Sources of uncertainty in gross primary productivity simulated by light use efficiency models: Model structure, parameters, input data, and spatial resolution

摘要

Accurate estimation of gross primary productivity (GPP) is essential for understanding ecosystem function and global carbon cycling. However, there is still substantial uncertainty in the magnitude, spatial distribution, and temporal dynamics of GPP. Using light use efficiency (LUE) models, we conducted a comprehensive analysis of the uncertainty in GPP estimation resulting from various sources: model structure, model parameters, input data, and spatial resolution. We first evaluated the influences of model structures, namely the fraction of absorbed photosynthetically active radiation (FPAR), water scalar (W-S), and temperature scalar (T-S), on site-level GPP estimates. We then used the Sobol' sensitivity analysis to quantify the relative contributions of model input variables to the uncertainty in GPP. In addition, we used different land cover and meteorological datasets to examine the effects of input data and spatial resolution on the magnitude and spatiotemporal patterns of GPP. We found that the model structures affected not only model performance but also model parameters in a manner that differed with vegetation type and region. Thus, proper model structures and rigorous model parameterization and calibration should be adopted in GPP modeling. The Sobol' sensitivity analysis showed that the meteorological drivers including photosynthetically active radiation (PAR) and daily minimum temperature (TMIN) had larger contribution to the uncertainty in simulated GPP than did the surface reflectance-based indices including enhanced vegetation index (EVI) and normalized difference water index (NDWI). At the regional scale, different land cover datasets had the largest impacts on GPP simulations, especially in heterogeneous areas, followed by the scale effects from different spatial resolutions; changing meteorological datasets had the smallest effects. Therefore, more accurate and finer-resolution land cover maps and meteorological datasets are essential for more accurate GPP estimates. Our findings have implications for improving our understanding of the full uncertainty in carbon flux estimates and reducing the uncertainty in carbon cycle simulations.