Abstract Recent advances have introduced diffusion models for probabilistic streamflow forecasting, demonstrating strong early flood‐warning skill. However, current implementations rely on Long Short‐Term Memory (LSTM) backbones and discrete‐time diffusion formulations. This leaves unresolved whether more advanced sequence‐modeling architectures and continuous‐time diffusion formulations can improve probabilistic streamflow forecasting. In this study, we developed HydroDiffusion, a diffusion‐based probabilistic forecasting framework integrating a decoder‐only state space model (SSM) backbone with a continuous‐time score‐based diffusion formulation. HydroDiffusion is evaluated across 531 watersheds in the contiguous United States (CONUS) in the CAMELS data set. We benchmark HydroDiffusion against two diffusion baselines with LSTM backbones, as well as the recently proposed Diffusion‐based Runoff Model (DRUM). Results show that HydroDiffusion achieves the strongest nowcast accuracy when driven by observed meteorological forcings. Moreover, HydroDiffusion delivers higher deterministic and probabilistic forecast skill than its LSTM‐based counterpart in operational forecasting. Attribution analysis further reveals that the SSM backbone excels in humid, densely vegetated, storage‐dominated watersheds with strong hydrologic memory, while its advantage becomes limited in arid regions and hydrologically complex systems such as karst and prairie pothole basins. These results establish HydroDiffusion as an effective generative modeling framework for medium‐range streamflow forecasting, providing both a new modeling benchmark and a foundation for future research on probabilistic hydrologic prediction at continental scales.

Abstract Groundwater is one of the most valuable natural resources, yet it is increasingly under pressure from climate change and human activities. Shallow groundwater depth (GWD) is a key indicator for evaluating groundwater availability and flood hazards, yet the measurements are often too sparse to capture its variability both in time and space. In this study, we evaluated hybrid data‐driven machine learning (ML) models that combine the strengths of existing hydrological simulations with the predictive power of ML methods for improved GWD estimation in Denmark. The hybrid model reduced mean bias of hydrological simulations from 1.72 to 0.22 m and decreased the Root Mean Square Error by about 1.64 m at national scale. Furthermore, we demonstrated that increasing the availability of GWD observations over time and space enhances model performance. We also illustrated the flexibility and effectiveness of the hybrid approach for GWD estimation across scales, and results showed that the hybrid model developed with coarse spatial resolutions can be easily used for high‐resolution GWD estimation without significant decrease of accuracy. This study highlights the potential of hybrid modeling as a practical solution for improving groundwater quantification accuracy and shows avenues for higher‐resolution estimates.

Abstract Intensifying wet extremes under continued warming threaten water security and ecosystem functioning by perturbing land–atmosphere water fluxes. Yet, the watershed‐scale controls on how wet‐event intensity and duration translate into changes in fractional evapotranspiration (fET) remain poorly constrained at global scales. Here, using monthly wet climate extreme indices (CEIs) and fET across ∼986 global watersheds, we quantify CEI–fET coupling and attribute its spatial variability to watershed properties using Random Forest models and SHapley Additive exPlanations (SHAP). Across event types, soil organic carbon (SOC) and groundwater table depth (GWT) consistently structure coupling regimes, indicating a robust role for soil state and subsurface storage. However, the dominant texture axis differs by extreme type: intensity‐based extremes are most strongly organized by SOC–GWT–clay regimes, consistent with pulse retention and buffering, whereas duration‐based extremes are organized by SOC–GWT–silt regimes and permafrost‐related constraints, consistent with persistence‐driven drainage and saturation sensitivity. Supported SHAP‐transition analysis identifies data‐constrained regime boundaries. Lagged correlations further reveal multi‐timescale behavior: coupling varies at short lags but tends to shift toward negative associations at longer lags, especially for duration‐based extremes. In a permafrost‐limited subset, vegetation state and topography become more prominent predictors, highlighting how cold‐region constraints alter which attributes dominate. Together, these results provide a globally consistent diagnostic framework for identifying watershed regimes where intensifying wet extremes are more likely to yield sustained evapotranspiration responses versus reduced fET sensitivity. By quantifying regime thresholds, the study offers a diagnostic framework that may inform targeted mitigation strategies and policy frameworks, such as IPCC adaptation guidelines.

