Interpretable AI predicts a 2026 summer dry anomaly in central China

Seasonal prediction of Western North Pacific tropical cyclone formation using convolutional neural network

MAUSAM: An Observations‐Focused Assessment of Global AI Weather Prediction Models During the South Asian Monsoon

Past evaluation of artificial intelligence (AI) weather prediction has primarily relied on reanalyses, which can obscure important deficiencies due to prevailing biases in reanalyses themselves. Here, we present MAUSAM (Measuring AI Uncertainty during South Asian Monsoon), an evaluation of seven leading AI‐based prediction systems—FourCastNet, FourCastNet‐SFNO, Pangu‐Weather, GraphCast, Aurora, AIFS, and GenCast—during the South Asian Monsoon, using ground‐based weather stations, rain gauge networks, and geostationary satellite imagery. The AI models demonstrate considerable forecast skill during the monsoon across a broad range of variables, ranging from large‐scale surface temperature and winds to precipitation and cloud cover. In addition, the models generate realistic zonal means and eddy statistics on subseasonal to seasonal timescales, even capturing key aspects of the abrupt Monsoon pause of 2025. However, the models still exhibit systematic errors in finer‐scale features, including the underprediction of extreme precipitation (“tail” events), divergent cyclone tracks, and disagreements in the mesoscale kinetic energy spectra, highlighting avenues for future improvement. A comparison against a combination of in situ and remote sensing observations reveals forecast errors 15%–45% larger than those relative to reanalysis and traditional HRES forecasts, indicating that reanalysis‐centric benchmarks can overstate forecast skill. Of the models assessed, AIFS exhibits the most consistent representation of atmospheric variables, with GraphCast and GenCast also showing strong skill. The analysis presents a framework for systematically evaluating global AI weather models on regional predictions. The results highlight both the promise and current limitations of AI weather prediction and the continued need for observations‐focused evaluation.

Identifying SST precursors for forecast skill windows of East Asian winter cold surges

Sources of Potential Predictability for July–August 2025 Extreme Precipitation Anomalies Over Northern China: Roles of the Western Pacific and North Atlantic–Barents–Kara Sea Sectors

Accurately predicting persistent extreme precipitation remains a major challenge for subseasonal‐to‐seasonal (S2S) prediction models. This study identifies two distinct sources of predictability for the July and August 2025 extreme precipitation anomalies over northern China (NC) and reveals how these remote signals improved the prediction. By separating S2S models into high‐skill and low‐skill groups, we show that the July predictability was mainly linked to tropical forcing over the western Pacific (WP), whereas the August predictability was primarily associated with midlatitude wave‐train activity from the North Atlantic (NA) and Barents–Kara Seas (BKS). In July, high‐skill models better captured typhoon‐induced convection over the subtropical WP and the associated northward‐propagating Pacific–Japan wave train. This circulation response favored westward moisture transport from the warm WP along the edge of the northward‐extended western Pacific subtropical high, thereby improving the prediction of NC precipitation. In August, high‐skill models more realistically represented eastward‐propagating midlatitude wave trains from the NA and BKS. These wave trains enhanced energy conversion over NC and contributed to a more accurate prediction of the August extreme precipitation anomaly. Although the S2S models showed clear indications of predictability, most models underestimated the precipitation intensity, highlighting persistent model deficiencies in representing the amplitude of extreme rainfall. Nudging experiments further quantified the contribution of these remote predictable signals. Correcting WP circulation and convection‐related anomalies explained approximately 78% of the July intensity improvement, while correcting NA and BKS wave‐train anomalies explained approximately 46% and 66% of the August improvement, respectively.

Toward robust fine-scale decadal precipitation forecasts through dynamically consistent subsampling

Seasonal precipitation skill in current climate models is due to initial tropical conditions alone

Using the model-analog forecast technique, we demonstrate that seasonal prediction skill of global precipitation anomalies in current climate models arises entirely from the initial sea surface temperature anomalies (SSTA) within the tropics. Model-analogs generate predictions by matching initial observed conditions to climate states drawn from pre-existing libraries of long model simulations; the subsequent time evolution of the matched states in the model forms a forecast ensemble that can replicate seasonal SSTA prediction skill of traditionally-initialized dynamical models. In this study, we show that model-analogs also capture the monthly-to-seasonal precipitation skill of current forecast models for almost all regions and seasons worldwide. Model-analogs are selected by matching monthly-mean SSTA from uninitialized CESM2 large ensemble runs to observations, between 60°S-60°N, for the years 1979–2021. At one season lead, the resulting model-analog hindcasts have global monthly precipitation skill that is statistically indistinguishable from the Seasonal to Multi-Year Large Ensemble (SMYLE), traditionally-initialized hindcasts using the same model and initialization dataset. This model-analog skill is not significantly improved by including observed 500 hPa geopotential height and/or soil moisture anomalies in model-analog selection. Additionally, multi-model ensemble-mean model-analog hindcasts based only upon tropical monthly mean SSTA, drawn from 33 CMIP6 model pre-industrial control runs, show seasonal precipitation skill comparable to (and for longer leads, even somewhat higher than) that of the operational North American Multi-Model Ensemble, particularly for the U.S. Southwest and Southeast regions. That is, if significant initial sources of extratropical seasonal precipitation skill exist beyond the tropics, current models appear unable to realize them.