HiClimR: Hierarchical Climate Regionalization

In HiClimR: Hierarchical Climate Regionalization

Description

HiClimR is a tool for Hierarchical Climate Regionalization applicable to any correlation-based clustering. Climate regionalization is the process of dividing an area into smaller regions that are homogeneous with respect to a specified climatic metric. Several features are added to facilitate the applications of climate regionalization (or spatiotemporal analysis in general) and to implement cluster validation with an objective tree cutting to find an optimal number of clusters for a user-specified confidence level. These include options for preprocessing and postprocessing as well as efficient code execution for large datasets and options for splitting big data and computing only the upper-triangular half of the correlation/dissimilarity matrix to overcome memory limitations. Hybrid hierarchical clustering reconstructs the upper part of the tree above a cut to get the best of the available methods. Multivariate clustering (MVC) provides options for filtering all variables before preprocessing, detrending and standardization of each variable, and applying weights for the preprocessed variables. The correlation distance for MVC represents the (weighted) average of distances between all variables.

HiClimR is the main function that calls all helper functions. It adds several features and a new clustering method (called, regional linkage) to hierarchical clustering in R (hclust function in stats library): data regridding (grid2D function), coarsening spatial resolution (coarseR function), geographic masking (geogMask function), contiguity-constrained clustering, data filtering by mean and/or variance thresholds, data preprocessing (detrending, standardization, and PCA), faster correlation function (fastCor function), hybrid hierarchical clustering, multivariate clustering (MVC), cluster validation (validClimR and minSigCor functions), and visualization of regionalization results, and exporting region map and mean timeseries into NetCDF-4 file.

Badr et al. (2015) describes the regionalization algorithms, features, and data processing tools included in the package and presents a demonstration application in which the package is used to regionalize Africa on the basis of interannual precipitation variability.

HiClimR is applicable to any correlation-based clustering.

Usage

HiClimR(

  # Input data matrix (N spatial elements x M observations)
  x = list(),

  # Geographic coordinates
  lon = NULL, lat = NULL,

  # Coarsening spatial resolution
  lonStep = 1, latStep = 1,

  # Geographic masking:
  geogMask = FALSE, gMask = NULL, continent = NULL, region = NULL, country = NULL,

  # Contiguity constraint:
  contigConst = 0,

  # Data thresholds:
  meanThresh = if (inherits(x, "list")) {
      vector("list", length(x))
  } else {
      list(NULL)
  },
  varThresh = if (inherits(x, "list")) {
      as.list(rep(0, length(x)))
  } else {
      list(0)
  },

  # Data preprocessing:
  detrend = if (inherits(x, "list")) {
      as.list(rep(FALSE, length(x)))
  } else {
      list(FALSE)
  },
  standardize = if (inherits(x, "list")) {
      as.list(rep(FALSE, length(x)))
  } else {
      list(FALSE)
  },
  weightMVC = if (inherits(x, "list")) {
      as.list(rep(1, length(x)))
  } else {
      list(1)
  },
  nPC = NULL,

  # Clustering options:
  method = "ward", hybrid = FALSE, kH = NULL, members = NULL,

  # Big data support:
  nSplit = 1, upperTri = TRUE, verbose = TRUE,

  # Cluster validation:
  validClimR = TRUE, rawStats = TRUE, k = NULL, minSize = 1, alpha = 0.05,

  # Graphical options:
  plot = TRUE, dendrogram = TRUE, colPalette = NULL,
  hang = -1, labels = FALSE, pch = 15, cex = 1

  )

