Abstract Understanding the hydrodynamics of bifurcation and confluence zones is essential for effective river management. This study investigated the complex flow structure in a large diffluence‐confluence unit (width‐depth ratio >100) of the Middle Yangtze River, utilizing systematic field observations of three‐dimensional velocities under low to high discharges (8,560–46,100 m3/s). In terms of streamwise flow, increasing discharges caused the bifurcation node to migrate downstream, and the flow diversion ratio of secondary branch rose as the incoming discharge increased. As for secondary current, it exhibited notable changes at bifurcation zone across different flow conditions. No significant, channel‐scale secondary flow cells were was observed at low discharge, and a single secondary cell formed at medium discharge; then dual face‐to‐face secondary cells developed at bifurcation zone at high discharge, with the surface flow directed inward and near‐bed flow moving outward. To summarize, the variation in flow patterns was collectively driven by streamline curvature, local geomorphological features, and topographic steering. Flow curvature generated centrifugal forces, altering the rotation direction of secondary flow along the reach. Abrupt changes in wetted cross‐sectional area caused flow separation, intensifying both downward and transverse motions and thereby enhancing secondary flow. Furthermore, the main flow preferentially entered the primary branch with lower bed elevation at low discharge; as the inflow discharge increased, the effect of topographic steering weakened, producing a more uniform velocity distribution and consequently raising the flow diversion ratio in the secondary channel. This provides a transferable mechanistic framework to understand flow dynamics in anabranching rivers globally.

Bank erosion in the densely populated Lower Yangtze River (LYR) poses a significant threat to riparian residents and industry. Different from general bank erosion types in previous researches, it typically has a arc shape in the LYR, develops rapidly within a few hours and results in an obvious erosion pond. However, modeling such erosion is challenging due to the continuously changing flow structure within the pond as the bank boundary retreats. Therefore, a three-dimensional model is proposed in the current research, coupling a hydro-and sediment dynamic module with a bank erosion module. The results show that: ① The proposed model successfully reproduced the hydraulic conditions and the arc-shaped bank erosion process in the study reach. The calculated bank erosion area in the tidal reach is around 98,370 m2 , close to the measured value of 84,060 m2 ; ② The proposed optimization strategies include raising the gate elevation by 2.5 meters and shifting the location of the shoulder guarding structures towards the main stream by approximately 35 to 63 m. Compared to the pre-optimization scenario, the optimized plan reduces the eroded bank area by 1493m2 , a reduction of 17.4%, and decreases the maximum flow velocity by an average of approximately 14.5%. Additionally, the area of intense turbulence within the ponds decreases by approximately 12.4%.

Human activities have led to global warming, triggering frequent natural disasters, especially floods, which have caused significant harm to people. Evacuation is an effective measure to enhance urban resilience against flood. Research on optimising evacuation routes, particularly for pedestrian evacuation, is limited. Metaheuristic route optimisation algorithms, such as Ant Colony Optimisation (ACO), Genetic Algorithm (GA), Particle Swarm Optimisation algorithm (PSO) and Sparrow Search Algorithm (SSA), are gaining attention due to their flexibility, high computational efficiency, and adaptability to various scenarios, yet they have not been applied to flood evacuation problems. This research improves metaheuristic algorithms to optimise flood evacuation routes by revising and including the flood risk associated with the route as a part of the objective function. The improved algorithms developed in this study include Improved ACO (IACO), Improved GA (IGA), Improved PSO (IPSO) and Improved SSA (ISSA). The improved algorithms were incorporated into a flood evacuation research framework comprising flood modelling, flood hazard rating, and evacuation route optimisation and applied to the 2023 flood event in York, UK. The improved metaheuristic algorithms effectively navigate road networks and narrow streets, optimising pedestrian evacuation routes and avoiding high-risk flood zones. The IACO and IGA produced the route with the lowest risk and fewest turns, respectively. The IPSO algorithm was the most computationally efficient, generating the routes at the highest speed, while the ISSA showed the slowest speed. This generalisable framework integrates flood risk with infrastructure connectivity to optimise evacuation routes that enhance urban resilience and sustainability.

