Project highlights

  • Research CO2 emissions; one of the most pressing environmental challenges of our time
  • Apply cutting-edge machine learning methods
  • Work with new ground-breaking satellite missions


Since the launch of the first dedicated satellites in 2009, satellite observations of CO2 have become a corner stone of carbon cycle research and upcoming missions are now designed to target anthropogenic CO2 emissions in support of the Paris agreement. One of the fundamental challenges is the development of methods that allow inferring CO2 from the satellite measurements with sufficient and guaranteed accuracy to allow pinpointing where CO2 and CH4 is exchanged between the atmosphere and the surface. The University of Leicester is one of the leading groups that has helped to develop sophisticated physics-based methods that are now employed in major initiatives like the ESA Climate Change Initiative to produce global CO2 records. The next generation of satellites is aimed at producing observations with incredible fine detail and current analysis methods based on traditional physical model-based inference practices are not suitable to deal with the dramatic increase in data volume which is set to rapidly grow further in coming years. We also need to extract information more quickly so that we can more rapidly diagnose emissions and respond to major events (such as the impact of COVID-19).

A fundamental re-think of the approaches to derive information from satellite measurement is needed. Machine Learning (ML) including neural networks has been successfully adopted in many areas for analysis of big data. Such methods have the potential to transform satellite remote sensing and our proof-of-principle research, carried out in the last 3 years, showed immense feasibility of the ML approach to this challenging problem. Building on our previous successful work, we will develop a new-generation ML method for CO2 retrievals from satellites and we tackle perceived limitations of ML that is limiting the uptake of these methods by the community, especially on the quantification of errors and certified robustness in ML. The new ML approach will be applied to data from current missions such as NASA OCO-2 and upcoming missions (French/UK MicroCarb mission and Copernicus CO2 mission) and we will be working closely with the mission teams.


University of Leicester


  • Climate and Environmental Sustainability


Project investigator

  • Hartmut Boesch, School of Physics and Astronomy



  • Ivan Tyukin, School of Mathematics

How to apply


In this project, we will develop ML methods based on neural networks (NN) for reliable CO2 remote sensing from satellites with guaranteed accuracy thus allowing to infer information on CO2 emission in near-real time.

The student will build on the core expertise of the Earth Observation Science (EOS) group in physics-based CO2 retrieval methods and of the Artificial Intelligence, Data Analytics and Modelling Centre (AIDAM) on error quantification, analysis, robustness, and guaranteed continuous learning of data-driven Artificial Intelligence systems.

We will adopt and test emerging methods for providing information for sensitivity and uncertainty of the NN which is a key step for reliable usage of inferred CO2 data. We will apply the NN method to current and upcoming CO2 missions to obtain global CO2 datasets which we will to evaluate models and to infer CO2 surface fluxes in collaboration with National Centre for Earth Observation (NCEO) partners.


Training and skills

The student will be part of the EOS group which provides an exciting cross-discipline environment. The student will obtain a wide range of skill and expertise in satellite remote sensing, carbon cycle and wider environmental science. Specific training will be provided by the supervisory team and partners with additional training provided via summer school (e.g. ESA summer school), workshops and University training (e.g. computing). The student will also benefit from the National Centre for Earth Observation which provides numerous training opportunities from data visualization to presentation skills. NCEO also provides ample opportunities to interact with researchers and PhD students.

Partners and collaboration

This studentship will be in collaboration with the National Centre for Earth Observation which will provide training and expertise to the benefit of the student and the student will work with NCEO partners on atmospheric modelling and surface flux inversions and there will be opportunities for visits. We will also collaborate with satellite missions’ teams, especially the NASA OCO-2 team at NASA JPL and Colorado State University and our partners for the French/UK MicroCarb mission at the Laboratoire des Sciences du Climat et de l’Environment LSCE in Paris.

Further details

It is strongly advised that you contact the supervisor Prof. Hartmut Boesch ( before applying

For more details about the EOS group, please see

Please visit the University of Leicester website for application guidance:

Possible timeline

Year 1

Training in satellite remote sensing methods and machine learning. Setup training framework for simulations of satellite observations. Apply neural network (NN) methods to problems with reduced complexity such as cloud screening retrieval.

Year 2

Tailor and train neural network for CO2 retrieval from OCO-2 and French/UK MicroCarb mission. Evaluate NN datasets against validation data and models. Apply methods to infer NN sensitivity and uncertainty.  Use inferred satellite CO2 data to constrain carbon surface fluxes in collaboration with NCEO partners.  Presentation at international conference.

Year 3

Investigate new hybrid approach by replacing identified processes in UoL physical retrieval model with NN approach thus maintaining the overall UoL retrieval framework. Demonstrate the use of NN for the upcoming Copernicus CO2 mission.  Presentation at international conference.

Further reading

Janssens-Maenhout, G., and Co-authors, 2020: Toward an Operational Anthropogenic CO2 Emissions Monitoring and Verification Support Capacity. Bull. Amer. Meteor. Soc.101, E1439–E1451,

ESA Climate Change Initiative (GHG-CCI) project:

Copernicus CO2 Report:


This studentship is in the area of data analysis and it does not require lab working and therefore is well suited to be carried out by working remotely. We have already adopted a number of measures for online working and the whole EOS group has very successfully transferred to remote working. We do not foresee any significant impact of COVID-19 on this studentship. Regular supervisory meeting, interaction with other team members and training is taking place via MS Teams and similar; an approach that is working successfully for current PhD students.