Classification of Environmental vulnerability regions of Marine Aquaculture using Neural Networks
Marine aquaculture is regulated and monitored through international and national legislation. However, there is still a need to improve the monitoring programmes, in particular for those related to the ecosystem approach at larger scales. Most monitoring programmes include examination of the benthic environment and some water quality. Benthic zone is the ecological region at the lowest level of a body of water such as ocean or a lake, including the sediment surface and some sub-surface layers.
There is great potential for marine pollution from marine aquaculture or cage farming fish from discharges of uneaten food and excreted material, resulting in organic enrichment of the sediments below the cages and hypernutrification and oxygen depletion in the water column. There are many models to predict these enviromental problems and also to locate fish cage farming activities in order to minimize these impacts. Recently, development in Geographic Information Systems (GIS) have enabled it to be used a tool for decision-making and policy formulation. There is considerable oppurtunity to develop new modeling techniques within GIS framework to classify and asses the suitability of coastal areas for development of sustainable marine aquaculture. However, locational data sets are often uncertain and incomplete. Therefore new models using soft computing methods such as neural networks, fuzzy logics are more suitable.
In Soft computing, neural networks are a powerful tool for solving complex regression. approximation. system indentification and time series problems. The approach has been applied to problems where domain knowledge is abundant but numerical data have been difficult to obtain. The neural networks have been applied in various and different domains, e.g. control, data analysis and decision support. Many ecological and environmental data are qualitative and combinations of quantitative and qualitative data are so difficul to incorporate in environmental modeling and classification schemes which produce numerical indices of environmental quality. But the use of neural networks can provide a consistent method for incorporating ambiguous quantitative and non-quantitative data into ecological studies.
A neural network system is trained using a sizeable amount of data set and then uses the knowledge to perform necessary prediction or classification. Several initial data inputs are required for model development, including information about hydrodynamic data sets and marine environmental parameters. These primary data are collected as cartographic reference data, hydrographic and bathymetric data and GPS positions of the cages. The seconday data includes physical environmental parameters such as mean current speed, tidal mixing/stratification, quiescence period, sediment granulometry and suitability parameters such as depth, protection zones, maximum current speed, slope which are derived from the hydrodynamic and GIS models.
Training data are selected such that sites with different hydrodynamic characteristics are included. The vulnerability category for these training sites data are classified maually in four different categories such as high vulnerability, medium/high vulnerability, medium/low vulnerability, low vulnerability. These categories are assigned by experts in this field. The training data are mostly obtained from a three-dimensional hydrodynamic models that are calibrated and validated using in situ data distribution, coupled with a GIS. The neural network thus trained is then used to classify new regions with different depth, current speed , quiescence and granulometry.
Thus neural network models, can be used as an effective tool to differentiate areas based on the level of vulnerability. Once identified, different vulnerable areas can then be subjected to use restrictions codes of practise or targeted for more detailed assessment.
