lat & long

[Jessica Junker]

Hi all

Just a quick question - does it make much sense to include latitude and
longitude as environmental layers for modeling species distributions (I
suspect they will be strongly correlated with climatic and other
environmental variables used in the model)?

Jessi

[Marnin Wolfe]

IMHO that would NOT make sense. Latitude and Longitude are at best proxies for true factors that influence your study organism. Things like Temperature and Precipitation vary with lat-lon for reasons including distance from an ocean or other water-body and distance from the sun. If you don't have climatic data like that (see www.worldclim.org) then Lat-Lon might be helpful.

I've used Lat-Lon as variables in a Principal Components Analysis and found that what you get out is a description of how the climate varies with geography, but this is not the same thing as what is done when modeling a distribution with Maxent.

Hope that was helpful.

Cheers,

Marnin

--
Marnin Wolfe
University of Pittsburgh
Department of Biological Sciences
Ecology & Evolution Program
wol...@gmail.com (or)
md...@pitt.edu
239-595-5081

[Jessica Junker]

Dear Marnin
Thank you very much for your input on this. Much appreciated.
Jessi

Marnin Wolfe wrote:
> IMHO that would NOT make sense. Latitude and Longitude are at best
> proxies for true factors that influence your study organism. Things
> like Temperature and Precipitation vary with lat-lon for reasons
> including distance from an ocean or other water-body and distance from
> the sun. If you don't have climatic data like that (see

> www.worldclim.org <http://www.worldclim.org>) then Lat-Lon might be

> helpful.
>
> I've used Lat-Lon as variables in a Principal Components Analysis and
> found that what you get out is a description of how the climate varies
> with geography, but this is not the same thing as what is done when
> modeling a distribution with Maxent.
>
> Hope that was helpful.
>
> Cheers,
>
> Marnin
>
> On Wed, Jun 10, 2009 at 6:05 AM, Jessica Junker
> <jessica...@eva.mpg.de <mailto:jessica...@eva.mpg.de>> wrote:
>
>
> Hi all
>
> Just a quick question - does it make much sense to include
> latitude and
> longitude as environmental layers for modeling species
> distributions (I
> suspect they will be strongly correlated with climatic and other
> environmental variables used in the model)?
>
> Jessi
>
>
>
>
>
> --
> Marnin Wolfe
> University of Pittsburgh
> Department of Biological Sciences
> Ecology & Evolution Program

> wol...@gmail.com <mailto:wol...@gmail.com> (or)
> md...@pitt.edu <mailto:md...@pitt.edu>
> 239-595-5081
>
> >

[Jonathan Koch]

I am curious. would you include the lat/long as a categorical variable in the environmental input? If so I would suspect that lat/long would be highly correlated with climatic variables.

--
Jonathan B. Koch
Department of Biology
Utah State University
5305 Old Main Hill
Logan UT, 84321-5305

[Jessica Junker]

Hi Jonathan
No, I included it as a continuous variable- but I understand that lat
and long are not predictor variables for species distributions when
climatic data are available.
Jess

[Milton Cezar Ribeiro]

Hi there,

 

"Eastitude" and "Northtitude" make sense on your "mind model" to explain Primary Producion (i.e. biomass production; biomass variability..")??. If so, why not incluid this on your model? Distance from coastal zone is not the same as "eastitude" because our continents are not paralel to our longitud zones :-).

So, I suggest you incluide this possible "mind model" on the modeling.Everything is a question os "scale"...

 

Cheers

 

milton

2009/6/10 Jessica Junker <jessica...@eva.mpg.de>

[Jessica Junker]

Hi
I'm a bit confused now as to whether to include lat long data or not.
So, I'm using NDVI, LAI, max and min temperature, precipitation, slope,
dem and cti among others as predictor variables. I'm not sure if
including Lat and long layers really have any predictive power?
Jess

Milton Cezar Ribeiro wrote:
> Hi there,
>
> "Eastitude" and "Northtitude" make sense on your "mind model" to
> explain Primary Producion (i.e. biomass production; biomass
> variability..")??. If so, why not incluid this on your model? Distance
> from coastal zone is not the same as "eastitude" because our
> continents are not paralel to our longitud zones :-).
> So, I suggest you incluide this possible "mind model" on the
> modeling.Everything is a question os "scale"...
>
> Cheers
>
> milton
>
> 2009/6/10 Jessica Junker <jessica...@eva.mpg.de

> <mailto:jessica...@eva.mpg.de>>

>
>
> Hi Jonathan
> No, I included it as a continuous variable- but I understand that lat
> and long are not predictor variables for species distributions when
> climatic data are available.
> Jess
>
>
>
> Jonathan Koch wrote:
> > I am curious. would you include the lat/long as a categorical
> variable
> > in the environmental input? If so I would suspect that lat/long
> would
> > be highly correlated with climatic variables.
> >
> > On Wed, Jun 10, 2009 at 4:05 AM, Jessica Junker
> > <jessica...@eva.mpg.de <mailto:jessica...@eva.mpg.de>

