IMPROVING WATER QUALITY FORECASTING USING DATA ASSIMILATION
MetadataShow full item record
Population growth increases agricultural, industrial and human activities which threaten the quality of water resources, especially water supply sources such as lakes and reservoirs. In potentially high-impact situations such as algal blooms, active measures such as controlled release from reservoirs may be necessary. To minimize release while meeting the water quality requirements, accurate short-range water quality forecast is necessary. Because watershed water quality models have a large number of state variables most of which are never observed, their initial conditions (IC) are subject to large uncertainties which may propagate into large forecast errors. In this research, a data assimilation (DA) algorithm is developed and evaluated which updates the ICs of the watershed water quality model, the Hydrologic Simulation Program – Fortran (HSPF), based on real-time observations of water quality and streamflow. The water quality observations include streamflow, water temperature (TW), ammonium (NH4), nitrate (NO3), phosphate (PO4), chlorophyll-a (CHL-a), total nitrate (TN), total phosphate (TP), total organic carbon (TOC), biochemical oxygen demand (BOD), and dissolved oxygen (DO). The DA technique used is maximum likelihood ensemble filter (MLEF) which combines the strengths of variational assimilation (VAR) and ensemble Kalman filter (EnKF). In this work, the resulting DA algorithm is developed into a plugin module, referred to as MLEF-HSPF, for the Water Quality Forecast System at the National Institute of Environmental Research (WQFS-NIER). To evaluate the MLEF-HSPF module, hindcast experiments were designed and carried out for a large number of catchments in the four major river basins in the Republic of Korea. To compare the performance of HSPF and DA with a purely data-driven approach, time series modeling was carried out for simulation and prediction for selected catchments. The results show that MLEF-HSPF consistently improves analysis and prediction of most of the water quality variables and streamflow over the DA-less results, but that the improvement varies significantly from catchment to catchment and from variable to variable. Comparisons with time series modeling and prediction show that the incremental value of water quality modeling and prediction using HSPF and DA is rather uneven; it varies significantly across catchments and variables. The findings suggest that there exists large room for improvement in HSPF modeling, including model physics and calibration. Also described toward that end are the factors limiting the performance of DA and the areas of improvement in the end-to-end forecast process to improve watershed water quality modeling and the performance of DA.