I doubt we can control something like a volcano, but the main example in the article, wildfires, can absolutely be controlled more than they are today.

If your dataset excludes natural fires, but doesn't exclude controlled burns, for example, controlled burns would end up looking more negative because they cause pollution and don't get "credit" for any prevention.

It sounds like we’re either measuring the wrong thing or putting too much weight behind a measurement that sums up an inherently complex system. This measurement matters enough to be able to petition days to be excluded, why can’t we just codify the exclusion of clearly out-of-normal datapoints? Ignoring everything else, a random spike caused by anything transient is not useful information to us, our goals need to be aligned to the trend over time.

In an ideal world, I don't think opinionated manipulation of the underlying dataset is a good idea. If you want to avoid outliers dominating your analysis, you should aggregate it in an outlier resistant method like using the median or other percentile based strategy.

If you allow for manipulating the data, especially on such a politically charged subject, the manipulation method itself will get gamed like it seems is happening here.

> We can’t regulate acts of nature

How can you be so sure? What about controlled burns?

> identified as non-human-caused pollution

The issue highlighted in the article is that human-caused pollution is being misclassified as non-human-caused pollution. Nothing more to it.