Three‐dimensional (3d) numerical models are state‐of‐the‐art for investigating complex hydrodynamic flow patterns in reservoirs and lakes. Such full‐complexity models are computationally demanding and their calibration is challenging regarding time, subjective decision‐making, and measurement data availability. In addition, physically unrealistic model assumptions or combinations of calibration parameters may remain undetected and lead to overfitting. In this study, we investigate if and how so‐called Bayesian calibration aids in characterizing faulty model setups driven by measurement data and calibration parameter combinations. Bayesian calibration builds on recent developments in machine learning and uses a Gaussian process emulator as a surrogate model, which runs considerably faster than a 3d numerical model. We Bayesian‐calibrate a Delft3D‐FLOW model of a pump‐storage reservoir as a function of the background horizontal eddy viscosity and diffusivity, and initial water temperature profile. We consider three scenarios with varying degrees of faulty assumptions and different uses of flow velocity and water temperature measurements. One of the scenarios forces completely unrealistic, rapid lake stratification and still yields similarly good calibration accuracy as more correct scenarios regarding global statistics, such as the root‐mean‐square error. An uncertainty assessment resulting from the Bayesian calibration indicates that the completely unrealistic scenario forces fast lake stratification through highly uncertain mixing‐related model parameters. Thus, Bayesian calibration describes the quality of calibration and correctness of model assumptions through geometric characteristics of posterior distributions. For instance, most likely calibration parameter values (posterior distribution maxima) at the calibration range limit or with widespread uncertainty characterize poor model assumptions and calibration.

This study proposes a method for downscaling the spatial resolution of daily satellite-based precipitation estimates (SPEs) from 10 km to 1 km. The method deliberates a set of variables that have close relationships with daily precipitation events in a Random Forest (RF) regression model. The considered variables include cloud optical thickness (COT), cloud effective radius (CER) an cloud water path (CWP), derived from MODIS, along with maximum and minimum temperature (Tx, Tn), derived from CHIRTS. Additionally, topographic features derived from ALOS-DEM are also investigated to improve the downscaling procedure. The approach consists of two main steps: firstly, the RF model training at the native 10 km spatial resolution of the studied SPEs (i.e., IMERG) using rain gauge observations as targets; secondly, the application of the trained RF model at a 1 km spatial resolution to downscale IMERG from 10 km to 1 km over a one-year period. To assess the reliability of the method, the RF model outcomes were compared with the rain gauge records not considered in the RF model training. Before the downscaling process, the CC, MAE and RMSE metrics were 0.32, 1.16 mm and 6.60 mm, respectively, and improved to 0.48, 0.99 mm and 4.68 mm after the downscaling process. This corresponds to improvements of 50%, 15% and 29%, respectively. Therefore, the method not only improves the spatial resolution of IMERG, but also its accuracy.

Physics-based hydrological models frequently estimate subsurface fluxes and storage behaviour in catchments from limited observations. As a result, simulations rely on oversimplified, static descriptors of subsurface processes. To address those limitations, data-driven approaches emerge as an alternative; nevertheless, many of these methods either rely on restrictive assumptions -such as discharge depends solely on storage derived from recession periods- or represent the internal state of the catchments implicitly, without an interpretable characterisation of the storage-discharge dynamics. Here, we introduce a data-driven framework for rainfall-runoff modelling that represents catchments as non-autonomous dynamical systems using a modulated discharge-storage sensitivity function. The approach implements the recession-based sensitivity function proposed by Kirchner (2009), which characterises the baseline drainage behaviour of groundwater-dominated catchments. In our formulation, the derived recession-based function serves as a limiting reference constraining a dynamic storage-discharge sensitivity function that is continuously modulated by net atmospheric forcing through an explicit state-forcing relationship. As a result, the storage-discharge relationship varies with different hydro-meteorological conditions and returns to the recession-based formulation when atmospheric forcings are negligible relative to the discharge. Our framework accounts for changes in the dynamical structure of watersheds during rising and recession periods without requiring calibration parameters and is primarily applicable to catchments where discharge is controlled by their storage-state dynamics. Initial tests show that our framework captures forcing-dependent variations in storage-discharge sensitivity functions and provides additional diagnostic insight into catchment behaviour during rising limbs and recession. Ongoing work evaluates the robustness of different hydro-climatic settings and explores the method’s potential to characterise storage-forcing interactions in groundwater-dominated catchments.