For a while now, I’ve been trying to figure out why fog frequency in BARRA-C2 (which uses CASIM bimodal cloud fraction scheme) is very different depending on whether you define fog using visibility or fog fraction (both are outputs). I think I may have solved it, but I was hoping I could get a sanity check (because I’m not a model developer or microphysics expert and this is like another language to me). I am trying to interpret this paper with particular focus on sections A6-A8.
My current interpretation is that fog fraction uses the cloud fraction scheme and visibility is based on information from the aerosol activation scheme. Now, because BARRA-C2 uses climatological aerosols (i.e. not super accurate input data for an hourly variable), the visibility parameter (based on the activation scheme) is affected by this poor input data but the cloud fraction scheme is unaffected.
I have modified the schematic in the paper to show what I mean
Some quick thoughts off the top of my head - happy to look into things in more detail with you or pass questions onto relevant people all the way up to Paul Field.
Although CASIM has the ability to be used in a mode where cloud droplet number is calculated from aerosol activation, in its operational configuration (the default RAL3p3 configuration, i.e. the default setting in rAM3 too) it uses a prescribed cloud droplet number concentration (this is the “CASIM-2M” configuration discussed in the paper). This means that the when condensation of water vapour to liquid occurs in the microphysics scheme, it is agnostic to aerosol - the same number of cloud droplets are created at the point of condensation at all grid boxes everywhere in the model that condensation occurs. (This of course means that you can overpredict cloud in supersaturated but aerosol-limited regimes).
I can’t see anything in the BARRA-C2 Bureau research report that says CASIM is coupled to aerosol. It would be first worth checking with Chun-Hsu what the CASIM settings in BARRA-C2 were, in order to rule your hypothesis in or out as a possibility.
I’m not familiar with the UM visibility parameterisation (happy to find out the relevant physics code and have a look into it after the Easter break), but I suspect what is going on is that you’re not comparing like-for-like quantities. Visibility is essentially a measure of the scattering of light due to all contributing scattering particles. One of the things that contributes to scattering will be low cloud (i.e. fog), but there are also other factors that reduce visibility e.g. aerosol content and species in unsaturated air (think dust!), humidity (humidity swelling of aerosol in unsaturated air leads to reduced visibility). So visibility and fog are not equivalent quantities. You should however be able to see fog fraction variability as a signal in visibility variability - something you could check?
Are the aerosol climatologies monthly? If so then you should also be able to see a step change in the visibility every month?
UPDATE: I’ve just had a quick search for the UM visibility scheme and this paper isn’t the original paper for the scheme but does a good job of describing the factors that in Section 1.1
So overall, my initial guess is that you’re seeing these differences because the fog fraction definition only depends on near-surface cloud, which is unaltered by aerosol quantities in the CASIM-2M configuration that I suspect was used in BARRA-C2, whereas the visibility calculation is a function of not just (liquid and frozen) cloud but also of pressure, temperature, specific humidity and aerosol mass.
Found the original visibility scheme paper, linking it here for completeness. (Fun fact, Pete Clark was my external examiner for my PhD thesis and viva!).
The visibility scheme (used in the ACCESS City model at BoM, but I assume is similar in BARRA-C2) is explained here. It says “the scheme operates by hydrating the provided aerosol field so that it is in equilibrium with the atmospheric humidity” which is where I am thinking the climatological aerosols come into it. Also the visibility time series at individual locations is very cyclical.
The problem is that the technical document I’ve linked above and other evaluation studies describe fog fraction as ‘the fraction of the grid box with visibility below 1km’ [section 3.4], which implies that visibility is used in the fog fraction calculation. I’ve spoken to people at BoM who said the same thing so this may have been true in earlier versions of fog fraction but it sounds like it’s not the case in RAL3? But yes, I should double check with Chun-Hsu
Check with Chun-Hsu which mode CASIM was run in for BARRA (prescribed CDNC or aerosol-activation), and
Get the BARRA-C2 suite and look at the code in the parameterisations in question, to find out exactly what is being calculated.
“the scheme operates by hydrating the provided aerosol field so that it is in equilibrium with the atmospheric humidity”
Aerosols do come into the visibility scheme for sure. (As relative humidity increases, aerosol particles absorb water and grow hygroscopically, which makes them larger optical scatterers ==> visibility is reduced. Once the RH reaches critical supersaturation, this particle is then activated into a cloud (or in this case, fog) droplet). So you’d expect visibility to have a daily cycle that follows diurnal RH variation.
Do you see a step change in the visibility field when the climatology changes month?
describe fog fraction as ‘the fraction of the grid box with visibility below 1km’
==> maybe fog fraction isn’t coming out of CASIM cloud fraction but is diagnosed elsewhere. We can track though the model code after the long weekend if we can get hold of the BARRA-C2 suite’s source.
