In many data assimilation applications, adding an error to represent forcing to certain dynamical equations may be physically unrealistic. Four-dimensional variational methods assume either an error in the dynamical equations of motion (weak constraint) or no error (strong constraint). The weak-constraint methodology proposes the errors to represent uncertainties in either forcing of the dynamical equations or parameterizations of dynamics. Dynamical equations that represent conservation of quantities (mass, entropy, momentum, etc.) may be cast in an analytical or control volume flux form containing minimal errors. The largest errors arise in determining the fluxes through control volume surfaces. Application of forcing errors to conservation formulas produces non-physical results (generation or destruction of mass or other properties), whereas application of corrections to the fluxes that contribute to the conservation formulas maintains the physically realistic conservation property while providing an ability to account for uncertainties in flux parameterizations. The results suggest that advanced assimilation systems must not be liberal in applying errors to conservative equations. Rather systems must carefully consider the points at which the errors exist and account for them correctly. Though careful accounting of error sources is certainly not an entirely new idea, this paper provides a focused examination of the problem and examines one possible solution within the 4D variational framework.
Monthly Weather Review
(2003). The Maintenance of Conservative Physical Laws Within Data Assimilation Systems. Monthly Weather Review, 131(11), 2595-2607.
Available at: http://aquila.usm.edu/fac_pubs/3152