Integrating a PhenoCam-derived vegetation index into a light use efficiency model to estimate daily gross primary production in a semi-arid grassland

Abstract

The accurate estimation of temporally-continuous gross primary production (GPP) is important for a mechanistic understanding of the global carbon budget, as well as the carbon exchange between land and atmosphere. Ground-based PhenoCams can provide near-surface observations of plant phenology with high temporal resolution and possess great potential for use in modeling the seasonal dynamics of GPP. However, due to the site-level empirical approaches for estimating the fraction of absorbed photosynthetically active radiation (fAPAR), a broad application of PhenoCams in GPP modeling has been restricted. In this study, the stage of vegetation phenology (Pscalar) is proposed, which is calculated from the excess green index (ExGI) derived from PhenoCam data. We integrate Pscalar with the enhanced vegetation index (EVI) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) in order to generate a daily time-series of the fAPAR (fAPARCAM), and then to estimate daily GPP (GPPCAM) with a light use efficiency model in a semi-arid grassland area from 2012 to 2014. Over the three continuous years, the daily fAPARCAM exhibited similar temporal behavior to the eddy covariance–measured GPP (GPPEC), and the overall determination coefficients (R2) were all > 0.81. GPPCAM agreed well with GPPEC, and these agreements were highly statistically significant (p < 0.01); R2 varied from 0.80 to 0.87, the relative error (RE) varied from -2.9% to 2.81%, and the root mean square error (RMSE) ranged from 0.83 to 0.98 gC/m2/d. GPPCAM was then resampled to 8-day temporal resolution (GPPCAM8d), and further evaluated by comparisons with MODIS GPP products (GPPMOD17) and vegetation photosynthesis model (VPM)–derived GPP (GPPVPM). Validation revealed that the variance explained by GPPCAM8d was still the greatest among these three GPP products. The RMSE and RE of GPPCAM8d were also lower than those of the other two GPP products. The explanatory power of predictors in GPP modeling was also explored; the fAPAR was found to be the most influential predictor, followed by photosynthetically active radiation (PAR). The contributions of the environmental stress indices of temperature and water (Tscalar and Wscalar, respectively) were less than that of PAR. These results highlight the potential for PhenoCam images in high temporal resolution GPP modeling. Our GPP modeling method will help reduce uncertainties by using PhenoCam images for monitoring the seasonal development of vegetation production.