Bachelor Thesis Co-Relator - Luigi Russo

Bachelor Degree in Electronic Engineering for Automation and Telecommunications, University of Sannio, Engineering Department, 2021

Development of a Machine Learning model based on the “categorical boosting” technique for the correlation between tropospheric NO2 and NO2 on the ground


Earth observation is based on the application of the principles of remote sensing to the monitoring of the earth’s surface, air and sea, mainly for scientific and environmental applications. Remote sensing generally refers to the detection and measurement of characters and quantities conducted from afar by means of sensors, generally located on aerial or space platforms. But this is a very modern definition.

Initially everyone’s thought, especially in times of war, was to be able to see “over” the hill. It is precisely for this reason that at the beginning the goal of the pioneers of remote sensing was to give their armies the opportunity to spy on the enemy, to know his weapons, to anticipate his moves. Balloons were used for this.

But all this did nothing but increase the curiosity of some people who began to see observation from above from other points of view. Photographers tried to go higher and higher to develop mind-blowing images and to have unique views of the world around them. The invention of the airplane increased this curiosity even more and it is precisely in this period that remote sensing also began to deal with environmental and territorial problems. It was understood that having an overview of a large area, remote sensing and the use of specific platforms could help find specific solutions.

And the birth of the first satellites has actually allowed us to have more and more extensive images with a good level of detail and with many possibilities of use in the most varied applications. In this thesis work we wanted to explore the application process of remote sensing and its techniques for the analysis of a case of interest.

In this regard, I would like to specify that part of this thesis relating to the study of remote sensing was conducted with my colleague Simona Reale. In particular, the work was divided into four fundamental chapters:

  • Chapter 1: after a detailed introduction to the discipline of remote sensing, the various techniques of earth observation, and of acquisition of quantities of interest, with the related sensory tools, parameters and solution applications have been analyzed.
  • Chapter 2: the Copernicus mission was discussed, and in particular the Sentinel-5P was discussed.
  • Chapter 3: the discussion concerning Machine Learning and the working principle of automatic learning has been developed. In particular, various techniques have been analyzed such as Random Forest, Support Vector Machine and Categorical Boosting. The latter in particular has played the role of founding tool to solve the problem faced in this thesis work.
  • Chapter 4: is the chapter concerning the application case. It fulfills the objective of the entire study, namely that of “translating” the data detected in the troposphere through the Sentinel-5P into the data available on the ground.