Traditional hydrodynamic models are often constrained by low computational efficiency，making it difficult to meet the demands of large-scale，high-accuracy flood forecasting and warning. To enhance the efficiency of urban flood simulation，this study develops a high-performance hydrodynamic model based on multi-GPU parallel acceleration. The model integrates three core computational modules: a 2-D surface runoff module，a surface-sewer flow interaction module，and an underground sewer flow module. To optimize the most computationally intensive 2-D surface runoff module，Metis graph partitioning was employed for spatial domain decomposition and load balancing，while MPI-OpenACC technology was implemented to achieve multi-GPU parallel acceleration. The model was applied to the Zhongshundawei (Zhongshan-Shunde Embankment) area in Guangdong Province to simulate the flooding process during the "24·5" extreme rainstorm event using nowcasting rainfall data. The results demonstrate that the developed model achieves high computational accuracy，with a coefficient of determination (R2) of 0.91 between the measured and simulated water depths at major waterlogging points. The multi-GPU parallel acceleration technology substantially improved the model’s efficiency，achieving a 26.38-fold speedup with 8 GPUs compared to a 64-core CPU setup. Notably，the model simulated a 6-hour flood process for an 811 km2 urban area (50.68 million computational meshes) at a 4-meter spatial resolution within 10 minutes. This study demonstrates the significant potential of the multi-GPU parallel hydrodynamic model for large-scale，high-resolution flood forecasting and warning，and the findings provide robust technical support for urban flood management.

The evolution of bifurcated reaches is complicated owing to flow and sediment diversion at bifurcation and subsequent differential branch deformation. Previous numerical studies have mostly focused on short-term bed deformation in bifurcated reaches, overlooking the process of bank erosion, and their practical application is often limited by high computational costs and the necessity for high-precision topographic data. Therefore, a conceptual model was proposed to simultaneously simulate bank erosion and bed deformation processes of different branches, calibrated and verified using the measurements from two typical bifurcated reaches in the Middle Yangtze River. Results show that: (i) the main branch experienced scouring, while the secondary branch showed deposition or slight scouring, with the calculated cross-sectional profiles closely matching the measurements; (ii) the mid-channel bar adjacent to the main branch exhibited more intense bank erosion due to a higher flow diversion ratio, with a bank retreat width of about 10 m, further promoting the development of the main branch; and (iii) the fluvial erosion intensity of incoming flow-sediment conditions showed a positive correlation with the deformation difference between the branches, and short branches with small bifurcation angles experienced stronger erosion compared with long branches. Besides, engineering interventions that rose the riverbed elevation at the main branch entrance by 1 m could switch the roles of the main and secondary branches, potentially promoting the transition from a two-branched system to a single-channel system.

Braided reaches are the most unstable channel type, prone to rapid and frequent migration. The construction of upstream dams significantly alters the downstream flow-sediment regime, leading to substantial variations in channel migration rates, which poses critical challenges for river management and training works. This study examines the braided reach of the Lower Yellow River, which exhibits intense channel adjustments and diverse migration patterns, serving as an ideal site to evaluate decadal-scale changes in thalweg and centerline migration rates. Based on long-term remote sensing imagery and cross-sectional profile measurements, results indicate that (i) a significant spatial and temporal reduction in channel migration rates has occurred. The average rate of thalweg migration in the braided reach reduced from 229 m/a during the pre-dam stage to 166 m/a during the post-dam stage, accompanied with the rate of centerline migration reducing from 122 to 76 m/a. The middle subreach was always the most active in the braided reach, but the spatial difference of channel migration rate reduced by around 20 % during the post-dam stage; (ii) the thalweg migration rate generally exceeded or equaled the centerline migration rate, but the discrepancy between these two types spatially varied. The smallest difference was observed in the upper sub-reach, and the largest in the middle sub-reach; and (iii) a reduction in channel migration rate was reconciled with the significant decrease in incoming sediment coefficient and the increase in bankfull depth caused by upstream damming, with the incoming sediment coefficient identified as the dominant controlling factor.

