Assimilating multi-source data into a three-dimensional hydro-ecological dynamics model using Ensemble Kalman Filter

Accurately predicting spatio-temporal patterns of algal bloom is important and also challenging. This study developed a three-dimensional hydro-ecological dynamics model (3DHED) to predict cyanobacterial biomass in lakes and applied Ensemble Kalman Filter to assimilate multi-source data into 3DHED for model improvement. The model was applied in Lake Taihu, using in-situ measurements and remote sensing (RS) retrievals. Two data assimilation experiments (named EnKF1 and EnKF2) were conducted. EnKF1 assimilated only in-situ measurements, while EnKF2 assimilated both in-situ measurements and RS data. The results revealed that 3DHED simulated the spatio-temporal patterns of cyanobacterial biomass in Taihu with an acceptable Index of Agreement (IOA). EnKF1 significantly improved the model fitness and increased the IOA of 85% measurement sites to 0.85, especially better captured the peak values. Compared with EnKF1, EnKF2 gave more improvements in spatial patterns besides model fitness, implying that assimilating multi-source data was helpful to improving the model performance.