Introducing Iterative Model Calibration (IMC) v1.0: a generalizable framework for numerical model calibration with a CAESAR-Lisflood case study

Abstract

Abstract. In geosciences, including hydrology and geomorphology, the reliance on numerical models necessitates the precise calibration of their parameters to effectively translate information from observed to unobserved settings. Traditional calibration techniques, however, are marked by poor generalizability, demanding significant manual labor for data preparation and the calibration process itself. Moreover, the utility of machine-learning-based and data-driven approaches is curtailed by the requirement for the numerical model to be differentiable for optimization purposes, which challenges their generalizability across different models. Furthermore, the potential of freely available geomorphological data remains underexploited in existing methodologies. In response to these challenges, we introduce a generalizable framework for calibrating numerical models, with a particular focus on geomorphological models, named Iterative Model Calibration (IMC). This approach efficiently identifies the optimal set of parameters for a given numerical model through a strategy based on a Gaussian neighborhood algorithm. Through experiments, we demonstrate the efficacy of IMC in calibrating the widely used landscape evolution model CAESAR-Lisflood (CL). The IMC process substantially improves the agreement between CL predictions and observed data (in the context of gully catchment landscape evolution), surpassing both uncalibrated and manual approaches.