Arguments

x	an (N rows by M columns) matrix of 'double' values: N objects (spatial points or stations) to be clustered by M observations (temporal points or years). For gridded data, the N objects should be created from the original matrix x0 using as.vector(t(x0)), where x0 is an (n rows by m columns) matrix, n = length(unique(lon)) and m = length(unique(lat)). Zero-variance rows (e.g., stations with zero variability) and/or missing values (e.g., years with missing observations) are allowed. The zero-variance rows and the columns with missing values will be removed. However, it is recommended to take care of both zero-variance rows and missing values before clustering. For Multivariate Clustering (MVC), x can be a list of nvar matrices for the nvar variables (one matrix for each variable). The matrixes in x list should have the same number of rows (objects: spatial points or stations) Data preprocessing is specified by lists of meanThresh, varThresh, detrend, and standardize with the same length of x where length(x) = nvar. Each variable is separately preprocessed to allow for all possible options. However, it is strongly recommended to standardize all variables since their magnitude range could be different. Note that: for gridded data, the rows of input data matrix for each variable is ordered by longitudes (check TestCase$x for more details).
lon	a vector of longitudes with length N. Longitudes takes values from -180 to 180 (not 0 to 360). For gridded data, the length may have the value (n) provided that n * m = N where n = length(unique(lon)) and m = length(unique(lat)).
lat	a vector of latitudes with length N or m. See lon.
lonStep	an integer greater than or equal to 1 for longitude step to coarsen gridded data in the longitudinal direction. If lonStep = 1, gridded data will not be coarsened in the longitudinal direction (the default). If lonStep = 2, every other grid in longitudinal direction will be retained.
latStep	an integer greater than or equal to 1 for latitude step to coarsen gridded data in the latitudinal direction. If latStep = 1, gridded data will not be coarsened in the latitudinal direction (the default). If latStep = 2, every other grid in latitudinal direction will be retained. lonStep and latStep are independent so that user can optionally apply different coarsening level to each dimension.
geogMask	a logical: if geogMask = TRUE, geogMask function will be called. Additional arguments are required. It requires the longitude and latitude vector together with a string (or array of strings) to specify continent, region name(s), or country ISO3 character code(s). If gMask != NULL, the provided gmask vector will be used for geographic masking without calling geogMask.
gMask	A vector of indices for the spatial elements to be masked, as required by HiClimR. This is typically an output vector from geogMask function. This helps in saving time when the same geographic mask will be used many times.
continent	NULL or a string (or array of strings) to specify continent name(s): only one of continent, region, or country should be specified. Valid list of continent names can be obtained by running geogMask().
region	NULL or a string (or array of strings) to specify region name(s): only one of continent, region, or country should be specified. Valid list of region names can be obtained by running geogMask().
country	NULL or a string (or array of strings) to specify country ISO3 character code(s): only one of continent, region, or country should be specified. Valid list of country ISO3 character code(s) can be obtained by running geogMask().
contigConst	0 or a positive value for contiguity constraint: This is used to apply weighted geographic distances on the initial dissimilarity matrix between the variable(s) in the data matrix x. It should be a positive real value (contigConst > 0, recommended between 0 and 1) where contigConst = 0 means that clustering is completely based on the data with no constraints for contiguity while contigConst > 0 gives priority to merging contiguous points. Geographic coordinates lon and lat are required to compute the geographic distances. regional linkage method does not support contiguity constraints.
meanThresh	NULL or a threshold for the temporal mean: This is used with varThresh to mask zero- and near-zero-variance data, Observations with mean less than or equal to meanThresh will be removed. If meanThresh = NULL, then the varThresh will be used either to mask zero-variance data by default or by increased variance threshold to mask near-zero-variance data. For Multivariate Clustering (MVC), meanThresh is a list of thresholds with the same length of x where length(x) = nvar. Each variable is separately preprocessed to allow for all possible options. However, it is strongly recommended to standardize all variables since their magnitude range could be different.
varThresh	zero or a threshold for the temporal variance: This is used with meanThresh to mask zero- and near-zero-variance data, Observations with variance less than or equal to varThresh will be removed. If varThresh = 0, then the zero-variance data will masked (default). For Multivariate Clustering (MVC), varThresh is a list of thresholds with the same length of x where length(x) = nvar. Each variable is separately preprocessed to allow for all possible options. However, it is strongly recommended to standardize all variables since their magnitude range could be different.
detrend	a logical: should the data be detrended before clustering? Detrending (removing the linear trend) is important when variations from temporal point to another is of interest (e.g., interannual variability). The columns of the data matrix x should be temporally ordered (constant step size) or have appropriate names (e.g., colnames(x) = years[1:M]). For Multivariate Clustering (MVC), detrend is a list of thresholds with the same length of x where length(x) = nvar. Each variable is separately preprocessed to allow for all possible options. However, it is strongly recommended to standardize all variables since their magnitude range could be different.
standardize	a logical: should the data be standardized before clustering? The standardized data makes use of the mean of equally-weighted objects within each cluster (cluster mean = mean of standardized variables within the cluster). Otherwise, the mean of raw data will be used (cluster mean = mean of raw variables within the cluster). The variance of the mean is updated at each agglomeration step. For Multivariate Clustering (MVC), standardize is a list of thresholds with the same length of x where length(x) = nvar. Each variable is separately preprocessed to allow for all possible options. However, it is strongly recommended to standardize all variables since their magnitude range could be different.
