Important: minor update to 4D-Var algorithms

[arango]

This is a minor update to the 4D-Var algorithms:

• The routines load_ADtoTL_tile and load_TLtoAD_tile in ini_adjust.F were updated to include the multiplication by the land/sea masking before manipulating the adjoint and tangent linear solutions, respectively. This is done to suppress a leak from the land/sea boundaries into the interior solution when the horizontal convolutions are applied. This is needed for W4DPSAS with RPCG to give the same solutions as I4DVAR. Many thanks to Andy Moore for bringing this to my attention.

• Added the missing arguments to several routines called in normalization.F when the option GEOPOTENTIAL_HCONV is activated. The horizontal convolution of the error covariance along geopotential is experimental and not fully tested. Please avoid using this option for now until we advise its usage.

• Added logic in read_asspar.F to make sure that Lprecond is false in input script s4dvar.in for the dual formulation 4D-Var drivers (W4D-PSAS and W4D-Var) when the RBLanczos minimization (RPCG) is activated.