Abstract Large bias exists in the shortwave cloud radiative effect (SWCRE) of general circulation models (GCMs), with unclear contributions from individual cloud properties. Here we present a machine‐learning approach, demonstrated using the FGOALS‐f3‐L GCM, to clarify the bias decomposition. A random‐forest model for calculating SWCRE was developed using observation and model data of cloud fraction (CFR), cloud‐solar concurrence ratio (CSC), cloud liquid/ice water paths (LWP/IWP), top of the atmosphere (TOA) upward clear‐sky solar flux (SUC), and solar zenith angle. Then, following the partial radiation perturbation method, we reveal that the global‐mean TOA SWCRE bias (in W m−2) is mainly contributed by CFR (+5.67), LWP (−5.92), CSC (+3.28), and IWP (−1.72). Regionally, the relative importance varies according to climate regimes. The large CSC contribution highlights the importance of cloud diurnal variation. The sensitivity to the observational data was discussed, emphasizing the necessity of reducing observational uncertainties for implementing effective model‐improving measures.
Observations from the Little Bunker
Last updated: August 18, 2025 04:13 PM
Atmospheric CO₂
425.4 ppm
+18.4
ppm/year
Atmospheric CH₄
1934.91 ppb
+9.11
ppb/year
Global Temperature
+1.48 °C
Peak: +1.73°C
World Population
8.1 billion
+67
million/year