are quietly revolutionising archaeological discovery.

What is airborne multispectral imaging?

Multispectral imaging (MSI) involves capturing data across several specific wavelengths of the electromagnetic spectrum — not just visible light, but also near-infrared and other spectral bands. Plants, soils, and moisture each reflect light differently depending on their condition and composition. This subtle variance, invisible to the naked eye, becomes gold dust to archaeologists. Buried walls, ditches, or pits disturb surface vegetation just enough to leave faint spectral traces — crop marks, soil marks — which MSI can detect with startling accuracy.

What makes the technique airborne is the use of drones, or UAVs, fitted with multispectral cameras. These nimble platforms allow rapid, high-resolution surveys of large areas without ever touching the ground. Non-invasive, efficient, and — frankly — much more fun than pacing fields with a magnetometer.

Matthias Lang’s contributions

Matthias Lang, representing the Bonn Center for Digital Humanities (BCDH), offered a rich account of how airborne MSI is being applied in the field — not as a gimmick, but as a robust research method.

In the Rhineland, Lang’s team collaborated with the Landschaftsverband Rheinland (LVR) to test UAV-mounted multispectral sensors. The goal: identify archaeological features through anomalies in vegetation growth. By developing a custom R-script, they streamlined the image analysis process, making it more accessible to non-specialists.

In Italy, projects in Paestum and Cerveteri took the approach further. By analysing seasonal patterns in plant growth across ancient landscapes, Lang’s group successfully mapped out buried Roman structures — including foundations, road traces, and subsurface alignments. It’s not quite X-ray vision, but it’s getting close.

Perhaps most impressively, Lang’s work doesn’t stop at data collection. It integrates remote sensing with machine learning and GIS, creating interdisciplinary toolkits that can detect, classify, and geolocate features with increasing precision.

A clear step up from traditional aerial archaeology

Compared to traditional aerial photography — often limited by altitude, weather, or resolution — drone-based MSI offers a serious upgrade. The resolution is remarkable: less than 10 centimetres per pixel, meaning small features like post holes or low ditches can be detected in the right conditions.

It’s also far more cost-effective. Hiring a plane and pilot to photograph your site is so 1997. Lightweight drones with open-source processing software are democratising the aerial view. And because MSI captures time-sensitive data — like how plants grow and change across seasons — researchers can observe subtle shifts that confirm the presence of archaeological features without ever turning a sod.

Not without challenges

Still, the technology comes with strings attached. Multispectral data is complex. Aligning different spectral bands, interpreting signatures, and distinguishing ancient features from modern agricultural noise requires serious processing muscle. Lang’s R-script helps, but it doesn’t replace the need for domain knowledge — or for ground-truthing.

False positives are also a risk. A natural patch of dry soil or a farmer’s drainage ditch can mimic a Roman trench. Distinguishing the genuinely ancient from the merely inconveniently recent takes experience.

Then there are the bureaucratic hurdles. In some parts of Europe, drone flights require special permits or are banned outright in certain airspaces. Nothing ruins an archaeological survey quite like a call from air traffic control.

Future directions

Lang’s keynote emphasised that MSI is only getting smarter. Artificial intelligence — particularly deep learning — is being used to automate the detection and classification of features across large datasets. This means less time sifting through pixels, and more time interpreting meaningful patterns.

Efforts are also underway to standardise how multispectral data is captured, processed, and archived across archaeology. Shared protocols would make it easier to compare findings, pool datasets, and avoid reinventing the spectral wheel for each new site.

And finally, there’s growing interest in underwater applications. While multispectral sensors were born for dry land, adaptations are underway to use similar techniques for submerged sites — lakes, riverbeds, and coastal ruins. Think sunken harbours, shipwrecks, or Roman fish tanks.

Is it suitable for the Netherlands?

Yes — cautiously, but enthusiastically. The Dutch landscape is practically begging for MSI. Flat, open, and rich in archaeological remains, the conditions are ideal for detecting crop marks and soil disturbances. In the lowlands and polders, subtle variations in vegetation often betray hidden ditches, foundations, or past human activity.

Drone-mounted sensors such as the Parrot Sequoia or DJI Mavic 3M can operate at high resolutions (as fine as 13 cm per pixel), which is perfect for identifying small, fragmented remains from Roman or medieval periods.

MSI also suits the scale of Dutch archaeology. With such a dense archaeological record and limited resources for ground surveys, MSI allows large areas to be quickly scanned and prioritised. Combined with GIS tools and Lang’s R-script methodology, automated feature detection becomes not only possible, but practical.

And the limitations?

Dutch weather, for one. Cloud cover and rain aren’t just inconvenient — they degrade the quality of optical data. Multispectral sensors need sunlight. Grey skies, while atmospherically Dutch, are not ideal.

Then there’s the soil. Peatlands and high groundwater can mute spectral contrast, making it harder to spot buried features. And in heavily farmed areas, ploughing may have already obliterated shallow remains — or confused the spectral signals so thoroughly that even an AI gives up.

Finally, while drones are cheaper than planes, high-end MSI sensors still cost money — and processing the resulting data requires significant computing power and technical expertise. This may be a barrier for smaller heritage institutions.

Conclusion

Airborne multispectral imaging is not a miracle tool — but it’s a powerful one when used wisely. It blends technology with archaeology in a way that respects both science and site. In the Netherlands, its application is particularly promising, especially when combined with ground-truthing and other remote sensing techniques like LiDAR.

Matthias Lang and his team at BCDH are not just flying drones; they’re flying blindfolds off the landscape. With continued innovation and thoughtful standardisation, airborne MSI could soon become a standard part of the archaeological toolkit — and a reminder that sometimes, to understand what lies beneath, we need only look more closely at what’s growing above.