A foremost difficulty in improving models is the inability to observe atmospheric process rates in real world conditions at model grid scales such that process rate parameterizations in the model can be robustly evaluated. Cloud processes including interactions with aerosols critical to climate prediction are a prime example. Thus, model evaluation has primarily relied on statistical comparisons of cloud and aerosol properties that are retrievable from observations. However, to proceed from model evaluation to improvement requires identification of the sources of model-observation discrepancies that are difficult to ascertain given many possible sources and feedbacks that obscure them. Most often, properties are compared between model output and observations individually, while a wide range of parameterization components are varied to assess impacts on comparisons, often leading to arbitrary tuning of models. Joint variable distributions in constrained atmospheric conditions are often controlled by a lesser number of relevant process parameterizations and thus provide valuable information on how to focus parameterization testing and development.
We compare observed and Energy Exascale Earth System Model (E3SM)-simulated 2- and 3-variable joint distributions over 5 years at the Atmospheric Radiation Measurement (ARM) Eastern North Atlantic (ENA) site using several different retrievals from surface and satellite remote sensing perspectives. This is done within the context of evaluating E3SM’s ability to simulate liquid cloud albedo susceptibility for overcast single layer liquid cloud conditions. In particular, the response of albedo to cloud droplet number concentration (CDNC; aka the Twomey effect), the liquid water path (LWP) adjustment to CDNC, and the response of CDNC to cloud condensation nuclei concentration (CCN) are evaluated. The simulated Twomey effect is underestimated, related to parameterized drop effective radii being smaller than observed for a given CDNC-LWP. However, CDNC increases with CCN more than observed, a relationship that offsets the lesser albedo sensitivity to CDNC to yield albedo responses to CCN that are similar to observed, though this indicates a possible bias in the aerosol activation parameterization. The observed net negative LWP adjustment to increasing CDNC is simulated and is potentially more negative than observed though observation-based retrievals are highly variable depending on the datasets used. LWP adjustment is further examined with respect to inversion strength, above inversion relative humidity, and precipitation rate to assess possible model-observation differences in autoconversion plus accretion vs. entrainment-driven evaporation.
The usage of different retrieval datasets for both observations and E3SM are essential in properly interpreting potential causes for model-observation differences. For example, the parameterized effective radius in E3SM alters CDNC-LWP-albedo relationships, but it is also divorced from the predicted size distribution moments that control microphysical process rates. Using multiple datasets is also critical for evaluating the robustness of comparisons as the spread between different observation-based retrievals at times exceeds the differences between those datasets and E3SM. Ongoing research is extending analyses to additional cloud regimes and model resolutions with incorporation into the open-source Earth System Model Aerosol-Cloud Diagnostics (ESMAC-Diags) evaluation package.