In dankelley/argoFloats: Analysis of Oceanographic Argo Floats

Abstract. Adjusted and unadjusted data are examined for cycle 153 of float 4904013 (downloaded from https://data-argo.ifremer.fr/dac/aoml/5904013/profiles/D5904013_153.nc). This cycle has 3 columns. The DATA_MODE is "DAD", meaning that columns 1 and 3 are in delayed mode, and column 2 is in adjusted mode. Some data are displayed in detail, but mostly the analysis centres on 4 diagrams. Figures 1 and 2 show the unadjusted and adjusted data^[Note that this meaning of 'adjusted' is not the same as for the DATA_MODE. Rather, we call the values in the netCDF variable PSAL as 'unadjusted', and those in PSAL_ADJUSTED as 'adjusted'.], with red dots for questionable data (flags not equal to 1=passed, 2=probably_good, or 5=changed). Figures 3 and 4 show unadjusted and adjusted data, omitting questionable data. On a quick reading of this document, start by comparing Figures 3 and 4, which shows that adjusted and unadjusted data are very similar, with the main apparent difference being in the near-surface salinity in the second profile.

Load the data and define support functions

library(oce)
url <- "https://data-argo.ifremer.fr/dac/aoml/5904013/profiles/D5904013_153.nc"
file <- gsub(".*/", "", url)
if (!file.exists(file))
    download.file(url, file)
d <- read.oce(file)

`DATA_MODE` and Flag overview

The data mode is

d[["dataMode"]]
ncol <- ncol(d[["pressure"]])

indicating r ncol columns, of type 'Delayed', 'Adjusted' and 'Delayed', respectively.

An overview of the data for these r ncol columns is

str(d@data, 1)

The data-quality flags are

str(d@metadata$flags, 1)

Define support functions

In oce, flags are described as follows (based on descriptions from an official Argo document): not_assessed=0, passed_all_tests=1, probably_good=2, probably_bad=3, bad=4, changed=5, not_used_6=6, not_used_7=7, estimated=8, and missing=9. The following functions consider all but categories 1, 2 and 5 as suspicious.

# accept if flagged with 1=passed, 2=probably good, 5=changed
flagIsAcceptable <- function(f1, f2=NULL, f3=NULL)
{
    if (is.null(f2))
        f2 <- rep(2, length(f1))
    if (is.null(f3))
        f3 <- rep(2, length(f1))
    (f1 %in% c(1, 2, 5)) & (f2 %in% c(1, 2, 5)) & (f3 %in% c(1, 2, 5))
}

# black if acceptable, red otherwise
flagCol <- function(f1, f2=NULL, f3=NULL)
{
    if (is.null(f2))
        f2 <- rep(2, length(f1))
    if (is.null(f3))
        f3 <- rep(2, length(f1))
    ifelse(flagIsAcceptable(f1, f2, f3), "black", "red")
}

# Show top and bottom of an item
headtail <- function(a, n=6)
    rbind(head(a,n), tail(a,n))

# Construct flag strings from oce flag vectors
flagToChar <- function(f)
{
    f[is.na(f)] <- " "
    paste(f, collapse="")
}

# Attempt to plot something, showing a box if not
P <- function(expr)
{
    t <- try(expr, silent=TRUE)
    if (inherits(t, "try-error")) {
        plot(0:1, 0:1, xlab="", ylab="", type="n")
        box()
        text(0.1, 0.5, "No valid data")
    }
}

An overview of the data

General overview

summary(d)

The QC tests are as follows

d[["historyAction"]]
d[["historyQCTest"]]
argoFloats::showQCTests(d, style="full")

As a matter of interest, are the flags for adjusted and unadjusted data different?

all(d[["salinityFlag"]] == d[["salinityAdjustedFlag"]])
all(d[["temperatureFlag"]] == d[["temperatureAdjustedFlag"]])
all(d[["pressureFlag"]] == d[["pressureAdjustedFlag"]])

Yes, they are the same for this ID/cycle (but not for all; see 13_useBest_02_b.Rmd).

Before moving on, we ought to double-check that oce is reading the netCDF correctly:

library(ncdf4)
n <- nc_open(file)
for (i in 1:ncol) {
    cat("checking column", i, "of", ncol, "\n")
    stopifnot(all.equal(ncvar_get(n, "PSAL_QC")[i], flagToChar(d[["salinityFlag"]][,i])))
    stopifnot(all.equal(ncvar_get(n, "PSAL_ADJUSTED_QC")[i], flagToChar(d[["salinityAdjustedFlag"]][,i])))
    stopifnot(all.equal(ncvar_get(n, "TEMP_QC")[i], flagToChar(d[["temperatureFlag"]][,i])))
    stopifnot(all.equal(ncvar_get(n, "TEMP_ADJUSTED_QC")[i], flagToChar(d[["temperatureAdjustedFlag"]][,i])))
    stopifnot(all.equal(ncvar_get(n, "PRES_QC")[i], flagToChar(d[["pressureFlag"]][,i])))
    stopifnot(all.equal(ncvar_get(n, "PRES_ADJUSTED_QC")[i], flagToChar(d[["pressureAdjustedFlag"]][,i])))
}

The lack of errors in the above indicates that the oce flag values match those in the netCDF file, in this case.

