COSP Satellite Simulator

COSP turns the model's cloud and precipitation fields into what a satellite instrument would have measured, so that a simulation can be compared with the satellite record on the instrument's own terms rather than through model-native cloud fraction. ClimaAtmos implements the CloudSat branch: a 94 GHz cloud radar, whose reflectivity depends on hydrometeor size and concentration and is attenuated by gases and by the hydrometeors themselves [80].

The simulator is off by default. It is switched on by giving the dt_subcol configuration key a finite value, which sets how often it runs, and it is configured through two further keys:

The CloudSat outputs need the hydrometeor masses and number concentrations of the one-moment scheme, so they require microphysics_model: "1M". With "2M" the subcolumns are still generated but no reflectivity is computed, and the outputs are left undefined; any other microphysics model is rejected when the simulator runs.

Why subcolumns

A grid column carries one cloud fraction per level and says nothing about how the cloudy parts of neighboring levels line up. A radar beam, by contrast, passes through one realization of that overlap. The simulator therefore samples each grid column into $N$ subcolumns, each of which is either fully cloudy or clear at every level, such that the average over subcolumns reproduces the grid-mean cloud fraction and the vertical alignment follows the chosen overlap assumption. The sampler is the SCOPS generator of COSP: a threshold recurrence that proceeds from the model top toward the surface, placing cloud where a per-level random draw falls below the cloud fraction. Under maximum overlap, cloud in adjacent levels is stacked; under random it is independent; maximum_random stacks contiguous cloudy layers and decorrelates across clear gaps.

Precipitation is sampled the same way, from selectors shared with the cloud draw so that rain and snow fall from the subcolumns that hold cloud. The random seed is fixed, so the subcolumns are reproducible across calls and across restarts.

Subcolumns are generated and used one at a time. The reflectivity of each is computed and folded into the running statistics before the next is drawn, so the memory footprint is that of one column, not of $N$ columns.

From hydrometeors to reflectivity

For each subcolumn the simulator assigns the cloud liquid, cloud ice, rain, and snow contents of the grid mean to the levels that the sampler marked cloudy or precipitating, then diagnoses a characteristic particle size for each species from the grid-mean state: a monodisperse radius for cloud liquid, from the prescribed droplet number, and the inverse Marshall–Palmer slope for cloud ice, rain, and snow, from the same size-distribution parameters the microphysics uses.

Radar optics follow Quickbeam, the COSP radar module. The one-way gas attenuation at 94 GHz is computed once per grid column from temperature, pressure, and water vapor, and integrated from the model top to give the two-way path attenuation at each level. The hydrometeor backscatter and attenuation are then evaluated per subcolumn, the path attenuation accumulated downward from the top, and the result expressed as the attenuated equivalent reflectivity $Z_e$ in dBZ.

Statistics

Three quantities are accumulated over the subcolumns of each grid column:

The detection window is CloudSat's: $-30$ dBZ is the minimum detectable signal, and $10$ dBZ the level above which the retrieval saturates. The CFAD bins follow COSP's hist1D convention, with lower edges inclusive and upper edges exclusive.

cloudsat_tcc and cloudsat_tcc2 are available as output diagnostics; requesting either in a run without COSP or without one-moment microphysics raises an error naming what is missing. cfadDbze94 is accumulated in the cache but not yet exposed as a diagnostic.

Where this is implemented

See Configuration Options for the keys named here and Computing and saving diagnostics for requesting the outputs.