Microphysics and Fog Group Meeting ‐ November 2nd 2023

Participants: Bjorg Jenny Engdahl, Emily Gleeson, Oskar Landgren, Rolf Heilemann Mhyre, Karl-Ivar Ivarsson, Panu Maalampi, Daniel Martín Perez, Gudrún Nína Peterson, Wim de Rooy, Laura Rontu, Metodija Shapkalijevski.

Presentation by Jenny on slides by Trude Eidhammer on parameter estimation using ML. Link to slides. This method helped with model tuning for WRF but could be used in NWP. Jenny is interested in trying it if there's a project she could be involved in. Several ML methods were tried but neural networks and a Gaussian method were best. Rolf pointed out that care needs to be taken regarding overfitting as many parameter sets can yield the same result. The method can show which parameters have the most impact and where on the globe - can be useful for SPP - Camiel from KNMI is working on something related.

Aristofanis (formerly KNMI now ECMWF) has a paper in preparation on perturbing parameters (correlation/anticorrelation), the type of PDFs used and keeping the values within a realistic range.

Laura Rontu Link to Slides showed slides related to aerosol work and the impact of different aerosol options (Tegen, CAMS climatology MMRS, CAMS climatology MMRs but only in radiation, 3D CAMS, 3D CAMS dust only). A paper on desert dust observations in Finland and SILAM simulations has been accepted. Karl-Ivar's RADSN/RADGR variables are used to include some of the precipitating snow and graupel in the calculation of ice crystal size, which is used in the radiation scheme. **NRADLP **is the liquid droplet effective radius (Re) - by default option 2 is used but option 4 is used to compute Re from the CDNC in the case where the default IFS radiation scheme is used. This setting influences the cloud droplet effective size, to which the radiation schemes are very sensitive. This is the secondary aerosol (Twomey) effect on shortwave radiation. In general, liquid droplet CDNC seems to be the key parameter that causes most of the differences between forecasts where it is derived from aerosol (n.r.t. or climatological) mixing ratios compared to where aerosols do not impact cloud microphysics and cloud-radiation interactions i.e. when Tegen or other AOD input is used (the default case). In ACRANEB2 the CDNC influences microphysics but not the effective radii used for radiation. Laura showed nice total AOD plots for the time when there was a dust intrusion over Finland - Tegen includes background aerosols, CAMS NRT - can see the impact of desert dust. There was less SW in the areas with dust but in general there was more SWD with NRT aerosols. Also showed that there's more snowfall when CAMS NRT are used (similar result to Daniel). The code for using climatological MMRs is only in Laura's CY46 branch. She pointed out that dev is constantly changing (with updates) so that's difficult for developers. Also mentioned that she got differences in convective cloud off the coast of Norway (due to branch differences) - thought it was due to a convection change by Wim but such changes are off by default.

Karl-Ivar Link explained what different cloud overlaps mean and the impacts he got for 2 cases. Note that this cloud overlap mainly affects postprocessing (and low clouds via Canari) and is not the same as what Daniel tested regarding radiation (NOVLP). For example, LRNUMX=T LACPANMX=T WMXOV = 0.5 means that the overlap is between maximum and maximum-random. He used RFRMIN(21) to determine how much solid water species should contribute to the output cloud cover fraction. He tested the reference (random overlap), something between random and max-random, and something between random and max-random but with RFRMIN(21)=1.8 (it's 0.6 by default). Gets a negative bias in total cloud cover when using something between random and max-random but using the new RFRMIN(21) setting mostly compensates for this. Suggests that low clouds are better with less random overlap but not sure if ice clouds should be compensated using RFRMIN(21). Has yet to test exponential overlap. Rolf asked for description of the RFRMIN variables RFRMIN link. Daniel mentioned that even if there's no clouds NOVLP seems to still have an effect - this is strange - needs to be checked. There's no general consensus on what should be used for the cloud overlap. Check what's used in IFS and what Ján Masek used in ACRANEB2. Karl-Ivar mentioned using Clara III data to evaluate clouds and Wim mentioned that we need to start using Cloudnet data.

Meto had done some comparisons of vertical cloud vs model output but did not show the results in this meeting Link.

Panu Link has been looking into the gl visibility parametrization. The current one uses the old CDNC values over land and sea along with the land/sea fractions. This should be used to use the latest values - Panu will look into this. Panu investigated how NRT aerosols could be included also - showed by ECMWF and the UKMO use. Karl-Ivar has also been looking into the visibility parametrization and will send Panu and Emily what he's done for testing. Panu mentioned the Sofog3D campaign and that the report/paper and field measurements will be available for use.

Daniel Link showed slides on the impact of a constant CDNC vs the profiles that are now default and how useful DDH is for identifying the processes that result in changed outputs. For example, there's more snow with the CDNC profile as there are fewer cloud droplets and there is therefore and increase in riming by cloud droplets (other tendancies as seen in DDH output are similar).