DSpace
 

KNUSTSpace >
Research Articles >
College of Science >

Please use this identifier to cite or link to this item: http://hdl.handle.net/123456789/11941

Title: A EMPIRICAL PERSPECTIVE ON ENVIRONMENTAL. ANTECEDENTS: PART 11- STOCHASTIC AR (1) MODELING FOR RETROSPECTIVE PREDICTION OF THE ENVIRONMENTAL QUALITY OF THREE RIPARIAN COMMUNITIES ASSOCIATED WITH Two DAMS IN KUMASI, GHANA
Authors: Tetteh, Isaac K.
Frempong, Emmanuel
Awuah, Esi
Issue Date: 2004
Publisher: Nova Science Publishers
Citation: Nova Science Publishers, Environmental Research Journal Volume 6, Number I
Abstract: This paper describes the systematic approaches used for making retrospective prediction (retrodiction) of the environmental quality (EQ) of three downstream riparian communities associated with two dams in Kumasi, Ghana. The first step consists in applying parametric and non parametric cross validations (CVs) as model selection tools to evaluate their predictive powers over specified time lags (i.e., 1957-+ 1958, 1966 -+ 1967, 1972-+ 1973, 1999-+ 2000), using the environmental quality indices (EQls) and operational activity indices (OAls) constructed in the Paper I, as essential ingredients. An attractive component of the entire modeling processes is that the use of the indices which had been generated using an orthogonalization process via a standard non-rotated empirical orthogonal function (EOF) analysis circumvents the problem of multicolinearity in the CV process. However, constrained by short data records, the model selection methods were enhanced using computer-assisted simulations (for generating "artificial data" based on random numbers), fed by the indices. The inclusion of "artificial data" was also intended to put the two models on the same evaluation platform. By limiting the number of predictors (OAls) to three, the "curse of dimensionality" is sidestepped, thus ensuring a good fit. The predictive powers of the parametric and nonparametric models were evaluated based on a part of the calibration datasets via a resampling CV technique, an approach recommended elsewhere for dealing with limited data. The CV outcomes indicated that the nonpararnetric model demonstrated a more capability than the parametric. and thus better fits the data. The inherent error detected in one of the scenarios for the parametric model signified that it would have been a superfluous activity had we further embarked to -evaluate the performance of the two models using the likelihood-ratio test we had planned for. The nonparametric outputs were then selected for the AR (I) model development. to evaluate the communities' EQ over the four scenarios. This model theoretically and consistently depicted highly positive first-order Markovian dependencies in the EQ of three out of the four lead year predictions, based on EQI (predictand) I baseline conditions. The magnitudes of these dependencies are a manifestation of deteriorations in the EQ (i.e., 1957---.1958, 1972---.1973, 1999---.2000), on the basis of high environmental persistence. These results further demonstrated that the communities' EQ, to a large extent. was driven by the dams' operational activities (OAs), the most important event flagged when the two dams operated simultaneously, which further supports our hypothesis in the Part I study. However, the non-correlation between 1966 and 1967 EQ depicted randomness, which could not be associated with the operational impacts of the dames). However. we suspect a dominant role of natural variability over the operational impacts of dames) as of that time. The lack of observational data to realistically evaluate the models was one of our challenges. However, the magnitudes of the model biases, coupled with low RMSEs. demonstrate acceptable model performances, which have helped us gauge, to a reasonable extent, the communities' EQ, under different time lags, and under different phases of the dames).
Description: An article published by Nova Science Publishers
URI: http://hdl.handle.net/123456789/11941
Appears in Collections:College of Science

Files in This Item:

File Description SizeFormat
ARTICLE_4.pdf9.11 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - Feedback