Climate variability modes, usually known as primary climate phenomena, are well recognized as the most important predictability sources in subseasonal–interannual climate prediction. This paper begins by reviewing the research and development carried out, and the recent progress made, at the Beijing Climate Center (BCC) in predicting some primary climate variability modes. These include the El Niño–Southern Oscillation (ENSO), Madden–Julian Oscillation (MJO), and Arctic Oscillation (AO), on global scales, as well as the sea surface temperature (SST) modes in the Indian Ocean and North Atlantic, western Pacific subtropical high (WPSH), and the East Asian winter and summer monsoons (EAWM and EASM, respectively), on regional scales. Based on its latest climate and statistical models, the BCC has established a climate phenomenon prediction system (CPPS) and completed a hindcast experiment for the period 1991–2014. The performance of the CPPS in predicting such climate variability modes is systematically evaluated. The results show that skillful predictions have been made for ENSO, MJO, the Indian Ocean basin mode, the WPSH, and partly for the EASM, whereas less skillful predictions were made for the Indian Ocean Dipole (IOD) and North Atlantic SST Tripole, and no clear skill at all for the AO, subtropical IOD, and EAWM. Improvements in the prediction of these climate variability modes with low skill need to be achieved by improving the BCC’s climate models, developing physically based statistical models as well as correction methods for model predictions. Some of the monitoring/prediction products of the BCC-CPPS are also introduced in this paper.

By performing perfect model predictability experiments in a coupled general circulation model from the point of view of initial error growth, it is demonstrated that there may exist a “summer prediction barrier” (SPB) in the predictions of Pacific Decadal Oscillation-related sea surface temperature anomalies (PDO-related SSTA), which refers to the phenomenon that initial errors exhibit a significant season-dependent evolution, with the largest error growth occurring in the August–September–October (ASO) season. Due to the effect of the SPB, the prediction error of PDO-related SSTA events starts to grow rapidly during the ASO season and yields a positive (negative) prediction error for the warm (cold) events, essentially inducing a delay of the events. The SPB may be one of the main factors limiting the predictability of North Pacific sea surface temperature. The physical and dynamical mechanisms of the SPB are explored from two aspects: thermodynamics and dynamics. In terms of thermodynamics, we demonstrate that the fastest error growth of PDO-related SST cold (warm) events during the ASO season is mainly due to the largest anomalous release (absorption) of the latent heat flux in this season; while for dynamics, the effect of the vertical gradient of climatological mean sea temperature on the anomalous upwelling (downwelling) dominates the contribution of the dynamical temperature advections to the error growth of the PDO-related SST cold (warm) events during the ASO season. The anomalous release (absorption) of the latent heat flux and anomalous upwelling (downwelling) of the ocean currents are both related to the anomalous northwesterly or cyclonic (southeasterly or anticyclonic) wind over the PDO-related SSTA region, which indicates that the error growth associated with the SPB of PDO-related SST cold (warm) events is mainly driven by anomalous wind stress. The error growth associated with the SPB for PDO-related SSTA may also explain why the SSTA in the Kuroshio–Oyashio Extension is much less predictable than in other regions of the North Pacific, as shown by some state-of-the-art climate models.

The seasonal predictability of sea surface temperature anomalies (SSTA) in the Kuroshio‐Oyashio Extension (KOE) is explored by performing perfect model predictability experiments from the viewpoint of initial error growth in a global coupled model. It is found that prediction errors of KOE‐SSTA always increase in the boreal summer and decrease in the boreal winter. This leads to smaller (larger) prediction errors and higher (lower) prediction skills in boreal winter (summer). This seasonal characteristic of the KOE‐SSTA error growth implies a season‐dependent predictability that is lower in summer and higher in winter. The mechanism responsible for error growth associated with seasonal predictability is also explored. The error increase in summer and error decrease in winter in the KOE‐SSTA are both largely attributed to the seasonal evolution of latent heat flux error and mean temperature advection by vertical current error in the KOE region, both of which are forced by the prediction error of 1 month leading zonal wind stress per unit mass for the mixed layer over the KOE region. The shallowest (deepest) mixed layer in summer (winter) amplifies (reduces) the forcing of zonal wind stress errors on the error growth of KOE‐SSTA, thereby causing the seasonal evolution of prediction errors of KOE‐SSTA and ultimately resulting in the season‐dependent predictability of the KOE‐SSTA, i.e., low in summer and high in winter.

Gaofen-1/Gaofen-6 (GF1/GF6) wide-field-view (WFV) imagery provides repeated wide-area observations for regional surface-water monitoring, but its use is complicated by fragmented water patterns, vegetation–water mixtures, terrain and cloud shadows, and cross-scene radiometric variability. This study presents a seven-channel nested U-Net (UNet++) framework with a residual logit refinement branch (RLRB) for binary surface-water extraction. The input combines Blue, Green, Red, near-infrared, normalized difference water index, normalized difference vegetation index, and digital-elevation-model-derived slope channels. Five complete GF1/GF6 scenes were locked before dataset partitioning, and an 8192 m exclusion buffer was used to exclude spatially overlapping or adjacent non-test tiles from model development, thereby reducing the risk of spatial leakage. The resulting quantitative test set contained 2783 valid 512 by 512 pixel tiles derived from the five scenes withheld at scene level. The UNet++ model with RLRB and edge-consistency supervision achieved an mIoU of 0.941, Water IoU of 0.889, and Water F1 of 0.941. A controlled $2\times 2$ experiment separated the effects of residual refinement and edge-consistency supervision, while feature, loss-component, random-seed, rule-based-baseline, and complete-scene analyses further characterized the framework. Complete-scene sliding-window inference required approximately 3.7 min per scene on a single NVIDIA GeForce RTX 5060 Laptop GPU, demonstrating the feasibility of georeferenced water-mask generation under the specified computational setting.