In dankelley/argoFloats: Analysis of Oceanographic Argo Floats

Abstract

Argo floats undergo a series of testing to ensure the data provided is a reliable source. Understanding the Quality Control (QC) completed on profiles can be problematic, however. The creators of argoFloats therefore developed a way to deal with flags and adjusted data. When dealing with flags, a three step process exists involving plot(which='QC'), showQCTests()and applyQC(). In addition to flag control, argoFloats also deals with adjusted data with the useAdjusted() function.

Introduction to Flags

Core, biogeochemical (BGC), and Deep Argo data all undergo testing to ensure the data found at the Data Assembly Centers are as accurate as possible. More specifically, testing is done in three levels: 1: Real time that complete checks on all measurements and assigned a quality flag within a 24-48 hour time frame, 2. Delayed mode, and 3. Regional scientific analyzes of all float data with other available data. The procedures for regional analyzes are still to be determined.

At first glance, it can be difficult for users to understand the Quality Control (QC) completed on profiles. To address this issue when dealing with Argo data, argoFloats created a way to deal with flags and adjusted data. Firstly, the creators of argoFloats designed a three step process to deal with flags. Step one is plot(which='QC'), which shows the quality and mean of parameters. Step two is showQCTests(), which demonstrates which QC tests were performed and failed. Step three is applyQC(), which allows the users to get rid of 'bad' data if they wish. In addition to flag control, argoFloats also deals with adjusted data in useAdjusted(), which allows users to use adjusted data if available.

Flags

As described by the Argo user’s manual [2], data that are given a flag value of 1, 2, 5, or 8 are considered 'good' data. If data are given a flag value of 9 (or missing value) they are not used, and flags 3, 5, 6, and 7 are considered 'bad' data.

argoFloats has taken this idea, and has incorporated a 3 step process when considering the quality control of Argo flags. The three step process encourages users to always analyze the data they are considering. The three step process is as follows:

colText <- "darkblue"
colCode <- "black"
library(graphics)
textInBox <- function(x, y, text, cex=1, pos=4, center=TRUE, family="Times New Roman", col="black", tweakx=0)
{
    w <- graphics::strwidth(text)
    h <- graphics::strheight(text)
    if (center)
        x <- x - w/2
    text(x+tweakx, y, text, cex=cex, pos=pos, family=family, col=col)
    rect(x+tweakx, y-h, x+tweakx+1.1*w, y+h, border=col)
    invisible(list(w=w, h=h))
}
omar <- par("mar")
par(mar=c(0,1,0,0))
plot(c(0, 1), c(0, 1), type="n", xlab="", ylab="", axes=FALSE)
x0 <- 0.25
y0 <- 0.9
dy <- 0.3
wh <- textInBox(x0, y0, "    What is the quality of the data?", col=colText)
h <- wh$h * 1.2
xarrow <- x0 + 0.01 # not sure why we need to move it
arrows(xarrow, y0-h, xarrow, y0-dy+h, length=0.1, col=colText)
y0 <- y0 - dy
textInBox(x0, y0, "   What tests were performed?", col=colText)
arrows(xarrow, y0-h, xarrow, y0-dy+h, length=0.1, col=colText)
y0 <- y0 - dy
textInBox(x0, y0, "   Set low-quality data to NA.", col=colText)
x0 <- 0.8
y0 <- 0.9
wh <- textInBox(x0, y0, "plot(which='QC') ", family="sans", col=colCode)
h <- wh$h * 1.2
xarrow <- x0 + 0.01 # not sure why we need to move it
arrows(xarrow, y0-h, xarrow, y0-dy+h, length=0.1)
y0 <- y0 - dy
textInBox(x0, y0, "showQCTests() ", family="sans", col=colCode)
arrows(xarrow, y0-h, xarrow, y0-dy+h, length=0.1)
y0 <- y0 - dy
textInBox(x0, y0, "applyQC()  ", family="sans", col=colCode)
par(mar=omar)

This vignette will walk the users through the three step process using the float with ID 1901584 near the Bahamas.

Step One: QC Plot

Our argoFloats package uses the flags of profiles to determine "good" and "bad" data. We then created a plot type to demonstrate the quality of data, the QC plot. The QC plot is a plot of parameter quality and parameter mean. This only works if x is an object that was created by getProfiles(). The user must also provide the parameter of interest. If the user is uncertain about what parameter exists within a certain float, argoFloats will prompt the user with the appropriate parameters.

An example of how to get the QC plot for temperature, or "TEMP" for the float with ID 1901584 is as follows:

library(argoFloats)
data("index")
subset <- subset(index, ID='1901584')
cycles <- getProfiles(subset)
argos <- readProfiles(cycles)
plot(argos, which='QC', parameter='temperature')

Exercise One: Use the previous code to determine the QC of temperature for float number 4900845 in synthetic data.Explain your results.

Step Two: showQCTests()

The showQCTests() is a newly added function. It's purpose is to use integer values from the hexToNibble() function to internally convert hex digits. In this context, the hex digits come from HISTORY_QCTEST of a single argoFloats object that was created by readProfiles(). The showQCTests() is then used to indicate which QC tests were performed and/or failed.

Using the same example as shown in plot(which='QC'), we can also determine which QC tests were performed on each cycle. This is completed by showQCTests() in the following code:

library(argoFloats)
data('index')
subset <- subset(index, 1)
cycles <- getProfiles(subset)
argos <- readProfiles(cycles)
argos1 <- argos[[1]]
showQCTests(argos[[1]])

In section 3.11 Reference table 11: QC test binary IDs of the Argo User's Manual [2], we see that test two is the "Impossible data test".

