slv.inversion.coverage.covered_state

slv.inversion.coverage.covered_state(problem, percentile)

The state without its poorly covered cells: those below percentile of coverage.

A cell’s coverage is its absolute sensitivity, summed over every obs and time, per obs location (divided by their number). Cells below the percentile-th percentile of all cells’ are left out at every time, which keeps the prior error’s Kronecker structure; every other block of the state is kept whole. The problem of the kept state is problem.restrict(keep), which holds the left-out cells at their prior (their modeled enhancement moves into the offset); post.estimate.combine_first(problem.prior) puts the estimate back on every cell.

Parameters:

• problem (Problem) – The problem, with a flux block in its state and concentration in its obs.

• percentile (float) – The coverage percentile below which a cell is left out (10 drops the least-sensitive 10%).

Returns:

The state to keep.

Return type:

Space