Dam-induced channel adjustments often reduce low-water levels but can paradoxically increase flood levels. The Middle Yangtze River (MYR) may serve as a representative example of dam-induced downstream fluvial responses. The cumulative channel scour volume in the MYR was 2.8 billion m3 after the operation of the Three Gorges Project. The flood levels exhibited a pronounced rise in some years owing to the combined effects of bed incision, increased channel roughness and local base-level rise, while the low-water levels declined consistently due to channel degradation. The total scour volume is predicted to reach 2.8 billion m3 in the next 30 years, with the low-water levels declining by 3 m in the Jingjiang Reach, and the flood levels would remain almost unchanged due to the balance between intensive channel degradation and increased channel resistance. The findings underscore the necessity of incorporating century-scale channel evolution predictions into the analysis of water level variations.

Driven by continuous channel degradation，the water level-discharge relationship has undergone significant changes in the Lower Yellow River (LYR). Based on the hydrological and topographic data from six hydrometric stations in the LYR from 1986 to 2023，this study investigated the characteristics and mechanisms of flood-level changes in the LYR，and quantified the effects of channel morphological adjustment and channel resistance variation on flood levels at a given discharge. Results show that: ①Before the operation of the Xiaolangdi (XLD) Reservoir (1986-1999)，flood levels at all six stations showed an overall upward trend，with the values at a discharge of 4000 m3/s increasing by 1.43-1.78 m. After the XLD Reservoir operation (1999-2023)，flood levels at 4000 m3/s decreased by 1.51-3.40 m. ②The decline in flood levels at identical discharges was positively correlated with channel scour volumes. For each 100 million m of scour in the braided，transitional and meandering reaches，flood levels at 4000 m3/s dropped by 0.17-0.85 m at the corresponding stations. ③Numerical experiments were used to quantify the contributions of different factors on flood levels at 4000 m3/s between 1986 and 2020. It revealed that increased channel resistance raised flood levels of 0.45-0.91 m，whereas channel morphological changes lowered them by 0.59-1.58 m. Downstream of the Sunkou Station，these two effects largely offset each other，resulting in little net change. While in the upstream reach，the flood-level rise attributable to increased channel resistance was less pronounced than the decline caused by riverbed incision，but it still counteracted 24.3% of the total flood-level decrease.

Study region The Waiho River is a rapidly aggrading, levee-confined, glacier-fed braided river on the West Coast of New Zealand, crossing a steep proglacial alluvial fan where stopbanks protect the State Highway 6 lifeline corridor and adjacent community. Study focus This study quantifies how riverbed aggradation alters flood behaviour and levee performance and evaluates alternative levee strategies for the Waiho River. Using multi-temporal topographies (2016–2023), terrain-change analysis, and spatially distributed floodplain roughness derived from remote sensing, we apply a two-dimensional hydrodynamic model to simulate (1) breach sensitivity on both riverbanks under varying bed elevations and (2) levee reconfiguration options from partial to complete removal of the southern levee system. New hydrological insights for the region Results show that incremental levee heightening yields only short-term benefits; under continued aggradation, it elevates water levels against protected margins and exacerbates breach consequences. Breach behaviour is strongly conditioned by riverbed elevation, indicating that aggradation control and freeboard management must be planned jointly. Partial south-levee removal provides limited relief, whereas complete removal (or an equivalent setback) substantially lowers hydraulic loading on the opposite bank and redistributes flow and shear across the southern floodplain, promoting wider conveyance and sediment dispersion consistent with a reconnected system. Overall, the findings support an adaptive pathway prioritising restored floodplain connectivity, complemented by targeted reinforcements and risk-informed operations near critical assets (e.g., lifeline highway infrastructures).

