The interaction between natural marine aerosols, clouds and radiation in the Southern Ocean is a major source of uncertainty in climate models. We evaluate the Australian climate model using aerosol observations and find it underestimates aerosol number often by over 50 %. Model changes were tested to improve aerosol concentrations, but some of our changes had severe negative effects on the larger climate system, highlighting issues in aerosol-cloud interaction modelling.

In this study we present an evaluation that considers complex, non-linear systems in a holistic manner. This study uses XGBoost, a machine learning algorithm, to predict the simulated Southern Ocean shortwave radiation bias in the ACCESS model using cloud property biases as predictors. We then used a novel feature importance analysis to quantify the role that each cloud bias plays in predicting the radiative bias, laying the foundation for advanced Earth system model evaluation and development.

The paper addresses the crucial topic of hail damage quantification using radar observations. We propose a new radar-derived hail product that utilizes a large dataset of insurance hail damage claims and radar observations. A deep neural network was employed, trained with local meteorological variables and the radar observations, to better quantify hail damage. Key meteorological variables were identified to have the most predictive capability in this regard.

We propose a new method that should facilitate the use of weather radars to study wildfires. It is important to be able to identify the particles emitted by wildfires on radar, but it is difficult because there are many other echoes on radar like clear air, the ground, sea clutter, and precipitation. We came up with a two-step process to classify these echoes. Our method is accurate and can be used by fire departments in emergencies or by scientists for research.

Clouds play an important role in the Earth’s climate system as they modulate the amount of radiation that either reaches the surface or is reflected back to space. This study demonstrates an approach to robustly evaluate surface-based observations against a large-scale model. We find that the large-scale model precipitates too infrequently relative to observations, contrary to literature documentation suggesting otherwise based on satellite measurements.