Often in CTD profiling, the goal is to isolate only the downcast, discarding measurements made in
the air, in an equilibration phase in which the device is held below the water surface, and then the
upcast phase that follows the downcast. This is handled reasonably well by ctdTrim
with
method="downcast"
, although it is almost always best to use plotScan
to
investigate the data, and then use the method="index"
or method="scan"
method based on
visual inspection of the data.
ctdTrim(x, method, removeDepthInversions = FALSE, parameters = NULL,
indices = FALSE, debug = getOption("oceDebug"))
A ctd
object, i.e. one inheriting from ctd-class
.
A string (or a vector of two strings) specifying the trimming method, or a function to
be used to determine data indices to keep. If method
is not provided, "downcast"
is
assumed. See “Details”.
Logical value indicating whether to remove any levels at which depth is
less than, or equal to, a depth above. (This is needed if the object is to be assembled into a
section, unless ctdDecimate
will be used, which will remove the inversions.
A list whose elements depend on the method; see “Details”.
Logical value indicating what to return. If indices=FALSE
(the default),
then the return value is a subsetted ctd-class
object. If indices=TRUE
,
then the return value is a logical vector that could be used to subset the data
with subset,ctd-method
or to set data-quality flags.
an integer specifying whether debugging information is
to be printed during the processing. This is a general parameter that
is used by many oce
functions. Generally, setting debug=0
turns off the printing, while higher values suggest that more information
be printed. If one function calls another, it usually reduces the value of
debug
first, so that a user can often obtain deeper debugging
by specifying higher debug
values.
Either an object of ctd-class
or a logical vector of length
matching the data. The first option is the default. The second option,
achieved by setting indices=FALSE
, may be useful in constructing
data flags to be inserted into the object.
ctdTrim
begins by examining the pressure differences between subsequent samples. If
these are all of the same value, then the input ctd
object is returned, unaltered.
This handles the case of pressure-binned data. However, if the pressure difference
varies, a variety of approaches are taken to trimming the dataset.
If method[1]
is "downcast"
then an attempt is made to keep only data for
which the CTD is descending. This is done in stages, with variants based on method[2]
, if
supplied. Step 1. The pressure data are despiked with a smooth() filter with method "3R".
This removes wild spikes that arise from poor instrument connections, etc. Step 2. If no
parameters
are given, then any data with negative pressures are deleted. If there is a
parameter named pmin
, then that pressure (in decibars) is used instead as the lower limit.
This is a commonly-used setup, e.g. ctdTrim(ctd, parameters=list(pmin=1))
removes the top
decibar (roughly 1m) from the data. Specifying pmin
is a simple way to remove near-surface
data, such as a shallow equilibration phase, and if specified will cause ctdTrim
to skip
step 4 below. Step 3. The maximum pressure is determined, and data acquired subsequent to
that point are deleted. This removes the upcast and any subsequent data. Step 4. If the
pmin
parameter is not specified, an attempt is made to remove an initial equilibrium phase
by a regression of pressure on scan number. There are three variants to this, depending on the
value of the second method
element. If it is "A"
(or not given), the procedure is to
call nls
to fit a piecewise linear model of pressure as a function of scan,
in which pressure is
constant for scan less than a critical value, and then linearly varying for with scan. This is
meant to handle the common situation in which the CTD is held at roughly constant depth (typically
a metre or so) to equilibrate, before it is lowered through the water column.
Case "B"
is the same,
except that the pressure in the surface region is taken to be zero (this does not make
much sense, but it might help in some cases). Note that, prior to early 2016, method "B"
was
called method "C"
; the old "B"
method was judged useless and was removed.
If method="upcast"
, a sort of reverse of "downcast"
is used. This
was added in late April 2017 and has not been well tested yet.
If method="sbe"
, a method similar to that described
in the SBE Data Processing manual is used to remove the "soak"
period at the beginning of a cast (see Section 6 under subsection
"Loop Edit"). The method is based on the soak procedure whereby
the instrument sits at a fixed depth for a period of time, after
which it is raised toward the surface before beginning the actual
downcast. This enables equilibration of the sensors while still
permitting reasonably good near-surface data. Parameters for the
method can be passed using the parameters
argument, which
include minSoak
(the minimum depth for the soak) and
maxSoak
the maximum depth of the soak. The method finds
the minimum pressure prior to the maxSoak
value being
passed, each of which occurring after the scan in which the
minSoak
value was reached. For the method to work, the
pre-cast pressure minimum must be less than the minSoak
value. The default values of minSoak
and maxSoak
are 1 and 20 dbar, respectively.
If method="index"
or "scan"
, then each column of data is subsetted according to the
value of parameters
. If the latter is a logical vector of length matching data column
length, then it is used directly for subsetting. If parameters
is a numerical vector with
two elements, then the index or scan values that lie between parameters[1]
and parameters[2]
(inclusive) are used for subsetting. The
two-element method is probably the most useful, with the values being determined by visual
inspection of the results of plotScan
. While this may take a minute or two, the
analyst should bear in mind that a deep-water CTD profile might take 6 hours, corresponding to
ship-time costs exceeding a week of salary.
If method="range"
then data are selected based on the value of the column named
parameters$item
. This may be by range or by critical value. By range: select values
between parameters$from
(the lower limit) and parameters$to
(the upper limit) By
critical value: select if the named column exceeds the value. For example, ctd2 <-
ctdTrim(ctd, "range", parameters=list(item="scan", from=5))
starts at scan number 5 and
continues to the end, while
ctdTrim(ctd,"range",parameters=list(item="scan",from=5,to=100))
also starts at scan 5,
but extends only to scan 100.
If method
is a function, then it must return a vector of logical
values, computed based on two arguments: data
(a
list
), and parameters
as supplied to ctdTrim
. Both
inferWaterDepth
and removeInversions
are ignored in the function case. See
“Examples”.
The Seabird CTD instrument is described at
http://www.seabird.com/products/spec_sheets/19plusdata.htm
.
Seasoft V2: SBE Data Processing, SeaBird Scientific, 05/26/2016
Other things related to ctd
data: [[,ctd-method
,
[[<-,ctd-method
, as.ctd
,
cnvName2oceName
, ctd-class
,
ctdDecimate
, ctdFindProfiles
,
ctdRaw
, ctd
,
handleFlags,ctd-method
,
initialize,ctd-method
,
initializeFlagScheme,ctd-method
,
oceNames2whpNames
,
oceUnits2whpUnits
,
plot,ctd-method
, plotProfile
,
plotScan
, plotTS
,
read.ctd.itp
, read.ctd.odf
,
read.ctd.sbe
,
read.ctd.woce.other
,
read.ctd.woce
, read.ctd
,
setFlags,ctd-method
,
subset,ctd-method
,
summary,ctd-method
,
woceNames2oceNames
,
woceUnit2oceUnit
, write.ctd
# NOT RUN {
library(oce)
data(ctdRaw)
plot(ctdRaw) # barely recognizable, due to pre- and post-cast junk
plot(ctdTrim(ctdRaw)) # looks like a real profile ...
plot(ctdDecimate(ctdTrim(ctdRaw),method="boxcar")) # ... smoothed
# Demonstrate use of a function. The scan limits were chosen
# by using locator(2) on a graph made by plotScan(ctdRaw).
trimByIndex<-function(data, parameters) {
parameters[1] < data$scan & data$scan < parameters[2]
}
trimmed <- ctdTrim(ctdRaw, trimByIndex, parameters=c(130, 380))
plot(trimmed)
# }
# NOT RUN {
# }
Run the code above in your browser using DataLab