Using a Sampling Bias File

[swajud]

I'm excited about the (relatively) new option for using a sampling
bias file. I created a raster of sampling areas in ArcGIS with the
potential sampling areas assigned a value of "1" and all other areas
given no value. I then convert this raster to ASCII through the same
process I use for extracting, resampling, and converting my
environmental layers. However, I keep getting an error message that my
environmental layers and the bias file have different geographic
dimensions. When I open the ASCII files in word, I get the following
parameters:

For the environmental layers:
ncols 3863
nrows 1271
xllcorner 87.749664307
yllcorner 41.567638397
cellsize 0.00833
NODATA_value -9999

For the Bias File:
ncols 2424
nrows 663
xllcorner 87.870833333
yllcorner 45.974579550574
cellsize 0.00833
NODATA_value -9999

Again, I use the same "extract by mask" and Resampling methods for
both types of layers. Any ideas as to what is happening? Does it have
anything to do with the fact that only potential sampling areas are
assigned a value of "1" in my Bias raster? Should my sampling areas be
assigned a different value? Any other tips/sources on creating,
processing, and using a Bias file in Maxent?

Thanks!!

[Jesús Muñoz]

Swajud:
Perhaps you should use 1s for sampled areas and 0s for unsampled areas,
and check if the bias raster is then identical to the other rasters. Te
easiest way to have identical grids is set the analysis options to one
of the environmental layers, and then multiply the problem raster by 1
in the raster calculator. The resulting grid should be identical in
extent and cell size to the other grids.
Jesus


*************************************************************************
Jesús Muñoz e-mail: jmu...@rjb.csic.es
Real Jardín Botánico (CSIC)
Plaza de Murillo 2 tlf. +34 91 420 3017
E-28014 Madrid fax. +34 91 420 0157
ESPAÑA (Spain)

http://www.masterenbiodiversidad.org

http://www.rjb.csic.es/jardinbotanico/jardin/contenido.php?Pag=413&tipo=cientifico&cod=66

KRAKEN research group (environmental modelling, data processing & analysis):
http://www.unex.es/unex/grupos/grupos/kraken

SYNTHESYS: EU funds short visits to our institution:
http://www.synthesys.info/index.htm
*************************************************************************

swajud escribió:

[Alvar Carranza]

Swajud:
One more tip: When creating the bias raster, make sure to select the
same extent and cell size that your environmnetal layers (Raster
analysys and Raster storage settings in ArcView)

best

Alvar

--
Msc. Alvar Carranza
UNDECIMAR
Facultad de Ciencias, Universidad de la República
Iguá 4225
CP 11440
Montevideo, Uruguay
http://alvar.carranza.googlepages.com

[Jessica Junker]

Hi
Your number of rows and columns for the bias file and the environmental
variables needs to be the same!
Jessi

--
Save the great apes!
Sign the mAn: http://www.apesmanifesto.org

Jessica Junker
Max-Planck Institute for Evolutionary Anthropology
Department of Primatology
Deutscher Platz 6
04103 Leipzig
Germany

tel: +49 341 3550 217
fax: +49 341 3550 299
cell: +49 0160 90594996
email: jessica...@eva.mpg.de

A.P.E.S. Database
http://apes.eva.mpg.de/

[Bill Peterman]

Also, it is my understanding that your bias file must consist of all
positive values (i.e., you cannot have NODATA everywhere you have not
sampled)

> email: jessica_jun...@eva.mpg.de
>
> A.P.E.S. Database      http://apes.eva.mpg.de/

[Anna]

In relation to this I was wondering if anybody has experience in creating sample bias files within marine environments? I work on a deep sea environment and my presence points are based on video transects of submarines and remotely operated vehicles. Therefore I have a situation where my sampling effort is constrained to a transect of a couple of meters width, whereas the rest of the area is unsampled. I guess the best way would be to create a buffer around the transect with the cell value 1, and assign 0 to the rest of the cells, but so far I chose the radius of the this buffer rather arbitrary. Any input would be appreciated.
Cheers,
Anna

2009/8/27 Bill Peterman <bill.p...@gmail.com>

[swajud]

Thanks all for your help. I don't know how I looked past that since
I'm used to processing the environmental layer. When I coded the bias
file with ones (possible sampling location) and zeros (no sampling), I
got the following error: "Bias has zero or neagtive values, ignoring
bias grid". As Bill mentions, the bias file must consist of positive
values, so I instead coded it as 1 (absence) and 2 (presence).

I have two related questions I hope this group can field:

1. How can I be sure that Maxent "knows" that I have coded one as
absence and 2 as presence? It doesn't ask me at any point in the
process which makes me nervous.

2. My outputs aren't significantly different with and without bias,
just some expansion of 0.5 probability areas. This brings me back to
question # 1., but also makes me wonder what situations using the bias
file has the most impact under. For example, does it significantly
effect the outputs when dealing with very low (20) or very high
(1000s) numbers of occurences? Or does it depend on the environmental
layers? Or is it best used to cover your tail statistically and
assumption-wise? Any insight from those using the bias file would be
appreciated.

Thanks!

