Toward real-time GPP estimation using geostationary satellites

Atmospheric carbon dioxide uptake is a critical ecosystem service that is highly sensitive to extreme events with important consequences for the human and natural systems that rely on it. We often monitor carbon dioxide uptake via gross primary productivity (GPP) at the ecosystem scale using the eddy covariance method, typically over half-hourly intervals. These observations are often ‘upscaled’ to regional or global scales using satellite observations, typically on time scales of weeks to years. These relatively infrequent estimates are due in part to intermittent overpasses from the polar-orbiting satellites like Landsat and MODIS that are commonly used for GPP monitoring.

Geostationary satellites on the other hand have long observed the Earth’s surface and atmosphere on the order of minutes with a consistent viewing geometry. The new generation of imagers on many geostationary platforms like the Advanced Baseline Imager (ABI) onboard the GOES-R satellite series now have enhanced spectral resolution in the visible and near-infrared regions of the electromagnetic spectrum. This resolution is comparable to Landsat and MODIS and can now in principle estimate GPP using similar approaches. Therefore, geostationary satellites can now measure ecosystem carbon uptake in near-real time and radically improve our understanding of the interaction between the carbon cycle, climate, and extreme events.

Here, we demonstrate an approach to estimate GPP using geostationary satellite observations. After correcting for atmospheric attenuation and applying a bidirectional reflectance distribution function, a model that uses the near-infrared reflectance of vegetation (NIRv) as a saturating function of GOES-derived photosynthetic photon flux density (PPFD) with adjustment for atmospheric vapor pressure deficit outperformed other models for simulating the diurnal pattern of eddy-covariance estimated GPP in crop, grass, savanna, and forested ecosystems. This model also captured the seasonal trend in the diurnal centroid of maximum diurnal GPP as it responds to seasonal drought stress. We describe current progress in upscaling geostationary GPP including machine learning algorithms to maximize computational efficiency and predictive skill. We also describe approaches to respect the data sovereignty of Tribal Nations while working with Tribal land managers to understand the consequences of ecosystem disturbances on natural resources. International collaboration is required to provide near-real-time GPP estimates across the globe, and our approach is applicable to European satellite systems like SEVIRI and other geostationary satellite systems like Himawari-8&9.