To answer some questions of you, here is some research directed to:
1 radiation errors of various sensorscreens (Stevenson screen well above 1,5 K in an intercomparison with other sensorscreens)
2 The effect of low windspeed on modern shields
3 The effect of open and closed surroundings on measurements.
That is why it is important, because the total summ of the errors can easily become 2-3 C without any city effect to speak of.
And here research of the WMO coded DB 260 site by the KNMI, which answers how a WMO site is seriously in error and it is caused by a single tree barrier. It also shows you how theory and practice are real. A sheltered area warms up faster and cools down faster (not in cities btw):
I quote:
"The results show that a large tree barrier in the vicinity of the DB260 has a significant effect on the operationally observed temperatures. Compared to more open locations at the KNMI terrain, DB260 shows higher maximum temperatures and lower minimum temperatures.
In the summer half year the daily maximum temperatures are on average 0.28°C higher than those for the most open site and the daily minimum temperatures are on average 0.48°C lower. Individual daily differences may, however, be much larger. Consequently, the representativeness of the measurements for the surrounding area of the KNMI-terrain may be seriously questioned. For operational temperature observations, it is therefore recommended to consider a move of the observations from DB260 to the most open location in the experiment, indicated as Test4. However,
for climatological purposes a continuous record without artificial breaks is important. Therefore, the temperature observations at DB260 should be continued until a satisfactory transfer function is developed that can be used to reduce the temperature observations at Test4 to DB260.
And
"It is
well known that wind speed influences the magnitude of the errors made in air temperature measurements, especially for daytime radiation errors. The results here stress the importance of high-accuracy wind speed
measurements near the thermometer screen at screen height during both day and night. These wind measurements are an important measure for determining: differences in radiation errors between sites, local differences in atmospheric stability (resulting in differences in vertical temperature profiles near screens), and, eventually, for corrections. Consequently, this type of wind measurements may be important for objectively monitoring the homogeneity of temperature sites, and for developing improved weather dependent transfer functions in case of future inhomogeneities. Positioning of high-accuracy wind speed measurements (in particular at low wind speeds < 3.0 m/s) near the thermometer screen at screen height, in addition to the current operational wind measurements, is therefore strongly recommended. This refers especially to stations of climatological interest."
Here the whole research. Read it yourself
http://www.knmi.nl/publications/fulltexts/hisklim7.pdf
With a stevensonscreen, this error would be even larger as you can read in the following two links, which conclude that articifical aspiration (daytime especially) is the best solution. Look how even Gill plate screens (to be more clear: screens that look like Davis sensorscreens) warm up rapidly with less wind in high sun. The trees near Amareleja are far worse than those at De Bilt....Look what errors can be expected...
First the preliminar results of an interscreen comparison. A small and easy reed showing that the Stevenson screen does give way to large errors, even in the Netherlands. Expec them to become bigger in the sunny south.
http://www.knmi.nl/~meulenvd/wmo/TECO1998/tc98temp.pdf
A long read, thororough research on plated screens and low wind effect can be read here. With windspeeds decreasing below 2 m/s, temperatures start to rocket. Easy to see that a small fan that blows over the sensor reduces errors dramatically. Good read if you wnat to construct the sensor yourself.
http://journals.ametsoc.org/doi/pdf/10.1175/1520-0426%281999%29016%3C1862%3AMEAWMR%3E2.0.CO%3B2
In conclusion: no matter where you measure. If you want to measure the climate, artificial windbreaks and most certainly cities are to be avoided.
The use of a Stevensonscreen is the worst one for measuring the correct temperature in sunny climates.