Siland a R package for estimating the spatial influence of landscape

The spatial distributions of populations are both influenced by local variables and by characteristics of surrounding landscapes. Understanding how landscape features spatially structure the frequency of a trait in a population, the abundance of a species or the species’ richness remains difficult specially because the spatial scale effects of the landscape variables are unknown. Various methods have been proposed but their results are not easily comparable. Here, we introduce “siland”, a general method for analyzing the effect of landscape features. Based on a sequential procedure of maximum likelihood estimation, it simultaneously estimates the spatial scales and intensities of landscape variable effects. It does not require any information about the scale of effect. It integrates two landscape effects models: one is based on focal sample site (Bsiland, b for buffer) and one is distance weighted using Spatial Influence Function (Fsiland, f for function). We implemented “siland” in the adaptable and user-friendly R eponym package. It performs landscape analysis on georeferenced point observations (described in a Geographic Information System shapefile format) and allows for effects tests, effects maps and models comparison. We illustrated its use on a real dataset by the study of a crop pest (codling moth densities).


Introduction
We present here an illustration of a landscape analysis using the R package siland. We applied siland methods to analyze the effect of local (treatment) and landscape variables (organic and conventional orchards) on codling moth densities. This example was previously described and analyzed in Ricci et al. (2009) 1 . The first part presents analysis conducted with the Bsiland method (buffers method). The second part presents analysis conducted with the Fsiland method (based on Spatial Influence Function). This analysis was conducted using R version 3.6.2 and package siland version 2.0.

Data load
The data are available in the package siland.
The Bsiland.lik' function allows to detect whether the estimated minimum is a local mimimun (and not the general minimum). Here the landscape has a global significant effect (p.val= 1.562683e-11).
The function summary() provides significance tests of the intensity of the effects of the explanatory variables (local or landscape).

summary(resB)
## The buffer sizes for conventional and organic orchards were estimated at 1777.52 m and 201.02 m, respectively. The effect of the local variable was estimated at -0.1550 but not significant (p.val=0.308734). The intensity of the effect of conventional orchards in buffers of size 1777.52 m was estimated negative (-40.9131) and significant (p.val<0.0001). The intensity of the effect of organic orchards in buffers of size 201.02 m was estimated positive (87.5657) and significant (p.val<0.0001). We can consider that the effect of conventional and organic orchards did not start from the observation sites, but from the boundary of the orchard where the observations are located using the "border=T" argument.

Fsiland method
In the approach Fsiland, Spatial Influence Functions (SIF) are used to modelize the influence of landscape variables (decreasing with distance). In this approach, the scale of effect of a landscape variable is estimated by the parameter of the SIF. The function summary() provides parameters estimation and significance tests of the intensity of the effects of explanatory variables (local or landscape). The mean distances for SIF of conventional and organic orchards were estimated at 1423.9033 m and 216.0852 m, respectively. The effect of the local variable treatment was estimated at-0.012 but not significant (p.val=0.94446). The intensity of the effect of conventional orchards was estimated negative (-11.8561) and significant (p.val=0.00349). The intensity of the effect of organic orchards was estimated positive (24.8800) and significant (p.val<0.001). conv org F igure 9 : Observations map. The discs represent the significant and medium effect area associated to the estimated SIFs.

summary(resF)
We can consider that the effect of conventional and organic orchards do not start from the observation locations but from the border of the orchard where observations are located using the argument 'border=T'.