Brynx Alegarbes, Neil Que, Jeremy Tan
For our undergraduate thesis and Master's capstone project, we developed artificial neural networks (ANNs) using TensorFlow to predict Marikina River water levels along the Sto. Niño and Montalban water level gauging stations -- two key stations focused on by previous studies. The ANNs were trained on previous water level data and rainfall data from five rainfall gauging stations around the Marikina River Basin: Boso-Boso, Mt. Aries, Mt. Campana, Mt. Oro, and Nangka. We found that the Long Short-Term Memory models performed the best. It predicted water levels at Sto. Niño station using 6 hours of past data with a 0.3% error and predicted Montalban station water levels using 1 hour of past data with a 0.8% error. We also found out that rainfall at Mt. Aries station contributed the most to the rise in water level at Sto. Niño station while rainfall at Mt. Oro contributed the most to Montalban station water levels.