Hello Alex,

I found the package ecodist quite useful. It allows to work with several approaches for analysing spatial patterns. Among the others, MRM, Mantel, partial Mantel together with their relative correlograms that can help you to discriminate the significance of the isolation at various distance lags.

Hi Romain, to run Patchstat I need a matrix of presence/absence, how can I get it from my occurrences data ?

Hi Alex,

you migth measure the spatial autocorrelation of your data using the ripley's k or the Moran's I implemented in SPATSTAT PACKAGES.

Hi Alex,

If you have access to ESRI ArcMap, it also has an implementation of a local Moran's I, called an Outlier Analysis, where you can compare a spatial value to its local neighbors.

tks Gabriele, I try to take as much as possible from occurrences data but with simple fonction, to measure how isolation is structured within the distribution. I don not want to dive too deeply within spat stats because it is for a paper and we have much analysis. I tried this morning to compute the sum of geographic distances for each point and I mapped it. What do you think about this figure ?

In white low isolation, in black high isolation.

Hello Alex,

of that plant is?

Angiorsperma or gymnosperm?

Hi Alex,

A technique I used for a study on spatial distribution was to create a fishnet in ArcGIS, and then measure the average distance between plants (within a certain distance of each point), and the count per Grid of the fishnet. I then used the Getis-Ord Index to calculate the Z-score, which gives an indication of the degree of clustering. The way the results are presented in ArcGIS is that the stats gives you an indication of:

- a high clustering of low values (lots of grids with low sample density)

- a low clustering of low values

- low clustering of high values

- high clustering of high values

This type of analysis should take long to do, provided your dataset isn't too huge. I had 45000 sample points and it took about a day to analyse.

Good luck!

hello Salvatore, this plant is Arenaria provincialis (Caryophyllaceae), a narrow endemic. You can find our last publication about its niche and origin on my RG page.

alex

If you analyze the isolation as the sum of distances, the points in the perifery will be more isolated than the points in the centre simply because the distance between points in the left and right perifery is bigger. I think that a clustering measure as the one proposed by MIranda is better.

Hi Alex,

another simple possibility is the J function in spatstst (Jest command).

Values J(r) > 1 suggest regularity, and J(r) < 1 suggest clustering.

Hi Alex,

what about measuring the connectivity to the network of each site i with a formula like:

Sum over j of (e^(-alpha*Distance[i,j]))

where alpha is the mean dispersal distance, and could be a value from literature or simply 1 (and test for sensitivity over it). Look Moilanen, A., and M. Nieminen. 2002. Simple connectivity measures in spatial ecology. Ecology 83:1131–1145.)

Then, you compare these connectivity values for every site i with the distance to the shore or a main population (maybe the big one near Cassis).

Cheers,

Ale

hi, tks !

I take some time today to just apply a kernel method from the spatstats packahe in R; the density function. Here is the figure, it clearly shows that the abundant center is SW . I trim the data to conserve only occurencies distant from at least 100 meters then I apply the fucntion density.ppp with a sigma parameter of 250. Do you think that is enough to show that isolation is not homogeneous within the distribution ?

I can add that filed sampling effort was higher at periphery (and mostly eastern) than in SW area, so I am relatively confident with this.

About the connectivity formula (tks Alejandro) my problem is that this plant species has a very short distance dispersal (

Hi, What you say is true. Just remember that estimations of dispersal distance are spatial- and time-scale dependent. Some epiphytic lichens were found to disperse ~50m on a tree to tree scale, but ~500m on a larger time- and spatial- scale (reference hopefully to be published soon).

Regarding your figure, I ("me and myself") still would like to see a plot of the values you got vs a spatial variable you consider relevant.

Cheers

ok, please find below two figures :

(1) is the local functions of neighborhood for 591 points (function "localL of spatstats package", then (2) is the local densities for the distance value of 500, in black are marks above the mean and in white the marks below the mean.

In brief does it make sens ? can I use simply the kernel function "density.ppp" to show how abundance is changing throughout the distribution or must I use moe complicate analyses ?

And the local density within the distance value of 500

Getis-Ord General G (Spatial Statistics) is a High/Low Clustering tool which measures concentrations of high or low values for an entire study area. As an example, you can use this statistic to compare spatial isolation and dispersal of clusters . A high index value indicates that high values are clustered within the study area. A low index value indicates that low values tend to cluster. A Z-score is important to determine if the index value is significant.

Well, it depends on your aims.

In general, you should be critical on landscape matricies as calculated by Fragstats: there are scores of them and they may not be very meaningful or they are hard to interpret. Usually you see the difference in map scale or mapping quality, if you compare areas.

They do become relevant if you can compare different areas, or analyse where 'weak spots' are or preferential corridors e.g. in landscape connectivity. An example is the study we did for the Red deer, where we calculated a landscape connectivity measure which I guess is fairly similar to the one mentioned by alejandro (referring to Moilanen). We did this on a raster basis with Corine Land Cover (paper Groot Bruinderink et al.).

It requires however more information on the 'mobility' of species, as mentioned above by Alejandro, for lichen species.

It is important to assess the 'functional connectivity': in how far is a landscape connected (or: how isolated are populations). We did such an approach for Moor frog, and for forest plant species (Maianthemum bifolium - False lily?) where we calculated absolute distances between (sub) populations, we studied the landscape history, and therefore the actual isolation (when did these ponds become isolated by the highway) and related that to the genetic distance between populations, by DNA testing (see our papers on Maianthemum and Rana arvalis).

Somehow the paper was not attached.

The other 2 papers from Arens et al. can be found under my publications

Article Designing a Coherent Ecological Network for Large Mammals in...