Snow density is an important measure in hydrology used to convert snow depth to the snow water equivalent (SWE). A model developed by Sturm, Tara and Liston predicts the snow density by using snow depth, the snow age and a snow class defined by the location. In this work this model is extended to include location and seasonal weather-specific variables. The model is named Weather Snow Density Model (Weather SDM). A Bayesian framework is chosen, and the model is fitted to and tested for 4,040 Norwegian snow depth and densities measurements between 1998 and 2011. The final model improved the snow density predictions for the Norwegian data compared to the model of Sturm by up to 50%. Further, the Weather SDM is extended to utilize local year-specific snow density observations (Weather&ObsDensity SDM). This reduced the prediction error an additional 16%, indicating a significant improvement when utilizing information provided by annual snow density measurements.

Due to the changing climate, flash floods have been increasing recently and are expected to further increase in the future. Flash floods caused by heavy rainfall with snowmelt contribution due to sudden rises in temperature or rain-on-snow events have become common in autumn and winter in Norway. These events have caused widespread damage, closure of roads and bridges, and landslides, leading to evacuations in the affected areas. Hence, it is important to analyze such events. In this study, the rain-on-grid technique in the TELEMAC-2D hydrodynamic model was used for runoff modelling and routing using input of snowmelt, and precipitation partitioned on snow and rain was calculated via the hydrological model HBV. The results show the importance of including snowmelt for distributed runoff generation and how the rain-on-grid technique enables extracting flow hydrographs anywhere in the catchment. It is also possible to extract the flow velocities and water depth at each time step, revealing the critical locations in the catchment in terms of flooding and shear stresses. The rain-on-grid model works particularly well for single peak events, but the results indicate the need for a time-varying curve number for multiple peak flood events or the implementation of another infiltration model.

Flash floods can cause great geomorphological changes in ephemeral fluvial systems and result in particularly severe damages for the unprepared population exposed to it. The flash flood in the Storelva river in Utvik (western Norway) on 24 July 2017 was witnessed and documented. This study assessed the causes and effects of the 2017 flood and provides valuable information for the calibration and validation of future modelling studies. The flooded area at peak discharge, maximum wetted and dry areas during the entire event, critical points and main flow paths were reconstructed using on-site and post-event (i) visual documentation, such as photographs and videos, and (ii) aerial surveying, such as orthophotographs and laser scanning, of the lowermost reach. The steep longitudinal slope together with the loose material forming the valley and riverbed contributed to a large amount of sediment transport during this extreme event. Steep rivers such as the Storelva river have very short response times to extreme hydrologic conditions, which calls for exhaustive monitoring and data collection in case of future events, as well as modelling tools that can emulate the hydro-morphodynamics observed during events such as the 2017 flash flood.

This study presents results of observations and analysis of the flood event in Utvik on 24 July 2017. Observations during and after the event, hydrau

Accurate yet interpretable inflow forecasts are essential for hydropower operation. This study introduces modular Hydrologiska Byråns Vattenbalansavdelning-deep learning (mHBV-DL), a physics-guided hybrid deep learning framework that integrates the HBV model structure with neural networks, exposing intermediate hydrological states and fluxes alongside inflow predictions to enhance physical explainability. To systematically evaluate physics-DL integration strategies, four model configurations are constructed and compared, balancing learning flexibility with physical consistency. The best-performing mHBV-DL configuration achieved a Nash–Sutcliffe efficiency coefficient of 0.833, outperforming the standalone HBV model and remaining competitive with the architecture-matched black-box DL baselines. Its intermediate outputs also reflected expected hydrological patterns. The modular architecture and intermediate state and flux tracking enable systematic analysis of seasonal performance and model limitations. These findings establish mHBV-DL as an effective framework for combining predictive performance with physical explainability in catchment inflow modelling and provide a basis for future lead-time-based forecasting applications.