Daily GPP of natural vegetation from new generation vegetation indices. Comparison with MOD17

Gross Primary Production (GPP), the amount of CO₂ that plants absorb due to photosynthesis, is the biggest carbon flux between biosphere and atmosphere. The current study investigates the estimation of daily GPP of natural vegetation from new generation vegetation indexes (VIs) and compares their performance to the production efficiency model of MODerate resolution Imaging Spectroradiometer (MODIS), MOD17. Two VIs are considered: the kernel version of Normalized Difference Vegetation Index (kNDVI), and the near infrared reflectance of vegetation (NIR_V).

kNDVI exploits the higher order relations between the reflectance in the NIR and red regions by defining NDVI in Hilbert spaces and using the radial basis function reproducing kernel. It uses a length-scale parameter (σ) that can be defined conveniently for a specific purpose. NIR_V is the product of NIR reflectance and NDVI. It represents the proportion of pixel reflectance attributable to the vegetation in the pixel.

VIs are calculated from MODIS daily continuous surface reflectance in red and NIR (FluxnetEO dataset version 2). This dataset offers quality checked and gap-filled daily MODIS surface reflectance observations during the 2000-2022 period centered in 647 eddy covariance (EC) sites located around the world. Different linear models are trained using VIs alone and combined with photosynthetically active radiation (PAR) measured at EC sites. Observations from 34 EC sites during the 2016-2020 period are used to optimize regression parameters and σ for three different biomes: grasslands, deciduous broadleaved forests, and evergreen needleleaved forests.

The daily GPP estimates are added to 8-day periods according to MOD17 frequency. The three GPP series are validated against EC observations and their results are compared. Using VIs alone, kNDVI achieved correlation R ϵ [0.79,0.87], relative mean bias error rMBE (%) ϵ [-9,6], and relative root mean squared error rRMSE (%) ϵ [52,60]; NIR_VR ϵ [0.79,0.87], rMBE (%) ϵ [-8,7], rRMSE (%) ϵ [52,60]. In combination with PAR: kNDVI R ϵ [0.81,0.88], rMBE (%) ϵ [-9,14], rRMSE (%) ϵ [52,58]; NIR_VR ϵ [0.81,0.88], rMBE (%) ϵ [-9,22], rRMSE (%) ϵ [51,61]. MOD17: R ϵ [0.41,0.70], rMBE (%) ϵ [-34,18], rRMSE (%) ϵ [35,56].