Downstream EO - eddies and rapids

Downstream Earth Observation derived analytics: where everyday people use EO products.

I posted New Horizons, New Horizontals, a year ago. That article discusses how upstream Earth Observation (EO) companies might engage with a modern market. But how should we think about those downstream companies? Those companies that distill the EO sensor product into something a consumer might want to buy. The companies that have the opportunity to create net new markets for the EO sector at large.

How should we consider the dramatic friction between sensor data and human needs?

Ultimately, we build tools that humans use (at least, we do for now until the robots take over.) So, where do we all sit with inevitability?

A useful tool is to consider the future. If I am to think forward, then clearly, we will need EO analytics. When that will happen in terms of human needs and market forces, I don’t know, and I have been constantly (too) optimistic in my assessments. But, EO-derived analytics must happen because our communities absolutely need access to a consistent monitoring capability. 

For other thoughts on inevitability and alternative futures, have a look at Domineering an alternative reality

Now, I do not think that every monitoring use case will be EO-based; that would be magical thinking. Clearly, there are numerous applications for IoT and other ancillary, or alternative data products. But, EO has a built-in scale, which is enormously compelling.

EO-based analytics, then, are inevitable but market-dependent. That market will be driven by cost and need. The needs are increasing, and the cost is diminishing ergo EO-derived analytics must become a credible future. 

But what do these business models look like? Let me paint an idle picture of the present industry with an eye on the potential. 

Price drives availability

First, we have a reduced launch cost. That’s not completely true, but true enough for our downstream analysis. More things can be launched faster. Indeed, with Starship, the price to access LEO, assuming a small payload, should be vastly diminished. So, let’s assume we have a reduced launch cost. That price reduction has increased the interest and capacity to build spacecraft. So, more things in space. As discussed previously, satcoms and navigation are the primary uses of space presently, but EO comes in third. In a ham-fisted manner, I combine weather with surface sensors in my rough assessment. Critically, all these sensors are effectively monitoring the transfer of energy across our planet. 

So, let’s assume that sensor data is becoming cheaper, and we are lucky enough to have more phenologies of sensor data, which means we have a wide range of sensory capabilities: we can look at more kinds of things. 

In effect, the EO industry is providing more capability combined with more capacity. Ultimately, we have more pixels at a lower price. This is important because EO data has fluctuated frustratingly between free and wildly expensive. bringing the price of commercial imagery down means there can be a reasonable commercial EO sector.

Data Strategy

We have an embarrassment of data riches, but what does it buy us? And who should be buying?

My argument has always been that the EO companies, those that manage sensory capability - should focus on data delivery. I had heard individuals from satellite companies tell me that they wanted to own the entire value chain. But how can any one company know everything about everything? Because that is how EO-derived data should be considered in the market. This data need not be special or different but instead be considered an expectation.

EO data is a horizontal data source that can fuel numerous vertical capabilities.

So often, geospatial or EO is considered a separate organizational structure, but this separation breeds needless exceptionalism, which adds friction to the data’s consumption and use.

Indeed, Brian Timoney argues that this exceptionalism hurts geospatial practitioners financially too.

EO data is just another source of data from which to draw insight. Most major enterprises now identify in some way as data companies. In my inevitable future, any Chief Data Officer should have a geospatial strategy; that strategy should identify if EO data is relevant. I don’t believe that future is far off.

Analytics

As an analytics company, there may be some complications with data acquisition in the short term, but ultimately, there will be more options. The key problem to date has been scarcity: too few companies served too few customers. This had driven non-defence customers off with price and availability.

In Fruit Salad, I talk about some specific technology-oriented approaches to analytics.

However, more sensors must lead to more choice. But, critically, analytics companies should be completely consumed with the problem they are addressing and be willing to solve that problem with the most appropriate sensory capability. By focusing on the best way to solve the problem they want to solve at scale, they will serve both the market and the broader EO community through discerning success. However, it should be noted that analytics companies could quite reasonably switch between data providers if distribution or quality is not adequate. This is how a normal market operates.

For a little more about focus for geospatial product companies, take a look at The Geospatial Product Trap.

There are several ways one can think about analytics in geospatial. Again, I also tend to think about these from a data-centric perspective rather than a problem-centric perspective. But together, we can carve the typical workflows up.

Counting Problems. Simply put, counting objects in a place. this might provide an indication of activity, or traffic of something. The classic example of counting cars, for instance. Often, this is a machine-learning exercise which involves algorithmic tweaking for different geographies. Though this activity can be mapped, the vast majority of use cases for this activity result in a report, email or spreadsheet. Analytics companies should not be scared of the humble spreadsheet as a flexible UI, though feel free to call it a data API if it makes you feel better.

Mapping/Space Problems. Identifying impervious surfaces, identifying flood or fire risks, and measuring volumetrics are all examples of more mapping-oriented problems. These may result in a map output but could also simply be an intermediary step before attaching a risk or production score to an asset.

For more thoughts on assets, take a look at Assets: The Scale of Spatial Finance and An Embarrassment of Assets. I am clearly overusing that idiom; apologies!

