Fully Open Edge Cloud

Earth Observation AppStore

A presentation of the Big Data AppStore in the context of digital transformation strategy of multinational companies specialising in Earth Observation.
  • Last Update:2017-02-24
  • Version:001
  • Language:en


Nexedi World Map
  • Founded in 2001 - Largest publisher of Free Software in Europe

Nexedi is the largest published of Free Software in Europe. It was founded in 2001 in Lille (France) and has offices in Tokyo, Munich, Shanghai and Paris. It is protitable since day one.

Nexedi Track Record

Nexedi customers are large companies or midsize companies with complex IT problems that can not solved with common off the shelf software. Most customers use Nexedi to implement mission critical management software such as ERP, CRM, accounting, online sales that require a level of flexibility or durability that is higher than usual. Recently, Nexedi started to sell management and data processing software for the Internet of Things, also called Industrial Big Data. Nexedi is also the inventor of edge cloud in 2009 and provides French Ministry of Economu with a Big Data platform used by 50% of French multinationals.

Nexedi Full Stack - Libre Software

Nexedi has a large portfolio: ERP (ERP5), Big Data (Wendelin), Cloud (SlapOS) but also a distributed database (NEO), a Web IDE (Webrunner), Web storage middleware (JIO), Web UI framework (RenderJS), multimedia conversion (Cloudooo), resilient software defined network (re6st), HTML5 office suite (OfficeJS) and secure Web operaring system for lapyops (NayuOS).



Nexedi has some experience with earth observation since it is one of the suppliers of the TSXX mission at Airbus. Nexedi provids the ERP segement in charge of image ordering as well as various other business rules that make this ERP segment a fairly critical and complex system.

SlapOS and Big Data


Nexedi is the creator of SlapOS. SlapOS is the Infrastructure automation system of Teralab, the Big Data platform operated through the French Ministry of Economy and used by half of French multinational companies for experimental purpose. Teralab was recently awarded by the European Big Data Value Association. SlapOS provides security, isolation, higher performance and lower cost compared to traditional cloud platforms.

Wendelin and Python Big Data

    # ipython
            In [1]: root = dbopen('neo://dbname@master')
            In [2]: A = root['A']
            In [3]: for i in range(4):
            ....:     A.append( np.arange(4*1024*1024) )
            ....:     transaction.commit()
            In [4]: A.shape
            Out[4]: (16777226,)
            In [5]: A[:]
            Out[5]: array([      0,       1,       2, ..., 4194301, 4194302, 4194303])
  • Arrays bigger than RAM
  • Arrays bigger than disk
  • Concurrent, persistent, distributed, replicated, transactions
  • Solves cpython's memory duplication problem
  • Applies to any data structure besides arrays


Nexedi created a software called Wendelin.core that solves the old problem of out-of-core processing in python. Nexedi's approach is based on implementing a low level library that acts as a distributed virtual memory manager with memory deduplication. With this approach, a python environment based on multi-processing can efficiently use available memory. To our knowledge, no other solution exists on the market that can achieve similar result, and thus enable PyData for true Big Data processing.

Big Data Appstore


The main topic of our talk is about Big Data appstore applied to earth observation. A Big Data Appstore is actually a simple idea. Many large companies own a lot of data. Thousands of small companies have many ideas how to analyse data and create value. However, small companies can not access this data because large companies want to be sure that no copy of their data goes outside (unless one pays a very high price).

With a Big Data appstore, small companies upload their algorithms to an infrastructure operated by the large company. Results of their algorithm can then be accessed through an API. If some business can be created by small companies, revenue is shared with the large company. This concept can be applied to health industry or automative industry.

This leads to an ecosystem in which large companies play the role of a lake owner, startup companies play the role of a fisherman looking for fish, API act as a fishin cane, algorithms play the role of a hook and A.I. models discovered by startup companies are the fish. By keeping data within the company premise with some kind of monitoring VPN, large company can ensure that small companies pay for every model they find. This acts in the same way as a fence around the lake with a tolling barrier to leav the are.

Earth Observation Appstore

  • Example API: flight plan as a service
  • Customers: drone producer (Parrot, DJI, etc.)
  • Business model: revenue share

The idea of Big Data appstore applies to Earth Observation in a quite straightforward manner. The lake is made of image catalogs taken from all missions of companies such as Airbus, Beidous, Digital Globe, etc. APIs provide value added services such as flight plan as a service for drones, or predictive crop calculation, predictive unsold car inventories, etc. Startup companies can be drone producers that need some kind of value added earth observation as a service for their product or service. No image needs to be copied outside the catalogs owner by Airbus, Beidous, Digital Globe, etc. Business model can be based on renting access to images per time or area, by sharing revenue generated by API access, or by both.

