Aristotle University Of Thessaloniki (AUTH)
Soil organic matter prediction considering moisture content using spectral data and machine learning techniques
Abstract
dc:descriptionThis dissertation explores the potential of VNIR-SWIR spectroscopy as a cost-effective method for Soil Organic Matter (SOM) prediction in Mediterranean soils with low Organic Matter (OM) content. The research emphasizes the need for robust spectral libraries representing various soil strata and highlights the impact of different multivariate methods and spectral pre-processing techniques on SOM prediction accuracy. More specifically it was found that for laboratory measurements the results give high accuracies irrespective of the multivariate method used. However the preprocessing method plays an important role mainly for machine learning techniques rather than PLSR which provident more constant results with smaller deviations. The study also underscores the significant influence of soil moisture on spectral data and SOM predictions. It is suggested employing separate calibration models for different moisture levels and incorporating appropriate correction techniques to enhance prediction accuracy. It became evident that models that were developed from air dried and sieved samples could not be utilized to predict soil samples with moisture contents even if SM was relative low. The second part of the thesis is dedicated to SOM predictions utilizing hyperspectral sensors both airborne (HySpex) and the recently launched PRISMA satellite sensor. Results were very promising for both sensors with HySpex showing slightly better prediction accuracies. Regarding the model validation HySpex had more consistent results irrespective of the method used. On the other hand PRISMA showed better and more consistent results with the use of PLSR and Cubist. Furthermore, the study explores the potential of auxiliary variables from satellite radar data, (Sentinel_1) that marginally improved the models’ accuracies emphasizing the importance of a detailed analysis and larger datasets for their effective incorporation into soil property prediction models. In conclusion, the research suggests meticulous calibration, validation, and consideration of site-specific conditions in utilizing soil spectroscopy for accurate and reliable SOM predictions. Additionally, the study emphasizes the role of earth observation data, in understanding the spatial distribution of SOM, offering valuable insights for sustainable land management and addressing global challenges such as climate change and food security. Future research directions include evaluating the proposed approaches with larger datasets, incorporating additional soil-related variables, development of common protocols and exploring detailed soil mapping for agricultural monitoring.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Angelopoulou, Theodora
- Αγγελοπούλου, Θεοδώρα
Subjects
dc:subject × 16- Φασματοσκοπία ανάκλασης εδάφους
- Τηλεπισκόπηση
- Δεδομένα παρατήρησης γης
- Οργανικός άνθρακας
- Εξόρυξη δεδομένων
- Soil reflectance spectroscopy
- Remote sensing
- Earth observation data
- Organic carbon
- Data mining
- Γεωπονία, Δασολογία και Αλιεία
- Γεωπονικές Επιστήμες και Κτηνιατρική
- Γεωπονία, διεπιστημονική προσέγγιση
- Agriculture, Forestry and Fisheries
- Agricultural and Veterinary Sciences
- Agriculture, interdisciplinary
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/54987