weightMVC	a list of positive wights (weightMVC > 0) for Multivariate Clustering (MVC) with the same length of x where length(x) = number of variables. The filtered variables are weighted and combined by column (for each object: spatial points or stations) after preprocessing (detrending and standardization) and before PCA (if requested) and computing the correlation/dissimilarity matrix. The default weight is weightMVC = 1 for all variables.
nPC	NULL or number of principal components (PCs) to be retained. If nPC = NULL, then the raw data will be used for clustering. Otherwise, the data will be filtered and reconstructed using nPC PCs obtained from SVD-based PCA. The detrend and/or standardize options will be applied, if requested, before PCA.
method	the agglomeration method to be used. This should be (an unambiguous abbreviation of) one of "regional", "ward", "single", "complete", "average", "mcquitty", "median" or "centroid". The default is "ward" method.
hybrid	a logical: should the upper part of the tree be reconstructed using regional linkage clustering method? This adds hybrid hierarchical clustering feature to get the best of the available methods. It utilizes the pros of available methods, especially the better overall homogeneity in ward's method and the separation, contiguity, and objective tree cut of regional linkage method. If hybrid = FALSE, only the default clustering using the selected method will be used (i.e., no hybrid clustering).Otherwise, the upper part of the tree will be reconstructed above a cut of kH clusters using regional linkage method. Note: hybrid option is redundant when using regional linkage as the main clustering method.
kH	NULL or an integer for the number of regions/clusters in the upper part of the tree to be reconstructed with regional linkage method, if hybrid = TRUE. If kH = NULL, the tree will be reconstructed for the upper part with the first merging cost larger than the mean merging cost for the entire tree. If hybrid clustering is requested, the updated upper part of the tree will be used for cluster validation, and so kH should be greater than the final number of clusters k, if selected.
members	NULL or a vector with length size of d. See the ‘Details’ section.
nSplit	integer number greater than or equal to one, to split the data matrix into nSplit splits of the total number of columns ncol(xt). If nSplit = 1, the default method will be used to compute correlation matrix for the full data matrix (no splits). If nSplit > 1, the correlation matrix (or the upper-triangular part if upperTri = TRUE) will be allocated and filled with the computed correlation sub-matrix for each split. the first n-1 splits have equal size while the last split may include any remaining columns. This is used with upperTri = TRUE to compute only the upper-triangular part of the correlation matrix. The maximum number of splits nSplitMax = floor(N / 2) makes splits with 2 columns; if nSplit > nSplitMax, nSplitMax will be used. Very large number of splits nSplit makes computation slower but it could handle big data or if the available memory is not enough to allocate the correlation matrix, which helps in solving the “Error: cannot allocate vector of size...” memory limitation problem. It is recommended to start with a small number of splits. If the data is very large compared to the physical memory, it is highly recommended to use a 64-Bit machine with enough memory resources and/or use coarsening feature for gridded data by setting lonStep > 1 and latStep > 1.
upperTri	logical to compute only the upper-triangular half of the correlation matrix if upperTri = TRUE and nSplit > 1., which includes all required info since the correlation/dissimilarity matrix is symmetric. This almost halves memory use, which can be very important for big data.
verbose	logical to print processing information if verbose = TRUE.
validClimR	a logical: If validClimR = TRUE, validClimR will be called to compute validation indices including statistical summary for inter- and intra-cluster correlations. This is computationally expensive. It can also objectively cut the dendrogram tree for regional clustering method, if k = NULL.
rawStats	a logical: should validation indices be computed based on the raw data or PCA-filtered data?
k	NULL or an integer k > 1 for the number of regions/clusters. Only for regional linkage method, k = NULL is supported, where the "optimal" number of regions will be used at a user specified significance level alpha. It is required to specify number of clusters k for the other methods, since they are not based on inter-cluster correlation. If k = NULL for these methods (except regional) linkage, the validClimR with be aborted. One can use validClimR function to compute inter-cluster correlation at different number of clusters to objectively cut the tree for the other methods, which could be computationally expensive to cover the entire merging history for large number of spatial elements.
minSize	minimum cluster size. The regional linkage method tend to isolate noisy data in small clusters. The minSize can be used to exclude these very small clusters from the statSum statistical summary, because they are most likely noisy data that need to be checked in a quality control step. The analysis may be then repeated.
alpha	confidence level: the default is alpha = 0.05 for 95% confidence level.
plot	logical to call the plotting method if plot = TRUE.
dendrogram	logical to enable or disable dendrogram plotting.
colPalette	a color palette or a list of colors such as that generated by rainbow, heat.colors, topo.colors, terrain.colors or similar functions.
hang	The fraction of the plot height by which labels should hang below the rest of the plot. A negative value will cause the labels to hang down from 0.
labels	A character vector of labels for the leaves of the tree. By default the row names or row numbers of the original data are used. If labels = FALSE no labels at all are plotted.
pch	Either an integer specifying a symbol or a single character to be used as the default in plotting points. See points for possible values.
cex	A numerical value giving the amount by which plotting symbols should be magnified relative to the default = 1.

