Zihui Liu

Zihui Liu

Auto-generated profile, based on publicly available information

Zihui Liu

Department of Atmospheric Sciences University of Washington Seattle WA USA

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 August 2025 in AGU Advances
Verified
Journal Article
Open access
NCNathaniel Cresswell‐Clay
Nathaniel Cresswell‐Clay
BLBowen Liu
Bowen Liu
DDDale R. Durran
Dale R. Durran
ZLZihui Liu
Zihui Liu
ZEZachary I. Espinosa
Zachary I. Espinosa
MKMatthias Karlbauer
Matthias Karlbauer
A key challenge for computationally intensive state‐of‐the‐art Earth System models is to distinguish global warming signals from interannual variability. Here we introduce Deep Learning Earth System Model (DL ESy M), a parsimonious deep learning model that accurately simulates the Earth's current climate over 1000‐year periods with minimal smoothing and no drift. DL ESy M simulations equal or exceed key metrics of seasonal and interannual variability—such as tropical cyclogenesis over the range of observed intensities, the cycle of the Indian Summer monsoon, and the climatology of mid‐latitude blocking events—when compared to historical simulations from four leading models from the sixth Climate Model Intercomparison Project. DL ESy M, trained on both historical reanalysis data and satellite observations, is an accurate, highly efficient model of the coupled Earth system, empowering long‐range sub‐seasonal and seasonal forecasts while using a fraction of the energy and computational time required by traditional models.

26 January 2026 in Journal of Advances in Modeling Earth Systems
Verified
Journal Article
Open access
ZLZihui Liu
Zihui Liu
RERyan Eastman
Ryan Eastman
RWRobert Wood
Robert Wood
Improving cloud cover prediction is one of the fundamental challenges for large‐scale climate simulations due to the stochastic nature of clouds and their complex interactions with the atmospheric circulation. This study explores various machine‐learning (machine learning) approaches and utilizes Lagrangian air mass history to improve prediction accuracy. Here, we use 169,824 isobaric boundary layer trajectories over the eastern subtropical oceans with colocated meteorological data at 12‐hr intervals over 4 days. Satellite cloud cover data from MODIS are colocated with the trajectory points where available, resulting in 43,830 trajectories (26%) that are fully filled. All models received seven cloud‐controlling factors (CCF) at each timestamp to predict total cloud cover simultaneously. Several statistical models applied here are found to predict cloud cover with similar or better performance than the leading meteorological reanalysis. The best model using recurrent neural networks and cloud cover feedback achieves a correlation coefficient of 0.72 between predictions and MODIS measurements, compared to 0.65 for the reanalysis. Applications of these models are investigated. We determine sensitivities of cloud cover to cloud‐controlling parameters by adding different perturbations to CCFs and recording consequent changes. This sensitivity study reveals a nonlinear relationship between cloud cover and numerous CCFs.

8 February 2025
Verified
Preprint
Open access
ZLZihui Liu
Zihui Liu
RERyan Eastman
Ryan Eastman
RWRobert Wood
Robert Wood
Jabbrv-ltwa-all.ldf jabbrv-ltwa-en.ldf Improving cloud cover prediction is one of the fundamental challenges for large-scale climate simulations due to the stochastic nature of clouds and their complex interactions with the atmospheric circulation. This study explores various machine-learning (ML) approaches and utilizes internal and external methods to incorporate Lagrangian air mass history to improve prediction accuracy. Here, we use 169,824 isobaric boundary layer trajectories over the eastern subtropical oceans with colocated meteorological data at 12-hour intervals over 4 days. Satellite cloud cover data from MODIS are colocated with the trajectory points where available, resulting in 43,830 trajectories (26\%) that are fully filled. These are used to train models via a variation of Stochastic Gradient Descent (SGD) called Adaptive Moment Estimation (Adam). All models received 7 cloud-controlling factors (CCF) at each timestamp to predict total cloud cover simultaneously. Several statistical models applied here are found to predict cloud cover with similar or better performance than the leading meteorological reanalysis. The best model using recurrent neural network and cloud cover feedback achieves a correlation coefficient of 0.72 between predictions and MODIS measurements, compared to 0.65 for the reanalysis. Applications of these models are investigated. We determine sensitivities of cloud cover to cloud-controlling parameters by adding different perturbations to CCFs and recording consequent changes. This sensitivity study reveals a peculiar nonlinear relationship between cloud cover and numerous CCFs. By providing different data-driven perspectives, we begin to leverage the power and speed of ML models as an alternative to the challenge of cloud cover prediction in physical models.