Finite-amplitude meander growth is accompanied by higher-order planform distortion, including fore-aft asymmetry (skewing) and internal curvature redistribution (fattening and sharpening). Although these traits are widely observed in natural rivers, their systematic scaling and environmental modulation remain poorly constrained. Here, we analyze 1,697 individual bends from 12 freely meandering rivers spanning arid, highland, arctic, and tropical environments. Meander centerlines are fitted with a modified Kinoshita-type function to extract non-dimensional skewing and fattening coefficients, enabling consistent comparison across river sizes and environmental settings. Results reveal robust scaling relationships linking higher-order distortions to fundamental, first-order meander morphometrics, including inflection angle, sinuosity, and wavelength. Upstream skewing strengthens progressively with increasing bend amplitude and sinuosity, reflecting an autogenic outcome of nonlinear, curvature-driven meander growth. In contrast, fattening and sharpening are strongly controlled by normalized wavelength, indicating partially independent evolution of bend amplitude and wavelength. Allogenic forcings, including hydrological variability and riparian vegetation density, primarily modulate first-order morphometrics and the variability of skewing and fattening around scaling trends, producing systematic differences in meander geometry across river types. Together, our results define a continuous spectrum of natural meander planforms and provide a quantitative framework for synthesizing representative meander geometries and interpreting river evolution under contrasting environmental conditions.

Upstream damming usually alters bed-material composition and bedform morphology in the downstream reaches, causing significant impact on movable bed resistance—a key factor in flood control and river management. This study systematically quantified resistance components (grain resistance and bedform resistance) using field measurements (2003–2017) at five hydrometric stations in the Middle Yangtze River (MYR), employing the method of hydraulic radius decomposition. Results show that both grain and bedform resistances under equivalent discharges increased gradually after the operation of the Three Gorges Project (TGP), and the increases in the Manning's roughness coefficient ranged from 8% to 101% for grain resistance, and from 5% to 54% for bedform resistance; spatially, the amplification magnitude decreased as it propagates downstream, and bedform resistance remained as the dominant component, constituting about 85% of total movable bed resistance. Furthermore, it is found that bedform resistance showed a discharge-dependent variation, decreasing with higher flows due to enhanced sediment suspension (suspension index > 1.1) that suppressed the relative bedform height; meanwhile, bedform resistance increased annually under equivalent discharge as progressive bed-material coarsening after the TGP operation reduced sediment suspension index, thereby promoting the formation of larger bedforms. This study provides the first systematic quantification of movable bed resistance dynamics in response to dam-induced flow-sediment alterations and channel degradation, offering critical insights for predicting flood-level changes and informing river management strategies in dammed river systems.

Abstract Effective sediment management in reservoir–river coupled systems remains a global challenge due to the conflicting objectives of sediment evacuation from reservoirs and deposition reduction in downstream reaches. This study proposed a physics‐based modeling framework oriented to a reservoir–river coupled system, which integrates morphodynamic calculation and reservoir operation simulation via a two‐way feedback mechanism. Unlike simplified approaches, this framework captures complex transient flow‐sediment processes at the system scale. The study region covers the 1,000‐km Tongguan–Lijin reach in the Middle and Lower Yellow River, encompassing the Sanmenxia–Xiaolangdi cascade reservoirs. This region represents a typical reservoir–river coupled system facing an urgent need to balance the conflicting objectives. The validation results demonstrated that the proposed model achieved high‐fidelity modeling of flow‐sediment fluxes in reservoirs and downstream reaches. Multi‐scenario simulations and sensitivity analyses provide quantitative insights into how the overall sediment transport efficiency responds to varying reservoir operation parameters during the period of Water‐Sediment Regulation Scheme, including replenishing discharge and regulation water volume. In addition, this study identifies distinct optimal operation strategies for different deposition reduction targets: a sharp and short‐duration flushing mode can maximize the reservoir sediment evacuation, while a low‐discharge but long‐lasting release process can maximize the downstream channel erosion. Further simulations across various flow‐sediment regimes demonstrate the universality of the proposed operation strategies. The findings provide insights for adaptive selection of reservoir operation strategies when addressing the conflicts between different stakeholders at a flood‐event scale.

