sensitivity_ews is used to estimate trends in
statistical moments for different sizes of rolling windows
along a timeseries. The trends are estimated by the
nonparametric Kendall tau correlation coefficient.
a numeric vector of the observed
univariate timeseries values or a numeric matrix where
the first column represents the time index and the second
the observed timeseries values. Use vectors/matrices with
headings.
indicator
is the statistic (leading indicator)
selected for which the sensitivity analysis is perfomed.
Currently, the indicators supported are: ar1
autoregressive coefficient of a first order AR model,
sd standard deviation, acf1 autocorrelation
at first lag, sk skewness, kurt kurtosis,
cv coeffcient of variation, returnrate, and
densratio density ratio of the power spectrum at
low frequencies over high frequencies.
winsizerange
is the range of the rolling window
sizes expressed as percentage of the timeseries length
(must be numeric between 0 and 100). Default is 25% -
75%.
incrwinsize
increments the rolling window size
(must be numeric between 0 and 100). Default is 25.
detrending
the timeseries can be
detrended/filtered. There are three options:
gaussian filtering, loess fitting,
linear detrending and first-differencing.
Default is no detrending.
bandwidthrange
is the range of the bandwidth used
for the Gaussian kernel when gaussian filtering is
selected. It is expressed as percentage of the timeseries
length (must be numeric between 0 and 100). Default is
5% - 100%.
spanrange
parameter that controls the degree of
smoothing (numeric between 0 and 100). Default is 5% -
100%. see more on loessstats
degree
the degree of polynomial to be used for
when loess fitting is applied, normally 1 or 2 (Default).
see more on loessstats
incrbandwidth
is the size to increment the
bandwidth used for the Gaussian kernel when gaussian
filtering is applied. It is expressed as percentage of
the timeseries length (must be numeric between 0 and
100). Default is 20.
incrspanrange
Span range
logtransform
logical. If TRUE data are
logtransformed prior to analysis as log(X+1). Default is
FALSE.
interpolate
logical. If TRUE linear interpolation
is applied to produce a timeseries of equal length as the
original. Default is FALSE (assumes there are no gaps in
the timeseries).
Details
see ref below
Arguments:
Value
sensitivity_ews returns a matrix that contains the
Kendall tau rank correlation estimates for the rolling
window sizes (rows) and bandwidths (columns), if
gaussian filtering is selected.
In addition, sensitivity_ews returns a plot with the
Kendall tau estimates and their p-values for the range of
rolling window sizes used, together with a histogram of the
distributions of the statistic and its significance. When
gaussian filtering is chosen, a contour plot is
produced for the Kendall tau estimates and their p-values
for the range of both rolling window sizes and bandwidth
used. A reverse triangle indicates the combination of the
two parameters for which the Kendall tau was the highest
Dakos, V., et al (2008). 'Slowing down as an early warning
signal for abrupt climate change.' Proceedings of the
National Academy of Sciences 105(38): 14308-14312
Dakos, V., et al (2012).'Methods for Detecting Early
Warnings of Critical Transitions in Time Series Illustrated
Using Simulated Ecological Data.' PLoS ONE 7(7):
e41010. doi:10.1371/journal.pone.0041010