Advancing oceanographic forecasts in the Bay of Bengal: an AI-based spatiotemporal prediction for SST and MLD

Hybrid U-Net and ConvLSTM-based spatial forecasting of marine heatwave intensity and occurrence probability

Marine Heatwaves (MHW) are extreme sea surface temperature events that significantly affect marine ecosystems, fisheries, and coral reef environments, highlighting the need for accurate forecasting systems in vulnerable regions such as the Lesser Sunda Islands. This study aims to develop a hybrid deep learning-based MHW prediction system by integrating U-Net and ConvLSTM models over the regions of Bali, West Nusa Tenggara, and East Nusa Tenggara. The datasets used consist of NOAA OISST V2.1 daily sea surface temperature data for the period 1985–2024, along with Niño 3.4 and Dipole Mode Index (DMI) atmospheric indices. The U-Net model was applied to predict Sea Surface Temperature Anomaly (SSTA) intensity, while ConvLSTM was used to estimate the probability of MHW occurrence. The results demonstrate that the best forecasting accuracies of the U-Net intensity model for 1-, 3-, 5-, and 7-day lead times reached 0.8896, 0.8597, 0.8462, and 0.8674, respectively, with optimal thresholds of 0.92 °C, 0.56 °C, 0.51 °C, and 0.31 °C. The ConvLSTM probability model produced maximum FAR values of 0.9101, 0.8962, 0.8941, and 0.8854 with optimal probability thresholds of 0.25, 0.1, 0.1, and 0.1. RMSE evaluation increased gradually from 0.27 °C on day-1 to 0.94 °C on day-7 forecasts. Overall, the hybrid deep learning framework demonstrated robust and stable performance in representing both the intensity and probability of MHW events up to a seven-day forecasting horizon.

Subseasonal predictability of the Bay of Bengal summer monsoon onset tied to three-ocean sea surface temperature anomalies

Deep Learning-Based Statistical Downscaling of Sea Surface Temperature Using a Residual Corrective Neural Network

This study proposes a novel deep learning framework that uses a U-Net to generate an initial high-resolution SST estimate, which is subsequently refined using a residual corrective approach, and progressively refines initial U-Net predictions by incorporating dynamically scaled residuals at each step, enabling accurate capture of broad patterns and fine-grained features such as eddies and fronts.

A Deep Learning Model for Decadal Indian Ocean Dipole

The rapid and unprecedented warming of the Indian Ocean intensifies climate risks across socio‐economically vulnerable regions, highlighting the need of accurate prediction over the coming decade. This study first assesses the decadal predictability of the Indian Ocean Dipole (IOD) using the Decadal Climate Prediction Project (DCPP) models of Coupled Model Intercomparison Project Phase 6 (CMIP6), then develops a deep learning model based on a bidirectional gated recurrent unit (BiGRU) to enhance its predictive skill. The BiGRU model is designed to process multisource sequential data and is trained on the IOD index derived from DCPP simulations. Results show the MME achieves moderate skill on detrended IOD with an anomaly correlation coefficient (ACC) and mean squared skill score (MSSS) of 0.51 and 0.20 during 1963–2020, respectively. The BiGRU model improved the skill with an anomaly correlation coefficient (ACC) of 0.88 and a mean squared skill score (MSSS) of 0.72 during the testing period of 2009–2020, and accurately captured extreme IOD events. Based on the enhanced IOD index, we further used it to improve the prediction skill of the IOD‐related Australian rainfall, achieving ACC and MSSS values of 0.90 and 0.70 during 2009–2020, compared to 0.29 and −1.62 from the MME, underscoring the reliability of the BiGRU model for both IOD and related precipitation forecasts.

The Importance of Subsurface Thermal Effects to Indian Ocean Dipole Prediction Revealed by a Skillful Three-Dimensional Deep Learning Model

Deep learning for multi-step forecasting of stratospheric wind and temperature profiles over the Tibetan plateau

Accurate prediction of stratospheric wind and temperature profiles is essential for understanding regional atmospheric dynamics over the complex terrain of the Tibetan Plateau. However, conventional numerical weather prediction models are computationally expensive, while statistical time-series models are limited in capturing the nonlinear evolution of atmospheric variables across both temporal and vertical dimensions. To address these limitations, MTPV-HDRNet is proposed as a fixed-site, multi-level intelligent forecasting model driven by ECMWF ERA5 pressure-level reanalysis data, with the fixed site defined as a selected ERA5 grid cell rather than an observational station. MTPV-HDRNet jointly predicts zonal wind (U), meridional wind (V), and temperature (T) for the next 24 h across 11 ERA5 pressure levels from 100 to 1 hPa, which correspond to heights of approximately 16–48 km. The model adopts a hybrid encoder–decoder architecture that explicitly represents temporal evolution and vertical stratification in a decoupled but complementary manner, thereby enhancing its ability to capture multi-lead-time profile evolution and cross-level dependencies. The model was trained using ERA5 data from 2016 to 2022, validated using data from 2023, and independently tested using data from 2024 at the fixed site in the central Tibetan Plateau. Results demonstrate that MTPV-HDRNet consistently outperforms the baseline models at all forecast lead times. On the 2024 test set, the mean RMSEs are 3.48 m s − 1 for U wind, 3.54 m s − 1 for V wind, and 1.62 K for temperature. Compared with the strongest baseline, ConvLSTM, MTPV-HDRNet reduces the RMSE by 11.9%, 11.3%, and 7.4% for U wind, V wind, and temperature, respectively. The model also maintains strong correlation performance at longer lead times, demonstrating its robustness in fixed-site atmospheric profile forecasting over complex terrain.