This study evaluates how horizontal resolution (~ 25 vs. ~ 100 km) affects El Niño–Southern Oscillation (ENSO) simulation in the Chinese Academy of Sciences Flexible Global Ocean-Atmosphere-Land System (CAS FGOALS-f3) climate model. A reproducible, process-based framework reveals ENSO biases stem from resolution-dependent air–sea feedbacks and high-frequency atmospheric variability. This work informs future development for the FGOALS-f3 family and serves as a reference for CMIP6/CMIP7 evaluation.

Wildfires pose serious risks to natural environments and cities, making it essential to predict how smoke travels through the air. We studied how wind patterns affect computer models of fire spread. We simulated fires under both smooth and turbulent airflows. Our results show that while assuming smooth wind is acceptable for light breezes, it causes errors in moderate winds. Therefore, models must account for natural turbulence to accurately predict fire plume in realistic weather.

This study tested using machine learning to speed up detailed simulations in the SCREAM (Simple Cloud-Resolving E3SM Atmosphere Model) model. By training ML (machine learning) models to correct a simpler version of SCREAM, some results improved, but others did not. Technical challenges were addressed, and new tools were developed. The work shows promise for making simulations more efficient, though further improvements are needed.

Wet soil are difficult to model because water changes how sunlight interacts with soil surfaces and alters reflectance in different viewing directions. We developed a unified physical model to simulate soil reflectance under varying moisture conditions. The results show that soil moisture not only darkens soil but also changes their directional reflectance, especially in the near-infrared region. This work can improve soil moisture retrieval and future remote sensing applications.

Bidirectional flux models give a unified model for emission and dry deposition, but few studies have been conducted in which satellite observations are used to refine the parameters in these models. A new bidirectional flux model for ammonia was developed that was designed specifically for use with satellite observations. Ammonia satellite observations were used to refine bidirectional flux model parameters, which improved the agreement of the model with ammonia surface observations.

This study highlights challenges in estimating seasonal climate predictability using the perfect model framework, which assumes only initial conditions cause error. We show that forecasts can exceed the predicted limit, known as the Potential Predictability Limit (PPL), due to model imperfections in simulating physical processes. A new method is proposed to estimate PPL more accurately and avoid such paradoxes.

We develop a novel data assimilation system by integrating incremental machine learning with a localized ensemble Kalman filter to enhance the simulation and reanalysis of particulate matter < 2.5 µm chemical components. Compared to traditional chemical transport model-based data assimilation, our assimilation system has superior computational efficiency and simulation improvements. Comparisons with independent observations and reanalysis datasets validate the robust performance of our system.