In alluvial rivers, sediment particles exhibit a strong affinity to phosphorus (P), with the adsorption capacity varying across different particle sizes, and the experimentally derived adsorption parameters can be used in water quality models that consider the effect of sediment adsorption processes. However, there exist limited systematic experimental studies on how particle size influences P adsorption by sediment. To investigate the P adsorption characteristics of different-sized sediments, surface bed sediment samples were collected from two sections in the Middle Yangtze River (MYR), with particle size (D) ranging between 2 and 500 μm. These sediment mixtures were then separated into 7 size fractions (down to D < 8 μm), with the corresponding mean particle size () ranging from 5.9 to 424.2 μm. The native adsorbed P (NAP) amounts for different-sized sediment fractions were measured, and P adsorption experiments were conducted for the 7 sediment fractions under 3 different sediment concentrations (S) with a similar environment of pH and temperature when sampled from the MYR. Experimental results show that: (i) NAP values for sediment particles with D < 500 μm ranged from 0.15 to 0.47 mg·g−1, exhibiting a negative exponential relation with ; (ii) P adsorption by sediment primarily occurred within the first 4 h, with the highest adsorption rate, and both the amount and rate of P adsorption were negatively correlated with S and ; (iii) key adsorption parameters in the Langmuir isotherm and kinetic equations were calibrated using the experimental data, achieving high fitting accuracy. The maximum P adsorption amount (Qmax) obtained in isothermal experiments was well described by the equation considering the combined effects of S and : Qmax = 1.6189S-0.228 -0.283, with particle size having a greater influence than sediment concentration; and (iv) in addition, the kinetic adsorption data for most sediment fractions (D < 250 μm) were closely distributed and displayed a consistent linear trend, leading to the determination of unified adsorption kinetic parameters for nonuniform sediment under different S values: k1 = 0.257 L·mg−1·h−1 and k2 = 0.057 h−1.

Abstract Significant changes in flow‐sediment dynamics and channel evolution have occurred in the Lower Yellow River owing to upstream damming. Notably, the post‐dam stage has witnessed an unexpected increase in movable bed roughness and more frequent dune development, driven by reduced sediment load, coarsened bed material, and altered hydraulic conditions. However, existing formulas for movable bed roughness, mostly calibrated using pre‐dam or flume data, fail to capture these new trends. A new formula is developed to calculate movable bed roughness, considering the effects of flow condition and bedforms, based on a new criterion for flow regime partition. The spatiotemporal variations in bedforms and the contributions of different factors were discussed. Results indicate that: (a) Froude number and relative water depth are key factors of movable bed roughness. The formula shows high accuracy, with the determination coefficients under different flow regimes larger than 0.70. (b) Lower flow regime is more likely to develop after the reservoir operation, especially in the braided reach under low discharges. The frequency of lower flow regime (embodied with ripples and dunes) under the low discharge generally increased by 1.3 times during the post‐dam stage. But there is a lag in the peak frequency of lower flow regime between the braided and transitional reach. (c) Although a slight increase existed in the contributions of relative water depth to movable bed roughness during the post‐dam stage, the flow condition, was still a dominant factor, with the contributions of Froude number exceeding 60% during both pre‐ and post‐ dam stages.

