Understanding the Information Content in the Hierarchy of Model Development Decisions: Learning From Data

Process-based hydrological models seek to represent the dominant processes in a catchment. However, due unavoidable incompleteness of knowledge, construction “fidelius” process-based depends largely on expert judgment. We present systematic approach that treats as hierarchical assemblages hypotheses (conservation principles, system architecture, process parameterization equations, and parameter specification), which enables investigating how hierarchy model development decisions impacts fidelity. Each step provides information progressively changes our uncertainty (increases, decreases, or alters) regarding input-state-output behavior system. Following principle maximum entropy, we introduce concept “minimally restrictive equations—MR-PPEs,” us enhance flexibility with can be represented, thereby investigate important role architectural hypothesis (discretization into subsystem elements) plays determining behavior. illustrate explore these concepts synthetic real-data studies, using constructed from simple generic buckets building blocks, paving way for more-detailed investigations sophisticated models. also discuss proposed MR-PPEs bridge gap between current modeling machine learning. Finally, suggest need calibration evolve search over “parameter spaces” “function spaces.”