LETKF configuration

All runtime configurable settings for the LETKF are controlled by the letkf.nml file in the experiment root directory. See the file for further documentation, though some of the concepts are clarified here.

##Localization

####Horiztonal localization controlled with sigma_xxx_h. Two values are defined in the namelist, the first is the localization sigma length at the equator, the second is the length at the poles, with values linearly varying in between. This is especially important for the ocean which has a much smaller length scale near the poles.

Each domain can have different localization radii for atmospheric observations (sigma_atm_h) and ocean observations (sigma_ocn_h), since with strongly coupled DA we can assimilate all observations into both domains.

For weakly-coupled DA experiments, only sigma_ocn_h is used by the ocean, and only sigma_atm_h is used by the atmosphere.

####Vertical localization Controlled with sigma_ocn_v (meters) and sigma_atm_v (ln of pressure).

For weakly-coupled DA experiments, only sigma_ocn_v is used by the ocean, and only sigma_atm_v is used by the atmosphere.

For strongly-coupled DA, this is slightly more complicated. So, for example: In the ocean an ocean observation is still only affected by sigma_ocn_v. An atmospheric observation though will be affected by sigma_atm_v, comparing the distance between the observation and the ocean/atmosphere surface, and by sigma_ocn_v comparing the distance between the ocean model level and the ocean/atmosphere surface.

##Inflation

####Multiplicative This is controlled in the namelist by setting infl_mult > 1.0. This is a constant percentage applied at all levels. Can be dangerous with the ocean, causing spread in the deep ocean to increase unbounded.

####Adaptive (Miyoshi, 2011) Controlled by setting infl_mul < 0, the initial value used for inflation is abs(infl_mul) Currently disabled

####Relaxation to prior spread (RTPS) RTPS (Whitaker, 2012) operates by expanding the spread of the ensemble analysis a percentage of the way back toward the background spread. Typical effective values are around 90% . This method has several nice features:

• more inflation in areas of higher observation density
• all variable spread are inflated separately, as opposed to adaptive inflation, where all variable spread are inflated the same percentage if there is not variable localization