Details

HiClimR function is based on hclust, which now uses an optimized algorithm to deal with only the upper/lower triangular-half of the symmetric dissimilarity matrix instead of the old algorithm that uses the full matrix in the merging steps. It performs a hierarchical cluster analysis using Pearson correlation distance dissimilarity for the N objects being clustered. Initially, each object is assigned to its own cluster and then the algorithm proceeds iteratively, at each stage joining the two most similar clusters, continuing until there is just a single cluster. At each stage distances between clusters are recomputed by a dissimilarity update formula according to the particular clustering method being used.

All clustering methods in hclust are included. The regional linkage method minimizes inter-cluster correlations between cluster means (see Badr et al. 2015). Ward's minimum variance method aims at finding compact, spherical clusters. The complete linkage method finds similar clusters. The single linkage method (which is closely related to the minimal spanning tree) adopts a ‘friends of friends’ clustering strategy. The other methods can be regarded as aiming for clusters with characteristics somewhere between the single and complete link methods. Note however, that methods "median" and "centroid" are not leading to a monotone distance measure, or equivalently the resulting dendrograms can have so called inversions (which are hard to interpret).

The regional linkage method is explained in the context of a spatiotemporal problem, in which N spatial elements (e.g., weather stations) are divided into k regions, given that each element has a time series of length M. It is based on inter-regional correlation distance between the temporal means of different regions (or elements at the first merging step). It modifies the update formulae of average linkage method by incorporating the standard deviation of the merged region timeseries, which is a function of the correlation between the individual regions, and their standard deviations before merging. It is equal to the average of their standard deviations if and only if the correlation between the two merged regions is 100%. In this special case, the regional linkage method is reduced to the classic average linkage clustering method.

If members != NULL, then d is taken to be a dissimilarity matrix between clusters instead of dissimilarities between singletons and members gives the number of observations per cluster. This way the hierarchical cluster algorithm can be ‘started in the middle of the dendrogram’, e.g., in order to reconstruct the part of the tree above a cut (see examples). Dissimilarities between clusters can be efficiently computed (i.e., without hclust itself) only for a limited number of distance/linkage combinations, the simplest one being squared Euclidean distance and centroid linkage. In this case the dissimilarities between the clusters are the squared Euclidean distances between cluster means.

In hierarchical cluster displays, a decision is needed at each merge to specify which subtree should go on the left and which on the right. Since, for n observations there are n-1 merges, there are 2^{(n-1)} possible orderings for the leaves in a cluster tree, or dendrogram. The algorithm used in hclust is to order the subtree so that the tighter cluster is on the left (the last, i.e., most recent, merge of the left subtree is at a lower value than the last merge of the right subtree). Single observations are the tightest clusters possible, and merges involving two observations place them in order by their observation sequence number.

