| Title: | Spatial Analysis with Self-Organizing Maps |
|---|---|
| Description: | Application of the Self-Organizing Maps technique for spatial classification of time series. The package uses spatial data, point or gridded, to create clusters with similar characteristics. The clusters can be further refined to a smaller number of regions by hierarchical clustering and their spatial dependencies can be presented as complex networks. Thus, meaningful maps can be created, representing the regional heterogeneity of a single variable. More information and an example of implementation can be found in Markonis and Strnad (2020, <doi:10.1177/0959683620913924>). |
| Authors: | Yannis Markonis [aut, cre], Filip Strnad [aut], Simon Michael Papalexiou [aut], Mijael Rodrigo Vargas Godoy [ctb] |
| Maintainer: | Yannis Markonis <[email protected]> |
| License: | GPL-3 |
| Version: | 1.2.4 |
| Built: | 2025-02-20 03:59:39 UTC |
| Source: | https://github.com/cran/somspace |
cnet plots the canonical network map of a single classification scheme.
cnet(x, n, thres)cnet(x, n, thres)
x |
regs object. |
n |
number of regions. |
thres |
the cross-correlation threshold of the network. |
The cnet function estimates the cross-correlation matrix of the average time series of
each region and plots a map linking the regions with cross-correlations above the selected threshold.
plot object
dummy <- owda[Time <= 1600] inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(3, 3, "hexagonal")) my_regions <- somregs(my_som, nregions = 6) cnet(my_regions, n = 5, thres = 0.2)dummy <- owda[Time <= 1600] inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(3, 3, "hexagonal")) my_regions <- somregs(my_som, nregions = 6) cnet(my_regions, n = 5, thres = 0.2)
Reconstruction of European hydroclimate derived from tree-rings. The variable used is self-calibrated Palmer Drought Severity Index (scPDSI) at annual time step.
data(owda)data(owda)
An object of class data.table (inherits from data.frame) with 1355264 rows and 4 columns.
Markonis et al. (2018) Nature Communications 9(1):1767 (Nature Springer)
str(owda)str(owda)
Plots the time series of SOM nodes or regions mean
plot_ts(x, n)plot_ts(x, n)
x |
is either a |
n |
is either the set of nodes for |
In case of regs, all the regions are ploted.
plot object
Regions class
regsregs
An object of class regs of length 0.
The regs class contains:
A summary data.table which updates the somsp object with the region ids of all classification schemes
up to nregions. Each different classification scheme is stored as an individual region, e.g. regions.2,
regions.3, etc.
A data.table with the original data set, as in somsp.
It can be plotted by plot and plot_ts.
If plot is used, three additional arguments are needed; a set with the classification schemes
that will be ploted, number of rows and number of columns of the plotted panels.
plot_ts plots all the time series of a given classification scheme.
sominp transforms the data set from data.table to
somsp format, which can be used as argument in the somspa function.
sominp(x)sominp(x)
x |
The |
x should be in tidy format
with four columns: time, latitude, longitude and variable.
A sominp object. It contains:
a matrix that can be used as input for the som function of
the kohonen package.
a data.table with the with spatial point coordinates and a corresponding id.
a data.table with the original dataset.
dummy <- owda[Time <= 1510] inp_som <- sominp(dummy)dummy <- owda[Time <= 1510] inp_som <- sominp(dummy)
somregs applies hierarchical cluster analysis to the Self-Organizing Map
to form regions with homogeneous characteristics (classification scheme).
somregs(x, nregions, ...)somregs(x, nregions, ...)
x |
A |
nregions |
The maximum number of classifications schemes to be determined starting from 2. |
... |
Other arguments passed to methods from |
nregions must be at least two, i.e., a classification scheme with two regions, and smaller than
the number of SOM nodes. In the latter case, each SOM node corresponds to a region.
The resulting regs object can be plotted by plot and plot_ts.
If plot is used, three additional arguments are needed; a set with the classification schemes
that will be ploted, number of rows and number of columns of the plotted panels.
plot_ts plots all the time series of a given classification scheme.
A regs object, which contains:
A summary data.table which updates the somsp object with the region ids of all classification schemes
up to nregions. Each different classification scheme is stored as an individual region, e.g., regions.2,
regions.3, etc.
to their corresponding winning unit, the number of points of each node, as well as the median
latitude and longitude of each node coordinates and their standard deviation.
The original time series which created the SOM as a data.table, as in somsp.
dummy <- owda[Time <= 1600] inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(4, 4, "hexagonal")) my_regions <- somregs(my_som, nregions = 9) plot(my_regions, regions = c(2, 4, 6, 8), nrow = 2, ncol = 2) plot_ts(my_regions, n = 4)dummy <- owda[Time <= 1600] inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(4, 4, "hexagonal")) my_regions <- somregs(my_som, nregions = 9) plot(my_regions, regions = c(2, 4, 6, 8), nrow = 2, ncol = 2) plot_ts(my_regions, n = 4)
Spatial SOM class
somspsomsp
An object of class somsp of length 0.
The somsp objects are created by somspa function and contain:
A summary data.table with the coordinates of each SOM node, the distances of objects
to their corresponding winning unit, the number of points of each node, as well as the median
latitude and longitude of each node coordinates and their standard deviation.
A Self-Organizing Map object (see also kohonen).
The sominp object used as input for the SOM, with an id number coressponding to
location and a node number to the classification group of SOM.
They can be plotted by plot and plot_ts functions or summarized by summary.
somspa creates a Self-Organizing Map from spatial data.
somspa(x, ...)somspa(x, ...)
x |
A |
... |
Other arguments passed to methods from |
x should be created by sominp.
The output somsp objects can be plotted by plot and plot_ts functions or summarized by summary
A somsp object, which contains:
A summary data.table with the coordinates of each SOM node, the distances of objects
to their corresponding winning unit, the number of points of each node, as well as the median
latitude and longitude of each node coordinates and their standard deviation.
A Self-Organizing Map object (see also kohonen).
The sominp object used as input for the SOM, with an id number coressponding to
location and a node number to the classification group of SOM.
dummy <- owda[Time <= 1600] #toy example inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(3, 3, "hexagonal")) my_som$summary my_som$som plot(my_som) plot_ts(my_som, n = 3) plot_ts(my_som, n = c(1, 2, 4, 9)) plot_ts(my_som, n = 1:max(my_som$summary$node)) #plots all somsdummy <- owda[Time <= 1600] #toy example inp_som <- sominp(dummy) my_som <- somspa(inp_som, rlen = 100, grid = somgrid(3, 3, "hexagonal")) my_som$summary my_som$som plot(my_som) plot_ts(my_som, n = 3) plot_ts(my_som, n = c(1, 2, 4, 9)) plot_ts(my_som, n = 1:max(my_som$summary$node)) #plots all soms