Changes in global net radiative imbalance 1985–2012

Combining satellite data, atmospheric reanalyses, and climate model simulations, variability in the net downward radiative flux imbalance at the top of Earth's atmosphere (N) is reconstructed and linked to recent climate change. Over the 1985–1999 period mean N (0.34 ± 0.67 Wm−2) is lower than for the 2000–2012 period (0.62 ± 0.43 Wm−2, uncertainties at 90% confidence level) despite the slower rate of surface temperature rise since 2000. While the precise magnitude of N remains uncertain, the reconstruction captures interannual variability which is dominated by the eruption of Mount Pinatubo in 1991 and the El Niño Southern Oscillation. Monthly deseasonalized interannual variability in N generated by an ensemble of nine climate model simulations using prescribed sea surface temperature and radiative forcings and from the satellite-based reconstruction is significantly correlated (r∼0.6) over the 1985–2012 period.

(1) 72-day 60 • S-60 • N mean WFOV radiative fluxes are matched in time with daily data from ERAI ( Fig. S3a) and both records are deseasonalized with respect to their 1985-1999 base periods.
(2) Mean anomaly biases are computed as ERAI−WFOV; the bias is linearly interpolated in time across gaps in the WFOV record (Fig. S3b) and assumed constant after the final WFOV data point in 1999. All 72 days in each segment are assumed to have the same bias and the remaining days at the end of each year are assumed to have the same bias as the final 72-day segment of each year.
(3) The 60 • S-60 • N daily anomaly bias is added to the global ERAI data at each grid point over the period 1985-1999 and integrated to monthly data. Deseasonalized global monthly anomalies (relative to 2001-2005) are displayed in Fig. S3c. Since the 60 • S-60 • N anomalies may not be representative of global mean anomalies we examined both 60 • S-60 • N mean and global mean anomalies using the unadjusted ERAI dataset. Calculating the global mean anomaly as a fraction of the 60 • S-60 • N mean anomaly, the median scaling factor is 0.98 for OLR, 0.89 for ASR and 0.89 for N and the standard deviation of global anomalies are approximately 90% of the standard deviation of the 60 o S-60 o N anomalies in all cases. Therefore 60 • S-60 • N anomalies overestimate global mean anomalies only marginally for ASR and NET. Since we do not consider this a large effect, compared to other assumptions made, we do not scale the 60 • S-60 • N mean anomalies. Finally, the reconstructed data is subjected to a homogeneity adjustment as described in the main text. The reason for this is that inaccuracies may be present during the period influenced by the gap between WFOV and CERES measurements in 1999-2000 and potentially also during a gap in the WFOV record during 1993 [Trenberth, 2002]. Since there is no way to know the true changes during these periods, we use the following method.
We compute changes in OLR, ASR and N from the UPSCALE ensemble mean simulation over the following two periods: 1994-1995 minus 1992-1993 and 2000-2001 minus 1998-1999. The reconstructed fluxes are then adjusted prior to January 2000 and January 1994 so that changes in global mean radiative fluxes agree with UPSCALE simulations. While the UPSCALE data are climate model simulations, they use realistic radiative forcings and sea surface temperature/sea ice fields as boundary conditions and are unaffected by the changing observing systems used within the data assimilation of reanalyses such as ERAI. The UPSCALE simulations are also high spatial resolution and contain the most up-to-date parametrizations [Walters et al., 2011;Mizielinski et al., 2014] and so we consider that simulated flux changes are likely to be realistic July 15, 2014, 2:15pm GEOPHYSICAL RESEARCH LETTERS, 2014GL060962R 2014, DOI:10.1029/, over these relatively short periods of the record although will be affected by inaccuracies in boundary conditions (sea surface temperature/sea ice and radiative forcings).
To characterize the uncertainty in N relating to internal variability of the climate system and the boundary conditions used, we also compute changes in radiative fluxes simulated by the 9 CMIP5 climate model simulations. Details of the uncertainty estimate and deseasonalized interannual anomalies of radiative flux for OLR, ASR and N are discussed in the main text and for the amip simulation from each CMIP5 climate model in Table 1. While a distinct signal of decreased N in the east Pacific and increased N in most other tropical regions is evident in Fig. 3b, the picture is more complex amongst individual models. This is partly explained by the internal atmospheric variability but also by differences in physical processes and feedbacks represented by each model simulation. In particular there are substantial differences over the eastern sub-tropical stratocumulus regions: while some models show increases in N over these regions, particularly away from coastal areas (e.g. CNRM, GISS, MIROC5) other models display decreases in N (e.g. CanESM2, HadGEM2, IPSL, MRI). It is also notable that the majority of models simulate increases in N over Europe (e.g. CanESM2, HadGEM2, IPSL, MIROC5, NorESM1) as discussed in the main text. which also dominates N . The link between MEI and ASR is less clear (Fig. S7b) and there is also a residual difference between AMIP5 and CMIP5 simulations during the Pinatubo volcanic eruption which may reflect unrealistic ocean heat uptake following volcanic eruptions in the coupled simulations.

Idealized experiments with the simple energy balance climate model
To further understand the links between changes in N , T s and ocean heating a simple global mean energy balance climate model with a two layer diffusive ocean is used [Allan et al., 2014;Watanabe et al., 2013]. This assumes that N is determined by changes in effective radiative forcing ∆F and a climate response (Y ∆T s ) dependent upon the feedback parameter, Y (W m −2 K −1 ); the temperature anomaly ∆T s in a mixed layer ocean is computed from N minus the diffused with heat capacity of the mixed layer (C m = 4.2 × 10 8 JK −1 m −2 ) and deep layer (C D = 3.8 × 10 9 JK −1 m −2 ), Y varies from 1 to 2 W m −2 K −1 and ∆T D is the deep ocean temperature anomaly.
∆F is prescribed from the IPCC Climate System Scenario Tables [Prather et al., 2013]  Making the assumption that observed ∆T s approximates the mixed layer temperature changes, we next infer changes in heat flux to the deeper layer using N from OBS (D = N −C m (d∆T s /dt)).
The aim is not to calculate precise values but to provide an indication of the tendency in deep ocean heat flux over the period inferred from our reconstruction. greater heat flux more recently. Although this could reflect errors in the reconstruction of N used in this calculation, it is broadly consistent with conclusions based upon ocean reanalyses [Balmaseda et al., 2013;England et al., 2014] and highlights the potential role of ocean circulation variability in contributing to the surface warming trends on decadal time-scales. However, a more in-depth treatment, within a climate modelling framework is required to understand the roles of radiative forcing and ocean heat uptake in determining current climate variability.
In conclusion, based upon the simple energy balance model, it does not appear possible to reconcile trends in N and ∆T s since 2000 without assuming changes in heat flux below the mixed layer that are independent of the radiative forcings for the model parameters used. Therefore the relatively stable ∆T s since 2000 appears more likely to be explained by internal variability of the climate system than by changes in radiative forcings, although a combination of factors remains plausible. Applying experiments with atmosphere-only and fully coupled comprehensive climate model simulations is required to test this hypothesis.