A tornado leveled a Starbucks in Kokomo, Ind., on Aug. 24. A customer at a restaurant across the street recorded video of the twister as it demolished the coffee shop. (Storyful)

Tornadoes raked across Indiana and Ohio on Wednesday afternoon and evening. Our analysis suggests more than 20 tornadoes touched down over the course of seven hours. The strongest blew through Kokomo, Ind. — population 57,000 — around 3:30 p.m., which the National Weather Service rated an EF-3 on a preliminary basis.

But residents in this region didn’t necessarily know a tornado outbreak was possible. They may not have even been prepared for thunderstorms.

“We’re all very grateful that there were no fatalities and no serious injuries,” Kokomo Mayor Greg Goodnight said at a news conference. It certainly could have been worse.

On Wednesday morning, the Storm Prediction Center did not indicate any probability for tornadoes in the affected region. At noon, the center issued a slight risk of severe weather for the southern half of Indiana and none for Ohio. A tornado watch was issued for parts of Indiana at 3:13 p.m., but by that time, three tornado warnings had already come down from the National Weather Service in Indianapolis, the first of which was issued at 2:37 p.m.

The question being asked today: Why did the Storm Prediction Center miss this outbreak? And does it matter?

The tornado warning for Kokomo was issued about 20 minutes in advance, John Kwiatkowski, the science operations officer at NWS Indianapolis, told The Post. He said he feels like they did their jobs on Wednesday.

“Nobody was killed or seriously injured with a tornado that was quite strong and struck a major city,” Kwiatkowski said. “If that isn’t success, then nothing is.”

The National Weather Service is surveying the damage. It will probably take at least a few days to know exactly how many tornadoes touched down. What we do know is that there are 35 “filtered” tornado reports. Filtered reports attempt to remove duplicates, but it’s not the tornado number. Quite often the actual number is much lower. In some cases, it’s higher.

From our analysis of radar data and storm reports, there were probably around two dozen tornadoes. We know this because there were multiple, cyclic supercells in two main batches. The first was a line of storms that broke up into supercells — something that’s pretty unusual in itself. The second batch was made up of isolated storms.

We can also look at the rotation track analysis produced by the National Severe Storms Laboratory. This isn’t an exact depiction of where the tornadoes touched down, but it gives us an idea of where the storms with the strongest rotation tracked.

August is considerably more active than, say, December for tornadoes. But the big outbreaks — like the one we saw last night in Indiana and Ohio — are extremely uncommon in August, especially when they’re not triggered by a tropical storm. Indiana’s average August tornado count is just one. Ohio’s is two.

Quite frankly, this outbreak is one of the worst in all of 2016. If we use the filtered report count of 35 from the Storm Prediction Center, it would be the second-largest outbreak this year based on the preliminary reports. The National Weather Service still needs to review these numbers, and some, particularly for May 24 in Kansas, could increase.

For Indiana and Ohio, Wednesday was probably the largest tornado outbreak on record in the month of August. By a lot. The second-largest outbreak in those states prior to this year is just nine, although just last week the NWS Indianapolis area saw seven tornadoes touch down.

It’s a lot of tornadoes for a day that appeared “minimal risk” at the outset.

Tornadoes are a science not settled. That much is clear from Wednesday’s events. It’s surprising that a large outbreak occurred with no risk area issued by the Storm Prediction Center in the hours prior. It hardly happens anymore, but that doesn’t mean it can’t.

“We’ve had great successes in past forecasts, but some days the forecast situation proves difficult,” Russell Schneider, director of the Storm Prediction Center, told The Post in an email. “In this case, the hazard threat was not clear days in advance, but a tornado watch was issued as soon as the threat was apparent.”

Perhaps it was the very early tornado warnings, or maybe the success of information sharing in the social media age, that we can thank for Wednesday’s results. It’s remarkable that no one was injured or killed in these storms, despite their strength and location.

It makes us wonder whether the Storm Prediction Center risk outlooks are actually that useful to the public. If warnings can be reliably issued 20 minutes in advance, that might be all the time people need to find shelter and stay out of harm’s way.