Calibration of a Landscape Evolution Model in an Alpine Region: on the value of in-situ observations and transferability

Abstract

Landscape evolution models (LEMs) are physics-based simulations that evaluate the influence of geomorphic processes on regional landscape development over longer time periods. These processes are represented in a simplified manner, and their parameterization cannot always be readily linked to real-world quantities. This makes their calibration and application challenging, especially as the calibrated parameter values are imposed to the entire model domain and extrapolated in time. This study introduces a data-driven calibration using historical observations from different stations in the Swiss Alps, aiming to automate and streamline the calibration process.

Using a Sobol sequence method allowed comprehensive exploration of the parameter space, identifying behavioral parameter sets that were then tested for their transferability across different spatial and temporal scales. Calibrating parameters on smaller catchments significantly reduced computational demands, streamlining the calibration process and making it more practical. Results from temporal transferability tests indicated consistent performance in discharge simulations, although sediment transport remained highly sensitive and variable over extended periods. Spatial transferability showed promising potential, suggesting that parameter sets calibrated on smaller catchments could be effectively applied to larger areas.

However, the study also highlighted significant challenges, particularly in simulating hydrological extremes and accurately capturing sediment transport processes. The use of a yearly timestep limited the model's ability to reflect seasonal dynamics, and the current model setup lacked sufficient data and complexity to properly simulate key processes such as glacier dynamics and temperature effects. These limitations underscore the need for further refinement in both data inputs and model structure to enhance the accuracy and robustness of LEMs.