Abstract: Objectives: Against the background of global warming and increasingly frequent drought disasters, the sustainable utilization of regional water resources and socioeconomic development are facing growing challenges. Accurate identification of individual drought events and quantitative characterization of their development, spatial migration, and overall severity are essential for drought risk assessment and water-resources management. The Gravity Recovery and Climate Experiment Drought Severity Index (GRACE-DSI) reflects integrated terrestrial water storage deficits, including variations in surface water, soil moisture, and groundwater. However, conventional GRACE-based drought studies generally analyze temporal variation or spatial distribution separately, making it difficult to distinguish multiple adjacent drought events from long-term gridded datasets while preserving their complete spatiotemporal continuity. This study therefore combines GRACE-DSI with a three-dimensional (3D) connectivity recognition method to identify hydrological drought events in the Yangtze River Basin and systematically characterize their duration, affected area, intensity, cumulative deficit, spatial migration, and occurrence frequency. Methods: Monthly terrestrial water storage data from April 2002 to December 2024 were obtained from the CSR RL06.3 Mascon product, which is provided on a 0.25° × 0.25° grid. The data gap between the GRACE and GRACE Follow-On missions from July 2017 to May 2018 was reconstructed using a backpropagation neural network. Precipitation, evapotranspiration, root-zone soil moisture, groundwater runoff, and shallow surface-water storage derived from the Global Land Data Assimilation System were used as model inputs, while GRACE-derived terrestrial water storage was used as the target output. For the 2002- 2010 modeling period, the correlation coefficient, Nash-Sutcliffe efficiency coefficient, and root mean square error were 0.93, 0.90, and 1.02, respectively. Corresponding values for the independent evaluation period from April 2010 to June 2017 were 0.90, 0.85, and 1.14, confirming the applicability of the model to the mission gap. GRACE-DSI was then calculated by standardizing terrestrial water storage against the long-term mean and standard deviation for the corresponding calendar month. For cross-validation, 1-, 3-, 6-, 9-, and 12-month Standardized Precipitation Evapotranspiration Index series were standardized using the same procedure to obtain SPEI-Z and were further processed using wavelet low-pass filtering. Grid cells with GRACE-DSI ≤ −0.8 were classified as drought cells. A 26-neighbor 3D connectivity scheme based on the Chebyshev distance was used to construct initial connected components in longitude-latitude- time space. Sensitivity experiments indicated that both the drought-occurrence area threshold and the overlap threshold between adjacent months should be set to 1.5%, which balances noise removal against the retention of drought signals in secondary sub-basins. For each identified event, drought duration, monthly and mean affected area, mean drought intensity, cumulative drought deficit, intensity-weighted drought-center migration, and drought occurrence frequency were calculated. Spatial trends in GRACEDSI were estimated using Sen’s slope. Results: After wavelet low-pass filtering, the 6- and 9-month SPEIZ series showed the strongest agreement with GRACE-DSI, with correlation coefficients of 0.70 and 0.69, respectively, whereas the correlation with the 1-month SPEI-Z was only 0.46. These results indicate that GRACE-DSI primarily represents the cumulative response of terrestrial water storage to sustained hydroclimatic deficits rather than short-term meteorological anomalies. A total of 82 drought events were identified during the study period, including 15 large-scale events whose mean drought-affected areas exceeded 25% of the basin. Their major affected regions, spatial expansion directions, and migration patterns were broadly consistent with historical drought records, although their identified onset and termination generally lagged the documented meteorological droughts by approximately 1-2 months. The 2009-2010 Southwest China drought and the 2022-2023 Yangtze River Basin drought demonstrated that the proposed method can track the expansion, migration, and recession of different types of drought events. The 2022-2023 event lasted 11 months and had a mean affected area of 47.00%, a mean drought intensity of 0.46, and a cumulative drought deficit of 6311.10, the largest among all identified events. Its mean intensity was not the highest, indicating that the severity of long-duration and large-area droughts cannot be adequately represented by mean intensity alone and should also be evaluated using cumulative deficit. Mild and moderate droughts occurred relatively frequently, whereas severe and more extreme droughts were characterized by intermittent and localized outbreaks. Approximately 70% of the basin exhibited a gradual negative trend in GRACE-DSI, indicating an intensification of terrestrial water storage deficits. The most pronounced negative trends occurred near the junction of Chongqing, Hubei, and Shaanxi, in the northwestern Qinghai region, and in the southeastern Hunan-Jiangxi region. Drought frequency was no more than 30% across most of the basin, while Yunnan and Sichuan provinces and the Dongting Lake and Poyang Lake basins were identified as drought-prone regions. Conclusions: Combining GRACE-DSI with 3D connectivity recognition enables the extraction of independent drought events from long-term gridded terrestrial water storage data while retaining their spatial extent, temporal persistence, and cross-month connectivity. The method is particularly suitable for identifying and tracking large-scale, persistent hydrological droughts and for evaluating cumulative water-storage deficits. The observed 1-2-month lag relative to meteorological drought records reflects the cumulative response and propagation of precipitation and evapotranspiration anomalies through the basin water-storage system. Nevertheless, the monthly temporal resolution and spatial smoothing of GRACE observations reduce sensitivity to short-duration, localized, and rapidly developing flash droughts. The proposed framework provides a useful basis for large-scale hydrological drought monitoring, event-based severity assessment, and long-term water-resources risk analysis in the Yangtze River Basin.