Articles | Volume 10, issue 7
Biogeosciences, 10, 4607–4625, 2013
Biogeosciences, 10, 4607–4625, 2013

Research article 11 Jul 2013

Research article | 11 Jul 2013

Improving North American terrestrial CO2 flux diagnosis using spatial structure in land surface model residuals

T. W. Hilton1,*, K. J. Davis1, K. Keller2, and N. M. Urban2,** T. W. Hilton et al.
  • 1Department of Meteorology, Pennsylvania State University, University Park, PA, USA
  • 2Department of Geosciences, Pennsylvania State University, University Park, PA, USA
  • *now at: Department of Biology, University of New Mexico, Albuquerque, NM, USA
  • **now at: Energy Security Center, Los Alamos National Laboratory, Los Alamos, NM, USA

Abstract. We evaluate spatial structure in North American CO2 flux observations using a simple diagnostic land surface model. The vegetation photosynthesis respiration model (VPRM) calculates net ecosystem exchange (NEE) using locally observed temperature and photosynthetically active radiation (PAR) along with satellite-derived phenology and moisture. We use observed NEE from a group of 65 North American eddy covariance tower sites spanning North America to estimate VPRM parameters for these sites. We investigate spatial coherence in regional CO2 fluxes at several different time scales by using geostatistical methods to examine the spatial structure of model–data residuals. We find that persistent spatial structure does exist in the model–data residuals at a length scale of approximately 400 km (median 402 km, mean 712 km, standard deviation 931 km). This spatial structure defines a flux-tower-based VPRM residual covariance matrix. The residual covariance matrix is useful in constructing prior fluxes for atmospheric CO2 concentration inversion calculations, as well as for constructing a VPRM North American CO2 flux map optimized to eddy covariance observations. Finally (and secondarily), the estimated VPRM parameter values do not separate clearly by plant functional type (PFT). This calls into question whether PFTs can successfully partition ecosystems' fundamental ecological drivers when the viewing lens is a simple model.

Final-revised paper