Figure 1: Time
series of cumulative anthropogenic CO2 emissions since 1850 (a);
emissions of CH4 (b); global mean emissions of SO2 (c) and black carbon
(BC; d) derived from NorESM2 ScenarioMIP simulations available within
ClimateBench, including the SSP245 test scenario (shown in black).
The aerosol (precursor) emissions are derived from the latest version of
the spatially resolved CEDS dataset and again summed over sectors and
months to produce maps of annual total emissions, as shown in Figure 2
for SO2 in different years. While the spatial distribution clearly
evolves over the historical period and into the future scenarios, the
emissions are fairly localised around industrialised regions and
dimensionality reduction can be used to reduce the size of these input
features (as discussed for the baseline emulators in Section 4). An area
preserving interpolation is performed so that the emission data are
provided on the same spatial grid as the NorESM2 output fields to
simplify its use in ML workflows. Again, as used for NorESM2 the
5-yearly data is interpolated to a yearly frequency for consistency.