IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group I: The Physical Science Basis

Appendix 5.A: Techniques, Error Estimation and Measurement Systems

5.A.1 Ocean Temperature and Salinity

Sections 5.2 and 5.3 report on the changes in the oceans using two different approaches to the oceanic part of the climate system. Section 5.2 documents the changes found in the most comprehensive ocean data sets that exist for temperature and salinity. These data sets are collected from a wide range of organisations and are a composite of heterogeneous measurement systems, including mechanical and expendable bathythermographs, research ship measurements, voluntary observing ships, moored and drifting buoys and Argo floats for recent years. The advantage of these composite data sets is the greater spatial and temporal coverage that they offer for climate studies. The main disadvantage of these composite data sets, relative to the research data sets used in Section 5.3, is that they can have more problems related to the quality and heterogeneity of the measurements systems. This heterogeneity can lead to subtle biases and artificial noise and consequently difficulties in estimating trends at small regional scales (Harrison and Carson, 2006). On the other hand, Section 5.3 described the changes found in detailed analyses of very specific research voyages that consist mainly of very tightly calibrated and monitored temperature and salinity measurements (and other variables). The internal consistency of these research data sets is much higher than the composite data sets, and as a consequence they have significant advantages in their ease of interpretation and analysis. However, research quality oceanographic data sets are only collected occasionally and are focussed more frequently on regional rather than global issues. This means that in the poorly sampled oceans, such as the Indian, South Pacific and Southern Oceans, observational records only cover a relatively short period of time (e.g., the 1960s to present) with some decades poorly covered and highly heterogeneous in space (see Figure 5.A.1).

Figure 5.A.1

Figure 5.A.1 The number of ocean temperature observations in each 1° grid box at 250 m depth for two periods: (a) 1955 to 1959, with a low density of observations, and (b) 1994 to 1998, with a high density of observations. A blue dot indicates a 1° grid box containing 1 observation, a green dot 2 to 5 observations, an orange dot 6 to 20 observations, and a red dot more than 20 observations.

An example of the distribution of ocean temperature observations in both space and time is shown in Figure 5.A.1. This figure shows the in situ temperature data distribution for two five-year periods used to create estimates of global heat content change (e.g., Figure 5.1), one with a low (a) and one with a high (b) density of observations. It is clear that parts of the ocean, in particular in the SH, are not well sampled even in periods of high observation density. Hence, sampling errors resulting from the lack of data are potentially important but cannot easily be quantified.

Several different objective analysis techniques have been used to produce the gridded fields of temperature anomalies used to compute ocean heat content and steric sea level rise presented in this chapter. The technique used by Levitus et al. (2005b), Garcia et al. (2005) and Antonov et al. (2005) in their estimates of temperature (heat content), oxygen and the thermosteric component of sea level change is based on the construction of gridded (1° latitude by 1° longitude grid) fields at standard depth measurement levels. The objective analysis procedure used for interpolation (filling in data-void areas and smoothing the entire field) is described by Boyer et al. (2002). At each standard depth level, all data are averaged within each 1° square, and the deviation from climatology yields the observed anomaly. From all observations within the surrounding region of diameter 888 km, the analysed value is computed. Features with a wavelength of less than 500 to 600 km are substantially reduced in amplitude; in regions without sufficient data, it is essentially the climatological information that is used. Ishii et al. (2006) employed similar techniques, with a smaller de-correlation length scale of 300 km and a least-squares technique for estimating corrections to the climatological field. Willis et al. (2004) used a two-scale covariance function, but also used altimetric data in areas where ocean observations were lacking.

There are some differences in the data used in these studies. In addition to ocean temperature profile data, Ishii et al. (2006) also used the product of climatological mixed layer depth and individual SST measurements in their estimates of ocean heat content. Southern Hemisphere World Ocean Circulation Experiment profiling float temperature profiles for the 1990s were used by Willis et al. (2004) that were not used by Levitus et al. (2005a) and Ishii et al. (2006). The similarity of the three independently estimated heat content time series shown in Figure 5.1 to within confidence intervals indicates that the differences between analysis techniques and data sources do not substantially influence the estimates of the three global ocean heat content time series.

All analyses are subject to statistical errors and sampling errors. Statistical errors are estimated in a straightforward way. For example, for the Levitus et al. (2005a) fields, the uncertainty at any grid point is estimated from the variability of observations that contributed to the analysed value. In this way, 90% errors for all analysed variables are computed as a function of depth and horizontal position, and correspondingly for integrated variables such as heat content. Both Ishii et al. (2006) and Willis et al. (2004) used the interannual variability of heat content as the basis for error analyses.