[R-sig-Geo] GRTS sampling - 2-level design

[Thierry Onkelinx]

Dear John,

A real life example is available at https://doi.org/10.5281/zenodo.2784012.
The idea is that the database returns a randomised set of points. You need
to overlay these points with your sampling framework. The final sample is
the set of points with the lowest ranking. The grtsdb package is a
reimplementation of my GRTS package (https://github.com/ThierryO/grts).
That package has a vignette describing GRTS via the Reversed Randomized
Quadrant-Recursive Raster strategy (
https://doi.org/10.1007/s00267-005-0199-x).

Best regards,

ir. Thierry Onkelinx
Statisticus / Statistician

Vlaamse Overheid / Government of Flanders
INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
FOREST
Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
thierry.onkelinx using inbo.be
Havenlaan 88 bus 73, 1000 Brussel
www.inbo.be

///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////

<https://www.inbo.be>


Op vr 8 okt. 2021 om 16:08 schreef John Wilson <jhwilson.nb using gmail.com>:

> Dear Thierry,
>
> Thank you so much for your reply. Yes, the loss of the spatial balance
> once the two-tiered approach is not accounted for was what was worrying me.
>
> The incorporation of region as a random effect has two issues - 1) the
> overall sampling area is a lake, and "regions" don't make sense in that
> context. 2) The analysis is a mark-recapture for fish (using the MARK
> software); I've never seen the incorporation of random effects in the
> Jolly-Seber / Cormark-Jolly-Seber models outside of Bayesian framework and
> outside of individual random effects... but even if I could do that - the
> regions just don't really make sense anyway (that I can see, anyway - maybe
> I'm not thinking about it the right way?)
>
> Thank you for the grtsdb suggestion. Do you have any examples of how this
> works? I couldn't find any vignettes or worked examples...
>
> Thank you so much,
> John
>
> On Fri, Oct 8, 2021 at 10:39 AM Thierry Onkelinx <thierry.onkelinx using inbo.be>
> wrote:
>
>> Dear John,
>>
>> Your procedure will create a spatially balanced level 1 sample (10
>> "regions") and within those regions a spatially balanced level 2 sample.
>> When you ignore the structure, there is no longer a spatial balance. So
>> you'll need to incorporate the two level sampling structure in your
>> analysis. E.g. by using region as random effect.
>>
>> I presume you are catching fish along rivers and assume that the rivers
>> are linear features. I'd consider drawing 10 samples using GRTS to define
>> the regions. Then use that location as the center point of 5 systematic
>> samples along the river (-2, -1, 0, +1 and +2 km).
>>
>> You might want to take a look at our grtsdb package. Available at
>> https://inbo.r-universe.dev/ It generates a full grid of master samples
>> and stores it in the database. So you can draw multiple samples from the
>> same master sample. This is useful in case of monitoring with a changing
>> population. You draw a sample and keep the lowest ranking locations that
>> are part of the population. If the population changes over time, then the
>> new sample will keep a proportion of the original sampling location
>> relative to the proportion of the population that remained stable. This
>> allows for repeated measures for stable locations while taking into account
>> the changes in population.
>>
>> Best regards,
>>
>> Thierry
>>
>> ir. Thierry Onkelinx
>> Statisticus / Statistician
>>
>> Vlaamse Overheid / Government of Flanders
>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE
>> AND FOREST
>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>> thierry.onkelinx using inbo.be
>> Havenlaan 88 bus 73, 1000 Brussel
>> www.inbo.be
>>
>>
>> ///////////////////////////////////////////////////////////////////////////////////////////
>> To call in the statistician after the experiment is done may be no more
>> than asking him to perform a post-mortem examination: he may be able to say
>> what the experiment died of. ~ Sir Ronald Aylmer Fisher
>> The plural of anecdote is not data. ~ Roger Brinner
>> The combination of some data and an aching desire for an answer does not
>> ensure that a reasonable answer can be extracted from a given body of data.
>> ~ John Tukey
>>
>> ///////////////////////////////////////////////////////////////////////////////////////////
>>
>> <https://www.inbo.be>
>>
>>
>> Op do 7 okt. 2021 om 15:54 schreef John Wilson <jhwilson.nb using gmail.com>:
>>
>>> Oh, sorry - I normally use the grts() function from the spsurvey package.
>>> My hacky approach was to make 10 balanced points with grts(), followed by
>>> imposing a 5 km buffer around each one, and either systematic sampling
>>> within the buffer circle, or running a separate GRTS for the 5 points
>>> within each 5 km buffer circle. Even writing this makes me cringe though,
>>> so hoping for something legitimate... I'll contact the authors if I don't
>>> get any solid leads on here.
>>>
>>> On Thu, Oct 7, 2021 at 10:40 AM Roger Bivand <Roger.Bivand using nhh.no>
>>> wrote:
>>>
>>> > On Thu, 7 Oct 2021, John Wilson wrote:
>>> >
>>> > > Hi everyone,
>>> > >
>>> > > I'm working on a sampling design using GRTS, but I'm running into a
>>> > > logistics problem. The field crew can set 5 nets per day, but only
>>> > within a
>>> > > 5 km stretch, due to travel time constraints. With 10 sampling days,
>>> > that's
>>> > > a total of 50 sites. The overall sampling area is huge, so running a
>>> > > regular GRTS design for 50 sites results, of course, in much larger
>>> > > distances between sampling points.
>>> > >
>>> > > Is there a legitimate way to create a 2-level GRTS design, where in
>>> step
>>> > 1
>>> > > we choose 10 spatially-balanced sampling points (one "core" point per
>>> > > sampling day), and then for each of these "core points", we create a
>>> grid
>>> > > of 5 sampling points that are constrained to all be within 5 km from
>>> each
>>> > > other? I can make that happen code-wise, but am not sure what the
>>> > > implications on spatial balance are, or if there's a built-in way to
>>> do
>>> > > this.
>>> >
>>> > Do you have a code example? Are you using BalancedSampling, SDraw or
>>> > Spbsampling or packages (probably SDraw)? Have you run any simulations
>>> to
>>> > try to get a first assessment on the impact of constraining your
>>> sample?
>>> > Might approach a package author also help?
>>> >
>>> > Roger
>>> >
>>> >
>>> > >
>>> > > Would appreciate any thoughts...
>>> > > John
>>> > >
>>> > >       [[alternative HTML version deleted]]
>>> > >
>>> > > _______________________________________________
>>> > > R-sig-Geo mailing list
>>> > > R-sig-Geo using r-project.org
>>> > > https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>> > >
>>> >
>>> > --
>>> > Roger Bivand
>>> > Emeritus Professor
>>> > Department of Economics, Norwegian School of Economics,
>>> > Postboks 3490 Ytre Sandviken, 5045 Bergen, Norway.
>>> > e-mail: Roger.Bivand using nhh.no
>>> > https://orcid.org/0000-0003-2392-6140
>>> > https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en
>>> >
>>>
>>>         [[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> R-sig-Geo mailing list
>>> R-sig-Geo using r-project.org
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>
>>

	[[alternative HTML version deleted]]