Retrogressive erosion, a critical process impacting river engineering structures, channel geomorphology, and reservoir sediment management, manifests primarily in two distinct forms: rotating and stepped bed deformation. The former type is dominated by sediment entrainment, whereas the latter is dominated by mass failure. Simulating these processes accurately, particularly predicting which form will dominate, poses significant challenges for existing numerical models. This study addresses this gap by developing and validating a numerical model capable of simulating both rotating and stepped retrogressive erosion with automatic adjustment between the two types. The model integrates governing equations for unsteady nonuniform flow and nonequilibrium sediment transport with specialized modules for each erosion type. The rotating erosion module incorporates sediment entrainment theories suitable for high flow velocities and steep slopes, accounting for shear dilatancy effects. The stepped erosion module employs force equilibrium analysis to predict the critical horizontal erosion distance at the step toe to induce mass failure. A key feature is the implementation of a criterion based on the ratio of shear stresses at the top and bottom of the overfall relative to the critical shear stress, allowing the model to adapt the simulation approach on the basis of evolving hydrodynamic conditions. The hydrodynamic and sediment transport equations are solved at each time step, and the criterion is applied to determine whether the rotating or stepped erosion model is used to further solve bed deformation. Model calibration and verification were performed via laboratory flume data covering various inflow discharges, initial step heights, and bed material properties. For rotating-type erosion simulations, the model demonstrated high accuracy, with Nash–Sutcliffe efficiencies (NSEs) for water surface and bed elevation calculations generally exceeding 0.9. The calculated cumulative erosion amounts also agreed well with the measurements, with relative errors mostly less than 10 % in the later stages. The maximum Froude number, which was located at the end of the foreset reach, increased from 1.72 to 3.40 during the entire test. The maximum sediment concentration was almost constant. For stepped erosion, the model successfully replicated the characteristic headcut migration, although the results were sensitive to the erodibility coefficient. The errors of predicted overall migration rate were within 12.5 % of the measured values. The developed model provides a robust tool for predicting retrogressive erosion dynamics and is uniquely capable of handling both rotating and stepped forms.

Bank erosion is a critical geomorphic process resulting from the complex interaction between flow-sediment dynamics and riverbed boundaries, posing significant threats to flood control and navigation safety. Accurate prediction of such events remains challenging due to the intricate coupling of river water, groundwater, and soil mechanics. To address this issue, this study improves a one-dimensional (1D) bed deformation and bank erosion model (1DBEM), which couples the modules for flow and sediment transport, bed deformation, bank erosion, and groundwater level calculation. The proposed model was applied to the Middle Yangtze River, and the simulated results were compared with those obtained using HEC-RAS to evaluate performance. The results show that: (i) 1DBEM achieves high accuracy in simulating discharge, water level, and suspended sediment concentration, with Nash-Sutcliffe efficiency values of 0.998, 0.995, and 0.850 at the Shashi hydrological station, outperforming the HEC-RAS (0.993, 0.990, and 0.572); (ii) the prediction precision of bank erosion locations reaches 60%, compared with 44% for HEC-RAS; (iii) the root mean square errors between the simulated and measured groundwater levels at the Xiangjiazhou and Beimenkou monitoring wells were 0.36 m and 0.39 m for 1DBEM, outperforming the 1.02 m obtained with HEC-RAS; and (iv) the two models exhibit consistent trends in bank slope stability factors, confirming the rationality of the 1DBEM approach. Overall, the improved model provides an effective tool for assessing bank erosion risk and managing flood safety.

Understanding the variation in bed material gradation provides a crucial perspective for assessing the functionality of river systems, including bed resistance, flood conveyance capacity and non-uniform sediment transport pattern, yet few studies have been conducted to simulate the adjustment processes of bed material gradation in a sand-bed reach. A new calculation method was developed that integrated active layer thickness with riverbed evolution and bedform geometry, which advanced the framework of active layer method for modelling bed material gradation adjustment in sandy riverbeds. The proposed method was incorporated into a section-scale morphodynamic framework, providing more detailed simulations of the dynamic adjustment processes of bed material gradation and riverbed evolution. The performance of this framework was evaluated through two laboratory experiments and field datasets at seven hydrometric cross-sections in the Lower Yellow River. Results demonstrated good agreement between the predicted and observed bed material gradations, with the mean absolute error less than 7.3%, and also captured the dynamic adjustment tendency of riverbed. Comparative analysis of typical methods for calculating the active layer thickness revealed that methods integrating hydraulic conditions and bed material properties, particularly the method proposed in this study, performed better in simulating both ultimate gradations and adjustment processes in sandy riverbeds. This framework provides a valuable tool for advancing the flow-sediment transport modelling and predicting the long-term geomorphic evolution trends in sand-bed river systems.

