Practical advice

Real survey data is messy

Distance sampling in the Real World

We've talked a lot about models
We've also talked about assumptions
Our example is relatively well-behaved
What can we do about all the nasty real world stuff?

Some days...

2 fire emoji, a computer emoji, 2 fire emoji

Aims

Here we want to cover common questions
Not definitive answers
Some guidance on where to look for answers

What should my sample size be?

What do we mean by "sample size"?

Number of animal (groups) recorded
- detection function
Number of segments
- spatial model
Number of segments with observations
- spatial model

Re-frame

How would we know when we have enough samples?

We don't
Heavily context-dependent
Go back to assumptions

"How many data?"

plot of chunk df-obs

Pilot studies and "you get what you pay for"

Designing surveys is hard
Designing surveys is essential
Better to fail one season than fail for 5, 10 years
Get information early, get it cheap
- Inform design from a pilot study

Avoiding rules of thumb

Think about assumptions
- Detection function
- Spatial model
Think about design
- Spatial coverage
- Covariate coverage

Spatial coverage (IWC POWER)

POWER

Covariate coverage

plot of chunk coverage

Sometimes things are complicated

Weather has a big effect on detectability
Need to record during survey
Disambiguate between distribution/detectability
Potential confounding can be BAD

weather or density?

Visibility during POWER 2014

Thanks to Hiroto Murase and co. for this data!

Covariates can make a big difference!

Disappointment

Sometimes you don't have enough data
Or, enough coverage
Or, the right covariates

Sometimes, you can't build a spatial model

@kitabet

"Which of options X, Y, Z is correct?"

Alternatives problem

When faced with options, try them.
Where does the sensitivity lie?
What's really going on?
What is your objective?

"How big should our segments be?"

Segment size

If you think it's an issue test it
Resolution of covariates also important
Maybe species-/domain-dependent?
(Solutions on the horizon to avoid this)

"Is our model right?"

Model validation

Some variety of cross-validation
Temporal replication
- Leave out 1 year, fit to others, predict, assess
Spatial “pseudo-jackknife”
- Leave out every \( n^{th} \) segment, refit, …
- (Maybe leave out 2, 3 etc…)

Modelling philosophy

Which covariates should we include?

Dynamic vs static variables
Spatial terms? Habitat models?

Getting help

Resources

Bibliography has pointers to these topics
Distance sampling Google Group
- Friendly, helpful, low traffic
- see distancesampling.org/distancelist.html

Advanced topics

This is a whirlwind tour...

...and some of this is experimental

Smoother zoo

Cyclic smooths

What if things “wrap around”? (Time, angles, …)
Match value and derivative
Use bs="cc"
See ?smooth.construct.cs.smooth.spec

A cyclic spline

Smoothing in complex regions

Edges are important
Whales don't live on land
Bad things happen when we don't account for this
Include boundary info in smoother
?soap

Example of smoothers versus the Antarctic peninsula

Multivariate smooths

Thin plate splines are isotropic
1 unit in any direction is equal
Fine for space, not for other things

Tensor products

\( s_{x,z}(x,z) = \sum_{k_1}\sum_{k_2} \beta_k s_x(x)s_z(z) \)
As many covariates as you like! (But takes time)
te() or ti() (instead of s())

Tensor product example

Black bears like to sunbathe

Slope-aspect interaction for black bears

Random effects

normal random effects
exploits equivalence of random effects and splines ?gam.vcomp
useful when you just have a “few” random effects
?random.effects

Making things faster

Parallel processing

Some models are very big/slow
Run on multiple cores
Use engine="bam"!
Some constraints in what you can do
Wood, Goude and Shaw (2015)

Summary

Lots of complicated problems
Lots of potential solutions
(see also “other approaches” mini-lecture)
Need to get simple things right first
Trade assumptions for data