Our project is a deep learning-based system for monitoring global CO2 emission and absorption balance. Its goal is to precisely monitor CO2 circulation (from natural and man-made sources), and its net worth all over the globe, with the usage of:

· Point Of Interest Sensors – mobile devices that measure local CO2 and O2 concentrations
· Earth Observation Satellites - used for imaging
· AI server – to process all of the gathered data

The core of our system is the 3-level Neural Network. The first level will analyse the satellite images, identifying and classifying Absorption And Emission Objects and determining their coordinates. The aim is to achieve an identification resolution of 1 square metre with around a thousand object classes. This will require a tremendous amount of computing power, which will be delivered within the next few years by upgrading the AI server.

The second level will take as an input the map of AAEO obtained in the first stage, with some additional pieces of information (established emissions, time, local biosphere, etc.). The CO2 and O2 measurements from POIS, with their location, will be used to train the network. This level will create a grid of nodal points, in which it will be possible to predict concentration courses over time as a function of nearby AAEO.

The final level, with the usage of meteorological data, air masses movement, and thermodynamics theorem, will create a map of CO2 and O2 concentration between nodal points (similarly to the finite element method). This map will be used to determine global concentration and trend.

As for cybersecurity: only POIS will log on to the server, using a unique HASH that is assigned to a particular device. Communication is done through the LORA network with the usage of the Helium blockchain. All of the data sent from the device will be encrypted. The computing server will be separated from the web network – the results of the analysis will be transferred to a separate WWW server, and published. The users of POIS will be able to get sensor’s readings through a mobile app connected to the device via inbuilt Bluetooth in its docking station. The pairing of devices excludes the demand for creating accounts and managing login data.

The POIS (with IP54, buttonless enclosure) will be as simple as it can be in terms of its usage. As it is pulled from the docking station, it will automatically enter a „cyclic measuring mode” acquiring parameters at a low frequency. When the embedded microcontroller receives a signal, it will enter a „continuous mode” rising measurement frequency. All of the data will be sent to a cloud database or saved on flash memory. The enclosure with its design allows POIS to be attached easily near various AAEO. The NN will correlate the location data from the sensor with the satellite images. The measurements will be made using commercially available components. POIS will use LiFePO4 14500 size cells with the capacity of 500-800mAh to power itself. That will provide it with around 2,5 hours of continuous work, or even hundreds of hours of work in a cyclic mode. The battery will be charged with the usage of a docking station with an optional solar cell. With the appropriate placement of the electrical components, passive cooling of the device will be achieved.

The Earth imagery will be required for the first NN level to work properly. For now, the data collected by the satellites that are already in orbit can be used. However, the goal is to design our own EOS with a specially prepared multimodal camera. Such a satellite will orbit at an altitude of several hundred kilometres with a velocity of around 8000 m/s. It is crucial to identify the proper solution to ensure the imaging with a satisfying resolution, density, and without motion blur, as well as develop an automatic flight control system to keep the satellite in its orbit, point the solar panels towards the Sun and damp the vibrations. The components may require the usage of jet gases to release the heat stored inside of the device – which will create a dependency relation between flight control and thermodynamic conditions. Due to the cost of building the satellite – this part of the project is optional.

Thanks to the non-monolithic structure of the software (AI system and sensor system are two separate entities) one of the systems can be easily scaled without the interference in the other. That is why it is possible to start small – with simpler network architecture and the usage of satellite images that are already on the web, but in time, with increasing expenditures, more precise results are to be achieved.

The plasticity of the input and the integration of global and local (specifying) data is the main reason why this project is innovative.

Why us? We strongly believe that saving the planet requires action on a global scale. Education, as well as presenting “the big picture” of the problem will have the strongest impact on people in order to change their attitudes.


Three factors brought us to this concept: machine learning, Amazonian forests, and The Orca. In 2019 Amazonian forests (called 'the lungs of the Earth') were afflicted by an outrage of fires that destroyed over 700 thousand hectares of the biosphere. This accumulation of fire was a birth point for the concept of using machine learning to estimate the Earth’s plant balance on a global and near real-time basis. The idea of combining AI and satellite imaging, which should result in open access green-areas counter, was made. Such a counter could become a warning to humanity, and a motivation to take action. In September 2021, our idea had evolved again. It was then that the Orca project was launched in Iceland. This installation is capable of absorbing around 2000 tons of CO2 per annum. This amount surely can make a great impression, and it is one of the reasons why 'artificial tree' became a media success. However, one must ask a question about whether it is sufficient. The estimated CO2 production in 2020 alone was 5.9 million tons. It is crucial to determine how many Orcas are needed and to what extent these estimates of production and absorption are accurate. The search for the answers to these questions led us here.


Voting is closed!