Assimilating Multiple Data Types in the Community Land Model (CLM) for Deciduous Forests in North America

Monday, 15 December 2014
Francesc Montane1, Andrew M Fox2, Timothy J Hoar3, Avelino F Arellano1, Yao Liu1, Gabriel Moreno1, Tristan L Quaife4, Andrew D Richardson5, Valerie Trouet1, M Ross Alexander1, Min Chen5, David Y Hollinger6 and David J Moore1, (1)University of Arizona, Tucson, AZ, United States, (2)NEON, Boulder, CO, United States, (3)Natl Ctr Atmospheric Res, Boulder, CO, United States, (4)University of Reading, Reading, United Kingdom, (5)Harvard University, Cambridge, MA, United States, (6)USDA Forest Service, Durham, NH, United States
Networks of eddy covariance towers like AmeriFlux provide the infrastructure necessary to study relationships between ecosystem processes and environmental forcing across a range of spatial and temporal scales. Recent syntheses of comparisons between observations from eddy covariance tower sites in North America and output from several Land Surface Models showed that the characterization of phenology was not accurate in most of the models. In order to improve phenological models, a continental-scale phenological observatory, the PhenoCam network, provides high-frequency observations of vegetation phenology, which can be used to derive a greenness index, the green chromatic coordinate (gcc). In this study we run the Community Land Model (CLM4.5) for 10 deciduous forests sites in North America, included in the AmeriFlux and PhenoCam networks, to assimilate multiple data types including one of the key variables in most ecosystem models, fPAR, the radiometric equivalent of Leaf Area Index (LAI). fPAR characterizes vegetation canopy function and energy absorption capacity and therefore it is important for estimating canopy photosynthesis. We use fPAR data from Moderate Resolution Imaging Spectroradiometer (MODIS), with a pixel resolution of 1 km x 1 km and a temporal resolution of 8 days. Data is assimilated in CLM with an Ensemble Kalman Filter, a sequential data assimilation technique, within the Data Assimilation Research Testbed (DART). In our study, we also compare observations available for Harvard Forest (LAI, NEE and gcc) with model output. The CLM output for LAI and NEE is sometimes located out of the observation space delimited by LAI and NEE measurements for Harvard Forest. After assimilating data, we compare observations and mean CLM model output from all the sites for a free run, an assimilation run and an assimilation run with inflation. We investigate the impact of assimilating these observations and the resultant model state updates on ecosystem carbon flux predictions.