Haoming Chen

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point matching is overly sensitive to minor errors, whereas spatial verification typically necessitates parameter tuning or heuristic rules derived from expert knowledge, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV's verification score. Experiments demonstrate that CLPFV outperforms traditional and spatial verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point methods are overly sensitive to minor errors, while spatial verification methods commonly require setting parameter and rule comprehensively, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area size) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV’s verification score. Experiments demonstrate that CLPFV outperforms traditional (POD, FAR, TS) and spatial (FSS, SAL) verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point methods are overly sensitive to minor errors, while spatial verification methods commonly require setting parameter and rule comprehensively, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area size) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV’s verification score. Experiments demonstrate that CLPFV outperforms traditional (POD, FAR, TS) and spatial (FSS, SAL) verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point methods are overly sensitive to minor errors, while spatial verification methods commonly require setting parameter and rule comprehensively, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area size) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV’s verification score. Experiments demonstrate that CLPFV outperforms traditional (POD, FAR, TS) and spatial (FSS, SAL) verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point methods are overly sensitive to minor errors, while spatial verification methods commonly require setting parameter and rule comprehensively, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area size) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV’s verification score. Experiments demonstrate that CLPFV outperforms traditional (POD, FAR, TS) and spatial (FSS, SAL) verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.

High-performance computing now enables a new generation of global kilometer-scale models. As part of the World Climate Research Programme (WCRP) Global Hackathon 2025 initiative, for the first time, multiple cutting-edge global kilometer-scale models have been run for an entire year. All of them have covered the summer of 2020, when East Asia experienced record-breaking precipitation and catastrophic floods, mainly driven by mesoscale convective systems (MCSs). Using an updated storm-tracking algorithm, this study investigated the performance of six global kilometer-scale models in simulating MCS characteristics during the record-breaking wet summer of 2020 in East Asia. Results revealed that all models generally reproduced MCS characteristics, including MCS size, duration, and key features of convection and precipitation. Models also generally captured finer characteristics such as diurnal variations and the frequency-intensity distribution of hourly precipitation. Among the models, Integrated Forecast System (IFS) performs best in capturing MCS rainfall spatial distribution, Nonhydrostatic ICosahedral Atmospheric Model (NICAM) excels in simulating MCS size, and Simple Cloud-Resolving E3SM Atmosphere Model (SCREAM) most accurately represents the land-sea contrast in MCS precipitation intensity. A common bias across models is the underestimation of rainfall area and overestimation of heavy precipitation intensity, indicating simulated convective cores are stronger than observed. Our results demonstrate that global kilometer‑scale modeling has reached a significant benchmark, yet persistent biases remain in MCS simulation. Continued improvements in these models will not only enhance the reliability of modeling but also to improve disaster risk reduction and climate change adaptation.

Accurately simulating mesoscale convective systems (MCSs) is essential for predicting global precipitation patterns and extreme weather events. Despite the ability of advanced models to reproduce MCS climate statistics, capturing extreme storm cases over complex terrain remains challenging. This study utilizes the Global–Regional Integrated Forecast System (GRIST) with variable resolution to simulate an eastward-propagating MCS event. The impact of three microphysics schemes, including two single-moment schemes (WSM6, Lin) and one double-moment scheme (Morrison), on the model sensitivity of MCS precipitation simulations is investigated. The results demonstrate that while all the schemes capture the spatial distribution and temporal variation of MCS precipitation, the Morrison scheme alleviates overestimated precipitation compared to the Lin and WSM6 schemes. The ascending motion gradually becomes weaker in the Morrison scheme during the MCS movement process. Compared to the runs with convection parameterization, the explicit-convection setup at 3.5-km resolution reduces disparities in atmospheric dynamics due to microphysics sensitivity in terms of vertical motions and horizontal kinetic energy at the high-wavenumber regimes. The explicit-convection setup more accurately captures the propagation of both main and secondary precipitation centers during the MCS development, diminishing the differences in both precipitation intensity and propagation features between the Morrison and two single-moment schemes. These findings underscore the importance of microphysics schemes for global nonhydrostatic modeling at the kilometer scale. The role of explicit convection for reducing model uncertainty is also outlined.

Accurate precipitation forecast verification (PFV) is essential for improving forecasting models and supporting disaster management. However, current PFV methods remain limited, point-to-point methods are overly sensitive to minor errors, while spatial verification methods commonly require setting parameter and rule comprehensively, which constrains their availability. To tackle these issues, we are inspired by the success of deep learning in image verification through extracting high-level features, and thus propose a self-supervised contrastive learning-based PFV method (CLPFV). First, CLPFV uses precipitations augmentation (displacement, intensity, area size) to simulate actual forecast errors and construct positive and negative training sample pairs. Subsequently, with a novel loss function proportionally penalizing forecast errors, a backbone network is trained in CLPFV to extract high-level precipitation features. Finally, the cosine similarity of features is calculated as CLPFV’s verification score. Experiments demonstrate that CLPFV outperforms traditional (POD, FAR, TS) and spatial (FSS, SAL) verifications in different degrees of forecast errors and aligns better with expert assessments. In general, CLPFV offers an efficient deep learning solution for PFV tasks.