Forecasting of Monthly Closing Water Level of Angat Dam in the Philippines: SARIMA Modeling Approach

Abstract

 The operation and management of Angat Dam as a multipurpose dam for domestic, irrigation, power, and flood control purposes, is governed by the operation rule curve of dam water level. This study was conducted to understand the behavioral pattern and to provide short-term forecast of the monthly closing water level of Angat Dam using the Seasonal Autoregressive Integrated Moving Average (SARIMA) model, considering the data from January 1990 to December 2021. Series decomposition revealed the absence of an overall trend component but the presence of seasonality in the dataset. An almost perfect partial correlation between the closing water level of a certain month and the preceding two months’ records was observed in the correlogram. Different SARIMA models were evaluated and subjected to diagnostic checking and based on minimum Akaike Information and Bayesian Information criteria, SARIMA (1,0,1) (0,1,1)12 model with estimated coefficients of φ1 = 0.8050, θ1 = 0.2278, and Θ1 = -0.999 was selected to forecast the monthly closing water level of Angat Dam. The model fits with a root mean square error (RMSE) of 4.79 meters, mean absolute error (MAE) of 3.45 meters and coefficient of determination (R2) of 0.93. On average, the forecast water levels of the best SARIMA model are off by around 1.8% of the actual value.

Keywords: SARIMA, autocorrelation, forecasting, Angat Dam, invertibility, stationary