Disentangling model complexity in green roof hydrological analysis: A Bayesian perspective.

Area of Science:

• Environmental Engineering
• Hydrology
• Sustainable Urban Development

Background:

• Green roofs (GRs) are crucial for urban stormwater management.
• Accurate and efficient numerical models are needed to increase GR adoption.
• Optimal model complexity for GR hydrological behavior remains undetermined.

Purpose of the Study:

• To provide a Bayesian framework for comparing GR hydrological models.
• To assess the information content of various observational data.
• To determine the most appropriate model complexity for GR analysis.

Main Methods:

• Bayesian comparison of one conceptual and multiple Richards-based mechanistic GR models.
• Probabilistic assessment of information content using Kullback-Leibler divergence.
• Analysis of marginal likelihoods to evaluate model fit.

Main Results:

• Conceptual and unimodal van Genuchten-Mualem models are the most suitable parameterizations.
• Increased model complexity beyond these is not supported by available measurements.
• Measured volumetric water content provides more information than outflow or tracer data.

Conclusions:

• Simpler models offer a robust and computationally efficient approach for GR hydrological analysis.
• Volumetric water content measurements are most informative for reducing predictive uncertainty.
• Findings guide experimental design for improved GR modeling and management.