Automatic Regionalization of Model Parameters for Hydrological Models

Abstract

Large-domain, spatially contiguous hydrological and land surface models are important tools for managing our water supplies. Hydrological information on the continental or global scale is needed to handle new emerging international and global water management challenges, which include topics like water allocation in international, national, and large river basins, operational flood forecasting services, global water security or the influence of climate extremes on water resources (Archfield et al., 2015). These applications are particularly challenging in areas without hydrologic measurements, which includes a majority of basins worldwide that are effectively ungauged (Hrachowitz et al., 2013). This results in the need for further development in large-domain hydrological modeling to simulate water fluxes and states in both gauged and ungauged basins in different climates in a spatially consistent manner (Rakovec et al., 2019).In 1982, Jim Dooge stated that "the parameterization of hydrologic processes to the grid-scale of general circulation models is a problem that has not been tackled, let alone solved" (Dooge, 1982) and shortly after that Leavesley et al. (1983) concluded that optimization of distributed parameters of hydrological models is an "ill-posed" problem due to the large number of degrees of freedom. Since then, model parameterization is still one of the major unsolved problems in hydrology (Blöschl et al., 2019). One way to potentially solve this problem is to relate hydrological model parameters/structures to landscape properties (e.g., K.

A Mass‐Conserving‐Perceptron for Machine‐Learning‐Based Modeling of Geoscientific Systems

“…Further, although “ parameter optimization ” is currently the main focus of model development using ML technologies, we reiterate our view (see Gharari et al., 2021) that the focus of model development should instead be shifting toward “ function space optimization. ” This points toward the potential for technologies such as symbolic regression (Udrescu & Tegmark, 2020) to be used for discovering plausible forms for the gating functions (Feigl et al., 2020; Klotz et al., 2017), with the further potential of being eventually able to regularize the model search to be consistent with physical‐conceptual information regarding hydro‐climatic context (Feigl et al., 2022).…”

Changes in the water retention of mountainous landscapes since the 1820s in the Austrian Alps

“…The weights in this work were estimated based on a frequently applied method from the literature (Becker et al, 2017;Vandecasteele et al, 2017;Lindén et al, 2021) and the applied optimization framework guarantees objectivity, transparency, and reproducibility, and some degree of comparability between different estimations of the WRI. Advances in the estimation of spatially distributed parameters in hydrological models, also using Machine Learning (e.g., Klotz et al, 2017;Feigl et al, 2020;Feigl et al, 2022) may be used in the future for calculating the weights of the WRI. For example, runoff data could be used for weight optimization.…”

Tradeoffs Between Temporal and Spatial Pattern Calibration and Their Impacts on Robustness and Transferability of Hydrologic Model Parameters to Ungauged Basins

“…As a results of increased availability of remotely sensed data sets combined with machine learning approaches and computational power, many gridded spatial products are now available (Belgiu & Drăguţ, 2016; Feigl et al., 2022). Despite the varying accuracy of both satellite data and machine learning approaches, these gridded data sets facilitate the spatial characterization of hydrologic variables and fluxes and enable spatial model evaluations.…”