The high nonlinearity and discontinuity of landsurface model and radiative transfer model hinder further practical operation of some advanced four dimensional data assimilation methods such as the Kalman filter and the variational method in the land data assimilation system. Accordingly, we develop a new data assimilation algorithm by employing a heuristic optimization approach named simulated annealing, which is capable of minimizing the four dimensional cost function without using the adjoint model. The method has advantages in dealing with the strong nonlinearity and discontinuity, and in finding the global minimal in the hilly structure of the cost function. Additionally, all the processes in the model operator and the observation operator can be kept because the method is independent on the cost function. The disadvantage of this method, when compared to the variational method, is its low efficiency. Therefore, we make efforts to improve the method by incorporating very fast simulated re-annealing(VFSA) algorithms into the data assimilation cycle.
    Based on the VFSA algorithm, we design a research purpose land surface data assimilation system which assimilates the in situ monitoring of soil moisture into a land surface scheme. The modified SiB2 with frozen soil parameterization is used as the model operator. We have implemented one dimensional offline test of the algorithm with the GAME-Tibet observations. The algorithm is compared with a control run without assimilating the observations. The results show that the cost value calculated from the optimized initial values is much smaller. In addition, the bias from observations is also significantly reduced.