DIY Big Data Appstore in 3 months with Wendelin

  • Data analysis: PyData (scikit-learn, scikit-image, etc.)
  • Storage: NEO (distributed transactional pickles)
  • Out-of-core processing: wendelin.core
  • Parallel processing: CMFActivity (as joblib backend)
  • Data access control: Restricted Python (Zope)
  • Accounting and billing: ERP5
  • Data ingestion: fluentd, embulk
  • Deployment automation: SlapOS
  • Global access: re6st / Grandenet


All components to create an Earth Observation Appstore already exist. PyData community has created numerous libraries for data analysis, machine learning or artificial intelligence. NEO database provide scalable transactional storage of binary data such as ndarrays. Wendelin.core lets data scientist transparently access data that is larger than RAM, no matter their data structure. CMFActivity provides active object programming, a concept that generalises approaches such as Map Reduce or actors and that supports parallel processing backends such as joblib. Restricted python ensures that multiple users can use the same data lake with strict access rules and data isolation, embedded inside the language runtime. ERP5 provides accounting and revenue sharing management, something that is actually much more complex that one usualing imagines and without which there is no business. Fluentd can be used for real time data ingestion. Embulk can be used for batch ingestion. Deployment can be fully automated with SlapOS and network access can be provided all over the world including in China thanks to Grandenet.

Google creates it first

  • Google P/L
    • Purchase more images: -A€
    • Rent images: B€
    • Share revenue: 20% * C€
  • Airbus/Beidou/DigitalGlobe P/L
    • Sell more images: A€

Since it is very easy - if one uses Wendelin - to build a Big Data appstore, let us now analyse the economic impact for Google (or Baidu in China) if this type of company becomes the first to start running such business.

Google uses their cash to purchase images (-A€). Google then rents images (B€) to startup companies. Once they find a good model, they share revenue (20%) on the business (C$) generated by the model.

On their side, Airbus/Beidou/DigitalGlobe sell more images thanks to this new business.

Airbus/Beidou/DigitalGloble Creates it First

  • Google P/L
    • Rent images: -X% * B€
    • Share revenue: 20% * X% - C€
  • Airbus/Beidou/DigitalGlobe P/L
    • Rent images: B€
    • Share revenue: 20% * (1 -X%) * C€

If Airbus/Beidou/DigitalGlobe run the appstore, they will rent images to Google and others (B€) instead of selling them. They will take 20% of the revenue (C€) of (1 -X%) of startups.

Google on the other hand with also rent images for its own purpose, on X% of the market. Google will also run its own appstore no matter what happens and take 20% of X% of the market of A.I. models (C€).

Airbus/Beidou/DigitalGloble Strategy ?

20% * (1 -X%) * C€ + B€ - A€ > 0 ?

Google acquired Skybox Imaging for $500 million during 2014

20% * (1 -X%) * profit(C)€ + profit(B)€ - profit(A)€ > 0 ?


20% * (1 -X%) * C€ + 4% * A€ - 0€ > 0 ?

the only future big profits is C

without rental, observation will no longer be profitable

If we compare the two scanrii, we see that what we need to compare is the sales of images (A€) with the rental of images (B€) and share of the new A.I. model market, equal to 20% * (1 - X%) * C€.

However, Google is also becoming on earth observation company, which means that it has ability to bring prices down to a point that Airbus/Beidou/DigitalGlobe barely profits. So, we shoudl rewrite the equation with A, B and C as profits.

Profits of A become close to zero because of competition. Rental of images has no reason to be much more profitable than rental of car or of realestate, that is a 4% of A€. We can even imagine a situation in which sales of images is losing moning and can only be made profitable through rental.

With this in mind, we clearly see that the only sustainable source of revenue is the 20% on the share of the market share (1 - X%) that Airbus/Beidou/DigitalGlobe will be able to take from Google in the field of A.I. models for earth obervation (C€).


  • The future industrial leader of Earth Observation will be the one who owns the Earth Observation Appstore
  • Due to competition, there will soon be no more profits in building satellites or operating satellites as such
  • Focusing only on B2B prevents the kind of progress that was brought by Google in earth observation
  • Teaming with Google is the same as for Nokia teaming with Microsoft
  • A Big Data Appstore takes 3 months to setup with Wendelin and SlapOS

A.I. market in 2025 is considered as a 36 billion dollar market at least. A value of 1 billion dollar for A.I. related to Earth Observation gives an order of magnitude of the market for an Earth Observation Appstore. Capturing 20% on 25% of that market represents 50 million dollar of additional profit per year. This compares for example to Astrium turnover (about 5 billion euros) and a profitability of about 4%, that is 200 million euros for a group of companies doing much more than just earth observation. It is thus reasonable to state that profits generated by A.I. of Earth Observation tomorrow are likely to compensate the loss of profits generated today by sales of Earth Observation images due to the competition from GAFA and Chinese suppliers.