Spatial downscaling of ERA5-Land 2-meter air temperature over Greece, using tree-based machine learning algorithms
| datacite.contributor.RelatedPerson | Καζαντζίδης, Ανδρέας | |
| datacite.contributor.RelatedPerson | Κιουτσιούκης, Ιωάννης | |
| datacite.contributor.Supervisor | Αργυρίου, Αθανάσιος | |
| dc.contributor.author | Καΐτσας, Ιωάννης | |
| dc.contributor.other | Kaitsas, Ioannis | |
| dc.date.accessioned | 2025-06-23T06:24:05Z | |
| dc.date.available | 2025-06-23T06:24:05Z | |
| dc.date.issued | 2025-06-20 | |
| dc.degree | masterThesis | |
| dc.description.abstract | This thesis presents a spatial downscaling method for gridded ERA5-Land 2-meter air temperature data over Greece using a tree-based machine learning regression approach. High-resolution air temperature estimates are crucial for regional climate impact assessments and environmental modeling at regional or local scales. The ERA5-Land reanalysis dataset is a promising source, offering high-quality climate data at a relatively coarse, 0.1° (~9 km) resolutio. However, many applications require finer spatial detail. Given the relatively sparse station coverage and the highly variable terrain across much of Greece, spatial refinement (i.e. spatial downscaling) of the ERA5-Land temperatures fields presents a compelling opportunity for generating high-resolution datasets. The proposed framework uses an Extremely Randomized Trees algorithm as the baseline statistical regression model, leveraging the principle of scaleinvariance between the ERA5-Land temperature field and topographic predictors, such as elevation and land cover. The scale-invariance principle assumes that the relationships between temperature and selected environmental predictors remain consistent across spatial scales, which enhances model generalizability and robustness in areas with sparse station data or complex topography. Model validation was performed using observational data from the HNMS meteorological stations across Greece. The results demonstrate moderate improvements in capturing local temperature variability, particularly in complex terrain (e.g. valleys), highlighting the effectiveness of statistical regression techniques, especially machine learning methods, in refining climate reanalysis products in data-sparse regions. | |
| dc.identifier.uri | https://hdl.handle.net/10889/29706 | |
| dc.language.iso | en | |
| dc.rights | Attribution-NonCommercial-ShareAlike 3.0 Greece | en |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc-sa/3.0/gr/ | |
| dc.subject | Statistical downscaling | |
| dc.subject | Machine learning | |
| dc.subject | Tree-based models | |
| dc.subject | ERA5-land | |
| dc.subject | Air temperature | |
| dc.title | Spatial downscaling of ERA5-Land 2-meter air temperature over Greece, using tree-based machine learning algorithms | |
| dc.title.alternative | Χωρική υποκλιμάκωση δεδομένων θερμοκρασίας αέρος 2 μέτρων ERA5-Land στην Ελλάδα, χρησιμοποιώντας αλγορίθμους μηχανικής μάθησης βασισμένους σε δέντρα αποφάσεων |