Persistence {s2dv}

Compute persistence

Description

Compute a persistence forecast based on a lagged autoregression of observational data along the time dimension, with a measure of forecast uncertainty (prediction interval) based on Coelho et al., 2004.

Usage

Persistence(
  data,
  dates,
  time_dim = "time",
  start,
  end,
  ft_start,
  ft_end = ft_start,
  max_ft = 10,
  nmemb = 1,
  na.action = 10,
  ncores = NULL
)

Arguments

data

A numeric array corresponding to the observational data including the time dimension along which the autoregression is computed. The data should start at least 40 time steps (years or days) before 'start'.

dates

A sequence of 4-digit integers (YYYY) or string (YYYY-MM-DD) in class 'Date' indicating the dates available in the observations.

time_dim

A character string indicating the dimension along which to compute the autoregression. The default value is 'time'.

start

A 4-digit integer (YYYY) or a string (YYYY-MM-DD) in class 'Date' indicating the first start date of the persistence forecast. It must be between 1850 and 2020.

end

A 4-digit integer (YYYY) or a string (YYYY-MM-DD) in class 'Date' indicating the last start date of the persistence forecast. It must be between 1850 and 2020.

ft_start

An integer indicating the forecast time for which the persistence forecast should be calculated, or the first forecast time of the average forecast times for which persistence should be calculated.

ft_end

An (optional) integer indicating the last forecast time of the average forecast times for which persistence should be calculated in the case of a multi-timestep average persistence. The default value is 'ft_start'.

max_ft

An integer indicating the maximum forecast time possible for 'data'. For example, for decadal prediction 'max_ft' would correspond to 10 (years). The default value is 10.

nmemb

An integer indicating the number of ensemble members to generate for the persistence forecast. The default value is 1.

na.action

A function or an integer. A function (e.g., na.omit, na.exclude, na.fail, na.pass) indicates what should happen when the data contain NAs. A numeric indicates the maximum number of NA position allowed to compute regression. The default value is 10.

ncores

An integer indicating the number of cores to use for parallel computation. The default value is NULL.

Value

A list containing:

$persistence

A numeric array with dimensions 'memb', time (start dates), latitudes and longitudes containing the persistence forecast.

$persistence.mean

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the ensemble mean persistence forecast.

$persistence.predint

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the prediction interval of the persistence forecast.

$AR.slope

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the slope coefficient of the autoregression.

$AR.intercept

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the intercept coefficient of the autoregression.

$AR.lowCI

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the lower value of the confidence interval of the autoregression.

$AR.highCI

A numeric array with same dimensions as 'persistence', except the 'memb' dimension which is of length 1, containing the upper value of the confidence interval of the autoregression.

Examples

# Case 1: year
# Building an example dataset with yearly start dates from 1920 to 2009
set.seed(1)
obs1 <- rnorm(1 * 70 * 2 * 2)
dim(obs1) <- c(member = 1, time = 70, lat = 2, lon = 2)
dates <- seq(1920, 1989, 1)
res <- Persistence(obs1, dates = dates, start = 1961, end = 1980, ft_start = 1,
                  nmemb = 2)
# Case 2: day
dates <- seq(as.Date(ISOdate(1990, 1, 1)), as.Date(ISOdate(1990, 4, 1)) ,1)
start <- as.Date(ISOdate(1990, 2, 15))
end <-  as.Date(ISOdate(1990, 4, 1))
set.seed(1)
data <- rnorm(1 * length(dates))
dim(data) <- c(member = 1, time = length(dates))
res <- Persistence(data, dates = dates, start = start, end = end, ft_start = 1)