Assimilation of satellite-derived profile observations for enhanced estimation of water quality variables in a deep...

Accurate and reliable estimation of the temperature and Dissolved Oxygen (DO) profiles in deep reservoirs is crucial for effective water quality management. Integrating incoming observations through a data assimilation scheme can ensure the precision and certainty of the predictions. As surface water quality variables cannot reflect the overall conditions of deep reservoirs, assimilating vertical profile observations is essential. This study pioneers the use of the ensemble Kalman filter to incorporate satellite-derived profile observations into CE-QUAL-W2, a two-dimensional hydrodynamic and water quality simulation model for deep reservoirs. This method was used to update the initial conditions of the temperature and DO vertical profile in the Wadi Dayqah Reservoir in Oman. Five different scenarios were evaluated to identify the optimal choice of state vector and assimilated observations and to evaluate the effect of cross-covariance between temperature and DO in multivariate data assimilation. Additionally, the approach was further refined by examining the performance of assimilating surface observations, observations from two stations, fewer data points in the vertical profile, and the effect of the ensemble size. The results indicate a significant improvement in model accuracy after data assimilation. The optimal scenario-updating and assimilating both the temperature and DO vertical profiles-reduced the root mean square error (RMSE) of the DO profile estimation by 44 %, from 1.37 to 0.77 in the segment near the dam. The RMSE decreased by 27 % for temperature, dropping from 0.66 to 0.48. Overall, across the entire reservoir, improvements were 11 % for temperature and 27 % for DO. Assimilating surface information was adequate only on dates when thermal stratification was absent. Additionally, as the correlation between consecutive vertical layers can be accurately estimated by ensembles, proper performance is achievable even with fewer observations in the vertical profile.