Longitudinal modelling of water levels of the Okavango River
Date
2016
Authors
Unandapo, Lazarus Pendapala
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In statistics, a model is as good as the data fed to it. Data about hydrological events continues
to grow rapidly over the years, with different variables being recorded on a continuous scale.
These variables can be interpreted and used in a different manner among disciplines. Thus,
choosing the right variables and interactions among variables is an important statistical step
in building a good and accurate model.
This dissertation involved the development of a statistical model which can be used to predict
weekly water level within the Okavango river in northern Namibia. The parameters of the
statistical mixed model were estimated based on two methods for longitudinal data, the Generalised
Estimating Equations (GEE) which is a well known method of parameter estimation
in longitudinal data analysis when the observed variables are correlated, and the Restricted
Maximum Likelihood Estimation (REML) which is a likelihood based approach method, unlike
the GEE. Using cross-validation and a simulation study, the GEE method of estimation was
found to be less accurate and inconsistent in terms of prediction of parameter estimation of
water level while the well known REML was found to predict the water level with a good degree
of accuracy, consistency and with lower variance. Parameters from a simulation study have also
shown less bias in REML method and predicted the cross-validation test-set with less bias.
Description
A dissertation submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in fulfilment of requirements for the degree of Master of Science. May 30, 2016.
Keywords
Citation
Unandapo, Lazarus P (2016) Longitudinal modelling of water levels of the Okavango River, University of Witwatersrand, Johannesburg, <http://wiredspace.wits.ac.za/handle/10539/21273>