Value

An object of class HiClimR and hclust, which describes the tree produced by the clustering process. The object is a list with the following components:

merge	an n-1 by 2 matrix. Row i of merge describes the merging of clusters at step i of the clustering. If an element j in the row is negative, then observation -j was merged at this stage. If j is positive then the merge was with the cluster formed at the (earlier) stage j of the algorithm. Thus negative entries in merge indicate agglomerations of singletons, and positive entries indicate agglomerations of non-singletons.
height	a set of n-1 real values (non-decreasing for ultrametric trees). The clustering height: that is, the value of the criterion associated with the clustering method for the particular agglomeration.
order	a vector giving the permutation of the original observations suitable for plotting, in the sense that a cluster plot using this ordering and matrix merge will not have crossings of the branches.
labels	labels for each of the objects being clustered.
method	the cluster method that has been used.
call	the call which produced the result.
dist.method	the distance that has been used to create d (only returned if the distance object has a "method" attribute).
skip	a vector of lonStep and latStep if coarseR = TRUE.
PCA	if nPC != NULL, the eigenvalues, explained variance, and accumulated variance will be stored here.
coords	an (Nc rows by 2 columns) matrix of 'double' values: longitude and latitude coordinates for the preprocessed data used for clustering, where NC is the number of spatial elements to be clustered after coarsening spatial resolution by lonStep and latStep. It will be returned only if lon and lat vectors were defined. It will be similar to the provided lon and lat if lonStep = 1 and latStep = 1.
nvars	number of variables used for multivariate clustering (MVC).
ncols	number of columns for each variable.
data	the preprocessed data used for clustering will be stored here. If rawData = TRUE and nPC != NULL, the preprocessed data without filtering (PCA) will be returned here.
mask	a vector of the indices of masked spatial points by both geographic masking and data thresholds.
treeH	An object of class hclust, which describes the upper part of the tree reconstructed by the hybrid clustering process if hybrid = TRUE.

If validClimR = TRUE, an object of class HiClimR, which produces indices for validating the tree produced by the clustering process, will be merged in the dendrogram tree above. This object is a list with the following components:

cutLevel	the minimum significant correlation used for objective tree cut together with the corresponding confidence level.
clustMean	the cluster means which are the region's mean timeseries for all selected regions.
clustSize	cluster sizes for all selected regions.
clustFlag	a flag 0 or 1 to indicate the cluster used in statSum validation indices (interCor, intraCor, diffCor, and statSum), based on minSize minimum cluster size. If clustFlag = 0, the cluster has been excluded because its size is less than the minSize minimum cluster size. The sum of clustFlag elements represents the selected number clusters.
interCor	inter-cluster correlations for all selected regions. It is the inter-cluster correlations between cluster means. The maximum inter-cluster correlation is a measure for separation or contiguity, and it is used for objective tree cut (to find the "optimal" number of clusters).
intraCor	intra-cluster correlations for all selected regions. It is the intra-cluster correlations between the mean of each cluster and its members. The average intra-cluster correlation is a weighted average for all clusters, and it is a measure for homogeneity.
diffCor	difference between intra-cluster correlation and maximum inter-cluster correlation for all selected regions.
statSum	overall statistical summary for interCluster, intraCor, and diffCor.
region	ordered regions vector of size N number of spatial elements for the selected number of clusters, after excluding the small clusters defined by minSize argument.
regionID	ordered regions ID vector of length equals the selected number of clusters, after excluding the small clusters defined by minSize argument. It helps in mapping ordered regions and their actual names before ordering. Only the region component uses ordered ID, while other components use the names used during the clustering process.

There are print, plot and identify (see identify.hclust) methods and rect.hclust() functions for hclust objects.

Author(s)

Hamada S. Badr <[e-mail]>, Benjamin F. Zaitchik <[e-mail]>, and Amin K. Dezfuli <[e-mail]>. HiClimR is a modification of hclust function, which is based on Fortran code contributed to STATLIB by F. Murtagh.

