Please use this identifier to cite or link to this item:
Title: Extracció atmosfèrica dels exoplanetes de la missió Ariel utilitzant mètodes de conjunt basats en arbres de decisió
Author: Ejarque Gonzalez, Elisabet  
Tutor: Ruiz Dern, Laura  
Fors Aldrich, Octavi  
Butturini, Andrea  
Others: Casas-Roma, Jordi  
Keywords: random forest
gradient boosting
ensemble models
random forest
gradient boosting
atmospheric retrieval
Ariel mission
Issue Date: 24-Jun-2023
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: In the last two decades up to 5272 exoplanets have been discovered, which have already begun to redefine our understanding of the formation and evolution of planetary systems. Recently, scientists have turned their attention from detection to the characterization of exoplanet atmospheres. Thanks to upcoming missions such as the recently launched James Webb Space Telescope, and the European Space Agency's Ariel Mission (due for launch in 2029), an unprecedented amount of atmospheric transmission spectra is to be obtained. These data will enable the characterization of the chemical composition and physical properties of these distant worlds. However, current state-of-the-art methods for interpreting atmospheric spectral data are computationally expensive and may pose a bottleneck when processing the expected volume of data to be generated in the coming years. In this context, the field of machine learning is emerging as a promising alternative due to its high flexibility and rapid inference time. Recently, the Atmospheric Big Challenge Database (ABC Database) has been released to the community. This data set simulates the quantity and quality of data that will be measured in the Ariel mission. Seizing the opportunity presented by this dataset, this study explores the use of ensemble models (Random Forest, Gradient Boosting) to retrieve temperature and chemical composition information from spectral data. The obtained results have demonstrated that this family of techniques, applied to the ABC Database, exhibit higher predictive capabilities compared to traditional Bayesian method Nested Sampling. Furthermore, all developed models have shown training times of up to 1.5 minutes, highlighting their computational eficiency. Consequently, ensemble methods based on decision trees emerge as a promising alternative to current methods in handling the large volume of data expected to be acquired in future missions dedicated to the atmospheric characterization of exoplanets.
Language: Catalan
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
TFM_EEjarque_2023.pdfMemòria del TFM2,96 MBAdobe PDFThumbnail
View statistics

This item is licensed under aCreative Commons License Creative Commons