Taal Lake is one important freshwater resource of the residents of the province of Batangas,island of Luzon, the Philippines. Aside from its inherrent livelihood and food source, the lake is an interesting subject of study because of its uniqueness. It is believed to have been created by volcanic eruptions during pre-historic period. It is actually a caldera which acts as a catch basin for the provinces of Batangas, Laguna, and Cavite. The water outflows through one and only river into the West Philippine Sea. In the middle of the lake sits an island called the Volcano Island. On the Volcano Island is a volcano named Taal Volcano. The volcano's creater formed another lake named the Yellow Lake (or the Main Crater Lake). Thus, there is a lake (the Yellow Lake) within a volcanic crater (Taal Volcano) on an island (the Volcano Island) in the lake (Taal Lake) in an island (Luzon). See this satellite image of the area courtesy of wikipedia (http://upload.wikimedia.org/wikipedia/commons/f/f0/Taal_Volcano_satellit...).
The residents whose livelihood rely on the marine ecosystem of the lake suffer from periodic fish kill events (FKE). Residents raise freshwater fishes Nile tilapia (Tilapia nilotica) and milkfish (Chanos chanos) in floating cages on the lake. What is very interesting is that FKE is so localized that it could affect a cage but nearby cages, as near as 5m between cage edges, will not be affected. Philippine limnologists, ecologists, marine scientists, and computer scientists have banded together to try to create a model that will predict the onset of FKE seven days before the event to provide the residents an early warning so that they can harvest the fish or move the cages to safer locations. They have recently found out that FKE are due to any or a combination of the following causes:
- Exposure below a certain concetration of dissolved oxygen (DO) beyond the fish' recovery time - Low DO may be caused by the mechanical stirring of the lake water due to the action of the wind on the lake surface or extreme water temperature difference. At the worst case, extreme weather may induce lake overturning. FKE have been recorded in higher frequency during the changing of the trade winds in May and in October each year: From northeasterly to southwesterly in May, and vice-versa in October. Higher frequency of FKE is also recorded during the coldest month of January wherein the colder water surface sinks to the bottom while the warmer water bottom floats up to the surface bringing along with it during its rise water with extremely low to zero DO.
- Sulfur upwelling from thermal vents located at the bottom of the caldera - Due to the highly dynamic nature of the local streamflows and eddies in the lake, fish cages above thermal vents are not necessarily affected by the extreme concentration of sulfur. The fishes die almost instantaneously upon contact with water with high concentration of sulfur. Similarly, thermal vents spew extremely high water temperature which causes lowering of DO concentration along the vent's path.
Because of the highly-dynamic nature of the lake and the very-localized FKE, scientists have resulted into modeling FKE through a Bayesian Network using weather data (i.e., lake overturn or stirring is induced mainly by weather). Other machine learning approaches are also being explored but the data scientists among them are faced with datasets of rare events. That is, the negative examples (no FKE) are far more frequent than the positive examples (with FKE). So the computer and data scientists are also looking at developing learning techniques for rare events modeling. The social scientists in collaboration with the computer and data scientists are also looking into "social sensors" exploiting the almost real-time availability of Twitter feeds from the residents of Taal. These scientists infer the polarity of sentiments in the tweets and say that when the negative polarity outnumbers the positive or neutral ones, there is almost a certainty that FKE will occur in less than a week. This is similar to what had been done in real-time tracking (a.k.a. nowcasting) of earthquake propagation in Japan, of paths of hurricanes and typhoons, and of expanse of propagation of allergy epidemics.
There may be computational and procedural techniques that others may have employed as adaptive responses to their own respective FKE, but this is so far what have been done in Taal Lake in the Philippines. Employing a 3D network of physico-chemical sensors in the lake whose readings maybe used as inputs to a process-based predictive model may actually give a definitive answer to the sought-for early warning system. This approach, however, is far more expensive annually than the losses incurred during an FKE, so Philippine scientists are hell-bent in creating ways to come up with models that do not require procuring, fabricating, operating, and maintaining expensive sensor networks.