Forecast of Algal Bloom Dynamics in a River Using the Ensemble Kalman Filter
Abstract
A forecasting framework of algal bloom in a river channel was developed by employing two numerical models coupled in serial order to simulate a watershed and the main river channel. The HSPF model simulates flow discharge and water quality from the watershed and the EFDC model takes the results as inputs to simulate river channel hydrodynamics and water quality. The ensemble Kalman filter (EnKF) was applied for data assimilation in the framework, linking uncertainties of inputs, model structures and observations. The input precipitation and water quality from point sources were perturbed to form ensemble members of state variables. The perturbation generated the ensemble of flow discharges and concentrations of water quality variables including Chl-a from HSPF model simulations, which were fed into the EFDC model. Stochastic forcing terms to simulate model structural errors were added for both models. Also an error distribution was used to address the measurement error. The framework was applied to theYoungsan River watershed, which is located in southwestern area of the Korean Peninsula (about 3,470 km2). The weather forecast data provided by Korean Meteorological Agency, consisting of combined data from the UM-Regional and UM-Global models, were used through the forecasting simulations. The HSPF model was calibrated for the observed flow and water quality data of 2009 to 2010 and the EFDC model for the data of 2008 to 2009 before they are used in forecasting. Data assimilation was conducted with weekly Chl-a data sampled along the river channel by updating Chl-a concentrations on the EFDC model grids. To keep mass conservation of water quality constituents, the other variables are also updated correspondingly, considering the changes due to growth, death and respiration of algae. The results show that EnKF works efficiently for updating spatial distribution of Chl-a concentrations. However, since the ensemble inputs at the boundary locations produced by the HSPF model are not updated, the effect of the update along the main channel is washed out after a couple of days by the inputs from tributaries, which is the limitation of the current study. This defect is expected to be decreased when flow residence time increases by two weirs recently constructed or when more frequent updating is possible by using real-time data from the automated water quality stations newly installed on the river.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2011
- Bibcode:
- 2011AGUFM.H53G1493K
- Keywords:
-
- 1871 HYDROLOGY / Surface water quality;
- 1910 INFORMATICS / Data assimilation;
- integration and fusion;
- 1922 INFORMATICS / Forecasting;
- 1952 INFORMATICS / Modeling