Skip to content

Another Set of Problematic Numbers For Sunset Dunes

The Counts from Two Surfline Cameras Have Several Big Problems

Update! The Creators of the SunsetDunesVisitors web-site Have Taken It Down!

A browser window showing sunsetdunesvisitors.com, which now reads "Thank everyone for their interest in this site." and says the authors have to pause the experiment for reasons beyond their control, hoping to have it back soon, followed by a list of things they say they learned.

You may have seen their posts on Nextdoor and on the SunsetDunesVisitors web-site (opens in a new tab). The Surfline cameras at Taraval and at Ulloa/Vicente are being used to produce counts of park visitors. These cameras were never intended to be used to do that. The counting software has major issues and the authors are making some very questionable assumptions about the data. Some of their calculations are clearly wrong.

A screenshot of the Sunset Dunes visitor site for Sunday, September 27. The hour menu for the Taraval camera is open, and every hour from 7 AM through 6 PM is listed as unavailable.
An Actual Screenshot of Sunday’s Data

They used software to count the number of walkers, runners, and cyclists. They then made assumptions about how the counts from one location relates to attendance for the entire road.

There are several problems with this:

  1. The data is highly cherry-picked from a single week.
  2. There are days when the camera views were unusable and almost all of the data was "estimated". These estimates appear to be much too high.
  3. Some of the adjustments on their calculator page are based on assumptions that are clearly wrong.
  4. The software is clearly overcounting the raw numbers.
  5. The video from the locations at Taraval and Ulloa is often unavailable or the quality is too poor to be used.

Let’s take these one at a time:

Cherry Picked Dates

As we all know, the best weather of the year in the Outer Sunset is in September and October. Attendance at the beach can be far higher than the average during a week when the weather is especially good. The weather during the period shown was extraordinary.

On Some Days the Data is Completely Made Up

On Tuesday and Wednesday, September 29th and 30th, the fog rolled in. The Ulloa data was not available at all. For most of the day, the Taraval data was unclear and the authors could not get counts from the video. Yet on these very foggy days, the site shows that the estimated total visitor counts are almost the same as on sunny weekdays.

Let’s look at Wednesday, 9/30. The Ulloa data was unavailable the entire day. Only two hours of the Taraval data was clear enough to be counted. The other ten hours are filled in with "estimates". We have no idea where these estimates come from. According to the web-site just 640 of the raw counts come from the Taraval camera while another raw visitor count of 3,188 is derived solely from "estimates" for a total of 3,828. There are some odd things about these estimates. The morning counts are as high or higher than the afternoon counts which is inconsistent with other weekday mornings.

A screenshot of the Sunset Dunes visitor site for Wednesday, September 30. The hour menu for the Ulloa camera is open, and every hour from 7 AM through 6 PM is listed as unavailable. The whole-park estimate panel for Wednesday, September 30, showing 5,569 visitors, circled in red and labelled "After Adjustments". Below it the breakdown reads 640 camera views counted from video and 3,188 estimated to fill gaps, including 10 hours 9 minutes of low-visibility video, with the two figures circled and labelled "Just 640 counts from the cameras!". A bar chart underneath shows mostly estimated rather than counted hours.
Click to Enlarge

These dubious counts are then multiplied as shown on the "Calculator" page. The raw count of 3,828 is multiplied by various factors to get a final count of 5,569. In other words, a dubious raw count is multiplied by dubious adjustments to get a dubious final count.

Tuesday has similar problems. Over half of the raw count is derived from estimates rather than direct camera counts. These estimated counts also appear to be too high.

Wrong Calculations And Dubious Adjustments

Some of the estimates on their calculator page are clearly wrong. The repeat view adjustment for walkers is way off resulting in a huge overestimate in total attendees.

Every walker who walks past the camera in one direction will walk back the other way. It’s the only reasonable way to get back. The raw count of walkers should be divided in half to account for this, but this is not being done. Instead there is a deduction of just 25%. Because of this alone, the final result is overestimated by a third.

There are other adjustments on the calculator page. The authors don’t say how they arrived at these. I have no idea if these assumptions are correct, but they appear to be based on guesswork.

The Raw Numbers From the Cameras are Being Overcounted

In each of these screenshots the people in blue have been misidentified as walkers on the road even though they were merely crossing to the beach. I did not have to look hard to find these. I saw them in the first few minutes of each video.

The software attempts to identify walkers, runners, and cyclists. There is no running count on the screen. If there were, we would know if it was counting correctly by watching it go up as each person is counted, but it is not present. Instead we have to depend on other cues to see if the software is counting correctly.

When the software counts a walker, it draws a blue rectangle around that person. According to their site it is not supposed to count people who are simply crossing the street to get to or from the beach, but it often does so, and it is not hard to find many examples of this.

Beachgoers Incorrectly Identified as Road Walkers

A frame from the Surfline camera at Taraval on Monday at 11 AM. Three people stand near the path to the dunes. The counting software has drawn a blue box around one of them, a person crossing toward the beach, recording them as a walker on the road.
Mon 11AM Hour
A frame from the Surfline camera at Taraval on Monday at 11 AM. Two people crossing toward the beach are each enclosed in a blue box and recorded as walkers on the road.
Mon 11AM Hour
A frame from the Surfline camera at Taraval on Monday at 3 PM. A person crossing toward the beach is enclosed in a blue box and recorded as a walker on the road, while a figure beside them is outlined in white and not counted.
Mon 3PM Hour
A frame from the Surfline camera at Taraval on Monday at 11 AM. Two blue boxes sit at the foot of the path: one around a beachgoer on the right, and a narrow one around the surfboard carried by the beachgoer on the left. Both are recorded as walkers on the road.

In this photo, the beach-goer on the right is being misidentified as a walker on the road. The software is misidentifying the surfboard of the beachgoer on the left as a walker on the road.

Miscounting walkers has major consequences. Because of the way the data is adjusted by the calculator, each miscounted walker adds 1.9 people to the final adjusted count.

There are other places in the video where the blue box disappears and then reappears. Does this mean the person is being counted twice? There is no way to know.

Data Is Frequently Unavailable

The video on the site is frequently marked as "Unavailable". The data is then filled from the camera at Ulloa adding either 20% or 40% to the totals. How did they arrive at this adjustment? The page again doesn’t say.

If the data is not available from either Taraval or Ulloa, the site fills the gaps with other numbers. The site fails to explain the source of these other numbers. Only 17% of Wednesday and just 46% of the Tuesday numbers were counted from the cameras.

A screenshot of the Sunset Dunes visitor site for Sunday, September 27. The hour menu for the Taraval camera is open, and every hour from 7 AM through 6 PM is listed as unavailable.

Conclusion

There is only one correct way to estimate crowd size and that’s by actually counting the people at all the entrances and exits. Parks and Rec and the No On G people simply didn’t do this. They are trying to fill in the missing data with numbers that are the result of cherry-picked data, bad data, and a lot of guesswork.

Leave a comment