Tabular overview

p <- d[["pressure"]]
pf <- d[["pressureFlag"]]
pa <- d[["pressureAdjusted"]]
paf <- d[["pressureAdjustedFlag"]]

S <- d[["salinity"]]
Sf <- d[["salinityFlag"]]
Sa <- d[["salinityAdjusted"]]
Saf <- d[["salinityAdjustedFlag"]]

T <- d[["temperature"]]
Tf <- d[["temperatureFlag"]]
Ta <- d[["temperatureAdjusted"]]
Taf <- d[["temperatureAdjustedFlag"]]

if (ncol != ncol(S))
    stop("p and S have differing numbers of columns")
if (ncol != ncol(T))
    stop("p and T have differing numbers of columns")
pDF <- data.frame(p1=p[,1], p2=p[,2], p3=p[,3],
    pf1=pf[,1], pf2=pf[,2], pf3=pf[,3],
    pa1=pa[,1], pa2=pa[,2], pa3=pa[,3],
    paf1=paf[,1], paf2=paf[,2], paf3=paf[,3])
SDF <- data.frame(S1=S[,1], S2=S[,2], S3=S[,3],
    Sf1=Sf[,1], Sf2=Sf[,2], Sf3=Sf[,3],
    Sa1=Sa[,1], Sa2=Sa[,2], Sa3=Sa[,3],
    Saf1=Saf[,1], Saf2=Saf[,2], Saf3=Saf[,3])
TDF <- data.frame(T1=T[,1], T2=T[,2], T3=T[,3],
    Tf1=Tf[,1], Tf2=Tf[,2], Tf3=Tf[,3],
    Ta1=Ta[,1], Ta2=Ta[,2], Ta3=Ta[,3],
    Taf1=Taf[,1], Taf2=Taf[,2], Taf3=Taf[,3])

options(digits=5)
headtail(pDF)
headtail(SDF)
headtail(TDF)

Plots

Unadjusted data, red if flagged as unacceptable

par(mfrow=c(4, ncol), mar=c(3,3,1,1), mgp=c(2,0.7,0))
ctds <- vector("list", ncol)
for (i in seq_len(ncol))
    ctds[[i]] <- as.ctd(S[,i], T[,i], p[,i])
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="S", col=flagCol(Sf[,i], pf[,i]), type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="T", col=flagCol(Tf[,i], pf[,i]), type="p", pch=20))
for (i in seq_len(ncol))
    P(plotTS(ctds[[i]], col=flagCol(Sf[,i], Tf[,i], pf[,i])))

Adjusted data, red if flagged as unacceptable

par(mfrow=c(4, ncol), mar=c(3,3,1,1), mgp=c(2,0.7,0))
for (i in seq_len(ncol))
    ctds[[i]] <- as.ctd(Sa[,i], Ta[,i], pa[,i])
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="S", col=flagCol(Saf[,i], paf[,i]), type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="T", col=flagCol(Taf[,i], paf[,i]), type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="sigmaTheta", col=flagCol(Saf[,i], Taf[,i], paf[,i]), type="p", pch=20))
for (i in seq_len(ncol))
    P(plotTS(ctds[[i]], col=flagCol(Saf[,i], Taf[,i], paf[,i])))

Unadjusted data, skipping unacceptable (to narrow scales)

par(mfrow=c(4, ncol), mar=c(3,3,1,1), mgp=c(2,0.7,0))
for (i in seq_len(ncol)) {
    accept <- flagIsAcceptable(Sf[,i], Tf[,i], pf[,i])
    ctds[[i]] <- as.ctd(S[,i][accept], T[,i][accept], p[,i][accept])
}
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="S", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="T", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="sigmaTheta", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotTS(ctds[[i]]))

Adjusted data, skipping if flagged as unacceptable (to improve scales)

par(mfrow=c(4, ncol), mar=c(3,3,1,1), mgp=c(2,0.7,0))
for (i in seq_len(ncol)) {
    accept <- flagIsAcceptable(Saf[,i], Taf[,i], paf[,i])
    ctds[[i]] <- as.ctd(Sa[,i][accept], Ta[,i][accept], pa[,i][accept])
}
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="S", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="T", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotProfile(ctds[[i]], xtype="sigmaTheta", type="p", pch=20))
for (i in seq_len(ncol))
    P(plotTS(ctds[[i]]))