Remote Sensing Driven Seasonal ARIMA Approach for Temporal Analysis and Forecasting of Water Quality Parameters

Water pollution is a major global concern, threatening ecosystem and public health. The National Capital Territory (NCT) of Delhi, India, is a prime example of this challenge, where water potability has become a critical concern. The Yamuna River, which traverses Delhi, along with various groundwater sources, is frequently found to be heavily polluted with industrial effluents, untreated sewage, and agricultural runoff, making the water unsafe for consumption and harming aquatic life. In this study, we used remote sensing technology to do a thorough assessment of water quality parameters over a multi-year period in an effort to address the serious issue of water pollution in Delhi. To extract important water quality parameters, we used satellite data from Sentinel-2, a popular platform for environmental monitoring. Turbidity, suspended sediment concentration (SSC), dissolved organic matter (DOM), chlorophyll-a concentration, normalized difference water index (NDWI), and normalized difference vegetation index (NDVI) are these parameters. Our objective in studying such indicators was to offer a comprehensive comprehension of the temporal and geographical fluctuations in the water quality throughout the area and making predictions based on this study. We used the Seasonal Autoregressive Integrated Moving Average (SARIMA) model, a popular machine learning method for time-series analysis. Future pollution levels were forecasted using SARIMA models, which provided insightful information about possible trends and aided in the development of water management plans. In order to assess the accuracy and dependability of the predictions, our method involved evaluating different combinations of SARIMA models using evaluation metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). In addition to these metrics, we utilized the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to compare the performance of different SARIMA models. Our findings demonstrate that the SARIMA models optimized for each water quality parameter exhibit varying levels of performance. For NDWI, the model with an order of (2, 1, 2) and a seasonal order of (1, 1, 1, 12) performed best, achieving the lowest MSE of 0.0019, RMSE of 0.0438, and MAE of 0.0339, with the lowest AIC of - 153.06 and the second lowest BIC. Similarly, the NDVI model, with the same order and seasonal configuration, showed strong results, with an MSE of 0.0034, RMSE of 0.0582, and MAE of 0.0458, along with the lowest AIC of - 125.78 and the second lowest BIC. The turbidity model, using an order of (1, 1, 1) and a seasonal order of (2, 1, 1, 12), achieved excellent performance with the lowest MSE of 0.00045, RMSE of 0.0212, and MAE of 0.0168, alongside the lowest AIC of -225.56 and the second lowest BIC. The chlorophyll-a model demonstrated superior accuracy with the lowest MSE of 0.00035 and AIC of -238.37. However, the DOM model, despite having the lowest AIC of -235.73, showed a wavy forecast trend with a decline over time, which deviates from the historical data. This suggests potential issues with overfitting or underfitting, indicating a need for further refinement or alternative modeling approaches to improve the reliability of DOM forecasts.