The manuscript describes a comparison between a statistically “realistic” 3D model with a more realistic and a parameterised model using a one day one city setting.

Given that the models are fed in with camera temperature data, and that the “true” values are based on the comparison amongst the models, would it not be better to feed the models with cloud cover and SW incoming data, calculate fluxes and temperatures, and contrast them to the temperature provided by the camera inputs? This approach would provide the possibility of a temperature forecast based on historic cloud and temperature forecasts, which would improve the ability of the modelling approach from just confirming past temperature profiles, to predicting possible extreme temperature scenarios. Given the availability of temperature profiles, I would suggest this modification to the study.

The manuscript requires a major revision in terms of readability.  Some suggestions are made here, but there are too many for a referee to make the manuscript readable.

One major aspect is how can the comparisons extended to LW bands after contrasting only the 10-micrometre wavelength? If this is the best approximation, there should be a section describing how is this approach affecting the evaluation of temperature changes.

Section 2.2 This paragraph uses too technical terminology that is not explained in the text.

Ln 75: Which two parameters?

Ln 37 This sentence needs to be rethought, as the verb “increased” does not match the syntax.

Ln 49 “need” instead of “need”

Ln 104 “makes” instead of “make”

Ln 124 Full stop missing.

The paper by Stretton et al. compares the longwave radiation component of SPARTACUS-Urban with the DART model and the model by Harman et al. (2004) for a central London (UK) domain. Different surface temperatures based on observations are prescribed. While the findings of this paper are import for the modelling community and the general approach is sound, the explanation of the derivation of the surface temperatures is confusing and important discussion points are missing. Thus, I recommend reconsideration of this work after major revisions.

Major issues:

1) In which way is DART suited to be the reference of a model evaluation? What are the expected errors of DART itself? The paper mentions an evaluation of DART for vegetation? Was there an evaluation in an urban area? DART was also used as a reference model for the shortwave part of SPARTACUS-URBAN. This does not make it automatically suited for the present longwave part, does it?

2) SPARTACUS-Urban is a multi-layer urban scheme. The paper misses to introduce more multi-layer schemes, in particular the widely used BEP (Martilli et al. 2002, implemented in WRF) and the schemes based on it. The approaches in the radiation exchange parametrization should be quickly compared. Multi-layer schemes are also not considered in the Discussion/Conclusion part of the paper. In which respect does the results of SPARTACUS-Urban are expected to differ from these more traditional, non-statistical urban schemes? Why should one prefer the new approach compared to the older, street-canyon view-factor based approach? The paper mentions that roof interception is a problem for the single-layer Harman et al. model because of missing radiation from higher walls. Other multi-layer models are able to consider this (e.g. Schubert et al. 2012).

3) The description of the averaging of the observed surface temperatures is not completely clear.

* In 3.2, it is stated that the surfaces were already separated into sunlit and shaded surfaces. Why is this process repeated with shortwave SPARTACUS-urban? Is it because the information of the observations is only known for a sub-area of the full analysis domain? Is it expected that the fractions of sunlit and shaded surfaces is considerably different in the sub-area and in full domain? 

* L198: How is a surface temperature range prescribed in DART?

* L215: Did you know for each observation whether it was sunlit or shaded? Then these values could have been averaged directly before calculating orientation specific values. With orientation-averages, you had to introduce a somewhat arbitrary averaging.

* I do not understand what was studied in 4.4. Did SPARTACUS-urban use the temperatures otherwise prescribed to the Harman model? What is the principal difference to the case in 4.2? In the end, it is just prescribed temperature anyway?

Minor issues:

4) The paper is only about the urban part of SPARTACUS-Surface. Thus, I recommend using SPARTACUS-urban instead of SPARTACUS-Surface in the title.

5) I recommend giving the computational times in Table 5 in relative units, for example in units of time needed for the Harman model. This absolute value can be given in the caption of the table. The factors in L336 should be without the unit "s".

6) L141: What does this sentence mean? How are the 25th and 75th percentile used?

7) The abstract does not give results of the comparison with the Harman model.

8) The figure captions are too short: 

* Add date to Figure 2. 

* Add what we see for Figure SM1. The y axis misses a label. 

* Add details to Figure SM2. 

References:

Martilli, A., A. Clappier, and M. W. Rotach. 2002. ‘An Urban Surface Exchange Parameterisation for Mesoscale Models’. Boundary-Layer Meteorology 104 (2): 261–304. https://doi.org/10.1023/A:1016099921195.

Schubert, Sebastian, Susanne Grossman-Clarke, and Alberto Martilli. 2012. ‘A Double-Canyon Radiation Scheme for Multi-Layer Urban Canopy Models’. Boundary-Layer Meteorology 145 (3): 439–68. https://doi.org/10.1007/s10546-012-9728-3.