Prediction of Philippine rainfall using artificial neural network model

Rainfall is a crucial component of the hydrologic cycle. Accurate prediction is thus essential in increasing the efficiency of agricultural production. This study was conducted to help predict the incoming monthly rainfall in various parts of the Philippines. This was done by relating rainfall data to the sea surface temperature and southern oscillation conditions in the tropical Pacific. Monthly rainfall data was obtained from 74 stations (preliminary analysis), and 37 selected stations (modified approach). Sea surface temperature and southern oscillation indices of the tropical Pacific were also used to estimate incoming rainfall depth in the Philippine through artificial neural network (ANN) models. Results of the preliminary analysis showed that clustering can be applied in the Philippine's rainfall situation as five rainfall patterns could be observed. However, the study revealed that clustering based only on rainfall patterns could be observed. However, the study revealed that clustering based only on rainfall patterns was not enough as basis for the grouping. Clusters obtained poor performances in the model evaluation thus resulting to poor rainfall prediction. Because of these, a modified approach method was used in the study which individually analyzed the stations' characteristics. Results of the modified approach showed that all the 37 selected stations had good performances in the model evaluation, 19 of which were found capable of predicting rainfall values which were close to the actual values obtained from January to December. The other 18 stations proved to have an improved predicting capability once the months with poor predicted values were removed. From these results, ANN models may thus be used to predict incoming monthly rainfall in various parts of the Philippines using sea surface temperature and southern oscillation indices of the tropical Pacific.