I'm curious where this opinion comes from? my understanding is the sampling takes a lot into account including the heat island effects. Your blanket claim does not come with any proof, so I'm curious where it comes from.
> I'm curious where this opinion comes from? ... my understanding is the sampling takes a lot into account
There is no sampling. We have an incomplete census of measured temperatures. Where humans decided to measure and record temperatures and which records survived were not determined by a sampling scheme.
Even if the record were complete, it would be subject to selection biases.
Just looking at the locations of temperature measurements in the definitive historical data set should drive this point home. Two, three, four hundred years ago, humans did not say "let's figure out a representative sampling scheme for accurately measuring global climate change." The current record is a function of where humans chose to live, work, produce, and travel.
Also when reigons decrease in data. After the end of the cold war russia looses 3/4ths of their datapoints, and then half a year later china looses about the same.
And, what happened to Australia?
https://www.youtube.com/watch?v=h95uvT67bNg
The data set is here: https://www.ncdc.noaa.gov/ghcnm/v3.php
IIRC, it took about a hundred lines of Perl to generate that animation of temperature station locations in the definitive global historical temperature data set.
When you are watching the animation, note when specific regions of the world make an appearance in the data set and the relative importance of measurements from specific countries.
I do not draw policy conclusions from these observations. However, I am frequently annoyed by official and unofficial visualizations which completely gloss over the sample selection and observational continuity problems in the historical record.