CST_WeatherRegimes {CSTools}

Function for Calculating the Cluster analysis

Description

This function computes the weather regimes from a cluster analysis. It is applied on the array data in a 's2dv_cube' object. The dimensionality of this object can be also reduced by using PCs obtained from the application of the #'EOFs analysis to filter the dataset. The cluster analysis can be performed with the traditional k-means or those methods included in the hclust (stats package).

Usage

CST_WeatherRegimes(
  data,
  ncenters = NULL,
  EOFs = TRUE,
  neofs = 30,
  varThreshold = NULL,
  method = "kmeans",
  iter.max = 100,
  nstart = 30,
  ncores = NULL
)

Arguments

data

An 's2dv_cube' object.

ncenters

Number of clusters to be calculated with the clustering function.

EOFs

Whether to compute the EOFs (default = 'TRUE') or not (FALSE) to filter the data.

neofs

Number of modes to be kept (default = 30).

varThreshold

Value with the percentage of variance to be explained by the PCs. Only sufficient PCs to explain this much variance will be used in the clustering.

method

Different options to estimate the clusters. The most traditional approach is the k-means analysis (default=’kmeans’) but the function also support the different methods included in the hclust . These methods are: "ward.D", "ward.D2", "single", "complete", "average" (= UPGMA), "mcquitty" (= WPGMA), "median" (= WPGMC) or "centroid" (= UPGMC). For more details about these methods see the hclust function documentation included in the stats package.

iter.max

Parameter to select the maximum number of iterations allowed (Only if method='kmeans' is selected).

nstart

Parameter for the cluster analysis determining how many random sets to choose (Only if method='kmeans' is selected).

ncores

The number of multicore threads to use for parallel computation.

Value

A list with two elements $data (a 's2dv_cube' object containing the composites cluster = 1,..,K for case (*1) or only k = 1 for any specific cluster, i.e., case (*2)) and $statistics that includes $pvalue (array with the same structure as $data containing the pvalue of the composites obtained through a t-test that accounts for the serial dependence.), cluster (A matrix or vector with integers (from 1:k) indicating the cluster to which each time step is allocated.), persistence (Percentage of days in a month/season before a cluster is replaced for a new one (only if method=’kmeans’ has been selected.)), frequency (Percentage of days in a month/season belonging to each cluster (only if method=’kmeans’ has been selected).),

Author(s)

Verónica Torralba - BSC, [e-mail]

References

Cortesi, N., V., Torralba, N., González-Reviriego, A., Soret, and F.J., Doblas-Reyes (2019). Characterization of European wind speed variability using weather regimes. Climate Dynamics,53, 4961–4976, doi: 10.1007/s00382-019-04839-5.

Torralba, V. (2019) Seasonal climate prediction for the wind energy sector: methods and tools for the development of a climate service. Thesis. Available online: https://eprints.ucm.es/56841/.

Examples

data <- array(abs(rnorm(1280, 283.7, 6)), dim = c(dataset = 2, member = 2, 
                                                 sdate = 3, ftime = 3, 
                                                 lat = 4, lon = 4))
coords <- list(lon = seq(0, 3), lat = seq(47, 44))
obs <- list(data = data, coords = coords)
class(obs) <- 's2dv_cube'

res1 <- CST_WeatherRegimes(data = obs, EOFs = FALSE, ncenters = 4)
res2 <- CST_WeatherRegimes(data = obs, EOFs = TRUE, ncenters = 3)