MODIS simulator detection limits (effective radius bounds)

[brandonduran]

Following #110 , changes to the MODIS effective radius joint histograms were implemented to account for large effective radii simulated by GCMs that may exceed values seen in observations. However, it appears that these new bounds will not fully capture all GCM clouds. Specifically, it appears like re_ice_max and re_water_max in cosp_modis_interface.F90 are set to the old bin boundaries (90 and 30 microns, respectively).

As far as I can tell, these bounds inform the range of values in trial_re_i and trial_re_w, which trickle down to the effective radius size determination in modis_simulator.F90. I'm wondering if re_ice_max and re_water_max should be adjusted to match the new changes introduced in #110 ? It is my impression that despite the updated bin ranges designed to capture large radii clouds potentially found in GCMs, the joint histogram will not report any cloud within this largest bin range due to the size determination process (interpolate_to_min on line 208 of modis_simulator.F90)

My own testing with CAM6 has shown that while cloud droplets rarely exceed the threshold of the old bin boundaries (30 microns), ice crystals above the old bin boundaries (> 90 microns) are not negligible. However, with the current setup of the MODIS simulator, 0% CF is reported in the new largest bin due to what I've outlined above. Somebody with a more detailed understanding of the MODIS simulator might be able to chime in on this more effectively than I can!

[alejandrobodas]

@brandonduran thanks for reporting this. I believe your interpretation is correct, but @RobertPincus should be able to confirm.
Given that we are extending the range, I think we may also need to increase num_trial_res in modis_simulator.F90 to keep the original size of the intervals used in the size retrieval.

[brandonduran]

@alejandrobodas @RobertPincus I think my suggestion for these changes would be to double num_trial_res to 30 and double re_ice_max and re_liq_max as well, as this would preserve the original size of the intervals used while extending to larger droplet size ranges. In theory, we could increase num_trial_res by any series of factors (double, triple, etc.) and change re_ice_max and re_liq_max, but a doubling should be sufficient to maintain the original code behavior and also handle potential edge cases in GCMs.

Happy to implement these changes if this is deemed a suitable solution.

[RobertPincus]

@brandonduran Thanks for continuing to press on this. I'm not sure that extending the range of drop sizes considered is the correct path here.

The main issue is that the MODIS retrievals do not report any values of ice particle radius larger than 60(?) microns. If we want the MODIS simulator to be consistent with the observations the simulator needs to respect this limit.

There's a technical issue, too, in that the polynomial fits for single scattering albedo used in the pseudo-retrievals (i.e. on line 616 of the MODIS simulator) were fit to the range used in the observations, so wouldn't necessarily be accurate at larger sizes

[brandonduran]

@RobertPincus thanks for chiming in.

Following the discussion from #109 and #110 , it seemed like recording all GCM-simulated clouds somewhere in the histogram was desirable behavior (and follows the setup of the ISCCP simulator). So these changes would go somewhere in the same direction for the MODIS simulator. My perspective is that when we want the simulator histograms to stay 1-1 with the observations, we can still have that, particularly by maintaining the same bin ranges as the observational histograms. Model-obs comparisons can still be conducted, while also pointing to where GCMs may be simulating unrealistic cloud distributions.

The technical problem does seem a bit challenging to work around, especially given that we couldn't use observations to calculate new fits since the satellite does not observe above the current bounds. Maybe we can brainstorm more on potential fixes if the above argument for extending the bounds is reasonable?

[alejandrobodas]

@brandonduran and @RobertPincus thanks for shedding some light on this. I agree that we should try and keep consistency with the observations and keep the limit to 60 micron (or whatever the exact limit is). I'd be in favour of somehow keeping a record of the GCM clouds with ice particle size retrieval. Perhaps there is already an existing cloud fraction variable that could be used to do that?