Exercise Two: Use the first float with ID 4900845 in exercise one,to determine which tests were performed and/or failed during quality control testing. EXPLAIN YOUR RESULTS.

Step Three: applyQC()

The applyQC() function examines quality-control within an argoFloats object that was created by readProfiles(). By default, it replaces all suspicious data with NA values, so they will not appear in plots or be considered in calculations. This is an important early step in processing, because suspicious Argo floats commonly report data that are suspicious based on statistical and physical measures.

Using the same example as shown in plot(which='QC') and showQCTests(), the user also has the ability to use applyQC() as shown below:

library(argoFloats)
# Contrast TS diagrams for raw and flag-handled data
data(index)
i <- subset(index, ID='1901584')
raw <- readProfiles(getProfiles(i))
clean <- applyQC(raw)
par(mfrow=c(1, 2))
plot(raw, which="TS")
plot(clean, which="TS")

Exercise Three : Use applyQC() on float ID with 4900845. Explain your results.

Adjusted Data

useAdjusted()

In addition to flags flagging bad data based on tests, argoFloats package has also incorporated the idea of <param>Adjusted vs <param>Unadjusted which considers "bad" data that has been adjusted [2].

Variables with original names indicating in the string <param>Adjusted are assigned nicknames names ending in Adjusted by readProfiles(), so that e.g. doxyAdjusted gets the nickname oxygenAdjusted, while doxy gets the nickname oxygen. useAdjusted() switches these, renaming the adjusted values, so that e.g. DOXY_ADJUSTED gets nickname oxygen and DOXY gets nickname oxygenUnadjusted. This is carried out for all data families, and also for the corresponding units and quality-control flags. See ?useAdjusted to understand when argoFloats chooses to use adjusted data.

An example of the significance of adjusted vs unadjusted can be shown below using the ID from the previous flag examples:

library(argoFloats)
data("index")
subset <- subset(index, ID='1901584')
cycles <- getProfiles(subset)
raw <- readProfiles(cycles)
adj <- useAdjusted(raw)
par(mfrow=c(1, 2))
oce::plotProfile(oce::as.ctd(raw[[1]]), xtype="temperature")
mtext("Raw data", side=3, line=-1, col=2)
oce::plotProfile(oce::as.ctd(adj[[1]]), xtype="temperature")
mtext("Adjusted data", side=3, line=-1, col=2)

Exercise Four Use useAdjusted() on the float with ID5903586, cycle=50 to compare <param>Adjusted vs <param>Unadjusted>.

Answers

Exercise One: Use the previous code to determine the QC of temperature for float number 4900845 in synthetic data. Explain your results.

library(argoFloats)
ais <- getIndex(filename = 'synthetic', age=0)
sub <- subset(ais, ID='4900845')
cycles <- getProfiles(sub)
argos <- readProfiles(cycles)
plot(argos, which='QC', parameter='temperature')

As shown in the plot, 100% of the temperature data is considered "good". This implies to the user that all data is considered okay to use.

Exercise Two: Use the first float with ID 4900845 in exercise one,to determine which tests were performed and/or failed during quality control testing. Explain your results.

library(argoFloats)
data('indexSynthetic')
sub <- subset(indexSynthetic, ID='4900845')
cycles <- getProfiles(sub)
argos <- readProfiles(cycles)
a1 <- argos[[1]]
showQCTests(a1)

As shown by the showQCTests() function, no quality control tests were actually performed on this float ID. This is a case where the user would have to self analyze their data to ensure all data seems reasonable.

Exercise Three : Use applyQC() on float ID = 4900845. Explain your results.

library(argoFloats)
# Contrast TS diagrams for raw and flag-handled data
data(indexSynthetic)
i <- subset(index, ID='4900845')
raw <- readProfiles(getProfiles(i))
clean <- applyQC(raw)
par(mfrow=c(1, 2))
plot(raw, which="TS")
plot(clean, which="TS")

As shown by the plot, the plot containing "good" data and the plot containing "bad" data are the same. This is the result of no tests being completed for QC, as shown by showQCTests().

Exercise Four Use useAdjusted() on the float with ID 5903586, cycle=50 to compare <PARAM>_ADJUSTED vs <param>Unadjusted.

library(argoFloats)
bai <- getIndex('synthetic')
s <- subset(bai, ID='5903586')
ss <- subset(s, 50)
cycles <- getProfiles(ss)
raw <- readProfiles(cycles)
adj <- useAdjusted(raw)
par(mfrow=c(1, 2))
oce::plotProfile(oce::as.ctd(raw[[1]]), xtype="oxygen")
mtext("Raw data", side=3, line=-1, col=2)
oce::plotProfile(oce::as.ctd(adj[[1]]), xtype="oxygen")
mtext("Adjusted data", side=3, line=-1, col=2)

References

1. Wong, A., Keeley, R. and Carval, T., 2009. Argo quality control manual. Version, 2(5), p.32.
2. Carval, Thierry, Bob Keeley, Yasushi Takatsuki, Takashi Yoshida, Stephen Loch, Claudia Schmid, and Roger Goldsmith. Argo User’s Manual V3.3. Ifremer, 2019. https://doi.org/10.13155/29825.

dankelley/argoFloats documentation built on April 18, 2024, 5:13 a.m.