UFO reports: mapping the hotspots and modeling the rhythm

The source is NUFORC, the National UFO Reporting Center, accessed through the public TidyTuesday mirror. Each row is one report: a timestamp, a city and state, latitude and longitude, a reported shape, a duration, and a free-text description. Below is a slice of the real log. Note the flagged cells, where the shape was logged as a placeholder like na or unknown. That is the kind of mess cleaning has to handle.

Before any analysis the pipeline keeps only rows that sit in a real US state with usable coordinates, and standardizes each timestamp into a year, month, hour, and weekday. The rest, mostly international reports and rows with a missing location, are set aside so the geography stays honest.

Two views of geography. A live map of the real geolocated sample, and the leaderboard of states. The outline of the country appears with no map underneath it, the signature of a population-driven process: reports trace where people live, the coasts, the Great Lakes, Texas, Florida.

The timing signals are the sharpest in the whole dataset. Reports peak in summer, peak on Saturday, and peak on the clock at 9 to 10 pm, the after-dinner, still-dark, still-awake window. And one calendar date towers over the rest.

Annual volume climbs steeply from the mid-1990s. The critical read is what this curve measures. NUFORC's web form arrived in the internet era, so the rise is mostly a story of access and awareness, not more objects overhead. A log-linear model is fit to 1995 to 2013, backtested on held-out years, then projected forward. It forecasts reporting behavior, nothing more.

The seasonal, weekly, and hourly patterns combine into a transparent likelihood model. Each dimension contributes a factor equal to its share divided by an even share, so 1.0 means average and 2.0 means twice as likely as a random hour. Pick a month, a weekday, and an hour to score it. This is the "find for Y" step: given a moment, how report-heavy is it, historically?