[WRF-UCM] Large bias in 2m temperature despite multiple physics configurations

[ZHOU Wan]

Dear WRF community,

I am running several WRF-UCM 4.6.1 simulations to evaluate near-surface air temperature at 2 meters. Despite testing multiple combinations of physics parameterizations, I still observe a significant positive bias compared to observations, especially during the daytime. The model tends to overestimate maximum temperatures and underestimates nighttime cooling.

Please see the attached figure below, which shows the diurnal variation of 2-meter temperature averaged over multiple stations. The black line represents observations, while colored lines represent four different WRF simulation scenarios.


My Questions:

1. What other physics options or settings would you recommend trying to reduce the overestimation of daytime temperature?
2. Is it possible that the PBL scheme is too diffusive during stable conditions (nighttime), and if so, which PBL scheme is more suitable for urban or semi-urban areas?
3. Are there any known combinations that particularly improve 2m temperature biases in warm or tropical environments?

Any suggestions or insights would be greatly appreciated!

 

[Ming Chen]

Hi Zhou,

What is your model resolution? How did you compare model results with observations (did you interpolate model data to obs sites?) Is this a single site result or regional average over multiple sites? What is the time period of your simulation?

The physics options you used are well tested options in WRF and I don't see any problems. However. WRF performance is often case -dependent, that is, one suite of physics options may perform well for a specific case, but the same suite may yield less realistic simulations for other cases.

 

[ZHOU Wan]

What is your model resolution? How did you compare model results with observations (did you interpolate model data to obs sites?) Is this a single site result or regional average over multiple sites? What is the time period of your simulation?

Hi Ming,

Thank you for your response and helpful comments.

Let me provide more details regarding your questions:
1. Model resolution: My simulation uses a three-domain nested setup with horizontal resolutions of 9 km (D01), 3 km (D02), and 1 km (D03). The results I shared are from the innermost domain (1 km).

2. Comparison with observations: I did not interpolate the model output to the exact station locations. Instead, I directly compared the observed station values with the corresponding nearest model grid points. Do you think this could significantly affect the bias, especially in complex urban environments?
3. Averaging method: The diurnal temperature curves are averaged over multiple observational stations within the innermost domain (D03).
4. Simulation period: The simulations cover the period from August 1 to August 5, 2022. The first day was treated as a spin-up period and excluded from the analysis.

I agree with your point that WRF performance can be case-dependent. I'm currently testing different combinations of physics schemes, but still struggling with the overestimation of maximum temperatures and underestimation of nighttime cooling, particularly in urban/semi-urban settings.

If you have any thoughts on which PBL or surface layer schemes might better represent urban heat dynamics in warm climates, I’d really appreciate your suggestions.

Best regards,
Zhou

 

[haiqingsong]

Turn off UCM，and have a try.

 

[Ming Chen]

The topography in your D03 area shows large east-west gradience, and the 1-km resolution implies that differences in terrain height might be large between adjacent grids. In this case, I have the following suggestions:

(1) when you compare observations with model simulations, please pay attention to terrain height difference between model grid and observation site elevation

(2) Please add the following options to your namelist.input, which mainly affects radiation:

slope_rad (max_dom) = 1, 1, 1 ! slope effects for solar radiation (1=on, 0=off)
topo_shading (max_dom) = 1, 1, 1 ! neighboring-point shadow effects for solar radiation (1=on, 0=off)
shadlen = 25000. ! maximum length of orographic shadow. You may need to adjust this value base don your case

(3) How is the synoptic weather background? If strong winds develop over your domain, the Foen effect may also affect the result.

Hope this is helpful for you.

 

[ZHOU Wan]

Dear Ming Chen,

Thank you very much for your valuable suggestions! I am trying these changes you recommended in the namelist.input.

Additionally, I would like to share the 10-meter wind field from our study area, where the maximum wind speed reaches around 8 m/s. Given the terrain complexity and the relatively strong wind in some regions, I would like to ask if any further treatment or parameter adjustment in WRF is necessary to improve the simulation accuracy?
Looking forward to your guidance.

 

[Ming Chen]

Surface wind simulation is kind of tricky and we are aware of the biases in WRF simulations, especially over complex terrain areas. Can you turn on the namelist option, topo_wind, which conducts topographic surface wind correction. Note that this option only works with YSU PBL scheme and its impact is still case-dependent.
Please let me know whether this option can improve your results. Thanks.