Streamflow data assimilation in SWAT model using Extended Kalman Filter

- The Extended Kalman Filter is applied to SWAT model for streamflow assimilation.
- A ratio method is suggested to reduce the computation burden.
- Opposite patterns are found in the soil moisture and curve number update.
- Case study shows the high flood forecast is improved moderately.
- A quasi-error update scheme based on EKF issues more accurate forecasting.

SummaryThe Extended Kalman Filter (EKF) is coupled with the Soil and Water Assessment Tools (SWAT) model in the streamflow assimilation of the upstream Senegal River in West Africa. Given the large number of distributed variables in SWAT, only the average watershed scale variables are included in the state vector and the Hydrological Response Unit (HRU) scale variables are updated with the a posteriori/a priori ratio of their watershed scale counterparts. The Jacobian matrix is calculated numerically by perturbing the state variables. Both the soil moisture and CN2 are significantly updated in the wet season, yet they have opposite update patterns. A case study for a large flood forecast shows that for up to seven days, the streamflow forecast is moderately improved using the EKF-subsequent open loop scheme but significantly improved with a newly designed quasi-error update scheme. The former has better performances in the flood rising period while the latter has better performances in the recession period. For both schemes, the streamflow forecast is improved more significantly when the lead time is shorter.