Other applications of the adjoint model
Singular vectors
- Lacarra and Talagrand (1988): identify those changes to the initial conditions dX0 that will lead to the most important growth at time t0 + T.
- Maximize
-
- Computation of the leading eigenvectors using the Lanczos algorithm
- Does not require an explicit form for L*L
- Requires only to provide L*L v, for any vector v
- Obtained by performing first an integration of the TLM and the results serves as 'initial conditions' for the backward integration of the adjoint model
- Accurate as long as the tangent linear approximation provides an accurate description of the evolution of dX(t)