Sampling bias
Historic data must be treated with caution as the number of observational data within the gridded data set is limited, both spatially and temporally at certain times and in certain regions. The open ocean in the Southern Hemisphere is histor- ically less observed than areas closer to land and sampling was seasonally biased,
being more confined to summer time at high latitudes due to weather restrictions. When assessing decadal trends in steric height, the irregular sampling of the sub- surface ocean impacts estimates of the trends, as shown by Lyman and Johnson
(2008), who state that the sampling locations between 1955 and 1966 were in- sufficient to reproduce the global trend during this time. Prior to the mid 1960s temperature profiles to around 300 m were taken using mechanical bathyther- mographs (MBTs). After the introduction of the expendable bathythermograph (XBT), MBTs were gradually phased out. XBTs sample temperature and pres- sure to depths of typically 460, 760 or 1800 m, with the two shallower depth ranges being most common, hence the depth range of certain gridded temperature and salinity products being 700m. Several studies have highlighted problems with the fall rate of the XBT (Gouretski and Koltermann, 2007, Hanawa et al., 1995,
Heinmiller et al.,1983) and corrections have since been applied (Ingleby and Hud-
dleston,2007,Wijffels et al., 2008). Temperature errors with XBTs have been 0.1
to 0.2o C. High precision data is available from research vessel based observations
from CTDs (conductivity, temperature, depth) where both temperature and salin- ity are available to an accuracy of 0.002o C and 0.005 psu respectively (Ingleby
and Huddleston,2007).
A large proportion of earlier sampling is confined to near coastal sites with large areas of the open ocean unobserved. ARGO floats have changed this completely to provide a more homogeneous coverage. ARGO floats are Lagrangian autonomous vertical profilers which compliment satellite altimetry by returning temperature, salinity and velocity data every 10 days from a depth of up to 2000 m, to the sur- face. From 1989 to the present day over 3500 floats have been placed in circulation in the global ocean by the Inernational ARGO Program (http://www.argo.ucsd. edu,http://argo.jcommops.org). Figure (4.2) highlights spatial differences be- tween historical and present day sampling by comparing observational reports from January in both 1958 and 2010. There are, however, still areas at very high latitudes which are less sampled by ARGO. The floats’ reliance on ocean currents means that those deployed at mid and low latitudes take time to travel to high lat- itudes. Deploying them at high latitudes is costly, and the seasonal growth of sea ice is a hindrance. These issues, however, are being addressed by the ARGO com- munity (Roemmich et al., 2009). Recent deployment of some deep ocean ARGO floats will one day provide long time series of data to 4000 m and pilot experiments are under way for some floats to reach 6000 m (http://www.argo.ucsd.edu/).
(a) January 1958
(b) January 2010
Figure 4.2: Observational subsurface temperature profile sampling for (A)
January 1958 (Ingleby and Huddleston, 2007) (Green X - MBT data (3508
reports), purple T - ocean station data (855 reports)) and (B) January 2010, Climate Prediction Centre, NOAA. (red x - XBT probes (1654 reports), green
+ - moorings (3487 reports), blue O - ARGO floats - (14,530 reports))
Presently, data is available to 2000 m and therefore the abyssal ocean changes are still only monitored by ship board CTD casts.
Instrumental error
decadal time scales there will be an inevitable change in the instruments used to collect the data. Comparison of one instrument’s record to another could lead to false assumptions unless they have been accurately calibrated. Since 1999 there was a dramatic increase in sampling, both at the surface and subsurface, within all ocean basins due to the introduction of ARGO floats. Around the same time the data also shows an increase in the heat content of some areas such as the Atlantic gyres and the Southern Ocean. Due to the short time scale of data recorded from ARGO floats their calibration is still ongoing and we have been careful not to make inferences about coinciding events in the data. The heat content estimates
of Smith and Murphy (2007), discussed below, were tested with and without the
ARGO data to confirm that the steric signals over the same time period were consistent (Williams et al., 2014) as regional thermosteric values can be sensitive to pressure bias from ARGO instruments (Barker et al., 2011). XBTs were the dominant source of ocean temperature profile data before ARGO floats. These have also been discovered to have instrumental error and all of the data sets that we use have been corrected for this (Gouretski and Koltermann, 2007, Ishii and
Kimoto,2009, Levitus et al., 2009, Wijffels et al., 2008)
Representational error
Unresolved observational points in the gridded data set are termed the represen- tational error. The length scale of a grid square using a 1o data set is around 100
km. Many oceanographic features, such as areas of high velocity currents, are of much finer scale (of order metres), and so are not represented by such a large grid. Many ocean models are now of a much finer scale to incorporate these features, however they are computationally expensive and are often on a regional scale for this reason. Representational error is often much larger than instrumental error (between 0.002o and 0.2o for temperatureIngleby and Huddleston(2007)) and the
observed error from the gridded data set will be a combination of the two in the form shown in equation 4.3
observation error=instrumental error+representational error (4.3)
Deep ocean warming
Song and Colberg(2011) examined the significance of deep ocean warming to sea
level trends by using a modelled result for the steric contribution below 700m com- bined with Ishii and Kimoto (2009) data for the surface layer, GRACE estimates for the mass contribution and altimetry for the sea surface height. Their results
indicate that the deep ocean (below 700 m) could have contributed up to 1.1 mm yr−1 to the global mean between 1993-2008. They show that ocean circulation
and dynamics distribute heat into the deep ocean, further explaining the unique regional patterns of absolute sea level trends from altimetry. If this steric value for the deep ocean is valid it would mean that observing and understanding changes in the deep ocean are vital for our interpretation and prediction of sea level trends.
Purkey and Johnson(2010) also highlight the importance of the abyssal ocean for
sea level. They report that their findings indicate a 0.1 mm yr−1 rise in global sea
levels from the deep ocean (1000 - 4000 m) between the 1990s and the 2000s, and warming in the Southern Ocean contributed 1 mm yr−1 locally. Their study of
abyssal ocean basin warming shows that ocean basins in the Southern Hemisphere have generally warmed more than those of the Northern Hemisphere. Advection from the source in the upper ocean through ocean circulation is one mechanism for the transportation of heat to deeper layers. Transportation time from Antarctic Bottom Water (AABW) to reach the North Pacific Ocean through advection is up to 1000 years, but this time is decreased to less than 50 years when propagated by Rossby or Kelvin waves (Nakano and Suginohara, 2002).
Steric estimates in the IPCC AR5 The IPCC AR5 report states that the correction for the XBT fall rate has been applied to the steric component of sea level since the IPCC AR4 report, and this correction has had the effect of doubling the estimated value of thermosteric sea level rise i the upper 700 m of ocean. This upper ocean warming is thought to have contributed 0.6 ±0.2 mm yr−1 between
1970 and 2009. ARGO floats during more recent years can provide a more global coverage to a depth of 1000 m. Trends using ARGO between 2005-2010 range from 0.2 to 0.8 mm yr−1 (Leuliette and Willis, 2011,Stocker et al.,2013, Vaughan
et al., 2013). The short time-scales of these estimates make them more uncertain
and subject to natural variability. The IPCC AR5 estimates that the thermosteric component of sea level trends was 0.7±0.3 mm yr−1 for the years 1993-2010 using
XBT reconstructions updated fromDomingues et al.(2008), (Stocker et al.,2013).