> <mailto:jessica...@eva.mpg.de

[Sam Veloz]

I think it is worthwhile to think about whether a species
distributions is actually constrained by lat/long. lat/long is just a
way people have come up with to describe locations on the earth. But you
can't take lat/long in a lab and measure a species physiological
tolerances to it. Other variables, annual/seasonal temperature,
precipitation, are actually variables that may constrain a species
distribution and they happen to be correlated with lat/long. So lat/long
is really just a "distal" (in Austin's terminology, sorry don't have the
reference on hand) variable for some variable that actually constrains a
species distribution. So my guess is that lat/long will not add much
predictive power *unless* it is correlated with some other variable that
you have not already included in your analysis. For that matter, the
same can really be said for elevation. Species undoubtedly have
elevational gradients in distribution, but are they responding to
elevation or another variable that is correlated with elevation? (And
all of the climate layers you are using were probably derived from an
elevation layer and are probably highly correlated with it.)
One other way to think about it. Suppose you developed a model for a
species response to latitude (or longitude, or elevation) and the model
output gives a predictive value of say 0.85 for a given degree latitude.
Now project that model into future climate space. The model will give
the same predictive value at the given degree latitude (or long, or
elevation) despite the fact that climate will have probably changed
there and the species tolerances for the climate there will probably
have changed as well. So what may be a good model now may be a poor
model in the future because you are not really modeling a variable that
determines a species distribution.
Hope that makes sense, probably drank way too much coffee before
replying.
Cheers,
Sam

[Michael Calkins]

I've found that by using them maxent will tend to create a box of suitable habitat around the occurrence records.

--
Michael Calkins, M.S. Graduate Student
Fish, Wildlife, and Conservation Ecology
New Mexico State University
2980 South Espina, Knox Hall 101
P.O. Box 30003, MSC 4901
Las Cruces, NM 88003-8003
Email: conserv...@gmail.com
Phone: 575-646-1353
Cell:  575-915-4798

[Gururaja K V]

I think lat-long will influence on spatial auto-correlation. It may not be so in Maxent. What about using Geographically Weighted Regression (GWR) and then bringing in relevant variables? I too need some insights on this.

with best wishes
gururaja

2009/6/10 Marnin Wolfe <wol...@gmail.com>

--
Dr. Gururaja K.V.
#233, Energy & Wetlands Research Group,
Centre for Ecological Sciences,
Indian Institute of Science, Bangalore - 560 012
INDIA
web: www.gururajakv.net Email: gur...@ces.iisc.ernet.in
Phone:+91-80-22932786-extn233

[Dani Villero]

Buenas,
only add to the arguments of Sam that if our data is geographically
biased, then lat long accentuate this bias in the models, very often
with unsatisfactory results.
Salut!

Dani Villero
Landscape Ecology Unit
Forest Technology Centre of Catalonia

On 10 Juny, 16:04, Jessica Junker <jessica_jun...@eva.mpg.de> wrote:
> Hi
> I'm a bit confused now as to whether to include lat long data or not.
> So, I'm using NDVI, LAI, max and min temperature, precipitation, slope,
> dem and cti among others as predictor variables. I'm not sure if
> including Lat and long layers really have any predictive power?
> Jess
>
> Milton Cezar Ribeiro wrote:
> > Hi there,
>
> > "Eastitude" and "Northtitude" make sense on your "mind model" to
> > explain Primary Producion (i.e. biomass production; biomass
> > variability..")??. If so, why not incluid this on your model? Distance
> > from coastal zone is not the same as "eastitude" because our
> > continents are not paralel to our longitud zones :-).
> > So, I suggest you incluide this possible "mind model" on the
> > modeling.Everything is a question os "scale"...
>
> > Cheers
>
> > milton
>

> > 2009/6/10 Jessica Junker <jessica_jun...@eva.mpg.de
> > <mailto:jessica_jun...@eva.mpg.de>>

>
> >     Hi Jonathan
> >     No, I included it as a continuous variable- but I understand that lat
> >     and long are not predictor variables for species distributions when
> >     climatic data are available.
> >     Jess
>
> >     Jonathan Koch wrote:
> >     > I am curious. would you include the lat/long as a categorical
> >     variable
> >     > in the environmental input? If so I would suspect that lat/long
> >     would
> >     > be highly correlated with climatic variables.
>
> >     > On Wed, Jun 10, 2009 at 4:05 AM, Jessica Junker

> >     > <jessica_jun...@eva.mpg.de <mailto:jessica_jun...@eva.mpg.de>
> >     <mailto:jessica_jun...@eva.mpg.de

[Marnin Wolfe]

One thought I had was that Maxent is likely to throw out Lat./Lon. automatically. That is to say, since it might give it a very low weight because Maxent only uses variables that add information. Lat./Lon. will only help maxent a very little. It would be very interesting to see if my theory is true!

