Aristotle University Of Thessaloniki (AUTH)
Assessment of soil quality indicators using novel Earth observation technologies and computational intelligence methodologies
Abstract
dc:descriptionTo safeguard natural capital of soils, in a continuously volatile landscape of reduced resources and environmental changes and pave the way for the development of evidence-based conservation recommendations for policies and sustainable services for relevant economic operators, it is essential to improve capacities for soil health monitoring by adopting multidimensional and integrated approaches. In this context, the advent of the Earth Observation (EO) big data era, spearheaded by free, full, and open spaceborne data policy, as well as the emergence of novel computational algorithms and cloud computing resources, pave the way to enhance the mature level of knowledge regarding the topsoil monitoring. The goal of this doctoral thesis is to develop methods able to estimate chemical and physical properties of exposed soils, exploiting EO imagery data and leveraging large soil spectroscopic archives in combination with advanced computational intelligence algorithms. More concretely, the thesis puts forward the development of a novel approach from the family of memory-based learning algorithms, able to utilize the combined spectral sources and geographical proximity for improved VIS-NIR-SWIR soil properties estimation. The proposed algorithm makes use of the complementary information contained within the different spectral sources that originate from them to accurately generate more relevant local calibration datasets and yield enhanced accuracy results. Following that, a novel spiked bottom-up approach for soil mapping is proposed which is designed with the rationale to enhance the predictive performance of the regression algorithms by integrating existing soil spectral libraries and EO imagery data. The proposed techniques are applied in a standardized large soil spectral library that has been developed to cover the significant gaps in the Balkans, Middle East and North African region. Subsequently, considering that deep learning models have currently been a game changer for Earth system monitoring, a set of convolutional neural networks are proposed. The proposed deep learning frameworks showcase concrete advantages over the other machine learning algorithms that have been proposed in the literature, since they are capable of combining the complimentary information contained in the pool of both optical and radar spectral information and those from auxiliary geographical coordinates. Moreover, the deep learning models have been developed in a way that allows a degree of interpretability to explain their reasoning process.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tziolas, Nikolaos
- Τζιόλας, Νικόλαος
Subjects
dc:subject × 30- Τεχνητή νοημοσύνη
- Βαθιά μάθηση
- Μηχανική μάθηση
- Φασματοσκοπία εδάφους
- Μεγάλα δεδομένα
- Δορυφορικά δεδομένα
- Ερμηνεύσιμη τεχνητή νοημοσύνη
- Συνελικτικά νευρωνικά δίκτυα
- Ψηφιακή χαρτογράφηση εδάφους
- Φασματικές βιβλιοθήκες εδαφών
- Artificial intelligence
- Deep learning
- Machine learning
- Soil spectroscopy
- Big data
- Satellite data
- Explainable artificial intelligence
- Convolutional neural networks
- Digital soil mapping
- Soil spectral libraries
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Γεωπονία, Δασολογία και Αλιεία
- Γεωπονικές Επιστήμες και Κτηνιατρική
- Εδαφολογία
- Computer and Information Sciences
- Natural Sciences
- Agriculture, Forestry and Fisheries
- Agricultural and Veterinary Sciences
- Soil Science
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/52114