The latent and sensible heat fluxes have a strong regional dependence, with typical values varying in the annual mean from close to zero to –220 W m–2 and –70 W m–2 respectively over strong heat loss sites (Yu and Weller, 2007). Estimates of these terms have many potential sources of error (e.g., sampling issues, instrument biases, changing data sources, uncertainty in the flux computation algorithms). These sources may be spatially and temporally dependent, and are difficult to quantify (Gulev et al., 2007); consequently flux error estimates have a high degree of uncertainty. Spurious temporal trends may arise as a result of variations in measurement method for the driving meteorological state variables, in particular wind speed (Tokinaga and Xie, 2011). The overall uncertainty of the annually averaged global ocean mean for each term is expected to be in the range 10 to 20%. In the case of the latent heat flux term, this corresponds to an uncertainty of up to 20 W m–2. In comparison, changes in global mean values of individual heat flux components expected as a result of anthropogenic climate change since 1900 are at the level of <2 W m–2 (Pierce et al., 2006).
Many new turbulent heat flux data sets have become available since AR4 including products based on atmospheric reanalyses, satellite and in situ observations, and hybrid or synthesized data sets that combine information from these three different sources. It is not possible to identify a single best product as each has its own strengths and weaknesses (Gulev et al., 2010); several data sets are summarised here to illustrate the key issues. The Hamburg Ocean-Atmosphere Parameters and Fluxes from Satellite (HOAPS) data product provides global turbulent heat fluxes (and precipitation) developed from observations at microwave and infrared wavelengths (Andersson et al., 2011). In common with other satellite data sets it provides globally complete fields, however, it spans a relatively short period (1987 onwards) and is thus of limited utility for identifying long-term changes. A significant advance in flux data set development methodology is the 1 × 1 degree grid Objectively Analysed Air–Sea heat flux (OAFlux) data set that covers 1958 onwards and for the first time synthesizes state variables (SST, air temperature and humidity, wind speed) from reanalyses and satellite observations, prior to flux calculation (Yu and Weller, 2007). OAFlux has the potential to minimize severe spatial sampling errors that limit the usefulness of data sets based on ship observations alone and provides a new resource for temporal variability studies. However, the data sources for OAFlux changed in the 1980s, with the advent of satellite data, and the consequences of this change need to be assessed. In an alternative approach, Large and Yeager (2009) modified NCEP1 reanalysis state variables prior to flux calculation using various adjustment techniques, to produce the hybrid Coordinated Ocean-ice Reference Experiments (CORE) turbulent fluxes for 1948–2007 (Griffies et al., 2009). However, as the adjustments employed to produce the CORE fluxes were based on limited periods (e.g., 2000–2004 for wind speed) it is not clear to what extent CORE can be reliably used for studies of interdecadal variability over the 60-year period that it spans.
Analysis of OAFlux suggests that global mean evaporation may vary at inter-decadal time scales, with the variability being relatively small compared to the mean (Yu, 2007; Li et al., 2011; Figure 3.6a). Changing data sources, particularly as satellite observations became available in the 1980s, may contribute to this variability (Schanze et al., 2010) and it is not yet possible to identify how much of the variability is due to changes in the observing system. The latent heat flux variations (Figure 3.6b) closely follow those in evaporation (with allowance for the sign definition which results in negative values of latent heat flux corresponding to positive values of evaporation) but do not scale exactly as there is an additional minor dependence on SST through the latent heat of evaporation. The large uncertainty ranges that are evident in each of the time series highlight the difficulty in establishing whether there is a trend in global ocean mean evaporation or latent heat flux. The uncertainty range for latent heat flux is much larger than the 0.5 W m–2 level of net heat flux change expected from the ocean heat content increase (Box 3.1). Thus, it is not yet possible to use such data sets to establish global ocean multi-decadal trends in evaporation or latent heat flux at this level. The globally averaged sensible heat flux is smaller in magnitude than the latent heat flux and has a smaller absolute range of uncertainty (Figure 3.6b).