References

Hamada S. Badr, Zaitchik, B. F. and Dezfuli, A. K. (2015): A Tool for Hierarchical Climate Regionalization, Earth Science Informatics, 8(4), 949-958, doi: 10.1007/s12145-015-0221-7.

Hamada S. Badr, Zaitchik, B. F. and Dezfuli, A. K. (2014): Hierarchical Climate Regionalization, Comprehensive R Archive Network (CRAN), https://cran.r-project.org/package=HiClimR.

Wilks, D. S. (2011): Statistical methods in the atmospheric sciences, Academic press.

Gordon, A. D. (1999): Classification. Second Edition. London: Chapman and Hall / CRC

Becker, R. A., Chambers, J. M. and Wilks, A. R. (1988): The New S Language. Wadsworth & Brooks/Cole. (S version.)

Murtagh, F. (1985): “Multidimensional Clustering Algorithms”, in COMPSTAT Lectures 4. Wuerzburg: Physica-Verlag (for algorithmic details of algorithms used).

Hartigan, J. A. (1975): Clustering Algorithms. New York: Wiley.

Everitt, B. (1974): Cluster Analysis. London: Heinemann Educ. Books.

Anderberg, M. R. (1973): Cluster Analysis for Applications. Academic Press: New York.

Sneath, P. H. A. and R. R. Sokal (1973): Numerical Taxonomy. San Francisco: Freeman.

McQuitty, L.L. (1966): Similarity Analysis by Reciprocal Pairs for Discrete and Continuous Data. Educational and Psychological Measurement, 26, 825-831.

Source Code: https://github.com/hsbadr/HiClimR

See Also

HiClimR, HiClimR2nc, validClimR, geogMask, coarseR, fastCor, grid2D, minSigCor. identify.hclust, rect.hclust, cutree, dendrogram, and kmeans.

Examples

require(HiClimR)

#----------------------------------------------------------------------------------#
# Typical use of HiClimR for single-variate clustering:                            #
#----------------------------------------------------------------------------------#

## Load the test data included/loaded in the package (1 degree resolution)
x <- TestCase$x
lon <- TestCase$lon
lat <- TestCase$lat

## Generate/check longitude and latitude mesh vectors for gridded data
xGrid <- grid2D(lon = unique(TestCase$lon), lat = unique(TestCase$lat))
lon <- c(xGrid$lon)
lat <- c(xGrid$lat)

## Single-Variate Hierarchical Climate Regionalization
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = FALSE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## For more examples: https://github.com/hsbadr/HiClimR#examples

## Not run: 

#----------------------------------------------------------------------------------#
# Additional Examples:                                                             #
#----------------------------------------------------------------------------------#

## Use Ward's method
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = FALSE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## Use data splitting for big data
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = FALSE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = TRUE, kH = NULL,
    members = NULL, nSplit = 10, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## Use hybrid Ward-Regional method
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = FALSE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = TRUE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)
## Check senitivity to kH for the hybrid method above


#----------------------------------------------------------------------------------#
# Typical use of HiClimR for multivariate clustering:                              #
#----------------------------------------------------------------------------------#

## Load the test data included/loaded in the package (1 degree resolution)
x1 <- TestCase$x
lon <- TestCase$lon
lat <- TestCase$lat

## Generate/check longitude and latitude mesh vectors for gridded data
xGrid <- grid2D(lon = unique(TestCase$lon), lat = unique(TestCase$lat))
lon <- c(xGrid$lon)
lat <- c(xGrid$lat)

## Test if we can replicate single-variate region map with repeated variable
y <- HiClimR(x=list(x1, x1), lon = lon, lat = lat, lonStep = 1, latStep = 1,
    geogMask = FALSE, continent = "Africa", meanThresh = list(10, 10),
    varThresh = list(0, 0), detrend = list(TRUE, TRUE), standardize = list(TRUE, TRUE),
    nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## Generate a random matrix with the same number of rows
x2 <- matrix(rnorm(nrow(x1) * 100, mean=0, sd=1), nrow(x1), 100)

