kaitlyngaynor
commented
2 years ago From e-mail to Alex 3 Jan 2022:
Continuing to work through this. I removed all points outside of the HREC bounding box, which got rid of the tracks when hunters left the property and also extreme GPS outliers. Now we want to remove steps that are unrealistic distances apart within HREC (the max corresponds to a speed of 60 mph, clearly incorrect—but what is the line here?)
What speed threshold makes sense for you? Here, step length refers to distance traveled in 3 minutes. I'm trying to remember how quickly people drive around those roads when hunting... do you know if there's a stated speed limit for hunters, or HREC in general? we could use that to justify, perhaps
- 10 mph / 0.80 km step length / 2% of steps (1619 steps)
- 15 mph / 1.21 km step length / 0.4% of steps (295 steps)
- 20 mph / 1.61 km step length / 0.17% of steps (124 steps)
- 25 mph / 2.01 km step length / 0.10% of steps (76 steps)
- 30 mph / 2.41 km step length / 0.06% of steps (47 steps)
For more context, here is a histogram of the speeds for all steps that are >10mph. Seems like 15mph may be a reasonable cut-off since we see a big drop-off there.
From e-mail to Alex 28 Dec 2021:
I am actually looking into a strategy that may solve most of our problems. I think two simple steps will solve issues with a) hunters leaving the property, b) long periods of operation in which loggers aren't working, and c) major outliers, and not require you to do any digging into the times that hunters stopped/started hunting.
Step 1: Clip the points to the HREC property
Step 2: After formatting analysis for HMM (calculating step length & turn angle), just drop the top 5% of step lengths, or otherwise determine an "unrealistic" step length, and drop any steps above this. And then we can just move forward with the analysis with the remaining steps.
In theory, this filter should address issues where hunters left the property for hours, or the logger wasn't working for an hour, or the point jumped across the site due to GPS error despite only 3 minutes passing, or the car was driving really fast on the county road. We just need to figure out what this threshold should be (how many meters, realistically, would a hunter travel in 3 minutes? we can look at distribution of data and see what we're dealing with).