Water temperature modeling through regression and artificial neural network using short-term data for a Philippine River

Abstract

ABSTRACT Predicting river water temperature is vital for monitoring water quality and managing water resources, yet tropical countries such as the Philippines often lack adequate monitoring systems. This study addresses this gap by developing predictive models using linear regression and artificial neural networks (ANNs) with short-term data from the Sapian River. The multiple linear regression model, which included same-day air temperature (AT), AT from 2 days prior, and the time of year, performed better than simpler models. Results show Nash–Sutcliffe efficiency (NSE) of 0.63 and root mean square error (RMSE) of 0.56°C, suggesting that incorporating lagged AT enhances model accuracy compared to the best simple linear model, using only same-day AT, with NSE of 0.61 and RMSE of 0.57°C. The ANN model, which utilized daily AT, the previous day's temperature, and time of year, was the most effective, achieving the highest NSE of 0.69 and the lowest RMSE of 0.51°C. This demonstrates its capability to capture non-linear relationships in the data. This research pioneers river water temperature modeling in the Philippines, offering baseline data and methodologies for future studies. The findings highlight the importance of comprehensive approaches and improved monitoring systems to better manage river ecosystems.