Abstract:
Surface water quality plays a significant role in maintaining healthy ecosystems and supporting both human and aquatic life, making reliable monitoring and forecasting essential due to increasing pollution and environmental changes. The objective of this study is to forecast the Weighted Arithmetic Water Quality Index (WAWQI) at 20 sampling sites along the River Thames using three time series models: Single Exponential Smoothing (SES), Holt’s (Double) Exponential Smoothing (HES), and Holt-Winters (Triple) Exponential Smoothing (HWES), and to evaluate and compare the forecasting performance using three accuracy metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) to identify the most suitable model for sites with varying temporal patterns (stable, trending, or seasonal). Water quality data collected from 20 sampling sites along the River Thames between March 2009 and September 2017 were arranged chronologically. Each model was applied after linearly interpolating missing data and splitting the series into training and testing sets in an 80:20 ratio. The results revealed that model performance varied depending on the temporal patterns of WQI data. SES performed well at sites with stable conditions, such as TC8, TC12, TC13, and TC17. TC8 recorded the lowest RMSE (1.49), MAE (1.12), and MAPE (1.68%), indicating high forecasting accuracy. The HES model accounts for both the level and trend components of time series data and generally does not outperform the SES model for most sites, since most of the sites lack trend components. At TC20, the HES model showed the highest accuracy, with RMSE of 2.11, MAE of 1.38, and MAPE of 1.96%.
HWES achieved the best performance across the majority of monitoring sites, particularly those exhibiting clear seasonal patterns in WQI fluctuations. In contrast, volatile sites (TC15) resulted in higher forecast errors (MAPE >15%) regardless of the model applied. These findings suggest that model selection should consider the underlying temporal characteristics of WQI behavior at each site: HWES for seasonal patterns, HES for trending series, and SES for stable conditions. These insights can aid water management authorities in proactive pollution control and sustainable resource planning by enabling accurate water quality forecasting.
