Integration of flux and remote sensing data and modelling approaches to quantify water and carbon fluxes from regional to continental scale Group Members: Eva van Gorsel 1 , Gab Abramowitz 2 , Damian Barrett 3 , Jason Beringer 4 , Pep Canadell 1 , Brad Evansx 5 , Vanessa Haverd 1 , Christopher Pickett-Heaps 1 , Colin Prentice 5 , Stijn Hantson 6 , Alex Held 1 , Alfredo Huete 7 , Lindsay Hutley 8 , Dong Gill Kim 9 , Natascha Kljun 10 , Matt Paget 1 , Youngryel Ryu 1 , Ying-Ping Wang 1 , Marta Yebra 1 Project Objectives Terrestrial ecosystems annually sequester about one quarter of the anthropogenic CO2 emissions (Canadell & Raupach 2008). It is of great importance to understand the drivers of carbon uptake and release on timescales ranging from sub-diurnal to multi-annual. We explore the improvement in Land Surface Models (LSMs) that can be achieved through integration with remote sensing and flux tower observations by (i) using remote sensing data to quantify model parameters in space and time, and (ii) using footprint weighted remote sensing data and model output to benchmark the model with flux measurements and to explore the interannual variability in the carbon dynamics in continental Australia by analyzing 30 years of model output using the CSIRO Atmosphere Biosphere Land Exchange model (CABLE) and a much simpler model, based on light use efficiency, to analyse trends, covariance with climate drivers and indices (e.g. ENSO, IOD, Monsoon) and spatially attribute these anomalies to the underlying processes. Leaf Area Index (LAI) and the Fraction of Photosynthetically Active Radiation absorbed by vegetation (fPAR or FAPAR) are key biophysical variables controlling the exchanges of energy, carbon and water between the Earth surface and atmosphere (Ganguly et al., 2008). These variables can be generated from various satellite sensors at different spatial and temporal resolution. This leads to several issues in remote sensing science, including scaling and sensor related topics (e.g. choice of radiative transfer model, mixing of soil and vegetation types and spectral characteristics, spatial resolution, calibration, measurement geometry etc.). The group aimed to: · Explore the improvement in Land Surface Models (LSMs) that can be achieved through integration with remote sensing and flux tower observations by: (i) using remote sensing data to quantify variables used in LSMs in space and time; and (ii) using footprint weighted remote sensing data and model output to benchmark the model with flux measurements. · Explore the interannual variability in the carbon dynamics of continental Australia by analysing 30 years of model output using a research version of the CSIRO Atmosphere Biosphere Land Exchange model (CABLE) in the BIOS2 framework (see Notes) and a much simpler model, based on light use efficiency, to analyse trends, covariance with climate drivers and indices (e.g. ENSO, IOD, Monsoon) and spatially attribute these anomalies to the underlying processes. · Assess the uncertainty that arises from using various LAI/fPAR products when computing carbon and water fluxes. Future work will compare the satellite products with fractional cover datasets from TERN and SLATS (the Statewide Landcover and Trees Study of the Queensland government) transects. Methods CABLE/BIOS2 (Haverd et al., 2012) were run with local meteorology for LAI/fPAR values derived from various sensors and hence pixel sizes to: (i) assess which product leads to the best agreement between weighted model output and flux tower observation; and (ii) assess the uncertainty associated with using the AVHRR LAI/fPAR (which is available for 30+ years). For model validation we agreed to use data from two towers with >10 years of time series (Tumbarumba and Howard Springs) and data of 14 towers with 1-4 years time series, which provides good spatial and temporal coverage. The degradation of the modelling results using gridded meteorological drivers rather than local meteorological data in conjunction with the AVHRR data will be assessed elsewhere. A covariance analysis was conducted between climate indices, climate drivers and the time series of remote sensing and measured fluxes. We assess the temporal scale at which we observe highest linearization between the major drivers and the fluxes by determining where the change in the covariance matrix reaches a minimum. This is the temporal scale at which simpler models are expected to perform best. Two simple models are used for testing with MODIS and AVHRR data: BESS, a coupled-process model which estimates gross primary productivity and evapotranspiration (Ryu et al., 2011); and a simple model, the Ecosystem Production in Space and Time (EPiSaT) model which is based on light use efficiency and is still under development. After establishing the model uncertainties, we ran the models for the last 30 years and collected further GPP and ET products (NASA-AMES, Max Plank Institute, etc.). We analysed trends, covariance with climate drivers and indices (e.g. ENSO, IOD, Monsoon) and spatially attributed these anomalies to the underlying processes. Having multiple model products allows for an ‘ensemble-view’ for quantifying the effects of climate models on the fluxes. The sites used and data availability is shown in Table 1. Where gap filling is necessary, a standardized approach was applied (Reichstein et al., 2005).