A novel application of a surrogate based optimization model for CTSM in Alaska (EGU 2021)

Posted by Yifan Cheng on April 29, 2021

In the EGU 2021 vPICO presentation, I presented the initial work of an application of a surrogate based modeling framework for the state-of-the-science Community Terrestrial Systems Model (CTSM) in Alaska. This work is part of an interdisciplinary research project, i.e., Navigating the New Arctic, funded by National Science Foundation. The motivation of this parameter estimation work is to provide a historical run with high fidelity.


To reduce the computational cost, we performed an optimization in four representative basins and two were shown here for initial analysis, including Kuparuk and Tanana. CTSM is a complex, physically based land surface model that includes complex vegetation and canopy representation, a multi-layer snow model, as well as hydrology and frozen soil physics necessary for the representation of streamflow and permafrost. We updated the soil textures in CTSM using the high resolution SoilGrid dataset. We included hillslope hydrology to further improve representation of subgrid spatial variability. To select the parameters with high sensitivity, we will take advantage of an on-going effort, called the Perturbed Parameter Ensemble experiment. This experiment examines mean state response to perturbation of parameters. ASMO is an optimization method recently developed by Qingyun Duan’s group. Compared to the widely used shuffled complex evolution algorithm, ASMO is more computationally efficient.

Here I showed the initial results for the CTSM-ASMO framework with one flow-related objective. The model performance gets slightly better and it shows that the CTSM-ASMO framework does work. And in order to improve the performance, we will dig deeper in parameter selection and tuning the hyper parameters for the surrogate models.