Significant attention has recently been paid to deep learning as a method for improved catchment modeling. Compared with process‐based models, deep learning is often criticized for its lack of interpretability. One solution is to combine a process‐based hydrological model with a residual error model based on deep learning to give full scope to their respective advantages. In classical residual error models, Bayesian inference via Markov chain Monte Carlo (MCMC) is commonly used to provide an estimation of the uncertainty. However, deep neural networks tend to have excessively large numbers of parameters, making MCMC an unsuitable approach. Here, we introduce an alternative to Bayesian MCMC sampling called stochastic variational inference (SVI) which has recently been developed for Bayesian deep learning in Natural Language Processing. We implement SVI in a Long Short‐Term Memory (LSTM) network and construct residual error models in process‐based hydrological models. This approach is examined in the contrasting geographical and climatic characteristics of two catchments from China, the Tangnaihai catchment and the Shiquan catchment. Compared with the Bayesian linear regression model, the Bayesian LSTM provides better uncertainty estimates. Specifically, the proposed method improves the Continuous Ranked Probability Score (CRPS) by over 10% in both two catchments. In the Tangnaihai catchment, it provides more than 10% narrower uncertainty intervals in terms of Sharpness with slightly superior Reliability. In the Shiquan catchment, it provides comparable uncertainty intervals with better Reliability. Further, our study highlights the scalability of SVI to high‐dimensional parameter spaces in hydrological applications (e.g., distributed hydrological models, groundwater models).

Flood forecasting in semiarid regions is always poor, and a single-criterion assessment provides limited information for decision making. Here, we propose a multicriteria assessment framework called flood classification–reliability assessment (FCRA) that combines the absolute relative error, flow classification and uncertainty interval estimated by the hydrologic uncertainty processor (HUP) to assess the most striking feature of an event-based flood: the peak flow. A total of 100 flood events in four catchments of the middle reaches of the Yellow River are modeled with hydrological models over the period of 1983–2009. The vertically mixed runoff model (VMM) is compared with one physically based model, the MIKE SHE model (originating from the Système Hydrologique Européen program), and two conceptual models, the Xinanjiang model (XAJ) and the Shanbei model (SBM). Our results show that the VMM has a better flood estimation performance than the other models, and the FCRA framework can provide reasonable flood classification and reliability assessment information, which may help decision makers improve their diagnostic abilities in the early flood warning process.

Intense sub-daily precipitation on the Tibetan Plateau (TP) can trigger a cascade of natural hazards (flash floods, debris flows, etc.), causing significant environmental impacts. Current precipitation products for the TP often lack sufficient spatial or temporal resolution, hindering accurate characterization and mitigation of precipitation-related hazards. Here, we carry out 10-year (2012–2021, June to September) convection-permitting model simulations based on the Weather Research and Forecasting (WRF) model over the TP. We adopt four one-way nested domains, with the innermost domain providing a 1-km, hourly resolution over the southeastern TP. Model performance is enhanced using a spectral nudging scheme. Our simulated precipitation data are compared against in-situ rain gauge observations and state-of-the-art gridded precipitation products (i.e., HAR v2, ERA5-Land, and IMERG) over the southeastern TP. Our precipitation dataset demonstrates superior accuracy in mean absolute error, root mean square error, and bias, compared to the other three products. It effectively captures the key feature of the diurnal precipitation cycle and the non-monotonic dependence of precipitation on complex topography over the southeastern TP.

With the increasing frequency of extreme rainfall events, enhancing urban drainage systems’ regulation capacity is crucial for mitigating urban flooding. Existing studies primarily analyze infrastructure impacts on peak flow delay but often lack a systematic exploration of time-lag mechanisms. This study introduces the time-lag parameter, using the hysteresis curve of the water level–flow rate relationship to quantify drainage system dynamics. An SWMM-based drainage model was developed for the Rongdong area of Xiong’an New District to evaluate the independent roles of green, gray, and blue infrastructures in peak flow reduction and time-lag modulation. The results indicate that green infrastructure extends the horizontal width and reduces the vertical height of the hysteresis curve, prolonging time lag and making it effective for small-to-medium rainfall. Gray infrastructure enhances drainage efficiency by compressing the hysteresis curve horizontally and increasing its vertical height, facilitating rapid drainage but offering limited peak reduction. Blue infrastructure, by lowering outlet water levels, improves drainage capacity and reduces time lag, demonstrating adaptability across various rainfall scenarios. This study systematically quantifies the role of each infrastructure type in time-lag regulation and proposes a collaborative optimization strategy for urban drainage system design.