Improved sensible and latent heat flux estimation of community land model by using ensemble Kalman filter assimilation

Accurate estimation of variability of the surface heat fluxes is very important in the hydrological, meteorological, and agricultural applications. Community land model (CLM) may be used to continuously predict the temporal and spatial sensible and latent heat flux, however, its output is contaminated by uncertainties of the model’s parameters, model structure and forcing data. The aim of this paper is to improve the sensible and latent heat flux prediction of CLM by using data assimilation schemes with Ensemble Kalman Filter (EnKF) algorithm. The data assimilation results are compared against eddy-covariance observations collected at three sites (Arou, Guantan and Yingke) in the northwest of China including grassland, forestland, and cropland cover types. The CLM usually overestimates the sensible heat flux while underestimates the latent heat flux at the three observation sites. The comparison results indicate that data assimilation method improves the estimation of surface sensible and latent heat fluxes from the model, with clear reduction in the resulting uncertainty of estimated fluxes. The average reductions in the RMSE and MAE values of all sites are 40.62 and 25.83 W/m2 while the average decline of MAE values were 33.80 and 26.92 W/m2 for sensible and latent heat fluxes, respectively, while the most significant reductions in the RMSE values are 67.70 and 30.40 W/m2 for sensible and latent heat flux with EnKF algorithm, respectively. Although this study clearly implies that the assimilation of sensible and latent heat fluxes EnKF algorithm has the potential to improve the surface heat fluxes predictions of CLM, further research is required to make definitive conclusions when assimilation of sensible and latent heat fluxes derived from real remote sensing data into CLM. Furthermore, the good approximation of the model and measurement errors and the assimilation multi-source data simultaneously into the CLM may produce better results.