Thread: Time Series? Regular sampling for 1 week per month

Hello,
I am a masters student studying the effects of underwater noise on harbour porpoise presence. I have collected noise samples (3min=1 noise level) every 30 min for ~1 week each month over the past year, as well as porpoise presence every minute for the same time periods.

I would like to 'stitch together' the week long data sets for each site (there are 4 sites) and analyse each site separately as a time series, however I have no experience with them and am not sure this is appropriate. I read that you can use one time series to explain the variation in another, so maybe this would be a better approach.

Ultimately, I need to determine whether noise levels have any impact on whether porpoise are present in an area. I appreciate any guidance.

Hey cetacean.

How exactly do you relate your noise data to dependency of a porpoise existing? (As an example, low noise? specific noise changes over time? Noise relative to time of day? etc).

Hi Chrio,

I am using the broadband (0.25-16 kHz) sound pressure level every half an hour, and relating it to porpoise presence within a shifting window around that half an hour. For example, I am starting by saying 'At 10:30 there was a noise level of 110 dB (re 1uPa), and from 10:30-11:00 there were porpoise present.' This will lead to a binomial test, however I also have the number of echolocation click trains within each minute, so I can also relate the 110 dB to an average echolocation activity of 500 click trains per minute in the following half hour.

I do have time of day as a covariate because I expect noise to be higher during the day (more ships on the water during working hours), and porpoise echolocation activity to be higher at night (chasing herring close to shore).

Thank you for your interest!

So just to be clear, you have porpoise data (times of day) and noise data (times of day) and you want to figure out based on the noise data whether its likely that a porpoise exists at some time interval?

In other words, you want to relate the noise time series with a "spiked" time series for porpoises (something like say a time series with 0's when porpoise isn't there and a 1 when they are there or say x number of porpoises at a given time in the time series): Is this true?

Yes, you're on the right track. I want to know if there is a relationship between noise level and porpoise presence (2 analyses - 0/1 and #click trains per min) on multiple time scales. I'm starting with linking porpoise presence 30 min after a noise sample, but will continue with a shifting window of 30 min around and before that noise sample. However, since I have porpoise presence each minute, I can look at other time scales.

However, to keep it simple, I'd like to start with porpoise presence for the 30min after each noise sample.

This sounds like it could get messy.

It also sounds like you might want to look at data mining tools for time series. Are you familiar with data mining at all?

No, I am not familiar with data mining. And yes - it is messy! I have 4 sites and ~10 trials per site, so I think I'm going to have to analyze each trial separately.

You might want to take a look at this:

Welcome to the SAX