On Aug 27, 6:39 am, Bill Peterman <bill.peter...@gmail.com> wrote:
> Also, it is my understanding that yourbiasfile must consist of all

> positive values (i.e., you cannot have NODATA everywhere you have not
> sampled)
>
> On Aug 27, 2:30 am, Jessica Junker <jessica_jun...@eva.mpg.de> wrote:
>
> > Hi

> > Your number of rows and columns for thebiasfile and the environmental

> > variables needs to be the same!
> > Jessi
>
> > swajud wrote:
> > > I'm excited about the (relatively) new option for using a sampling

> > >biasfile. I created a raster of sampling areas in ArcGIS with the

> > > potential sampling areas assigned a value of "1" and all other areas
> > > given no value. I then convert this raster to ASCII through the same
> > > process I use for extracting, resampling, and converting my
> > > environmental layers. However, I keep getting an error message that my

> > > environmental layers and thebiasfile have different geographic

> > > dimensions. When I open the ASCII files in word, I get the following
> > > parameters:
>
> > > For the environmental layers:
> > > ncols         3863
> > > nrows         1271
> > > xllcorner     87.749664307
> > > yllcorner     41.567638397
> > > cellsize      0.00833
> > > NODATA_value  -9999
>
> > > For theBiasFile:
> > > ncols         2424
> > > nrows         663
> > > xllcorner     87.870833333
> > > yllcorner     45.974579550574
> > > cellsize      0.00833
> > > NODATA_value  -9999
>
> > > Again, I use the same "extract by mask" and Resampling methods for
> > > both types of layers. Any ideas as to what is happening? Does it have
> > > anything to do with the fact that only potential sampling areas are

> > > assigned a value of "1" in myBiasraster? Should my sampling areas be

> > > assigned a different value? Any other tips/sources on creating,

> > > processing, and using aBiasfile in Maxent?

[swajud]

I just remembered one more question. If a few of your occurences fall
outside of what the bias file marks as "potential sampling" (i.e. what
I mistakenly called "present" in my last post), how does Maxent
respond? This is a current problem for me, but I'm always adding
occurences to my data sets and it just takes more time to re-make the
bias file each time. If this does cause an error, perhaps that is how
Maxent tells the different between "present" (sampling location) and
"absent" (not a sampling location).

On Aug 27, 6:39 am, Bill Peterman <bill.peter...@gmail.com> wrote:

> Also, it is my understanding that yourbiasfile must consist of all

> positive values (i.e., you cannot have NODATA everywhere you have not
> sampled)
>
> On Aug 27, 2:30 am, Jessica Junker <jessica_jun...@eva.mpg.de> wrote:
>
> > Hi

> > Your number of rows and columns for thebiasfile and the environmental

> > variables needs to be the same!
> > Jessi
>
> > swajud wrote:
> > > I'm excited about the (relatively) new option for using a sampling

> > >biasfile. I created a raster of sampling areas in ArcGIS with the

> > > potential sampling areas assigned a value of "1" and all other areas
> > > given no value. I then convert this raster to ASCII through the same
> > > process I use for extracting, resampling, and converting my
> > > environmental layers. However, I keep getting an error message that my

> > > environmental layers and thebiasfile have different geographic

> > > dimensions. When I open the ASCII files in word, I get the following
> > > parameters:
>
> > > For the environmental layers:
> > > ncols         3863
> > > nrows         1271
> > > xllcorner     87.749664307
> > > yllcorner     41.567638397
> > > cellsize      0.00833
> > > NODATA_value  -9999
>
> > > For theBiasFile:
> > > ncols         2424
> > > nrows         663
> > > xllcorner     87.870833333
> > > yllcorner     45.974579550574
> > > cellsize      0.00833
> > > NODATA_value  -9999
>
> > > Again, I use the same "extract by mask" and Resampling methods for
> > > both types of layers. Any ideas as to what is happening? Does it have
> > > anything to do with the fact that only potential sampling areas are

> > > assigned a value of "1" in myBiasraster? Should my sampling areas be

> > > assigned a different value? Any other tips/sources on creating,

> > > processing, and using aBiasfile in Maxent?

[Bill Peterman]

Not sure I can address all of your questions, but I'll at least add to
the discussion. It is my understanding that the bias file is
essentially down-weighting regions that have been extensively sampled.
So in my mind, what you really want is a continuous layer that ranges
from very small (0.0001; i.e. no sampling effort) to some scaled
measure of your maximum sampling effort, whether it be days sampled,
hours sampled, etc. Since most of us don't have great knowledge of
this kind of effort, a quick and effective work around is to use all
known/available sample points for your taxa as this can serve as a
proxy for sampling effort. For instance, you may be interested in
modeling a widely distributed plant species in the Appalachian
Mountains of the US, but you have an excessive clustering of points in
Smoky Mountain Nat. Park. If you were to pull all the records for
plants sampled in the Appalachians, you will likely see that there has
been much greater sampling effort within parks and/or near
universities. By using all of these points as your background sample,
you are effectively accounting for the bias in the sampling effort.

-Bill-

[swajud]

Interesting. Should my sampling bias raster then be a continuous
probability range from 0.00001 to 1, with 1 being sites I actually
sampled at (i.e. repetitive of species occurences)? Do you know what
kind of documentation/tutorial Phillips et al. (or others) have on
building bias rasters? I've looked over their most recent papers that
introduce the use of sampling bias files, which only seem to mention
that it exists and their results from using them in the analysis.
Phillips et al. 2008 refers to Dudik 2005, Ponder 2001, and Anderson
2003, but I am wondering if there is a paper or tutorial out there
that discusses the methods of building such a sampling probability
layer.