Join us at 11am this Wednesday 15th June for the latest installment of the EarthByte seminar series, featuring Sydney Uni’s own Matt Boyd! The seminar will be held over Zoom at the following link: https://uni-sydney.zoom.us/j/82071832536. For more details, see below:
Optimising and evaluating parameters in landscape evolution models
Landscape evolution models are routinely used to evaluate landscape responses to different forcing conditions (tectonics, climates, soil properties). Their calibration is challenging (1) because nonlinear relationships exist between the different physical components that are used to simulate sediment erosion, transport, & deposition, and (2) due to the scarcity of available datasets across geological time scales.
Here I present a new workflow with 2 example Badlands model sets based on a design of experiment library (DoEgen) capable of optimising and efficiently evaluating the parameters space of the landscape evolution model Badlands.
To better understand how the interaction of sediment supply and accommodation affects stratigraphy in rift basins I select key parameters to evaluate. In this case the strategy involves examining the parameter space around erosion, sediment transport and the rate of accommodation change. The optimal number of experiments with varying combinations of parameters are automatically generated and the Badlands experiments run.
The resulting experiments are then evaluated for their similarity to observations at well locations using two methods. One method searches for the closest matching total thickness amongst the experiment set. The other method compares proxies for depositional environment between experiment and observation from wells. In this case gamma ray logs are a useful proxy for water depth in the fluvial-lacustrine system under investigation and are compared with the lake level values obtained from the set of Badlands experiments.
The workflow automatically determines which combinations of parameters produce results most closely matching observed properties (thickness and stratigraphy) at well locations.
I find that the parameters that have the most obvious effect on stratigraphy and thickness are erodibility and m/n ratio. The rate of accommodation change and pitfilling parameters affect the experimental results to a lesser but still significant extent. I obtain a good calibration between experimental and observed stratigraphy and thicknesses.
These results demonstrate the usefulness of the design of experiment (DoE) approach in Badlands experiments and more broadly any landscape evolution models. Numerically relevant design of multiple alternative models produced for each experiment using DoE principles allows the relative effects of variations in experiment parameters to be determined.
