Abstract: The Pearl River Delta (PRD) is experiencing increased hydroclimatic variability alongside intensive human regulation, resulting concurrent risks of floods, droughts, and saltwater intrusion. This study examines multi-scale changes in extreme flow events and quantifies reservoir regulation, with attention to the combined effects of upstream karst–surface water interactions and downstream tidal dynamics. We compiled daily discharge and precipitation records from seven representative stations across upstream, midstream, and downstream reaches, along with operational data from one key reservoir, covering the period from 2006 to 2017. Methods include the Mann-Kendall test with linear slope estimation for trend detection, seasonal indices for flood−drought allocations, extreme value models (Gumbel distribution for flood peaks and Gamma distribution for low flows), and a physically interpretable reservoir regulation ratio R. This ratio isolates the operational contribution of the reservoir under varying hydrological regimes. Trend diagnostics indicate significant increases in annual flow at Makou (the Xijiang River) and Sanshui (the Beijiang River), while Boluo (the Dongjiang River) shows no monotonic trend. Estimated slopes are approximately 171.4 and 50.3 m³·s⁻¹ at Makou and Sanshui, respectively. Seasonal indices reveal that upstream stations retain a pronounced flood-season amplification consistent with natural precipitation variability; midstream stations exhibit stabilization due to regulation; and downstream stations, influenced by tidal backwater and diversions, display more balanced flood-drought allocations. Frequency analysis reveals spatially contrasting hazards. At Makou, the 10-year flood is about 5.9×10⁴ m³·s⁻¹, reflecting strong basin-scale concentration and operational constraints during heavy rainfall. For drought risk, the 20-year low flow at Sanshui is 53 m³·s⁻¹, indicating elevated supply risk during prolonged deficits. The Gamma model for low flows avoids nonphysical negative values and preserves the lower-tail behavior needed for drought design. Overall, flood hazard scales with contributing area (Boluo<Sanshui<Makou), whereas drought vulnerability is station-specific, shaped by both regulation and tidal backwater. Downstream, days with flow reversal (upstream-directed velocity), which diagnose tidal dominance and salt-intrusion propensity, have increased over the study period. Reversal-day counts rise by 12% on average and cluster in the dry season (December-March), when upstream inflows are minimal. This trend is consistent with sea-level rise and dredging-altered hydraulics that enhance tidal penetration. The regulation ratio R captures operational fingerprints across regimes. During flood seasons, R<1 dominates, indicating peak shaving; typical events yield between 0.3 and 0.9. In dry seasons, R>1 occurs frequently, with support ratios above 0.8 in most years, showing active augmentation of downstream flows for ecological and municipal needs. The R series exhibits spikes during basin-wide droughts (emergency releases) and troughs during intercepted flood peaks, directly evidencing dual mitigation by storage.Synthesizing the results, climate change and human activities jointly polarize the PRD hydrologic regime, resulting higher flood peaks and lower low flows upstream, and more frequent tidal reversals downstream. Large reservoirs serve as effective tools for managing the flood-drought trade-off by reducing peak discharges while maintaining environmental flows, but they also alter natural hydrographs and sediment-morphodynamic processes. The proposed R metric provides a concise, physically grounded indicator for diagnosing and comparing regulatory efficacy across events and years, as well as for communicating operational targets under multi-hazard constraints. Although the analysis is limited by the 2006 to 2017 record and idealizations in tributary corrections, the framework is transferable to other mixed karst/non-karst river systems facing compound hazards.