The annual count of reported tornadoes in the United States has risen over the decades. Interpreting that trend requires separating what the atmosphere did from what observers could see.
A tornado has to be witnessed or leave damage
Historically, a tornado entered the record because someone saw it or because it damaged something that was later surveyed.
A tornado crossing empty rangeland at night with no witness and no structures in its path simply did not appear in the count.
The record therefore describes detected tornadoes, and detection has always depended on where people and property happen to be.
Doppler radar changed what could be seen
Nationwide deployment of Doppler radar allowed forecasters to identify rotation within storms rather than waiting for a report from the ground.
That capability prompted warnings and directed spotters toward storms that would previously have passed unremarked in sparsely populated areas.
Dual polarization added the ability to detect lofted debris, which confirms a tornado is on the ground even when nobody can see it.
Weaker tornadoes account for most of the increase
The rise in counts is concentrated in the weakest damage ratings, which are exactly the events most likely to have been missed before.
Counts of the strongest tornadoes, which leave unmistakable damage across a wide path, show no comparable increase across the long record.
That asymmetry is the strongest evidence that improved detection rather than changed weather drives much of the trend.
Rating practice complicates the comparison further, since damage scales have been revised over time and a survey today weighs construction quality more carefully than earlier ones did.
Reporting practice and population both shifted
Storm chasing, mobile phones and social media have vastly increased the number of people able to document and transmit a sighting immediately.
Suburban expansion has placed structures across land that was previously open, giving tornadoes something to damage and survey teams something to rate.
Verification procedures have also become more systematic, with damage surveys conducted more consistently than in earlier decades.
Researchers work around the inhomogeneity
One approach is to analyze only the strongest ratings, on the assumption that those were rarely missed even early in the record.
Another is to count tornado days rather than tornadoes, since an outbreak producing many separate touchdowns inflates counts more than it inflates days.
Both methods point toward changes in the clustering and geography of tornado occurrence being more informative than the raw annual total.
The wider lesson applies to any long hazard record: a count reflects the observing system as much as the phenomenon, and the two have to be separated before a trend means anything.