Newer
Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
#'Compute the correlation coefficient between an array of forecast and their corresponding observation
#'
#'Calculate the correlation coefficient (Pearson, Kendall or Spearman) for
#'an array of forecast and an array of observation. The correlations are
#'computed along time_dim, the startdate dimension. If comp_dim is given,
#'the correlations are computed only if data along the comp_dim dimension are
#'complete between limits[1] and limits[2], i.e., there is no NA between
#'limits[1] and limits[2]. This option can be activated if the user wants to
#'account only for the forecasts which the corresponding observations are
#'available at all leadtimes.\cr
#'The confidence interval is computed by the Fisher transformation and the
#'significance level relies on an one-sided student-T distribution.\cr
#'
#'@param exp A named numeric array of experimental data, with at least two
#' dimensions 'time_dim' and 'memb_dim'.
#'@param obs A named numeric array of observational data, same dimensions as
#' parameter 'exp' except along memb_dim.
#'@param time_dim A character string indicating the name of dimension along
#' which the correlations are computed. The default value is 'sdate'.
#'@param memb_dim A character string indicating the name of member (nobs/nexp)
#' dimension. The default value is 'member'.
#'@param comp_dim A character string indicating the name of dimension along which
#' the data is taken into account only if it is complete. The default value
#' is NULL.
#'@param limits A vector of two integers indicating the range along comp_dim to
#' be completed. The default is c(1, length(comp_dim dimension)).
#'@param method A character string indicating the type of correlation:
#' 'pearson', 'spearman', or 'kendall'. The default value is 'pearson'.
#'@param pval A logical value indicating whether to compute or not the p-value
#' of the test Ho: Corr = 0. The default value is TRUE.
#'@param conf A logical value indicating whether to retrieve the confidence
#' intervals or not. The default value is TRUE.
#'@param conf.lev A numeric indicating the confidence level for the
#' regression computation. The default value is 0.95.
#'@param ncores An integer indicating the number of cores to use for parallel
#' computation. The default value is NULL.
#'
#'@return
#'A list containing the numeric arrays with dimension:\cr
#' c(nexp, nobs, all other dimensions of exp except time_dim).\cr
#'nexp is the number of experiment (i.e., memb_dim in exp), and nobs is the
#'number of observation (i.e., memb_dim in obs).\cr
#'\item{$corr}{
#' The correlation coefficient.
#'}
#'\item{$p.val}{
#' The p-value. Only present if \code{pval = TRUE}.
#'}
#'\item{$conf.lower}{
#' The lower confidence interval. Only present if \code{conf = TRUE}.
#'}
#'\item{$conf.upper}{
#' The upper confidence interval. Only present if \code{conf = TRUE}.
#'}
#'
#'@keywords datagen
#'@author History:\cr
#'0.1 - 2011-04 (V. Guemas, \email{vguemas@bsc.es}) - Original code\cr
#'1.0 - 2013-09 (N. Manubens, \email{nicolau.manubens@bsc.es}) - Formatting to R CRAN\cr
#'1.1 - 2014-10 (M. Menegoz, \email{martin.menegoz@bsc.es}) - Adding conf.lev argument\cr
#'1.2 - 2015-03 (L.P. Caron, \email{louis-philippe.caron@@bsc.es}) - Adding method argument\cr
#'1.3 - 2017-02 (A. Hunter, \email{alasdair.hunter@bsc.es}) - Adapted to veriApply()
#'3.0 - 2020-01 (A. Ho, \email{an.ho@bsc.es}) - Adapt multiApply feature
#'@examples
#'# Load sample data as in Load() example:
#'example(Load)
#'clim <- Clim(sampleData$mod, sampleData$obs)
#'ano_exp <- Ano(sampleData$mod, clim$clim_exp)
#'ano_obs <- Ano(sampleData$obs, clim$clim_obs)
#'runmean_months <- 12
#'dim_to_smooth <- 4
#'# Smooth along lead-times
#'smooth_ano_exp <- Smoothing(ano_exp, runmean_months, dim_to_smooth)
#'smooth_ano_obs <- Smoothing(ano_obs, runmean_months, dim_to_smooth)
#'leadtimes_per_startdate <- 60
#'corr <- Corr(smooth_ano_exp,
#' smooth_ano_obs,
#' comp_dim = 'ftime', #Discard start dates which contain any NA ftime
#' limits = c(ceiling((runmean_months + 1) / 2),
#' leadtimes_per_startdate - floor(runmean_months / 2)))
#'
#'@rdname Corr
#'@import multiApply
#'@importFrom stats cor pt qnorm
#'@export
Corr <- function(exp, obs, time_dim = 'sdate', memb_dim = 'member',
comp_dim = NULL, limits = NULL,
method = 'pearson', pval = TRUE, conf = TRUE,
conf.lev = 0.95, ncores = NULL) {
# Check inputs
## exp and obs (1)
if (is.null(exp) | is.null(obs)) {
stop("Parameter 'exp' and 'obs' cannot be NULL.")
