Anticipation of Hyperspectral

The growth of the commercial Earth observation imagery market since the early 2000s has been mainly geared around high-resolution satellite data. This morphed into high-resolution data with high revisit – a key defense requirement, and enabled by the development of satellite constellations. Companies attracted hundreds of millions of dollars in investment to support constellation development. First this focused on optical satellite solutions, and then SAR. The driving application (whatever the actual vertical market) being to explore the potential of high-frequency change detection applications. Only more recently, however, has the importance of spectrum become more of a conversation topic.

In some ways it could be surprising that this has taken a while; the ability to image across the spectrum being one of the fundamentals of satellite-based remote sensing. Industry workhorse Landsat-7 has eight spectral bands; there are plenty of environment monitoring satellites which have more. Commercial optical satellites however have tended to focus on visible channels, with perhaps a band or two in the near-infrared [NIR] or red-edge to support vegetation monitoring/land-use classifications. There are several factors as to why commercial operators have stayed in this area: cost to build satellites with more bands, and lower-hanging fruit in building applications focused mainly in the visible part of the spectrum, especially in key markets such as defense. Much can even be done with one panchromatic band. ?

Multispectral imaging is collected data in pre-selected spectral bands. Hyperspectral imaging is contiguous data collection across areas of the electro-magnetic spectrum. The output is a data cube. The number of bands is then a function of spectral spacing.

So WHAT is commercial hyperspectral imaging all about?

Firstly, hyperspectral technology is relatively proven. For instance, it is already utilized with sensors on board UAVs/ airplanes (such as AVIRIS) to support agricultural applications, coastal zone mapping, forestry etc. NASA - National Aeronautics and Space Administration 's Hyperion sensor also showed how this could be achieved from space. In Earth-science applications there is a clear understanding of the various spectral signatures of various materials and substances, and how to measure these – the algorithms are there to be utilized. Creating an operational satellite system with sufficient ground resolution, signal-to-noise ratio, revisit etc., however, has proved to be a challenge. Data capture is a further test – the size of the hyperspectral data cube means much higher downlink, storage and processing costs compared to a multispectral sensor.

There is a common misconception as to what hyperspectral actually is – which can be to its detriment – and that is hyperspectral is all about being able to image in hundreds of bands. Whilst this may be a result of the technology, hyperspectral imaging is contiguous data collection across areas of the electro-magnetic spectrum. The number of bands is then a function of spectral spacing (basically, 25nm spectral spacing across the VNIR [400-900nm] = 20 bands.) Trying to reduce the spectral spacing can have an impact on the signal-to-noise ratio, reducing overall image quality. Therefore, there is a trade-off: being able to image in the areas you need [nominally based on user requirements] whilst maintaining image quality. It also means “having as many bands as possible” may not be the best solution to fulfill market demand, especially if overall data quality is compromised. Most commercial EO applications do not need to achieve science-grade spectroscopy for them to be valuable.

To think about this in a different way, most commercial revenues from satellite imagery today is based on just five spectral bands (RGB plus may be one or two bands in the NIR); multiplying the number of bands does not multiply the overall demand for solutions. I spent the early part of my career working in R&D on satellites which had 15-30 spectral bands; outside of very few people within the EO science community, there aren’t too many people who know what to do with all this data.

It is not expected that hyperspectral imaging needs to get to the same levels of ground resolution as multispectral in order to be successfully commercial either: 5m ground resolution would suffice to produce local area monitoring e.g. precision agriculture requires getting to 5m ground resolution for there to be a commercial interest. Being able to get to 5m in the Shortwave-infrared (SWIR) at low-cost could well be a significant breakthrough to support various sectors, including agriculture, as well as in defense, mining, forestry, emissions and so on.

RENDVI is the narrowband equivalent of NDVI. This index takes advantage of slight changes in sensitivity of vegetation in the red edge to examine changes in foliage and senescence. RENDVI also has the advantage of not "saturating" to the extent that NDVI does. Sentinel-2 imagery is 10m GSD compared to Wyvern Dragonette's 5m GSD. More information at: ?

https://knowledge.wyvern.space/#/datasets_and_use_cases/agriculture/plant_health_indices?id=background

So WHY IS hyperspectral important?

Think of this in three parts. What hyperspectral can do now, what it can do in the future, and where the research could take us:

