This function allows to evaluate the reconstruction skills for a given proxy time series in split-calibration approach.
skills(
object,
target = NULL,
model = "ols",
calibration = "50%",
timespan = NULL
)'skills' returns an 'object' of class '"tc_skills"'.
An object of class '"tc_skills"' is a list containing at least the following components:
the call made to function 'skills'
the target used for reconstruction
the coefficient of correlation for the calibration timespan
the coefficient of correlation for the complete data set
regression coefficients for the calibration model
regression coefficients for the full model
significance for the calibration model
significance for the full model
reduction of error statistic
coefficient of efficiency statistic
Durbin-Watson statistic
The root mean squared error for the prediction interval
the complete calibration model (an object of class 'lmodel2')
the complete full model (an object of class 'lmodel2')
an object of class "tc_dcc", "tc_dlm", or "tc_seascorr"
a treeclim selection modifier specifying the climate target to be reconstructed, see below for details
one of "ols" or "rma"
which part of the data shall be used as calibration subset? Given as either a range of years, an integer corresponding to the first or last number of observations, or a percentage as character string corresponding to the part of the data set to be used as calibration subset.
timespan to be used to truncate the data
The result of a call to dcc, dlm, or
seascorr can be used as object for the
function. The required data is then taken from this object and no
further processing of the tree and climate variables has to be
done for by the user. This reflects the flow of analysis, where
first general climate/growth relations are explored, and then the
strongest ones are deployed for reconstruction purposes.
target is an aggregation modifier (one of
.mean, .sum, and
.range). The user should be aware of the fact that
in case the aggregation modifier evaluates to more than one
variable (e.g., summer means for both temperature and
precipiation), a warning message is issued, and only the first
variable is taken into consideration for evaluating the
reconstruction skills. If not specified, the selection from the
original call to dcc, seascorr, or
dlm is used.
The type of regression model (ordinary least squares or errors-in-variables via reduced major axis regression) can be selected.
The part of the data to be used as a calibration subset can be specified in three different ways: 1) as a range of years, these are then taken as calibration period; 2) as a single integer, if positive, this number of observations at the recent end of the data set is taken as calibration set, if negative, this number of oldest observations is taken; and 3) as a character string giving a percentage of values, e.g., "-40%" would select the 40% oldest observations, while "55%" would select the 55% most recent ones.
The relationship between climate and tree-ring data is evaluated for the calibration period and the complete data set. Frequently used verification statistics are computed: reduction of error (RE), coefficient of efficiency (CE), and the Durban-Watson statistic (DW) (Cook et al. 1994, Durbin and Watson, 1951).
Cook E, Briffa K, Jones P (1994) Spatial regression methods in dendroclimatology: A review and comparison of two techniques. International Journal of Climatology, 14, 379-402.
Durbin, J, Watson, GS (1951) Testing for serial correlation in least squares regression. Biometrika 38:159-78.
dc <- dcc(muc_fake, muc_clim, .mean(6:9, "temp") + .sum(6:9,
"prec"))
sk <- skills(dc)
sk
plot(sk)
Run the code above in your browser using DataLab