}
if (!is.numeric(exp) | !is.numeric(obs)) {
stop("Parameter 'exp' and 'obs' must be a numeric array.")
}
if (is.null(dim(exp)) | is.null(dim(obs))) {
stop(paste0("Parameter 'exp' and 'obs' must be at least two dimensions ",
"containing time_dim and memb_dim."))
}
if(any(is.null(names(dim(exp))))| any(nchar(names(dim(exp))) == 0) |
any(is.null(names(dim(obs))))| any(nchar(names(dim(obs))) == 0)) {
stop("Parameter 'exp' and 'obs' must have dimension names.")
}
if(!all(names(dim(exp)) %in% names(dim(obs))) |
!all(names(dim(obs)) %in% names(dim(exp)))) {
stop("Parameter 'exp' and 'obs' must have same dimension name")
}
## time_dim
if (!is.character(time_dim) | length(time_dim) > 1) {
stop("Parameter 'time_dim' must be a character string.")
}
if (!time_dim %in% names(dim(exp)) | !time_dim %in% names(dim(obs))) {
stop("Parameter 'time_dim' is not found in 'exp' or 'obs' dimension.")
}
## memb_dim
if (!is.character(memb_dim) | length(memb_dim) > 1) {
stop("Parameter 'memb_dim' must be a character string.")
}
if (!memb_dim %in% names(dim(exp)) | !memb_dim %in% names(dim(obs))) {
stop("Parameter 'memb_dim' is not found in 'exp' or 'obs' dimension.")
}
## comp_dim
if (!is.null(comp_dim)) {
if (!is.character(comp_dim) | length(comp_dim) > 1) {
stop("Parameter 'comp_dim' must be a character string.")
}
if (!comp_dim %in% names(dim(exp)) | !comp_dim %in% names(dim(obs))) {
stop("Parameter 'comp_dim' is not found in 'exp' or 'obs' dimension.")
}
}
## limits
if (!is.null(limits)) {
if (is.null(comp_dim)) {
stop("Paramter 'comp_dim' cannot be NULL if 'limits' is assigned.")
}
if (!is.numeric(limits) | any(limits %% 1 != 0) | any(limits < 0) |
length(limits) != 2 | any(limits > dim(exp)[comp_dim])) {
stop(paste0("Parameter 'limits' must be a vector of two positive ",
"integers smaller than the length of paramter 'comp_dim'."))
}
}
## method
if (!(method %in% c("kendall", "spearman", "pearson"))) {
stop("Parameter 'method' must be one of 'kendall', 'spearman' or 'pearson'.")
}
## pval
if (!is.logical(pval) | length(pval) > 1) {
stop("Parameter 'pval' must be one logical value.")
}
## conf
if (!is.logical(conf) | length(conf) > 1) {
stop("Parameter 'conf' must be one logical value.")
}
## conf.lev
if (!is.numeric(conf.lev) | conf.lev < 0 | conf.lev > 1 | length(conf.lev) > 1) {
stop("Parameter 'conf.lev' must be a numeric number between 0 and 1.")
}
## ncores
if (!is.null(ncores)) {
if (!is.numeric(ncores) | ncores %% 1 != 0 | ncores < 0 |
length(ncores) > 1) {
stop("Parameter 'ncores' must be a positive integer.")
