95th ESA Annual Meeting (August 1 -- 6, 2010)

OOS 43-3 - Real time forecast of forest carbon dynamics using ensemble Kalman filtering method

Thursday, August 5, 2010: 8:40 AM
401-402, David L Lawrence Convention Center
Shenfeng Fei, Department of Botany and Microbiology, University of Oklahoma, Norman, OK, Zhongda Zhang, Department of Computer Science, University of Oklahoma, Norman, OK and Yiqi Luo, Department of Microbiology and Plant Biology, University of Oklahoma, Norman, OK
Background/Question/Methods   Successfully real time ecosystem forecasting would greatly facilitate natural resource management under ongoing global change. The FLUXNET is a worldwide network collecting carbon flux, water flux, and energy density data with very short time interval. These data sets provide us great opportunity to evaluate the possibility of real time ecosystem forecasting. In this study, we developed the ecosystem forecast based on the data assimilation using Ensemble Kalman Filter (EnKF). The observed fluxes of carbon dioxide and environmental driving factor data from an Ameriflux site in Harvard Forest, Massachusetts, USA, were assimilated into a biogeochemical model. EnKF had been proved to be an efficient and effective data assimilation method in ecosystem carbon dynamics studies. The model parameters were estimated simultaneously with state variables in EnKF by being combined together with the state variables to form a joint state vector. In order to evaluate the uncertainties caused by temporal evolution of model parameter values, the parameters were considered having temporal variation in the model which combining a kernel smoothing technique with EnKF. The short term forecasts of state variables such as NEE, ecosystem respiration, etc. that indicate forest carbon dynamics were made for 6 hours, 12 hours, 24 hours, till as long as 14 days, after the last observed data by using weather forecasts as driving variables. Long-term forecasts were made till half year by using mimic climate forecasts data, which were derived from re-sampling of the driving data from previous years. In order to evaluate the uncertainties caused by various sources, such as model structure, model parameter, observational data noise, and weather forecasts, etc., the forecasted state variables were stored to be re-analyzed in comparison with the observed data collected afterwards.

Results/Conclusions   This study demonstrated that data assimilation with EnKF can be successfully used to make short time and long time ecosystem carbon dynamics forecasting based on eddy flux data and ecosystem biogeochemical models. The parameters estimation was greatly improved by the data assimilation method and provides valuable information for other ecosystem modeling studies. The uncertainly analysis from this study provide the information for further improvements for the ecosystem forecasting.