Modelling of environmental parameters in lake environmnets using neural networks

• Κύριος συγγραφέας: Χατζησολωμού, Αικατερίνη
• Άλλοι συγγραφείς: Παπαθεοδώρου, Γεώργιος
• Μορφή: Thesis
• Γλώσσα: English
• Έκδοση: 2020
• Θέματα: Environmental parameters; Neural networks; Lakes; Περιβαλλοντικοί παράμετροι; Νευρωνικά δίκτυα; Λίμνες; 577.63

Aquatic pollution is one of the most serious environmental issues that the human kind has to deal, especially the last decades. Therefore, the aquatic environments are suffering a great environmental pressure as well. Since the lakes are having about the 90% of the Earth’s liquid freshwater, their proper management is a crucial task in order to maintain their good water quality. Eutrophication is one of the major problems that are observed in the limnetic environments. The eutrophic water is related with a series of environmental problems like anoxia, the existence of harmful cyanotoxins etc. Therefore, this study is focused on the assessment of the environmental parameters that are related with the phenomenon of the eutrophication. The mathematical models are capable to contribute positively to the restoration of the good water quality status of a limnetic environment. The mathematical models can act as a mean to describe several environmental processes with the use of mathematical relationships. Several types of mathematical models have been applied into the area of environmental sciences, including the artificial neural networks. During the last decades the artificial neural networks have been applied successfully into the field of limnology. It has been documented that the artificial neural networks are superior to other modelling techniques (e.g. multiple linear regression model). This is attributed to the fact that the artificial neural networks can simulate with good correlation the complex non-linear relationships that are describing the environmental processes. Therefore, several categories of artificial neural networks are applied in this study. The artificial neural networks can be divided into two main categories, those with supervised learning and those with non-supervised learning. In Application 1 and Application 2 supervised artificial neural networks are developed aiming to predict the chlorophyll parameter. Afterwards several sensitivity analysis (one way-sensitivity analysis) algorithms are applied in order to evaluate the environmental parameters with the biggest impact on the artificial neural network, which managed to simulate well (produced small error) the chlorophyll parameter. The synergistic effect of the parameters (two-way sensitivity analysis) is calculated with the use of the “PaD2” algorithm. In Application 3 a self-organizing map (SOM), which is a non-supervised artificial neural network, is presented. Based on that SOM neural network it is possible the limnological data to be clustered and at the same time the interactions among the environmental parameters to be examined. Furthermore, the statistical methods of Principal Components Analysis and Cluster Analysis are applied and their results are compared with the results produced by the SOM. The results of the Principal Components Analysis and Cluster Analysis methods are with agreement with those produced by the SOM, although the SOM is a superior modelling method because of its advanced visualization abilities to assess the parameters interactions.