A very fast simulated re-annealing (VFSA) approach for land data assimilation

We develop a new data assimilation algorithm by employing a heuristic optimization approach named very fast simulated re-annealing (VFSA), which is capable of minimizing the cost function without using the adjoint model. The method is independent on model operator and observation operator, and has advantages in dealing with strong nonlinearities and discontinuities. Based on the VFSA algorithm, we design a land data assimilation scheme which can assimilate both the in situ measurement of soil moisture, and the passive microwave remote sensing observations into a land surface model. The scheme is tested with GAME-Tibet observations and TMI brightness temperature. Results show that it works successfully with strongly nonlinear land surface model, SiB2 and the radiative transfer model of moist soil. The cost value calculated from optimized initial state is much smaller and the bias from observations are also significantly reduced. r 2004 Elsevier Ltd. All rights reserved.