maxent settings

[Jessica Junker]

Hi all

I have a few questions regarding some setting options in Maxent 3.3.0:

How exactly does the biasfile option work? I'm not sure if I understand
correctly, do I have to add a grid with my sampling locations (e.g. all
transect locations including presences and absences) here?
What is the difference bewteen the "add all samples to bachground" and
the "add samples to background" option. Can this have a large effect on
my output model?
What does the option "extrapolate" do? What exactly does "outside the
limits encountered during training" mean?
When should I activate the option "allow partial data"?

Perhaps this is explained in more detail in some technical document for
maxent?

Thank you!
Jess

[Bill Peterman]

Hi Jess--

I'll see if I can help with a few of your questions.
Bias File: The option to add a bias file in Maxent changes the
weighting lambdas in the calculation of the distribution; areas with
higher bias get downplayed in the final prediction. The bias file must
be a .asc file with same dimensions and resolution as your
environmental grids. All grid cells must a value >0. In my experience,
creating a legitimate bias file can be difficult. It is possible to
alleviate sampling through your selection of background points, and
this has been discussed in some detail in previous posts.

Adding Samples to Background: These options determine whether or not
your presence points used for modeling will be included with
background points. "Add Samples..." will add presence samples that are
unique and not yet included in the background points, while "Add All
Samples..." will add your presence points regardless if they are
identical to some of your background points (i.e., may lead to
redundant background points). If neither of these are selected, only
the background points in your background file will be used (or only
the 10,000 maxent generated points). I can't really comment as to how
changing these options will affect models, but would imagine that
changes would become more evident with a higher ratio of presence to
background points.

Extrapolate: Extrapolation occurs when environmental conditions are
encountered during projection of the model that were not encountered
during training in either the presence or background points (either
higher or lower). In my experience this occurs mostly with small
samples from a very restricted range, and projection then occurs
across a large range. It will also frequently occur when projecting to
future climate scenarios. This now brings in the idea of clamping.
Because Maxent does not really "know" the response curve to fit beyond
the maximum/minimum conditions encountered during training, clamping
occurs to effective "cap" the prediction at the level of the highest/
lowest observed values in training. In other words, future climate
scenarios may predict that temperatures will be 5 degrees warmer, but
if extrapolation ids allowed, all of these areas that are warmer than
currently observed will have the same probability assigned to it as
the highest observed temperatures seen during training. I hope I have
not butchered this too much, or confused you more.

Allow Partial Data: I have not used this personally, but understand it
to be useful if you have missing environmental data for some of your
records.

Hope this helps, good luck--Bill