Systematic Bias and Model Independence in Land Surface Models
Primary Author: Abramowitz, Gab
Systematic Bias and Model Independence in Land Surface Models
Gab Abramowitz, Macquarie University/CSIRO Marine and Atmospheric Research
This presentation will discuss a neural network based technique for identifying and correcting systematic bias in land surface model(LSM) flux predictions. As a result of the correction, per time-step RMSE in latent heat, sensible heat and NEE is reduced by as much as 45%, translating to reductions of as much as 90% on annual time scales when tested against a range of Fluxnet sites. Three LSMs are considered. By manipulating the relationship between neural network training and testing sets, we show that LSM bias is relatively independent of vegetation characteristics, and further that the nature of bias is shared between some LSMs. This type of result has the potential to reveal which areas of model parameterisation are weak as well as to provide a metric for model independence. The implications for parameter estimation, model validation and model independence in ensemble simulations will be discussed.
For questions or comments regarding this website, please contact the Webmaster.