## Multivariate Hierarchical Climate Regionalization
y <- HiClimR(x=list(x1, x2), lon = lon, lat = lat, lonStep = 1, latStep = 1,
    geogMask = FALSE, continent = "Africa", meanThresh = list(10, NULL),
    varThresh = list(0, 0), detrend = list(TRUE, FALSE), standardize = list(TRUE, TRUE),
    weightMVC = list(1, 1), nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)
## You can apply all clustering methods and options

#----------------------------------------------------------------------------------#
# Miscellaneous examples to provide more information about functionality and usage #
# of the helper functions that can be used separately or for other applications.   #
#----------------------------------------------------------------------------------#

## Load test case data
x <- TestCase$x

## Generate longitude and latitude mesh vectors
xGrid <- grid2D(lon = unique(TestCase$lon), lat = unique(TestCase$lat))
lon <- c(xGrid$lon)
lat <- c(xGrid$lat)

## Coarsening spatial resolution
xc <- coarseR(x = x, lon = lon, lat = lat, lonStep = 2, latStep = 2)
lon <- xc$lon
lat <- xc$lat
x <- xc$x

## Use fastCor function to compute the correlation matrix
t0 <- proc.time(); xcor <- fastCor(t(x)); proc.time() - t0
## compare with cor function
t0 <- proc.time(); xcor0 <- cor(t(x)); proc.time() - t0

## Check the valid options for geographic masking
geogMask()

## geographic mask for Africa
gMask <- geogMask(continent = "Africa", lon = lon, lat = lat, plot = TRUE,
    colPalette = NULL)

## Hierarchical Climate Regionalization Without geographic masking
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = FALSE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## With geographic masking (you may specify the mask produced above to save time)
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = TRUE,
    continent = "Africa", meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## With geographic masking and contiguity contraint
## Change contigConst as appropriate
y <- HiClimR(x, lon = lon, lat = lat, lonStep = 1, latStep = 1, geogMask = TRUE,
    continent = "Africa", contigConst = 1, meanThresh = 10, varThresh = 0, detrend = TRUE,
    standardize = TRUE, nPC = NULL, method = "ward", hybrid = FALSE, kH = NULL,
    members = NULL, nSplit = 1, upperTri = TRUE, verbose = TRUE,
    validClimR = TRUE, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL, hang = -1, labels = FALSE)

## Find minimum significant correlation at 95
rMin <- minSigCor(n = nrow(x), alpha = 0.05, r = seq(0, 1, by = 1e-06))

## Validtion of Hierarchical Climate Regionalization
z <- validClimR(y, k = 12, minSize = 1, alpha = 0.01,
    plot = TRUE, colPalette = NULL)

## Apply minimum cluster size (minSize = 25)
z <- validClimR(y, k = 12, minSize = 25, alpha = 0.01,
    plot = TRUE, colPalette = NULL)

## The optimal number of clusters, including small clusters
k <- length(z$clustFlag)

## The selected number of clusters, after excluding small clusters (if minSize > 1)
ks <- sum(z$clustFlag)

## Dendrogram plot
plot(y, hang = -1, labels = FALSE)

## Tree cut
cutTree <- cutree(y, k = k)
table(cutTree)

## Visualization for gridded data
RegionsMap <- matrix(y$region, nrow = length(unique(y$coords[, 1])), byrow = TRUE)
colPalette <- colorRampPalette(c("#00007F", "blue", "#007FFF", "cyan",
    "#7FFF7F", "yellow", "#FF7F00", "red", "#7F0000"))
image(unique(y$coords[, 1]), unique(y$coords[, 2]), RegionsMap, col = colPalette(ks))

## Visualization for gridded or ungridded data
plot(y$coords[, 1], y$coords[, 2], col = colPalette(max(Regions, na.rm = TRUE))[y$region],
    pch = 15, cex = 1)

## Export region map and mean timeseries into NetCDF-4 file
y.nc <- HiClimR2nc(y=y, ncfile="HiClimR.nc", timeunit="years", dataunit="mm")
## The NetCDF-4 file is still open to add other variables or close it
nc_close(y.nc)


## End(Not run)

HiClimR documentation built on Jan. 21, 2022, 1:08 a.m.