National Technical University of Athens (NTUA)
Land cover and crop type mapping at national scale from multitemporal high resolution satellite data
Abstract
dc:descriptionAccurate and regularly updated land cover and crop type maps are considered essential for several scientific communities as well as for public and regional authorities. Currently, satellite Earth Observation data are considered the main input for the production of such maps thanks to the high spatial and temporal coverage, systematic monitoring and access to inaccessible areas, they can offer. The last decade, open data policies have given free access to an unprecedented volume of high spatial, temporal and spectral resolution satellite data, bringing a revolution in research and operational applications. Nonetheless the seamless exploitation of such Big Data for the production of accurate and reliable maps requires the developmental of efficient, robust and cost-effective pre-processing, classification and mapping frameworks. Towards this end, in this dissertation the subject of land cover and crop type mapping using multitemporal multispectral satellite data of high spatial resolution is thoroughly studied, assessed and discussed. The dissertation’s main contribution is the development, implementation and validation of a mapping framework exploiting Sentinel-2 data for the production of annual land cover and crop type maps at national scale for Greece. For data pre-processing, we have designed, developed and validated an efficient pipeline for the creation of radiometrically, geometrically and temporally consistent analysis-ready time series, while we have also further investigated the contribution of different applied methods and techniques. Various aspects of the mapping framework, i.e., machine learning algorithms, classification features, spatial stratification and validation approaches, have been experimentally assessed towards finalizing the proposed scheme. For map production we implemented a framework employing the Random Forest classifier, selected spectral and auxiliary features and a highly reliable training dataset, created under the scopes of this dissertation for 42, customized to the Greek Landscape, categories. Internal validation of the national classification experiments has documented a very high overall accuracy rate of over 91%. The generalization capacity of the classifier in predicting areas where it has not been trained on was validated with a high overall accuracy rate over 83%, as well. Finally, the produced country-scale land cover and crop type map at 10m with 42 classes, was externally validated by an independent rigorous accuracy assessment procedure that reported overall map accuracy at 82%. To the best of our knowledge there is no other available land cover map for Greece at 10m spatial resolution with such a thematic analysis and that high accuracy and reliability levels. Suggested ideas for the enhancement and further development of the proposed framework raised significant expectations for its expansion and use towards operational land cover and crop type map production at the country scale at regular time intervals, for supporting national and European actions, initiatives and mapping programs, such as the CORINE Land Cover inventory and the EU Common Agricultural Policy partnership.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Karakizi, Christina
- Καρακίζη, Χριστίνα
Subjects
dc:subject × 25- Τηλεπισκόπηση
- Ταξινόμηση εικόνας
- Μηχανική μάθηση
- Παρακολούθηση της γης
- Οπτικά δεδομένα
- Αξιολόγηση ακρίβειας χάρτη
- Remote sensing
- Image classification
- Machine learning
- Earth observation
- Optical data
- Sentinel-2
- Map validation
- Γεωεπιστήμες και Επιστήμες Περιβάλλοντος
- Φυσικές Επιστήμες
- Παγκόσμια και πλανητική αλλαγή, κλιματική αλλαγή
- Ηλεκτρονικοί υπολογιστές στις γεωεπιστήμες
- Περιβαλλοντική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Earth and Related Environmental Sciences
- Natural Sciences
- Global and planetary change, climatic change
- Computers in Earth Sciences
- Environmental Engineering
- Engineering and Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/51551