adjoint model

A model composed of adjoint equations that maps a sensitivity gradient vector, ∇_xJ(t₀) = 𝗟^T∇_xJ(t₁) , from a forecast time, t₁, to an earlier time, t₀, which can be the initial time of a forecast trajectory.

J is some scalar measure of the forecast, 𝗟^T is a linear adjoint operator, and x is the model state vector. An adjoint model can provide a first-order (tangent linear) approximation to sensitivity in a nonlinear model.
See adjoint equation, adjoint sensitivity, tangent linear equation.