UTCI Sensitivity to variables

Has anyone determined a method to assess the sensitivity of a given climates UTCI results, and obtain a clearer understanding as to which variable (solar radiation, wind speed, humidity) is the highest contributor to the results observed?

Thanks in advance

Hi @jwoodall,
Not a sensitivity analysis but I have conducted several UTCI maps. I observed that the direct radiation from the Sun is the variable that affects UTCI more than any other. I believe this should not be a surprise at all.

@devang thanks for the feedback. So I take it that you are using comparative graphs of different metrics and gauging correlations? I was just curious to see if there was a way of being able to quantify the sensitivity of the UTCI map to each variable?

You can use Ladybug_Outdoor Solar Temperature Adjustor and the Ladybug_Outdoor Comfort Calculator to conduct such a sensitivity analysis.

hi @jwoodall

have a look at this one, it’s a good read by the allstar roster.


Thanks for posting the paper @OlivierDambron :slight_smile:

@jwoodall, I think Figure 9 in particular might give you some insight into the impact of each variable.

I just wanted to add that, since the publishing of that paper, I have tested a few climates where the presence / absence of wind ends up having a stronger impact on thermal comfort than sun / shade (aka. sky heat exchange in Figure 9). Typically, these types of climates are in mid-to-higher latitudes with relatively high cloud cover (ie. some climates in Ireland, Seattle, Sydney Australia, etc.). However, more often than not, the sun has a larger impact than the wind as @devang notes.

The effect of urban surface temperatures and urban heat island consistently have a smaller impact than the direct effect of wind and sun. The one exception might be if you narrow your analysis to only a few specific hours of the day (ie. right after sunset when urban heat island tends to be strongest).


Hi @jwoodall.
Love this question and its one I’ve been considering with for some time.
My approach to sensitivity in the past, mainly using UTCI as a metric, has been a somewhat manual process of removing certainly variables and tracking the impact on UTCI. This is pretty easy to do with wind and sun, simply by disconnecting those inputs into the UTCI calculation and tracking the resultant UTCI temperature. In the end, I’ve creating graphs like the one below to help study the relative impact of solar and wind. Granted this graph represents the average monthly condition, it may help show the benefit of providing shading in the summer (does the removal of solar wave provide comfortable UTCI temperature?) or wind shading in the winter. Although these bars are only made up of three test points, perhaps one can extrapolate between points to determine the relative benefit of shade? So ,perhaps this may only be appropriate for early phase climate analysis.

I would love to hear of other approaches or critiques on this method.


Thanks all for the comments and thoughts.

@OlivierDambron & @chris - the paper was very interesting to read more into the work carried out related to UTCI, and also the relationship that heat island effect has in proportion to wind and solar.

@KitElsworth your method seems the closest to what I had conceived in my head - this is a fantastic method to begin visualising sensitivity of UTCI to variables. I’m keen to advance this a little further, as one of the questions I ask myself at early stages of microclimate assessment is not only what the greatest contributor to the prevailing UTCI condition is, but also whether there is an ability to quantify variables sensitivity so as to gauge comparison.

Any further thoughts much appreciated.

I wonder how one can say “the effect of sun is more than wind” while they are completely two different parameter with different units!! Its very confusing! Is 500wh/m2 sun radiation more than 4m/s wind velocity? Please help me to understand this comparison since it might change the whole data mining methods.

My relatively naive mind tells me that this should be possible through a % sensitivity result on the comparative influence each variable has over the resultant UTCI. Although creating a script to determine this would take time

I really dont want to disappoint you but UTCI is a kind of “Weighted arithmetic mean” . You may consider several “IF” conditions to see one variable effect like; if Tdb=28 and RH=60% and Tmrt=25 what does happen if one increase/ decrease air velocity? no problem at all!
The “influence each variable has over the resultant UTCI” varies in different conditions. Anyway, as long as you are dealing with different units this theory is mathematically wrong. You may introduce a new unit for all parameters!

@Navid have you explored the method of “centering and scaling” various elements in a multiple linear regression? Seems like it may have some application here based on your observation. If you were to perform a MLR on the UTCI analysis, the weighting of coefficients could be used to determine their respective influence. What do you think?

@KitElsworth “centering and scaling” is a must to do for almost any statistical model. I agree that “weighting of coefficients” should be used for this purpose. The problem here is exactly the “weighting of coefficients” itself. How can be determined for various parameters with different unit? To simplify this please think about this question "Is 500wh/m2 sun radiation more than 4m/s wind velocity? "


Kit’s method here should be mathematically correct, it’s simply a type of global sensitivity analysis.

If I understand your criticism correctly, you believe this visualization is conflating different, linearly independent values (wind, solar) together in such a way that the resulting UTCI outputs aren’t accounting for how different variables impact that output. Hence your criticism that wind and solar values cannot be compared against each other, as they are based on different units.

I believe we can prove that this is not the case, by thinking about this visualization in 3d space, where each dimension corresponds to a wind, solar, UTCI value. The reason this is helpful, is that it allows us to intuitively “see” that values along each axis are linearly independent of each other - no linear combination of wind vectors alone can be used to produce a vector along the solar axis, and vice versa. Thus they are not being compared against each other, or conflated, just as you are saying.


A 3d vector in this space illustrates the respective contributions of wind, and solar and specific values to the UTCI output. So, if we do a bunch of simulations and generate different UTCI values, with different solar, and wind values (and some default, fixed values for other variables), all of that is clearly tracked along the different axis. Importantly, we can quantify the impact of solar radiation on UTCI, relative to a wind baseline, and vice versa. This is a mathematically valid way of representing the relationship between our inputs and outputs, and is subsequently mapped to the monthly analysis done by Kit in a way that correctly preserves the linear independence of our input parameters.

Specifically, this is achieved by taking the wind and solar dimensions and representing them by colour, rather then axis geometry. Colour preserves the linear independence of all the input parameters, and thus is able to communicate the respective contributions of solar and wind. The benefit here is that we can then do this for every month along a second analysis, and represent 4 dimensions of information in 2d (wind, solar, UTCI, month).

You can even the derive the coloured visualization from the 3d space I describe. Just convert our wind and solar coordinates to an appropriate colour (something like this):


And then remove those two axis. This effectively projects our 3d vectors to the 1d UTCI. Obviously the actual visualization is a little more simplified then this, more like this:

Blue: UTCI where Wind < Wind Baseline, and Sun Baseline (and baseline values for all other inputs)
Red: UTCI where Sun < Sun_Baseline, and Wind Baseline (and baseline values for all other inputs)

Hence, we can say that solar radiation has more of an impact on UTCI, then wind speed, given the underlying baseline conditions.

Make sense?


1 Like

@SaeranVasanthakumar, I appreciate your description of the design space and design “vectors.”

For those who want to work like @SaeranVasanthakumar is proposing, I think the best and fastest way is to build a parametric model of the problem, in this case a UTCI calculation, then iterate over all variable parameters using the Colibri plugin. Outputs can be easily reviewed using TT’s Design Explorer. Examples files explaining this workflow can be found on hydrashare. Thanks @MingboPeng for the plugin and the examples.

For more advanced analysis the Design Space Explorer plugin by MIT’s Digital Structures group can be used to compare the sensitivity of the various parameters. See attached pdf for explanation. The DSE plugin can also be used to generate the space, but I prefer Colibri because it’s so simple.tools_readme_v1_4_reduced.pdf (1.1 MB)