Urban Cooling, UHI effect

Good morning I am a PhD student and for my research I am trying out the application of the Urban Cooling InVEST model.
I have some doubts regarding the data to use when calculating the UHI Effect. According to the manual, it is defined as “the difference in temperature between the rural reference area and the maximum temperature observed in the city.” When I use the difference between the maximum temperature in the urban area and the maximum temperature in the rural area, I get a negative result, which suggests that the urban heat island effect is not present. However, for this results could be several explanations, ventilation conditions, the different placement of stations on tall buildings, materials closer to the station that capture the heat, etc…
However, if I consider the minimum temperatures, the difference is positive. In the urban area, the minimum temperature is 2.2°C higher than in the rural area for the same year and month.
Is it acceptable to use the difference between minimums? Has anyone used the difference between the minima as input data?

Hi @NicolettaDenaro ,

This is an interesting question. InVEST is meant to be flexible so that users can modify how models are used, even if that’s somewhat different than how they’re intended.

The Urban Cooling model is agnostic to time, meaning all of your input data should cover the same period of interest, whether that’s something like average daytime conditions during summer months or average annual night time conditions. When you talk about minimum temperatures, I don’t imagine it’s a stretch to assume that those tend to occur overnight (or in the very early morning hours), in which case you should choose to use the night time, or “factors”, method when running this model, but that may not provide the results you’re most interested in. The daytime method it typically more desirable and informative when modeling urban heat.

Personally, I think it would be okay to use the difference in minimum temperatures if you are looking to estimate the UHI effect on low temperatures. However, I’m suspicious of the negative temperature difference you’re calculating between maximums for rural versus urban areas. Are these data definitely representing the same temporal coverages? For example, are they something like averages over the same period of months and years? I recommend carefully checking that this is true. If so, I would question the sources of data for the urban and rural area’s maximum temperatures. Individual weather stations are certainly prone to aberrations based on factors like some of those you listed. An alternative source for the difference in UHI effects in different cities across the globe is linked from the InVEST User Guide here and is available as a web app on Google Earth Engine. I suggest you check that out.

Please let us know if you have more questions,
Jesse

Good morning Jesse, first of all thank you for your feedback and I am sorry that I could read and respond to you only today.
I’ll give you some more information about the data I’m using as input data: I calculated the UHI using summer minimum temperature data, for the month of August for the city of Cuneo in Italy.
The data I am using refer to the same period, August 2023, and for two stations:

• CUNEO CAMERA DI COMMERCIO alt- 540 m urban area
• DRONERO alt-575 m rural area

  1. UHI = Minimum temperature in urban area in august - Minimum temperature rural area in august

I used the factor method, so for what you are confirming to me and for what I had already read in the manual, it is essential that the data are consistent with each other but not necessarily referring to maximum temperatures, which is only one example.
However, I am left with a doubt in case I want to use the data for the daytime and therefore the maximum temperatures. The manual says:

Magnitude of the UHI Effect: i.e. the difference between the maximum temperature in the city and the rural reference (baseline) air temperature.

It would be:

  1. UHI = Maximum temperature in urban area in august - Average temperature in rural area in august

or

  1. UHI= Maximum temperature in urban area in august - Maximum temperature in rural area in august

Because it is by following this third formula that I get negative values that led me to choose the difference between the minimum temperature values (Formula 1. UHI).
I wanted to use local data rather than global data because I am setting the model to a resolution scale of 1x1 m and I would like to be as accurate as possible. But I will now check the method that was used for the global data you sent me in order to understand if could be better to use them in any case.
As an output I have such a map related to heat mitigation: the fluid geometric shape with the mean-high HM value corresponds to the LC of a natural extensive grassland. The result seems to be coherent as the HM varies gradually around having an influence on the other LC as well.
HM MAP


On the other hand, the trees (values in red) do not have the slightest influence on the HM of the contiguous pixel, corresponding to another use. It is a net change of values, e.g. Between HM for a tree is 0.75 and the next pixel is 0.45 and is constant for the entire LC categories. The same thing occurs with T. air nomix.
T air nomix. map of estimated air temperature values.

Could it be an error caused by the input data? Or do I have to calibrate the model in some way? Sorry if I am asking more questions not related to the initial topic. But I am very curious about the functioning of the InVEST Urban Cooling model and would like to use it to the best to obtain accurate results.

Thanks again for your help :slightly_smiling_face:

Hi @NicolettaDenaro ,

Firstly, when you are using summer minimum temperatures, which tend to occur overnight (or at dawn), you should choose the “Intensity” method rather than “Factors”. Factors rely on shade, albedo, and ET, which are processes much more relevant to daytime than they are to night. For instance, everything is shaded when the sun is down (on the other side of the Earth).

I do recommend trying out the model with the daytime (maximum) temperatures and the “Factors” method. Have you varied the factor values at all from their defaults?

To compare apples to apples, you should be subtracting maximum rural temperatures from maximum urban temperatures. Be very careful that you are not comparing temperatures at significantly different elevations. A 35m difference in elevation does not sound like much, but it could make a difference, especially near the escarpment of the Alps which likely experiences strong convection currents during summer heat. If this approach yields negative values, some other dynamics are at play for your area compared to most cities and/or you should vet the observed data more closely and potentially seek other sources which include more data points. Perhaps you can figure out how to pull some historic data from open, crowd-sourced weather stations, such as those available on WeatherUnderground’s Wundermap.

The results are largely dependent on the LULC map and its corresponding parameter values for each associated class. Are “trees” a class in the LULC map? If so, how are they weighted for shade, albedo, and ET compared to those weights for other, surrounding LULC classes? This drives the results. But it may also be that 1m pixels are too small to have a considerable impact. Also, note that I do see different values displayed in the screenshot you shared for T_air_nomix.tif. 24.06°C represents quite a bit more heat than does 22.832°C. Be sure you are evaluating the values closely, and not simply relying on the color rendering drawn by GIS software.

Please remember that for all InVEST models, without performing careful calibration, any values in the results should only be viewed as relative spatial patterns. At a resolution of only 1m, it’s even more important that you are precise. Since you say that you “would like to be as accurate as possible”, if you’re expecting to put confidence in absolute results, you must certainly calibrate the model for August in Cuneo, including carefully choosing the best values for all the different parameters (e.g.: factors weights, maximum air blending distance, etc.). Modeling absolute temperatures at a 1m resolution does not sound like an easy task to perform properly, so it’s imperative to gain as many observed data points as possible and to diligently vet every data input.

I hope this helps,
Jesse

Here are some other potential data sources for historic climate data in Cuneo: