= 2009 07 17 GLORIA =
This set of simulations was started exporatory to determine if the 3D simulations also converge when using an initial guess unequal to the true value.

== base ==
{{{
KERNEL_RECOMP = 5
CONV_ITMAX = 50
CONV_DMIN = 1e-4

CG_ITERMAX = 2000
CG_REL_TOL = 1e-3

# Quick error analysis...
ERR_QUICK = 1
ERR_QUICK_REL[*] = 0.3

# Noise error...
ERR_NOISE_ABS[*] = 0.10e-5

# A priori...
ERR_Q[5] = 0
ERR_Q_CZ[5] = 1
ERR_Q_CH[5] = 200

INITIAL_GUESS = ig_val

LAT = 44
LON_START   = -5
LON_END     = 5
LON_SPACING = 0.3

RES_HOR = 3
RES_VER = 64

AZI_LEFT = -45
AZI_RIGHT = 45
AZI_DELTA = 6
}}}

== cz large ==
{{{
< ERR_Q_CZ[5] = 1
---
> ERR_Q_CZ[5] = 10
}}}

== ch large ==
{{{
< ERR_Q_CH[5] = 200
---
> ERR_Q_CH[5] = 2000
}}}
== c large ==
{{{
< ERR_Q_CZ[5] = 1
< ERR_Q_CH[5] = 200
---
> ERR_Q_CZ[5] = 10
> ERR_Q_CH[5] = 2000
}}}

== qerr large ==
{{{
< ERR_Q[5] = 0
---
> ERR_Q[5] = 10
}}}

= Ergebnisse =
Insegesamt schlecht. Gewisse Strukturen ähnlen sich, werden aber i.a. von Artefakten überdeckt.
Im Bereich 52-56 Grade Breite klebt es auf dem initial guess fest.

Bei grossen Höhen funkt die Version mit qerr_large deutlich besser. Bei kleinen Höhen macht es kaum einen Unterschied.

Größere Korrelationslängen habe vermischte Effekte.