Pixel-level parameter optimization of a terrestrial biosphere model for improving estimation of carbon fluxes with an efficient model–data fusion method and satellite-derived LAI and GPP data

[摘要] Inaccurate parameter estimation is a significant sourceof uncertainty in complex terrestrial biosphere models. Model parameters mayhave large spatial variability, even within a vegetation type. Modeluncertainty from parameters can be significantly reduced by model–datafusion (MDF), which, however, is difficult to implement over a large regionwith traditional methods due to the high computational cost. This studyproposed a hybrid modeling approach that couples a terrestrial biospheremodel with a data-driven machine learning method, which is able to considerboth satellite information and the physical mechanisms. We developed atwo-step framework to estimate the essential parameters of the revisedIntegrated Biosphere Simulator (IBIS) pixel by pixel using thesatellite-derived leaf area index (LAI) and gross primary productivity (GPP)products as “true values.” The first step was to estimate the optimalparameters for each sample using a modified adaptive surrogate modelingalgorithm (MASM). We applied the Gaussian process regression algorithm (GPR)as a surrogate model to learn the relationship between model parameters anderrors. In our second step, we built an extreme gradient boosting (XGBoost)model between the optimized parameters and local environmental variables.The trained XGBoost model was then used to predict optimal parametersspatially across the deciduous forests in the eastern United States. Theresults showed that the parameters were highly variable spatially and quitedifferent from the default values over forests, and the simulation errors ofthe GPP and LAI could be markedly reduced with the optimized parameters. Theeffectiveness of the optimized model in estimating GPP, ecosystemrespiration (ER), and net ecosystem exchange (NEE) were also tested throughsite validation. The optimized model reduced the root mean square error(RMSE) from 7.03 to 6.22 gC m −2  d −1 for GPP, 2.65 to 2.11 gC m −2  d −1 for ER, and 4.45 to 4.38 gC m −2  d −1 for NEE.The mean annual GPP, ER, and NEE of the region from 2000 to 2019 were 5.79,4.60, and −1.19  Pg yr −1 , respectively. The strategy used in this studyrequires only a few hundred model runs to calibrate regional parameters andis readily applicable to other complex terrestrial biosphere models withdifferent spatial resolutions. Our study also emphasizes the necessity ofpixel-level parameter calibration and the value of remote sensing productsfor per-pixel parameter optimization.

[学科分类] 土木及结构工程学