National and Kapodistrian University of Athens
Parameter estimation methods in multivariate distribution families with applications in remote sensing agriculture and biostatistics
Abstract
dc:descriptionThe topic of this dissertation is parameter estimation methods in multivariate distribution families with applications in remote sensing agriculture and biostatistics. In order to provide a seamless reading experience, the text is divided into three parts. Part I focuses on the methodological advances made in the Dirichlet and multivariate gamma distribution families. Part II demonstrates the developed methodologies in three distinct case studies, two agricultural and one biostatistical. Finally, Part III is dedicated to software implementation, detailing the development of two R packages designed to make the proposed methodologies readily available to the broader research community. Chapter 1 covers the foundational framework in parametric statistics that deals with independent and identically distributed (iid) observations, presumed to be drawn from a distribution family with some unknown parameters requiring estimation. Chapter 2 covers the parameter estimation methods for the Dirichlet distribution family. The new score-adjusted moment estimators are presented, and their asymptotic variance is compared both analytically and numerically to that of the MLE and the ME. Two versions of the classic moment estimators are described and compared against each other. Chapter 3 covers the parameter estimation methods for the multivariate gamma distribution family. The new score-adjusted moment estimators are presented, and their asymptotic variance is compared numerically against that of the MLE and the ME. Some bias correction results are also included. Finally, some preliminary results for the matrix gamma distribution family are presented as well. Part I is concluded with Chapter 4, in which the cumulative link mixed-effects model (CLMM), a particular type of ordinal generalized mixed-effects linear model is revisited and adapted for use in the specialized application settings of Part II.Chapter 5 introduces the first and most extended application of the dissertation, which concerns the estimation and prediction of crop progress, i.e. the percentage of crops that have reached a certain phenological stage over a large area, using satellite data. In this context, two applications of the methodology developed in Part I are demonstrated: a constant in-time estimation of crop progress with the Dirichlet SAME, and a full growing season modeling with the CLMM. Chapter 6 highlights an application of the multivariate gamma SAME in a similar small-scale application, which monitors the growth of Arabidopsis Thaliana plants using camera imagery. Part II is concluded with Chapter 7, which concerns a one-year cohort study conducted in Greece, tracking the quality of life (QoL) of patients who have experienced a stroke at three time points: initially (one week after the stroke), in the post-acute phase (six months after the stroke), and in the chronic phase (one year after the stroke). A number of important underlying health conditions like diabetes and hypertension are tracked by the study, along with several demographic characteristics such as marital status and education level. The study's aim is to present a clear and concise methodological approach by employing binomial and ordinal logistic mixed-effects linear models in order to quantify the impact of both the treatment and these factors. This research is a collaboration with the Nursing Department of the University of West Attica. Chapter 8 focuses on the joker package, which provides a comprehensive set of features for probabilities and mathematical statistics. It extends the range of available distribution families and facilitates the computation of key parametric quantities, such as moments and information theoretic measures. The main focus of the package is parameter estimation through maximum likelihood and moment-based methods under an intuitive and efficient framework. All package features are available both in a stats-like syntax for entry-level users, and in an S4 object-oriented programming system for more experienced ones. New distribution families, such as Dirichlet and multivariate gamma, not included in the stats package are made available. Parameter estimation is performed analytically if possible, while numerical optimization of the MLE (whenever required, e.g. the beta and gamma distribution families) is performed with computational efficiency, taking advantage of the score equation system to reduce the dimensionality of the optimization problem. Finally, focusing on research reproducibility, Chapter 9 introduces the Ages of Man R package ecosystem, which streamlines the crop and satellite data download and editing processes. This project can be used to apply the CLMM methodology to all available crops for all US states.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Oikonomidis, Ioannis
- Οικονομίδης, Ιωάννης
Subjects
dc:subject × 14- Οικογένεια Κατανομών Dirichlet
- Εκτιμητές Ροπών Score Adjusted
- Μοντέλο Αθροιστικού Συνδέσμου Μικτών Επιδράσεων
- Πρόοδος Καρπών
- Dirichlet Distribution Family
- Score-Adjusted Moment Estimators
- Cumulative Link Mixed-Effects Models
- Crop Progress
- Μαθηματικά
- Φυσικές Επιστήμες
- Στατιστική και Πιθανότητες
- Mathematics
- Natural Sciences
- Statistics and Probability
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/59145