[p2] Ideas for new applications


The role of background and observation error correlations in improving a model forecast of forest carbon balance using four-dimensional variational data assimilation

Nancy Nichols (University of Reading), Ewan Pinnington (University of Reading), Eric Casella (Forest Research UK), Sarah Dance (University of Reading), Amos Lawless (University of Reading), James Morison (Forest Research UK), Matthew Wilkinson (Forest Research UK), Tristan Quaife (University of Reading)


Forest ecosystems play an important role in sequestering human emitted carbon-dioxide from the atmosphere and therefore understanding their response to climate change is of great importance. In models of forest ecosystems the use of variational data assimilation has been limited, with the background and observation errors treated as independent and uncorrelated. Here we implement a 4D-Var scheme for joint parameter and state estimation in a simple model of carbon balance and assimilate daily observations of Net Ecosystem CO2 Exchange (NEE) taken at the Alice Holt forest in Hampshire, UK. We derive novel techniques for including correlations in the background error statistics using a set of previously postulated dynamical constraints and for including time correlations in the observation error statistics using a Gaussian correlation function. In our experiments we compare the results using our new estimated error covariance matrices with those using diagonal error covariance matrices. We find that using the new correlated error correlation matrices reduces the root mean square error in the 14 year forecast of daily NEE by 44%.