Take a look at their figure, especially in May 2020—the average appears lower, but, more significantly, there is much less variability in May 2020 compared to earlier years.
The authors' model quite strongly includes their preferred confound (secular decrease in PM2.5) but doesn't explore what other covariates could explain the differences between years.
It's fine to say that one should be skeptical, but one contrary report doesn't invalidate an antecedent report, and the structure of a linear model strongly influences an outcome.
Given the physical mechanisms involved it is implausible that pollution did not decline. And if you look at their data you see a marked drop in 2020 at day 70
This is March 10 or thereabouts, I think. And there are ZERO high pm 2.5 days for a 20 day stretch or so. This isn't seen in other years. The vast bulk of days are below the trend.
And then for the rest of the year there are some days above the trend line but no high pm 2.5 days.
This fits with people being extra cautious in the early days and then relaxing a bit as things went on.
Now, I'm eyeballing this so I could be incorrect. But:
1. The effect was found in other cities
2. The physical mechanism makes it highly expected that there would be a drop
The study was about the slope of the regression modal, but if you had scrambled the years I'm fairly confident I could have picked 2020 out of the set.