Reducing Forecast Uncertainty of Ecosystem Changes in Climate Models by Analysis and Parallel Global Optimisation for Model Parameter Calibration

Climate change research has a linked system of models of the ocean, atmosphere and land. This talk, presented by Dr Yanyan Cheng and Prof Christine Shoemaker focuses on the land model used in climate change forecasting. As a main part of the Earth’s vegetation, tropical forests play a critical role in regulating carbon and water cycle dynamics both regionally and globally.  The climatic feedbacks from biological systems on earth surface, particularly from the vegetation, are a new focus in the Earth System models (ESMs) development, and these new vegetation models are called ecosystem demographic (ED) models. The ED models allow the authors to represent the vegetation component more explicitly in ESMs and advance the understanding of vegetation feedbacks to the climate. An overview of the components of the ESMs and the factors influencing interactions between vegetation and climate was briefly discussed.

The complex ED models are much more computationally expensive, which makes it difficult to calibrate the model parameters against observations. In this study, they propose a calibration framework that uses a parallel surrogate global optimisation algorithm (Xia & Shoemaker, 2021) to compute (at National Supercomputer Center) model parameter values for ED models. They applied the framework to simulate early and late successional tropical forests in Panama in an ED model that simulates coexistence for different tropical plant species. The authors compare the results with observed carbon and water cycle measurements concurrently so the analysis includes ecological and hydrologic impacts.

Hence, the authors show that by using improved optimisation methods, they are able to reduce the uncertainty in the ESM prediction by improving the calibration of model parameters to data.