Flooding driven by climate change increasingly threatens cities, placing subterranean infrastructure at risk. Existing assessment methods may fail to fully capture the associated flood risks. In particular, they may underestimate the impact on pedestrian evacuation safety in underground spaces. In this study, an integrated multidimensional hydrodynamic framework is proposed to investigate urban flood evolution and intrusion processes in underground spaces. The model is applied to an urban area in Wuhan that is susceptible to severe flooding, incorporating pedestrian risk assessment under extreme rainfall scenarios. Spatiotemporal analysis of floodwater depth and velocity reveals that flooding in underground spaces escalates quickly. The first basement level (B1) becomes submerged to a depth of 0.3 m within the first 5 min. Floodwaters subsequently overflow into the second basement level (B2), reaching the critical floodwater depth by 15 min. The distribution of risk levels shows a significant increase in flood hazard for both adults and children. By 15 min, over 90 % of the B1 area and 96 % of the B2 area are classified as medium or higher risk. Notably, children experience higher and earlier flood risks compared to adults. By the 10-min mark, 70 % of the area for children is classified as high or extremely high risk. In contrast, only 13 % of the same area poses a similar risk level for adults at that time. These findings underscore the urgent need for improved flood risk management and early warning systems to protect urban infrastructure and vulnerable populations, especially children.

Sub-cloud evaporation is a critical aspect of the hydrological cycle, reducing surface precipitation totals and altering the stable isotopic composition as raindrops fall from the cloud base towards the surface. However, isotopic modelling of sub-cloud evaporation in humid climates and its implications for hydrological processes remain poorly understood, posing challenges for regional water resource management and ecological conservation. In this study, a comprehensive assessment was conducted to understand the influence of sub-cloud evaporation on precipitation isotopes in Poyang Lake, the largest freshwater lake in China. Using 4-year hourly meteorological observations from 11 national meteorological stations, we found that there was significant sub-cloud evaporation during the precipitation process in humid regions. The remaining fraction of evaporated raindrops varied between 81% and 95%, with the lowest values occurring in September and the highest in February. The monthly average triangle delta 2H, triangle delta 18O and triangle d-excess values ranged from 2.9 parts per thousand to 7.0 parts per thousand, 0.7 parts per thousand to 1.8 parts per thousand, and -7.7 parts per thousand to -2.8 parts per thousand, respectively, and the sub-cloud evaporation effect during the rainy season was more intense than that during the dry season. By modifying the sub-cloud evaporation effect, precipitation isotopes monitored at the surface and estimated at the cloud base were confirmed to exhibit consistent temporal patterns on both monthly and daily scales. Sensitivity analysis revealed that precipitation isotopic changes were more sensitive to fluctuations in relative humidity and precipitation intensity under varying meteorological scenarios. The underrepresentation of low-intensity precipitation events was found to lead to a statistical underestimation of precipitation isotopic changes, and when the low-intensity events (<= 1.0 mm/h) were excluded, the average triangle delta 18O and triangle d-excess values shifted from 1.25 parts per thousand and -5.25 parts per thousand to 0.63 parts per thousand and -2.72 parts per thousand, respectively. These findings contribute to a better understanding of hydrological cycle processes in Poyang Lake and other regions with similar humid climate characteristics, especially for the interpretation of regional paleohydrological records and ecohydrological mechanisms using stable isotopes.