~Marnin

[Jessica Junker]

Hi Marnin
Actually, I found that especially longitude was given relatively high
weight in my model. It ranked third among 12 environmental predictor
variables.
Jessi

Marnin Wolfe wrote:
> One thought I had was that Maxent is likely to throw out Lat./Lon.
> automatically. That is to say, since it might give it a very low
> weight because Maxent only uses variables that add information.
> Lat./Lon. will only help maxent a very little. It would be very
> interesting to see if my theory is true!
>
> ~Marnin
>
> On Thu, Jun 11, 2009 at 3:03 AM, Dani Villero <dani.v...@ctfc.cat>
> wrote:
>
>
> Buenas,
> only add to the arguments of Sam that if our data is geographically
> biased, then lat long accentuate this bias in the models, very often
> with unsatisfactory results.
> Salut!
>
> Dani Villero
> Landscape Ecology Unit
> Forest Technology Centre of Catalonia
>
> On 10 Juny, 16:04, Jessica Junker <jessica_jun...@eva.mpg.de

> <mailto:jessica_jun...@eva.mpg.de>> wrote:
> > Hi
> > I'm a bit confused now as to whether to include lat long data or
> not.
> > So, I'm using NDVI, LAI, max and min temperature, precipitation,
> slope,
> > dem and cti among others as predictor variables. I'm not sure if
> > including Lat and long layers really have any predictive power?
> > Jess
> >
> > Milton Cezar Ribeiro wrote:
> > > Hi there,
> >
> > > "Eastitude" and "Northtitude" make sense on your "mind model" to
> > > explain Primary Producion (i.e. biomass production; biomass
> > > variability..")??. If so, why not incluid this on your model?
> Distance
> > > from coastal zone is not the same as "eastitude" because our
> > > continents are not paralel to our longitud zones :-).
> > > So, I suggest you incluide this possible "mind model" on the
> > > modeling.Everything is a question os "scale"...
> >
> > > Cheers
> >
> > > milton
> >
> > > 2009/6/10 Jessica Junker <jessica_jun...@eva.mpg.de
> <mailto:jessica_jun...@eva.mpg.de>

> > > <mailto:jessica_jun...@eva.mpg.de

[Marnin Wolfe]

That's curious, but maybe not surprising. I suppose the point that has been made in this thread is that if it weights Longitude very highly it is because there is some environmental variable correlated with it that makes it a useful proxy in modeling. 

Can this be a good model then, assuming your interest is only in predicting the distribution, not understanding the underlying climatic factors?

wol...@gmail.com (or)
md...@pitt.edu
239-595-5081

[Samuel Veloz]

[Samuel Veloz]

"Can this be a good model then, assuming your interest is only in predicting the distribution, not understanding the underlying climatic factors?"

This is an important point. If you only want a prediction of the current distribution maybe using long is ok. But wouldn't it be nice to know what variables are really important for predicitng  a species distribution? You might want to look at the correlations between your predictor variables. If long is highly correlated with any variables, to me it makes more sense to include the other variable because that variable is probably more meaningful. If it isn't highly correlated with other variables, than perhaps it is actually acting as a surrogate for some variable you have not included and it is worth leaving in. Sorry for the previous message. Am on the road and having email issues. Sam

[Jessica Junker]

Wow! Thank you all for replying and for sharing your thoughts on this.
How exactly do I check for correlations between my predictor variables?
Can Maxent do this?
Jessi

[Milton Cezar Ribeiro]

Hi Jessica,

 

one easy way of do this is you load your layers on R,

choose a large number of random points over the

region of interest, get the value of each layer and

run some correlation test. May be pca could also

be interesting.

 

cheers

 

milton

brazil=toronto

2009/6/12 Jessica Junker <jessica...@eva.mpg.de>

[Jessica Junker]

I will do that. Thank you Milton!
Jessi

Milton Cezar Ribeiro wrote:
> Hi Jessica,
>
> one easy way of do this is you load your layers on R,
> choose a large number of random points over the
> region of interest, get the value of each layer and
> run some correlation test. May be pca could also
> be interesting.
>
> cheers
>
> milton
> brazil=toronto
>
> 2009/6/12 Jessica Junker <jessica...@eva.mpg.de

> <mailto:jessica...@eva.mpg.de>>

[Marnin Wolfe]

ArcGIS Spatial Analyst has a nice, easy function to randomly sample points on a map and also to extract data from them. You can use it to make a dataset of your variables and test the correlations. You can easily do 10,000 or more points and have a ridiculous amount of statistical power.

Cheers,

Marnin