Presence/Absence Problems. Is something there or not? This is a more specific variation on the above basic analytic tools and comes with a pleasantly probabilistic UI. But presence/absence could also be a mineral monitoring project or identifying if an asset is switched on or not.

Change Problems. Finally, change. I left this for last because it is the most important but potentially the knottiest problem. Change is a deeply contextual problem, and the sensor is rarely created with that context in mind. But change at scale should probably be considered the gold standard of opportunities EO-analytics presents. As an easy starter, we could consider anomaly detection before change: Something has changed, but our algorithm is too general to know what.

In Transitions, I discuss the idea of measuring between the lines on a map. If we think about change, we need those lines to move and our maps to live.

Time. To measure change, however, we also need to be considering time. In fact, time has become the new central vector of geography. For any modern application of EO analytics, we need to understand time because the delta, the difference, is often more important than the absolute measurements. This is convenient for EO companies because absolute measurement from space sensors is exceedingly difficult. However, deltas are much, much easier.

Archive. What, then, are we doing with all our archive of imagery and sensor data, that trove of pixels? The deep catalogues of data that have been captured, and will continue to be captured into the future. With apologies to our SAR friends:

What of the lonely pixel?

Captured by a CCD in the sky, but

Never looked at by a human eye?

Seriously though, every pixel not used meaningfully is a value trap. It is a waste of space, quite literally, in terms of the satellite time and launch cost, but also cloud storage. Unlocking the value of these archived pixels will lead to new business models. I suspect that training Large or Foundational Geospatial Models may hold to key to opening this trap. Present LLMs lack a robust understanding of space or geography, but future models would be able to ingest this archive and understand landscape patterns in a far more scalably profound way than we can now.

While it’s great for an LLM to guess the next word, it would also be interesting for a Large Geospatial Model to guess the colour of the next pixel and all that might imply…

Value Chains

Sensors. If we consider the entire market, we can start to see emerging patterns and operating models. At the top, literally and figuratively, we have the sensor companies which capture data. Hopefully, that data is available as an Application Programming Interface (API) to be consumed by whomever. Umbra has been the most vocal in this discussion recently, helping to raise the consuming market's expectations. This is not always the case, but this level of sophistication is now the market expectation.

Note that an S3 bucket is just a modern FTP site, not an API - IYKYK.

Marketplaces. Beyond the API endpoint, there are a variety of marketplaces which act as common data aggregators: Up42, SkyFi and Skywatch, notably. These companies have provided a critical service to the EO community in delivering common APIs for multiple data sources and thus lowering the barriers to entry for data consumption. These marketplaces have also provided APIs where none existed, encouraging a greater level of distribution sophistication from the EO companies.

As the market stands today, these marketplaces or “data API proxies” are enormously valuable to our community. Each of these companies should not be just seen as channel partners for the EO companies but as enablers. Satellite companies have traditionally been fiercely protective over their channels and distribution networks, to the point of self-flagellation.

So, these marketplaces have taken time to flourish. Each of the companies listed above has its own attributes, and I suspect all will do well as we see increased differentiation with the market evolving.

Analytics Providers. After the marketplaces, we have the analytics providers. These organizations accept API endpoints and create a business product from them. These products can take various forms with various attributes. Ultimately, data is received, an algorithm is applied, and a result is presented to a consumer. Ideally, the consumer pays for the product, and that value climbs back up the value chain.

These products could be anything from a wedding photo to a physical risk analysis to a military assessment to a vegetation inventory. This goes back to my original thesis that an EO company should not attempt to create all its own analytics. The scope of product delivery is so broad, yet the data is so specific that everything about everything can’t possibly be known, let alone marketed or sold. Instead, a team of domain experts with some EO competence should focus on their domain and build the most appropriate data product. If a sensor company subsequently acquires that team, then so be it!

Again, this is in line with The Geospatial Product Trap.

The value chain I outline here moves from the EO sensor company down through marketplaces and into analytics companies. Of course, there are variations here, and analytics companies could buy directly from the sensor company. The central point here is that each link in the chain allows each company in question to focus. Either on creating an amazing sensor product, providing excellence in data distribution, or building a domain-specific data product. Each company remains focused on building its own excellence. The EO analytics market is still far too unsettled for meaningful vertical integration to make sense. Integration of what?

Interfaces

I wrote about interfaces in Interfacing recently, and throughout this post, I have been referencing APIs and value chains. The nature of what I am proposing here is a collaborative ecosystem of companies. I do not expect any kind of happy-clappy utopia, but I do see that domain expertise can lean on others’ domain expertise and differentiate appropriately. What I see is a more functional EO analytics market.

Key Ideas

There are several ideas in this article, here are three:

1. The EO-analytics market is still unsettled, but reduced price of data will act as a settling influence.

2. There is a distinct value chain from the sensor companies through to end users. While there are variations in this chain, focusing in on value and reasonable expectations benefits all.

3. Interfaces between data products, companies, and markets will drive greater adoption. Clean APIs and clean sales practices will ease transport of data and margin.