An Atlas of Hydrological Model-Structure Preferences: Improving Model-Selection Guidelines through Automatic Model Structure Identification (HMP-Atlas)

• Applicant: Dr. Diana Spieler
• Subject Area: Hydrogeology, Hydrology, Limnology, Urban Water Management, Water Chemistry, Integrated Water Resources Management
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 582077272

Choosing an appropriate hydrological model for a specific modelling purpose, catchment, and data scenario is a critical step in the hydrologic modelling chain, with direct consequences for the reliability of modelling results and decision-making. Yet, despite numerous model intercomparison studies, the field lacks generalizable a priori guidance that links catchment characteristics to suitable model structures. This project aims to develop such guidance by applying and improving the Automatic Model Structure Identification (AMSI) approach across a large-scale cross-continental domain. AMSI automates the model selection process by searching through many model "building blocks" and assembling model combinations that best represent the observed data. It does so by simultaneously optimizing which processes and process equations are used in the model, along with the parameters associated with these equations. Implemented by coupling the modular modelling framework Raven with the mixed-integer optimization algorithm Dynamically Dimensioned Search (DDS), AMSI efficiently evaluates large sets of model structure hypotheses and parameter combinations to build conceptual models that meet user-defined evaluation criteria. This project extends AMSI with a multi-objective calibration strategy tailored to three different modelling purposes (flood, drought, and water-balance modelling) by using multiple calibration metrics and data sources to reduce equifinality and increase the reliability of the identified model structure preferences. Applied across ~3,000 catchments in Germany and North America, the project will produce the first data-driven atlas of conceptual model-structure preferences over a large geographical domain. The atlas will identify, for each of the three modelling purposes, relevant hydrological processes and the preferred process equations that yield hydrologically realistic representations of catchment behaviour. By statistically relating these preferences to catchment and climate traits, the project will uncover generalizable patterns that explain where and why particular structures or process equations work, and where they fail or require new formulations. These outputs will enable scientists and practitioners to assess model appropriateness without time-intensive testing, reduce subjectivity in model structure choice, and guide future model development by highlighting gaps and redundancies in current process equations.

• DFG Programme: Research Grants