An Automatic Framework for Sensitivity Analysis and Parameter Optimization of the Ecosystem Model Biome-BGC
Abstract
The terrestrial ecosystem model Biome-BGC is an important tool for studying the ecosystem's matter and energy cycles. Currently the parameter values of the ecosystem model are determined according to statistics of observation data as well as experts' experience. The uncertainty of the parameterization scheme has significant influence to the model simulation results. In this research, we developed a python-based common framework that can automatically build Biome-BGC model with FLUXNET2015 data, carry out sensitivity analysis to identify important parameters and optimize important parameters with multi-objective optimization. The data of US-MMS station from FLUXNET2015 dataset was used to build the Biome-BGC model. Four model output variables: latent heat (LE), net ecosystem exchange (NEE), gross primary production (GPP) and ecosystem respiration (RECO), were involved in the sensitivity analysis and optimization. Three sensitivity analysis methods: MOAT (Morris One At a Time), MARS (Multivariate Adaptive Regression Splines), SPC (Sparse Polynomial Chaos) and three optimization methods: NSGA-II, NSGA-II and MO-ASMO (Multi-Objective Adaptive Surrogate-Modeling based Optimization) were compared. In this research, the NT (Number-Theory) based mixture design was integrated in the sampling processes to uniformly evaluate the parameter space with less sample points which have constrain a+b+c=1 (such as the proportion of liable, cellulose and lignin in leaf litter). The MO-ASMO algorithm was updated to integrate NSGA-III in order to solve many-objective optimization problems (typically has 3-4 or more objects). The automatic framework was tested with the data in US-MMS station It can be found that the results calculated by different sensitivity analysis methods are relatively consistent. For 3 or 4 objects problems, NSGA-III has significant advantage. The flux tower data and leaf C:N ration data were used together to validate the optimization result.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMH195.0003G
- Keywords:
-
- 0430 Computational methods and data processing;
- BIOGEOSCIENCES;
- 1805 Computational hydrology;
- HYDROLOGY;
- 1846 Model calibration;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY