Ali, Mohd Zahit (2007) The application of the artificial neural network model for river water quality classification with emphasis on the impact of land use activities: a case study from several catchments in Malaysia. PhD thesis, University of Nottingham.
Several methods of river water quality assessment techniques have been introduced. Among the most commonly used are the water quality index system and classification scheme. These two systems are designed to simplify the huge amount of water quality data down to its simplest form, while retaining the essential meaning of the information. They offer the means for measuring the effectiveness of pollution abatement strategies by comparing the status of water quality both temporally and spatially. In this way, it is useful for management purposes, especially in determining priorities for resource allocation and planning of new development areas.
The water quality index system and the classification schemes currently available, however have some limitations in their structural design. They often exhibit inherent loss of information, are complex and may involve subjective judgement in their interpretation. However, because of the critical issues on water pollution and the scarcity of water resources, these systems are being applied despite of these limitations. The current situation is that, different countries are applying different models of water quality assessment system. Based on the limitations of the existing assessment systems, it is appropriate to explore other approaches that can be more flexible, robust to noisy data, and adaptable to new form of environmental data, in order to provide direct and prompt results for classifying of river water quality. One avenue for research is that based on Artificial Neural Network (ANN).
Artificial Neural Network comprises of several techniques. One of this technique that is widely being used is the Back-Error Propagation (BEP). BEP of ANN was used in this research in conjunction with the Interim National Water Quality Standard (INWQS) data for Malaysia. The findings of the study shows that the classification results based on the evaluation of the water quality variables were good when compared with the results obtained from other water quality classification models, which include: the Department of Environment Water Quality Index (DOE-WQI), the Harkins'-WQI, Mahalanobis Distance Classifier, Maximum Likelihood Distance Classifier and the Decision Tree Classifier. The accuracy for BEP of ANN was found to be 86.9% and correlated well with all of these five models. The highest correlation was, with the Mahalanobis Distance Classifier and the DOE-WQI. The analysis on sensitivity shows that the BEP of ANN was sensitive to Dissolved Oxygen, a condition similar to the DOE-WQI model.
Comparisons were made with two types of BEP of ANN architecture, a simple network with less number of hidden nodes and a relatively complex network with more hidden nodes. It can be concluded from the analysis that a small and simple network performed well with large samples and with test data that are widely distributed than the complex network with more hidden nodes.
Using the same model, different approaches were used in evaluating the classification of water quality were applied, such as the used of the land use variables and hydrological features (LUVHF) to replace the water quality data. Using these variables, the performance of the BEP of ANN in classification of water quality was low (24% and 31%). However, its performance can be improved, if more samples with wider range of LUVHF were used.
Throughout this study, the BEP of ANN model has shown some remarkable achievements. In view of these, several knowledge contributions have been made. The first contribution is the flexibility of the system approach and operationally simple to perform. Secondly, it provides a practical approach in classification of river water quality, such that through a single network computation of a sample, the results are presented promptly as the probability value and the class grade value. The third contribution is that the water quality can also be classified based on the land use variables and hydrological features, without dependence on water quality data. This approach is suitable for remote areas, where accessibility is relatively difficult.
|Item Type:||Thesis (PhD)|
|Uncontrolled Keywords:||water quality, rivers, Artificial Neural Network, ANN, Interim National Water Quality Standard, INWQS|
|Faculties/Schools:||UK Campuses > Faculty of Social Sciences, Law and Education > School of Geography|
|Deposited By:||Ms. K EVANS|
|Deposited On:||16 Mar 2011 09:33|
|Last Modified:||16 Mar 2011 09:33|
Archive Staff Only: item control page