I applied the transmissivity difference over the entire day to make sure it has a smooth profile. I did a run over April period. The run is not fully finished yet (11 to 22 is finished), but here are the results. The first two plots are original and modified Kdown.

from wrf-suews.

better to be resolved at the SUEWS side.

from wrf-suews.

from wrf-suews.

It turned out WRF overestimates Kdown in our London case as well; sorry I overlooked this part and thought it had been resolved.

Given the results about AOD and Kdown in Shanghai and Beijing, I suggest for our WRF-SUEWS evaluations we patch SuMin specifically for WRF-SUEWS instead of SUEWS for the following concerns:

1. we need to apply thresholds of transmissivity at the grid/site scale;
2. transmissivity has apparent seasonality (i.e., intra-annual variability, see below from Ao et al (2016)), suggesting specific thresholds should be applied for a short period (e.g., ~1 month);

3. SuMin is designed for WRF-SUEWS simulations of shorter simulation periods (e.g. ~10 days) compared with standard alone SUEWS, which is usually run for years.

As such, we can apply the thresholds in SuMin to correct the Kdown going into SUEWS kernel while keep current SUEWS as is.

However, for the long run, we may consider to introduce a chemical forcing file (e.g., hourly AOD measurements) that allows dynamic correction/adjustment of Kdown in a more responsive way.

Please let me know your thoughts.

If the SuMin approach sounds sensible, I'll start the implementation.

from wrf-suews.

I think we should first compare the transmissivity for the model and observations for the study periods. I.e. use the data we already have

from wrf-suews.

results of transmissivity for London during four simulation periods

month	transmissivity	diff=sim-obs
Jan
Apr
Jul
Oct

from wrf-suews.

If we use a Kdn threshold - what is needed to remove the switch in the morning hours? (ie. from negative to positive difference)

Please could you now plot the differences normalised by time after SR (i.e. SR to SS=1) for the four periods. Please plot the four different periods (with the shading ) with one normalised time access

from wrf-suews.

Hi @suegrimmond,
I'm exploring different options for Kdown threshold but the results are not ideal.
I guess your second part is to generalise diff_tau as a function of t_SR, which I'm working on now.

from wrf-suews.

from wrf-suews.

That's a good point! re: zenith angle.
My quick tentative conclusion for this is that a single cutoff might not be necessary to correct the modelled Kdown.
I'll keep on posting new results that are sensible for further discussions.

from wrf-suews.

Maybe we can use a zenith angle as the starting time of the correction.

from wrf-suews.

The overestimated Kdown by WRF seems to be a well-known issue that has been discussed in many papers, some of which are the following:
https://dx.doi.org/10.1002/2016jd025527
https://journals.ametsoc.org/doi/10.1175/MWR-D-15-0262.1

from wrf-suews.

London results

Note:

1. t^~ is the time since sunrise normalised by length of day.
2. the error bars show the confidence level (95%) of mean bootstrapped values.

Transmissivity

dif = sim - obs

from wrf-suews.

Swindon results

Note:

1. t^~ is the time since sunrise normalised by length of day.
2. the error bars show the confidence level (95%) of mean bootstrapped values.

Transmissivity

dif = sim - obs

from wrf-suews.

from wrf-suews.

I have started implementing this. However, I think we need a more sophisticated method to modify kdown rather than a threshold (for example after midday) because it leads a jump in kdown when we kick in the modification.

from wrf-suews.

I see this in a simple implementation I did

from wrf-suews.

The modification is done in the interface of WRF-SUEWS, so the original SUEWS code is not changed!

from wrf-suews.