Locally,
there has been little or no attempt to address the above-mentioned goal
in the recent past. Much of the focus has been into modeling the system
dynamics of red tide in itself and there has been little need to address
the problem of cost optimization until recently. Moreover, our research
revealed only a number of attempts by foreign institutions in reducing
the number of stations in their own monitoring areas.
In fact, the problems presented above have been dealt with from the past up to the present using the so-called "gut-feeling" approach. Presently, the Bureau of Fisheries and Aquatic Resources (BFAR) are looking into their data and they are manually going about reducing the number of monitoring stations in Manila Bay by 20%. A similar project was performed years earlier by a group of researchers in the Coast of Maine who managed to reduce the number of monitoring stations by up to 40%. Conceptually the procedure for the said reduction entails manually looking at large amounts of data and possibly noting correlation between stations and red-tide toxicity levels within the area of coverage. By how the data appears, the researchers hopefully would be able to identify redundant stations and gaps in the locations of the stations where an additional station should be placed. With the same method the researchers should be able to identify the priority in which the stations should be monitored given an increase in toxicity level noted from a certain monitoring station. The problem with this method is that it is not reliableand often it will take too much time even without the assurance of success at the end of the process. This is mainly due to the indefinitely large dimensionality of the set variables that can or should be considered in the analysis and the amount of data existing for all instances of each variable making it difficult to generalize and thereby arriving at a conclusion from the data.
We are now asked for probable options for solving the problem. First statistical analysis of the existing data set may reveal relationships between the variables under consideration. This can be done through the use of third-party data mining software capable of performing multi-dimensional analysis of the existing data. Generalization can then be performed by eye-looking at the results. We present a second method here in which we can use a software application utilizing Artificial Neural Networks (ANN?s) to perform the necessary generalization of the data. While statistical analysis offers a feasible solution to the problem, it requires an initial knowledge of the relationships between the variables involved. ANN?s on the other hand does not necessarily require knowledge of the variable relationships. Moreover, due to the dynamic nature of the red-tide system, relationships between some variables are not guaranteed to hold on all observation areas implying that different approaches be used when generalizing data from different locations. Whereas ANN?s offers the flexibility of adaptation allowing us to use our software on different test areas with minimal modifications. As a plus, while data mining can, in nature, be used to come up with a predictive models for mapping the progression of red tide over a test area given an ideally complete data set, ANN?s offers the same capability with a higher degree of assurance while not requiring ideally complete data sets and with less human intervention.
The
above mentioned procedures in Manila Bay are taken from an interview with
selected staff members of BFAR and information on the procedures
done in the Coast of Maine were provided by Mr. Don Anderson. As of this
writing, detailed documentation on the above-mentioned manual reduction
procedures are being requested from the said institutions/individuals and
will be available upon receipt.