You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
quickbeam_subcolumn: hydrometeor attenuation at gate 2 (d_gate=1) neither accumulates gate 1 nor uses its own layer spacing #163
Location:src/simulator/quickbeam.F90, quickbeam_subcolumn, hydrometeor-attenuation accumulation for the d_gate==1 direction (master 5eb05e5, lines ~122–130):
if(d_gate==1) then
! dheight calcuations assumes hgt_matrix points are the cell mid-points.
if (k>2) then
! add to previous value to half of above layer + half of current layer
a_to_vol(pr,k)= a_to_vol(pr,k-1) + &
(kr_vol(pr,k-1)+kr_vol(pr,k))*(hgt_matrix(pr,k-1)-hgt_matrix(pr,k))
else
a_to_vol(pr,k)= kr_vol(pr,k)*(hgt_matrix(pr,k)-hgt_matrix(pr,k+1))
endif
For k = 2 the else branch executes, so gate 2's attenuation
does not accumulate gate 1's a_to_vol(pr,1), and
uses the layer spacing below it, hgt(2)-hgt(3), instead of hgt(1)-hgt(2).
This is inconsistent with (a) the trapezoid recurrence used for k>2, (b) the gas-attenuation sum immediately below, which special-cases only k=1 (if (k>1) accumulates from gate 2 onward), and (c) the d_gate==-1 branch, which also special-cases only the single starting gate (k<ngate accumulates). The expected value is
Impact: for a spaceborne radar with TOA-first input the first two gates are near the model top where hydrometeor attenuation is ~0, so the error is usually negligible in CloudSat products — but the offset it introduces propagates unchanged to every gate below whenever the top two layers do attenuate, and the surface-radar direction does not have the equivalent behaviour, so the two orientations are inconsistent.
Found while building a JAX translation of COSP (jax-cosp), whose default accumulates gate 2 with the trapezoid rule (a legacy switch reproduces the current behaviour for regression comparison).
Activity