国家气象局

Abstract Simulations of numerical weather prediction models indicate that the atmosphere possesses an intrinsic limit of predictability. Initial perturbations of tiny amplitude grow quickly in areas of convection and latent heat release, then spread out and move upscale, eventually affecting even the largest planetary scales after about 2 weeks. In this study, we investigate the ability of several state‐of‐the‐art AI‐based weather prediction models to reproduce this phenomenon, which is sometimes referred to as the “butterfly effect.” The AI results are compared to those of a conventional, physics‐based, weather prediction model run at various resolutions. Evaluating six key characteristics of this butterfly effect, we find that the behavior of the AI models can be separated into two groups. The first group did not reproduce any of the key characteristics, while the second group did reproduce some, in particular fast initial uncertainty growth and indication of an intrinsic limit. However, the behavior was physically inconsistent and based on the production of numerical noise, and for some models even dependent on whether the experiments were carried out on a CPU or a GPU. It seems likely that the inability of AI models to simulate the butterfly effect results from limitations in the analysis data used for training, since their size, design and architecture turned out to be largely irrelevant.

There is evidence that rainfall extremes have become more intense and frequent over the last few decades, but it is difficult to assess these changes due to the limitations of our short observational records. We use the UNprecedented Simulated Extreme ENsemble (UNSEEN) approach to (1) assess changes in extreme rainfall over Southern Africa and Southeast Asia over the last 40 years and (2) identify locations that have a high chance of breaking rainfall records. We find that extreme rainfall risk has already increased since 1981 during the rainy season in both regions, including a doubling of risk in some months for many major population centers such as Phnom Penh, Vientiane, Bangkok, Hanoi, Maseru, Johannesburg, Lilongwe, and Lusaka. The pattern of increasing risk of extreme rainfall is projected to increase further in the coming 20 years in the CMIP6 ensemble; yet UNSEEN estimates of changes from the last 20 years are already greater than these future projections in the Philippines, northern Mozambique, and northern Madagascar. Finally, we compare the UNSEEN ensemble to historical records to identify places that have “soft records” and are likely to see record-setting events. These places with increasing risks but no recent extremes are labeled as “sitting ducks” in today’s climate. We find that much of Mozambique, the Philippines, and Laos would be considered “sitting ducks” for extreme precipitation in at least one month of the year. Disaster risk managers should use these types of large ensembles when estimating the risk of extremes in today’s climate, in order to ensure that society is prepared for record breaking events. This approach can also be used for improving engineering design estimates of rainfall return periods and for stress-testing health system and disaster preparedness.

Light-absorbing particles on surface ice in ablation areas can accelerate glacier melting and shrinkage. A Single Soot Particle Photometer was used to measure black carbon (BC) mass concentrations (MBC) in the ablation area of Potanin Glacier, Mongolia during summer. Surface-ice MBC values (42–555 ng g−1) greatly exceeded those of surface snow (5–22 ng g−1), snow and rain (2–6 ng g−1), and surface melt water (2–11 ng g−1). Vertical profiles of MBC revealed high surface-layer concentrations, suggesting impurities trapped in the granular ice: the particularly low-density layer of the weathering crust surface. In the ablation area, MBC values of granular ice decreased with lower elevation: 134–601 ng g−1 at the 3317 m site and 8–96 ng g−1 at the 3078 m site. The fraction of residual surface BC to BC contained in lost water over a year, R, was calculated using the yearly BC deposition flux and water ablation weight Aw. Average R values were 0.17 and 0.011, respectively, at 3317 and 3078 m. Aw were 246 and 325 gw.e.cm-2, suggesting that the granular ice retains BC particles best in the upstream ablation area, showing concomitantly less capability with increasing ablation. Enriched BC on the ablation area surface comprises recent BC deposits and BC from the glacier's lower layer after rising during decades or more. Those BC emissions and deposits can therefore affect both future and present ablation area melting processes.

As artificial intelligence (AI) systems transition from research prototypes to operational tools in Earth system science and forecasting, establishing trust in their predictions becomes increasingly important. Although model inputs and outputs are observable, the internal decision-making of modern AI models remains complex and hard to interpret, earning them the label “black boxes.” Explainable artificial intelligence (XAI) offers techniques to provide insight into these processes. However, most XAI methods were developed for classification tasks, raising questions about their suitability for the regression problems that dominate geoscientific applications. We review XAI approaches through this lens, organising them into a structured framework and examining both their theoretical foundations and practical behaviour. To ground this discussion, we apply a selection of methods to a machine learning emulator of the Lorenz 1963 system, an archetypal chaotic model that provides a tractable, physically meaningful setting for exposing the limitations and failure modes of general-purpose XAI in regression contexts. We then survey how these and related methods have been applied across a variety of Earth system sciences. We further situate XAI within the model development lifecycle, linking methodological choices to the needs of different stakeholder groups across operational Earth system science. We close by identifying gaps in existing methodologies and outlining a forward-looking research agenda, with practical recommendations for the responsible, effective use of XAI in regression applications of geoscientific modelling and forecasting.

Cold extremes continue to have considerable impacts on a wide range of sectors including health, energy, agriculture, and infrastructure. On a global scale, the frequency and intensity of cold extremes are declining due to anthropogenic climate change. This general decreasing trend in cold extremes is well captured by models for the historical period. However, there are strong regional and seasonal differences in cold extreme occurrence that can be attributed to the variability of large-scale dynamical drivers of cold extremes such as sea ice, the polar stratosphere, and ENSO. The uncertain future evolution of these large-scale drivers, as well as shortcomings in our understanding of the links between these drivers and cold extremes, make it difficult to constrain the magnitude and year-to-year variability of the projected decrease in cold extremes. This review reveals a range of unresolved questions pertaining to the dynamical forcing of cold extremes and their evolution in a changing climate.