Abstract Offshore freshened groundwater (OFG) is a significant yet underexplored component of global water resources, occurring widely along many continental margins. However, its dynamics remain poorly constrained due to limited offshore observations and the absence of robust evaluation models for unambiguous emplacement interpretation. To address this knowledge gap, we integrate existing porewater data and geological constraints from the Canterbury Bight with paleo‐hydrogeochemical modeling to investigate the origin, evolution, and biogeochemical processes related to OFG driven by sea‐level fluctuations. Our simulation results indicate that modern recharge from onshore aquifers contributes only a limited fraction of the present OFG. The majority of OFG has mean groundwater ages of less than ∼300 kyr, suggesting that it mainly originates from local meteoric recharge during sea‐level lowstands since the late Pleistocene. OFG volume and paleo‐submarine groundwater discharge along the continental shelf have varied periodically in response to sea‐level fluctuations throughout glacial‐interglacial cycles. The Canterbury Bight OFG system is currently in a non‐steady state, undergoing gradual salinization by downward intrusion from overlying seawater at an estimated rate of 0.06 cm yr −1 . The long‐term evolution of the OFG system has increased the complexity of biogeochemical processes within shelf sediments through freshwater–seawater mixing. By integrating paleo‐hydrogeological reconstruction, geological archives, and reactive transport modeling, this study provides a comprehensive framework for understanding OFG formation and evolution, advancing beyond previous models focused primarily on salinity.

Abstract In fractured media with rough fissure skins, ignoring the fissure skin effect limits the accuracy of parameter inversion. In karst aquifers, groundwater storage and movement are typically governed by dual‐porosity models, where the fissure skin plays a key role in water exchange between fissures and the rock matrix. This study develops a radial CRT model incorporating the fissure skin effect based on a quasi‐steady state approach, deriving an analytical solution in the Laplace domain and obtaining a real‐time domain solution via Stehfest method. The improved model deepens the understanding of drawdown responses in fractured aquifers. The ratio of fissure skin storage coefficient ( α ) and exchange coefficient ( β ) jointly control the shape of the drawdown curve. An increase in α enhances the storage capacity of system, significantly reducing drawdown in the later stages. When β > 1, the fissure skin strengthens the hydraulic connection between the matrix and fissures, causing the early‐stage drawdown curve to shift downward, while β < 1 exhibits a typical flow resistance effect. Validation through a constant‐rate pumping test (CRT) in a mining area of Inner Mongolia demonstrates that the new model provides a better fit than the classical dual‐porosity model. Automatic inversion further confirms improved transition‐stage fitting with stable parameter estimates, providing a more reliable basis for characterizing fissure–matrix transition zones in karst aquifers.

Abstract Hydraulic barriers (HBs) are commonly used for solute plume containment in polluted aquifers. Their long‐term sustainability depends on minimizing the cumulative pumping rate, , required to ensure plume capture. Lagrangian random‐walk particle tracking (RWPT) is an effective approach for simulating advective‐dispersive transport to wells. However, as this method typically requires a large number of particles, , RWPT‐based simulation–optimization (SO) can be computationally demanding. This work presents an SO framework that couples a warm‐start genetic algorithm with parallel RWPT to identify minimal pumping rates for HB containment in a computationally efficient manner. The framework was applied in a Monte Carlo study using stochastic hydraulic conductivity fields, with ranging from to . Results showed that optimal values were jointly conditioned by transport processes, particle discretization, and algorithmic initialization. Under advection‐only transport, was nearly insensitive to . In contrast, when local dispersion was included, and its variability increased with . The influence of particle discretization depended strongly on the adopted recovery criterion, becoming substantially weaker when incomplete particle recovery was allowed. Neglecting local dispersion can therefore underestimate pumping requirements, particularly under stringent recovery criteria and increasing particle resolution. Warm‐start initialization generally identified lower solutions than cold initialization while also reducing computational time. These findings demonstrate that particle discretization, local dispersion and the recovery criterion should be considered explicitly when optimizing HBs, and that warm‐start strategies can substantially improve computational efficiency in RWPT‐based SO workflows.

Abstract Global warming is altering sea‐surface salt concentration (SSC) through intensified evaporation, shifted precipitation patterns, and accelerated melting of polar ice. Subsurface freshwater in oceanic islands is typically shaped like a thin lens due to limited land area and recharge, often the primary freshwater source for local communities and highly vulnerable to seawater intrusion (SWI). However, the effects of SSC changes on freshwater lens (FWL) and SWI extent in island aquifers remain unknown, and there is little knowledge on how hydraulic conductivity (HK), local aquifer recharge (RCH) and island width ( L ) affect these effects. To end this, this study employs a simplified two‐dimensional, variable‐density island groundwater simulation model to investigate how SSC affects FWL and SWI extent across seven distinct hydrogeological scenarios, using six quantitative indicators: relative changes in (a) average water table elevation (WTE), (b) maximum WTE, (c) FWL depth, (d) FWL thickness, (e) fresh groundwater volume, (f) aquifer salt mass. Simulation results demonstrate that rising SSC increases aquifer salt mass and elevates WTE while shrinking FWL (reducing its depth, thickness, and freshwater volume), as enhanced freshwater‐saltwater density contrasts drive more vertical flow pushing FWL upward. Island aquifers with higher HK, lower RCH, or smaller L show minimal changes in WTE and aquifer salt mass but are more sensitive to SSC changes in terms of FWL depth, thickness and fresh groundwater volume. Moreover, WTE and FWL characteristics show greater sensitivity to SSC declines than to SSC increases.