It could also be worth sending an email to Belinda Roux to ask if she is aware of the difference between the diagnostic quantities.
Have you tried plotting a timeseries of visibility and precipitation? Visibility may include other things than fog, including rain (depending on which STASH code it is).
Ok, I tracked down the STASH code for visibility from the BARRA-C2 paper here.
Page 67 shows it is STASH m01s03i247. The description BoM give for this variable is as follows (note that it does not include precip, and there is caution for it’s use in Australia, that it should be considered “experimental” because it has not been tuned for Australia, and is not dynamic)
names['m01s03i247'] = {
'bom_name': "visibility",
'bom_description': "Estimated horizontal visibility at approximately 1.5m above-ground-level assuming no precipitation. Calculated from 1.5m T, q, qcl, qcf using droplet growth equations and assuming standard values for aerosol concentration, density, dry radius, etc. Please note that this diagnostic has not be tuned for Australian conditions and does not include dynamic atmospheric particulate densities, which is an important part of the algorithm used to generate this field. At this stage this field is considered experimental and should be used with due caution.",
'stash_long_name': "VISIBILITY AT 1.5M M",
'stash_code': 3247,
'stash_help': "Calculated from 1.5m T, q, qcl, qcf using droplet growth equations and assuming standard values for aerosol concentration, density, dry radius, etc, found in comdeck C_VISBTY. Further details in: Wright, B. J., 1997: Improvements to the Nimrod Visibility Analysis/Forecast System. FR-Div. Tech. Rep., No. 217. Wright, B. J., 1997: A New Visibility Analysis/Forecast System for Nimrod. Met. Office FR Tech Rep., No. 222. Available on all timesteps.",
'stash_description': "VISIBILITY AT 1.5M M",
I wouldn’t assume the visibility scheme matches between original ACCESS City and BARRA-C2, as there is quite a distance of time between the two systems.
Regarding precip, for defining foggy hours, I did filter timesteps when precip>0.2mm to exclude low visibility due to precip which other fog studies do too. But it looks like it’s done anyway in the scheme
Sorry I meant the visibility equation rather than the scheme.
I’m only evaluating it for this study rather than using it for applications because my results definitely agree with that statement.
There’s something even more important to note about the BARRA-C2 STASH code description. This description says that it is diagnosed from temperature, specific humidity, liquid cloud content and frozen cloud content.
This parameter is the maximum distance at which objects can be clearly seen, at 1.5 metre above the surface. The visibility can be influenced by fog, rain, snow and other weather phenomena.
Whereas the BARRA-C2 STASH code description says that no dependence on precipitating hydrometeors (rain and snow) is taken into account in the calculation.
In my mind these two pieces of BARRA-C2 information contradict each other.
@Kim_Reid, if you could plot those timeseries of visibility and precipitation that Mat mentioned, that would show if there’s any covariability or not.
Next up in my opinion is locating the visibility scheme and seeing what variables the code itself uses.
To provide closure to this discussion, I went on a deep dive through the BARRA-C2 production suite source code and the end finding is that the original discrepancies that Kim was seeing turned out to be an artefact of the analysis method applied to the BARRA data rather than discrepancies in the model code or its output fields.
i.e. there is no major discrepancy in fog frequency in BARRA-C2 depending on whether you define fog using visibility or fog fraction.
For anyone who lands here because they are searching for the details of the visibility scheme and how it works:
The visibility scheme sits external to the microphysics scheme, and is called from the boundary layer diagnostics module.
The calculations of the visibility field and the fog fraction field are done by the visibility scheme, and both calculations are passed the same set of the model’s prognostic fields: temperature, pressure, specific humidity, liquid cloud water content, and aerosol mass (in the case of BARRA, this is climatological aerosol).
The fog fraction calculation also uses critical relative humidity and a ‘visibility threshold’ (a distance), because the fog fraction scheme first calculates the total water content required for the visibility to be less than the visibility threshold, then uses that total water threshold to then calculate a grid-box fog fraction.
There are two visibility fields that the scheme calculates: visibility without precip (which is just based on the scattering coefficients calculated from the hydrated aerosol field) and visibility including precip, which takes scattering coefficients from rain into account as well as the hydrated aerosol.
In the BARRA-C2 production suite, only the former field (no precipitation effects) was written as model output. (So BARRA-C2 visibility shows no covariation with precip). This has now been noted on the BARRA parameter description webpage.
In contrast, the (current) default release of ACCESS-rAM3 writes out the visibility field that accounts for scattering by precip.
Referring back to Kim’s original diagram:
fog fraction is calculated in the visibility scheme, and is not done by the microphysics cloud fraction calculation.
visibility and fog fraction are calculated in the same place (the visibility scheme), are external to the microphysics scheme, and use the same set of input fields from the model, including the same aerosol information.
the visibility scheme is called from the boundary layer diagnostics module (i.e. during the boundary layer call).
