The extended Kalman filter for short term prediction of algal bloom dynamics

Abstract

An Extended Kalman Filter (EKF) is incorporated into a water quality model to assimilate high frequency field observations of algal bloom and dissolved oxygen dynamics in a marine fish culture zone (FCZ). The weakly-flushed fish culture zone is modeled as a well-mixed system with numerically determined flushing rate. The ecosystem model incorporates phytoplankton growth kinetics, nutrient uptake, photosynthetic production, nutrient sources from organic fish farm loads, and nutrient exchange with a sediment bed layer. Supported by the high frequency observations (sampling interval δt = 1 day, 1 hour, and 2 weeks for chlorophyll and dissolved oxygen, hydro-meteorological parameters, and nutrient, respectively), a number of algal blooms observed at Lamma Island of Hong Kong are used to assess the performance of the EKF. Daily chlorophyll levels estimated by the EKF are compared with field observations and the unfiltered deterministic model prediction for different algal bloom events. The data assimilation with different observation lead-times is also studied. It is found that the EKF estimate well captures the nonlinear error evolution in time and gives good predictions of short term algal bloom dynamics up to a lead-time of 2 or 3 days. The present study is the first time the Extended Kalman Filter is successfully applied to forecast an entire algal bloom cycle, suggesting the possibility of using EKF for real time forecast of algal bloom dynamics.