A Multi-Institutional Big Data Collaboration to Estimate Long Term Terrestrial Net Carbon Uptake from Remote Sensing and Hydrological Modeling
Friday, 19 December 2014
Recent aircraft measurements from scattered records have shown long-term, global, seasonal photosynthetic CO2 uptake over land accelerating over the past 50 years. The successful launch of the sun-synchronous Orbiting Carbon Observatory 2 (OCO-2) on July 2, 2014 is expected to provide global, high spatial and spectral resolution datasets of vertical CO2 concentrations with surface spectral resolutions capable of yielding accurate CO2 flux profiles. It is unclear whether the biosphere will continue to act as a sink for anthropogenic CO2 loading of the atmosphere. Since current climate models with detailed terrestrial ecosystems are unable to simulate the observed increase in net ecosystem production (NEP), we will conduct assimilation studies with the derived CO2 fluxes in the GSFC Land Information System hydrological model to validate the generated NEP uptake. Further, we plan to use the OCO-2 CO2 concentrations to train a neural network to enable the calculation of long term trends from a decade of AIRS CO2 concentration data to produce regional NEP. To address this important Big Data science issue, a multi-institutional collaboration was formed to leverage their combined resources and the expertise of the researchers at the NASA GSFC, the Lamont Doherty Earth Observatory and UMBC. We will employ a high speed 10Gb network to connect the collaborating researchers and provide them with remote access to dedicated computational hybrid multicore resources at UMBC, as well as access to an archive containing more than a decade of readily accessible continuous daily gridded AIRS data and ten years of daily MODIS data for each September. The status of the following research efforts is planned to be presented; (i) acquisition and processing of the expected CO2 profile data from OCO-2 for two test sites, a low latitude region over the Amazon and a Boral forest at high latitude, (ii) initial impact of 3-D data assimilation of CO2 fluxes with the advanced Goddard LIS hydrological surface model, (iii) preliminary results in training AIRS CO2 data. In addition, early results of innovative exploration on quantum annealing optimization for 3-D data assimilation, image registration and a Hopfield neural network for training the AIRS CO2 spectral data through UMBC remote access to the D-Wave system in Vancouver, CA, will be introduced.