• Firstly, the lower-hanging fruit. Effectively, hyperspectral imagery can mimic any multispectral data set if the ground resolution and data pricing is comparable. A hyperspectral sensor can copy the spectral band selection of Landsat, Sentinel-2 and even the SWIR channels of ASTER. In the case of Wyvern , it can now do this already in the VNIR from its initial satellites at 5m ground resolution, and will do this in the SWIR with its next generation. Therefore, commercial markets for multispectral imagery are equally as relevant for hyperspectral: agriculture, mining, water and forest resources monitoring etc. These markets mean an early route to revenue generation for hyperspectral satellite operating companies.
• Then, what can be done next. Getting to 5m SWIR will be an important milestone. Like other aspects of hyperspectral imaging, what can be done is already demonstrated from aerial imaging solutions or scientific EO missions. Being able to offer SWIR solutions at scale, however, has more significant market potential. Mining will benefit from SWIR – being able to detect a wider variety of minerals, this is already demonstrated with coarser resolution ASTER imagery. As is applicability for defense – being able to detect camouflaged objects, or even detection of chemical weapons usage. Emissions monitoring is a further area to explore in the SWIR – companies such as GHGSAT utilize the SWIR for methane detection, or in wildfire management. Getting to higher resolution SWIR is a challenge, and why several companies are first focusing on VNIR hyperspectral (to start generating revenues) despite the potential of SWIR imaging.
• And then, what can be done next. Being able to image 400-2500nm data cubes at 5m ground resolution at global coverage means being able to explore the potential of moving from research to commercial potential. For instance, will higher granularity data along the red-edge provide insights into crop growth stage monitoring, or exploring the SWIR to asses biophysical and chemical properties of vegetation, stress indicators caused by invasive species etc. Can emissions monitoring extend to Carbon Dioxide monitoring at ground level (a much bigger GHG contributor.) Moving into supporting monitoring of various environmental, societal and government (ESG) variables could well end up being the most significant commercial market for hyperspectral imaging.

Being able to image in fine spectral bands means being able to detect different mineral compositions to support exploration and mine site extension. This is especially evident when imaging in the SWIR as these (multispectral) ASTER images show at 30m ground resolution. The technology can also support mine reclamation and asset monitoring through stock-pile assessment.

New operators starting commercial operations

Certainly, there is significant potential for solutions built with hyperspectral imagery, at the right price and with the right services delivery mechanism. Four commercial companies, which are targeting higher resolution hyperspectral imaging, now have satellites in-orbit. This excludes specific emissions monitoring satellites which may carry hyperspectral sensors. Kuva Space’s launch is its first Hyperfield satellite in August added to Wyvern, Pixxel and Orbital Sidekick (OSK) with satellites already in orbit.

• Wyvern (Canada) has three VNIR satellites in orbit with 5m ground resolution. A fourth satellite in its Dragonette constellation is expected early 2025. It is able to delineate to about 30 spectral bands at this ground resolution. The company’s second-generation satellites are expected to carry SWIR imaging capabilities at a high resolution. Unlike the other commercial provides, Wyvern’s focus is on delivering imagery, not analytics, for which it is working with partners to develop. To get to orbit quickly, the company has gone through the satellite-as-a-service approach with AAC Clyde Space and Loft Orbital. Eventually the company aims to build sensors using its patented deployable optics technology which is currently in-development.
• Pixxel (India) has launched three technology demonstrator satellites in orbit, VNIR imagery at up to 10m ground resolution. The number of spectral bands it can delineate is measured is reportedly measured in the hundreds. It is the highest capitalized of the new hyperspectral operators with over $70 million raised. Pixxel is more vertically integrated, manufacturing its own satellites down to providing value-added solutions/analytics.? In addition to being a satellite operator is opened a satellite manufacturing facility in India. Data from its first two satellites was not made commercially available. It aims to launch its commercial Firefly constellation by mid-2025. It also targets getting to 5m ground resolution in both the VNIR and SWIR.
• The Kuva Space (Finland) launch of the Hyperfield-1 satellite is the first in its commercial constellation, having already launch two technology demonstration satellites. The company raised a Series A round at the end of last year. By 2030, the company plans to deploy as many as 100 satellites offering spatial resolution <10m. Its first generation of satellite are VNIR, with the second generation to include SWIR. It’s aim to build a larger constellation compared to the other companies points to trying to differentiate by being able to capture imagery over specific areas at greater frequency.
• Orbital Sidekick 's (U.S.) approach is a little different with focus on environmental/emissions solutions from the offset. It’s Global Hyperspectral Observation Satellite constellation, or GHOSt, now has six satellites in-orbit offering 8.3m hyperspectral imagery. It also carries a 3m panchromatic imaging capability – which is likely to be used to pan-sharpen the hyperspectral data. A final constellation of 14 to 20 satellites to provide weekly global monitoring is its goal. The number of spectral bands it can offer is measured in the hundreds, across the VNIR and SWIR.

In addition, other companies, such as Esper (Australia) – who recently announced a partnership with Wyvern – will aim to successfully deploy satellite capacity. Planet (through its Tanager constellation) and GHGSat will also explore the technology to support emissions monitoring and GHG quantification. As with other EO sensor typologies these companies all offer different approaches, the data sets are not exactly the same, they are more or less integrated, they may target specific vertical markets or act as a data shop, and so on. And as with other sensor types, there is plenty of room for multiple companies to take their position in the market.?

There is still the need for more project financing for most of these companies to come to full fruition. This said, all four of the detailed companies above have the funds to further develop their respective constellations. Over $250 million has been raised by hyperspectral companies to support technology development. Soon we shall be in a better position to really judge the potential for commercial hyperspectral imagery as it starts to become a reality.

*note on header image - Wyvern Dragonette-001. ICA + RGB, Barrax Spain. The color overlays represent Independent Component Analysis (ICA) processing result which highlights the spectrally distinct features in the hyperspectral imagery data by amplifying the signal and suppressing the noise. The ICA ?algorithm that decomposes data into independent signals and is powerful processing technique for hyperspectral data in order to unmix the statistical independent endmembers.