krige.bayes and krige.conv.output.control(n.posterior, n.predictive, moments, n.back.moments,
simulations.predictive, mean.var, quantile,
threshold, sim.means, sim.vars, signal, messages)n.posterior.lambda = 1
there is no transformation/back-transformation.
If lambda = 0 or lambda = 0.5 the moments are
back-tralocations of the main functions.
Defaults to FALSE but changed toprobabilities is included in the output.
This object contains, for each
prediction location, the probability that the variable is less
than TRUE, if simulations from the predictive are required.FALSE.output.control. See DETAILS
below.TRUE. This function is typically called by the krige.bayes and krige.conv
defining the output.
By default, krige.bayes sets signal = TRUE
and krige.conv sets signal = FALSE.
The underlying model $$Y(x) = \mu + S(x) + \epsilon$$ assumes that observations $Y(x)$ are noisy versions of a signal $S(x)$ and $Var(\epsilon)=\tau^2$ is the nugget variance.
If $\tau^2 = 0$ the $Y$ and $S$ are
indistiguishable.
If $\tau^2 > 0$ and regarded as measurement error, the
option signal defines whether the $S$ (signal =
TRUE) or the variable $Y$ (signal = FALSE) is to be
predicted.
For the latter the predictions will "honor" the data,
i.e. predicted values will coincide with the data, at data locations.
For unsampled locations and untransformed data,
the predicted values equals data
regardless signal = TRUE or FALSE, however
predictions variances will differ.
The function krige.conv has an argument
micro.scale. If $micro.scale > 0$ the error term is
divided as $\epsilon = \epsilon_{ms} + \epsilon_{me}$ and the nugget variance is divided into two terms: micro-scale variance
and measurement error.
If signal = TRUE the term $\epsilon_{ms}$ is
regarded as part of the signal and consequently the micro-scale variance is added to
the prediction variance.
If signal = FALSE the total error variance $\tau^2$
is added to the prediction variance.
krige.bayes and krige.conv.