Current corrections

General banter about diving and why we love it.
Post Reply
User avatar
YellowEye
I've Got Gills
Posts: 2726
Joined: Sun Oct 29, 2006 8:08 am

Current corrections

Post by YellowEye » Tue Oct 20, 2020 4:31 pm

Hi
Diving at certain sites like Keystone, Pt Hudson. Deception and Skyline are a bit of a guessing game when it comes to slack predictions. There are no nearby stations that are sufficient. Simple scalar rules like jump in 75 minutes before slack at Admiralty Inlet simply don't work consistently at these sites.

Divers have tried to come up with various rules of thumb and different heuristics for trying to predict things. Things like looking at two nearby stations and choosing one or the other depending on the height of the tide, or keeping historical records. These are hard to share with others, and hard to apply to new conditions.

I was thinking that if we could build a good data set of when slack actually occurs given a set of conditions, then we could utilize machine learning (ML) to build a model that can serve as a predictor for future diving. Applying an ML to a data set and then using the ML is actually very easy these days. The hard part is getting the data and deciding which variables you want in the data.

I'm looking for help:
1) Assume I want to get accurate to within 15 minutes. Do you think this is possible or does the concept of "currents are just predictions" throw this idea completely in the garbage bin? (I'd toss out bad weather days from consideration)
2) what input variables do you think should be in the model? Anything other than nearby current station(s) predictions with tide height, tidal exchange and current strength before/after the prediction, and weather? (would toss out the results for bad weather)
3) does anybody have historical data they could share? Could be for any divesite.
4) how to record slack could be a challenge. Perhaps people can record rough beginning and end of slack and where in the divesite it was observed. Thoughts?
5) Seems worth recording overall conditions. Like how long the site was divable for, but this seems more subjective.

If I can get a basic model working I can build a simple website around it for predictions and adding further data points.
Thank you!
Get out of cove 2 and explore!

Sharkb8
Dive-aholic
Posts: 229
Joined: Sun Jan 28, 2018 10:05 am

Re: Current corrections

Post by Sharkb8 » Wed Oct 21, 2020 8:45 am

Sounds like we will need to do some extensive field work for the next few months! Trying not to be tempted to join a scooter gang right now, but maybe next year.....we can start working on data from the weirder exchanges around Candy Land and Certain Death Pass
Party underwater

User avatar
Cmjstealth
Avid Diver
Posts: 76
Joined: Sat Jan 29, 2011 11:02 pm

Re: Current corrections

Post by Cmjstealth » Wed Oct 21, 2020 10:08 am

Happy to partner/collaborate and help if I can.

I had similar ideas (site specific ml models). Although I've so far been focused on other projects my long term idea was to develop a cheap 3d printed arduino microprocessor device that could be tied to a cinderblock at various sites for a month or so each. There was already a diver a while back that tried this with a tilt sensor although it didn't appear to have the resolution i'd want so was going to try a design with a magnetically coupled impeller/propeller, highly swept to prevent fouling. Could also capture current direction. Try and build on his existing work. Hopefully cheap and simple enough i could build a couple of them and maybe get some help from others to deploy and recover them.

Since it seems like exchange size might affect things I assumed we would need data across at least a month. Not sure if we would be able to infer seasonal differences from that alone or if we would need samples from thought the year. Also planned to pull historical weather data and perhaps the actual (not prediced) slack times observed at NOAA's co-ops stations. If the prior exchange was observed as being early or late perhaps the model could use that to inform the next cycle. Unfortunately if things like weather are significant variables It means we couldn't make predictions far in advance but I think worth exploring since anecdotally I've heard people mention it as a factor.

I haven't searched the academic literature yet but surely there is something that would help inform us about other possible variables we should be including. Either that or could try talking to the folks who maintain the NOAA prediction models?

I have some data already from Mike Snow who has kept a spreadsheet manually tracking current and slack times at skyline wall. Just want to make sure he gets credit for his work.

I also considered trying to do something with the reef.org dataset since it collects some limited current info and has tens of thousands of observations across pnw dive sites. We would probably need to reach out to surveyors and get more specific splash times from their dive logs, however.

Sent from my SM-G965U1 using Tapatalk


Post Reply