Abstract Previous research has pinpointed storm events as significant potential controls of microplastic (MP) export to the oceans. Hysteretic behavior of discharge ( Q )—concentration (c) relationships for suspended sediment (SS) and solutes has been found important in understanding the dynamic activation and transport of these substances, additionally providing better estimations of long‐term loads. Due to limited availability of high frequency event‐based MP concentration data their potential hysteretic behavior has not been studied, leaving a substantial knowledge gap regarding the fate and transport mechanisms controlling MP activation and retention within catchments. We pioneer the use of high frequency MP data collected during a storm event revealing the time variable activation of different sources and transport mechanisms within a peri‐urban catchment. Surface water samples were collected before, during and after the event and combined with discharge and turbidity data to quantify MP concentrations and fluxes. Analysis of the hysteretic behavior of the Q‐c relationship revealed a clockwise hysteresis pattern with MP concentration peaking before Q , suggesting potential MP supply limitation for the sampled event. Regression analysis indicates Q to be the primary driving force promoting MP flux, with strong correlations between Q and MP concentration in rising and falling limbs. A strong Q ‐ turbidity relationship was observed only in the rising limb, potentially indicating spatial differences in the sources of SS and MPs transported in the falling limb. Total MP flux increased by 17–217 times to that of baseflow, highlighting the significance of often‐overlooked dynamics when calculating yearly MP flux.

Abstract The 18 July 1942 Smethport, Pennsylvania storm produced US and world record rainfall accumulations of 165 mm in 10 min, 240 mm in 25 min and 780 mm in 270 min. Observations from the Smethport storm control sub‐daily Probable Maximum Precipitation (PMP) estimates over a large region of the eastern US. The 18 July 1942 storm is the archetypal example of a Smethport‐Like storm. A prominent feature of Smethport‐Like storms is that they are Northwest Flow events, a class of storm systems linked to development of the summer ridge over the US. The Northwest Flow setting for these storms results in characteristic properties of space‐time variability of rainfall extremes. Polarimetric radar analyses for recent Smethport‐Like storms illustrate the evolution of mesoscale convective systems producing extreme rainfall and provide insights into physical processes associated with peak rainfall rates from the Smethport storm. Smethport‐Like storms are on the extreme high end of the convective intensity spectrum, with storm rotation often contributing to large, intense updrafts. Synoptic scale, mesoscale and storm‐scale properties of Smethport‐Like storms combine to create a near‐optimal environment for extremes of water vapor convergence, a principal ingredient of record, short‐duration rainfall. The central element of the Smethport Challenge is to accurately simulate the climatology of rainfall extremes for Smethport‐Like storms using convection‐permitting climate models.

Abstract Understanding the temporal and spatial dynamics of critical shear stress τ c * is essential for predicting bedload transport rate in gravel‐bed rivers, especially under changing sediment supply conditions. This study conducted eight experimental runs using four sediment supply regimes and two feed textures to investigate how sediment supply history, including supply rate, magnitude, and texture, influences τ c * under non‐equilibrium conditions. The bedload transport rate, bed elevation, and sediment storage were measured with high temporal resolution, and τ c * was estimated using three classical bedload transport models and a slope‐dependent equation. The results show that τ c * is not constant, but varies dynamically with sediment input and bed adjustment. The higher the sediment supply rate and magnitude, the greater τ c *. Different variable sediment supply regimes would produce different τ c * trajectories, which are related to the specific experimental sequence of pre‐supply adjustment, variable sediment input, and post‐supply reorganization. Longitudinal differences also developed: localized deposition and bed slope steepening in the upstream reach caused τ c * values to be about 43% higher than those in the downstream reach. Grain selective transport further shaped system behavior: coarse gravels ( D 50 ‐ D 84 ) are preferentially stored while intermediate fractions ( D 16 ‐ D 50 ) are efficiently transported, generating counterclockwise hysteresis loops between sediment transport and storage. Overall, these findings demonstrate that τ c * should be treated as a dynamic parameter controlled by bed slope, sediment supply history, and bed surface adjustment. It also provides new insights for sediment transport modeling and river management under variable sediment regimes.