}
}
## exp and obs (2)
name_exp <- sort(names(dim(exp)))
name_obs <- sort(names(dim(obs)))
name_exp <- name_exp[-which(name_exp == memb_dim)]
name_obs <- name_obs[-which(name_obs == memb_dim)]
if(!all(dim(exp)[name_exp] == dim(obs)[name_obs])) {
stop(paste0("Parameter 'exp' and 'obs' must have same length of ",
"all dimension expect 'memb_dim'."))
}
if (dim(exp)[time_dim] < 3) {
stop("The length of time_dim must be at least 3 to compute correlation.")
}
###############################
# Sort dimension
name_exp <- names(dim(exp))
name_obs <- names(dim(obs))
order_obs <- match(name_exp, name_obs)
obs <- .aperm2(obs, order_obs)
###############################
# Calculate Corr
# Remove data along comp_dim dim if there is at least one NA between limits
if (!is.null(comp_dim)) {
if (is.null(limits)) {
limits <- c(1, dim(obs)[comp_dim])
}
pos <- which(names(dim(obs)) == comp_dim)
outrows <- is.na(Mean1Dim(obs, pos, narm = FALSE, limits))
outrows <- InsertDim(outrows, pos, dim(obs)[comp_dim])
obs[which(outrows)] <- NA
}
res <- Apply(list(exp, obs),
target_dims = list(c(time_dim, memb_dim),
c(time_dim, memb_dim)),
fun = .Corr,
time_dim = time_dim, method = method,
pval = pval, conf = conf, conf.lev = conf.lev,
ncores = ncores)
return(res)
}
.Corr <- function(exp, obs, time_dim = 'sdate', method = 'pearson',
conf = TRUE, pval = TRUE, conf.lev = 0.95) {
# exp: [sdate, member_exp]
# obs: [sdate, member_obs]
n_exp <- as.numeric(dim(exp)[2])
n_obs <- as.numeric(dim(obs)[2])
CORR <- array(dim = c(n_exp = n_exp, n_obs = n_obs))
eno_expand <- array(dim = c(n_exp = n_exp, n_obs = n_obs))
p.val <- array(dim = c(n_exp = n_exp, n_obs = n_obs))
# ens_mean
for (i in 1:n_obs) {
CORR[, i] <- sapply(1:n_exp,
function(x) {
if (any(!is.na(exp[, x])) && sum(!is.na(obs[, i])) > 2) {
cor(exp[, x], obs[, i],
use = "pairwise.complete.obs",
method = method)
} else {
CORR[, i] <- NA
}
})
}
# if (pval) {
# for (i in 1:n_obs) {
# p.val[, i] <- try(sapply(1:n_exp,
# function(x) {(cor.test(exp[, x], obs[, i],
# use = "pairwise.complete.obs",
# method = method)$p.value)/2}), silent = TRUE)
# if (class(p.val[, i]) == 'character') {
# p.val[, i] <- NA
# }
# }
# }
if (pval | conf) {
if (method == "kendall" | method == "spearman") {
tmp <- apply(obs, 2, rank)
names(dim(tmp))[1] <- time_dim
eno <- Eno(tmp, time_dim)
} else if (method == "pearson") {
eno <- Eno(obs, time_dim)
}
for (i in 1:n_exp) {
eno_expand[i, ] <- eno
}
}
#############old#################
#This doesn't return error but it's diff from cor.test() when method is spearman and kendall
if (pval) {
t <-sqrt(CORR * CORR * (eno_expand - 2) / (1 - (CORR ^ 2)))
p.val <- pt(t, eno_expand - 2, lower.tail = FALSE)
}
###################################
if (conf) {
conf.lower <- (1 - conf.lev) / 2
conf.upper <- 1 - conf.lower
conflow <- tanh(atanh(CORR) + qnorm(conf.lower) / sqrt(eno_expand - 3))
confhigh <- tanh(atanh(CORR) + qnorm(conf.upper) / sqrt(eno_expand - 3))
}
if (pval & conf) {
res <- list(corr = CORR, p.val = p.val,
conf.lower = conflow, conf.upper = confhigh)
} else if (pval & !conf) {
res <- list(corr = CORR, p.val = p.val)
} else if (!pval & conf) {
res <- list(corr = CORR,
conf.lower = conflow, conf.upper = confhigh)
} else {
res <- list(corr = CORR)
}
return(res)
}