Not OP but I'll give you a better example.
Let's say you are processing the temperature data of all of the US over 20 years, tens of thousands of weather stations with multiple readings per day. It's likely that some of those malfunction every now and then. How do you know which readings are correct and which are not? Well you can eliminate the outliers.
You take an area and you look at the temperature readings - if they are all around 75F plus or minus 3 degrees, but you have one that's showing 90F, there is pretty good change that one station is busted.
That's exactly how they deal with the urban island effect. Temperature readings in cities are higher due to high area of pavement/concrete, so instead they use the readings from a rural area nearby.
In the end you don't care about the ups and downs, you smooth those over and you look at direction of the average. I