Presentation on theme: "Using Flux Observations to Improve Land-Atmosphere Modelling: A One-Dimensional Field Study Robert Pipunic, Jeffrey Walker & Andrew Western The University."— Presentation transcript:
Using Flux Observations to Improve Land-Atmosphere Modelling: A One-Dimensional Field Study Robert Pipunic, Jeffrey Walker & Andrew Western The University of Melbourne Cathy Trudinger & Ying Ping Wang CSIRO Marine and Atmospheric Research Supported by an Australian Postgraduate Award Scholarship and University of Melbourne – CSIRO Collaborative Research Support Scheme
Synthetic Twin Experiments Pipunic et al., 2007. Remote Sensing of Environment, In Press.
Kyeamba Creek Experimental Site 3D sonic anemometer & open path gas analyser for LE & H, 3m above ground: 10Hz measurements, 30min averages recorded Barometric pressure sensor: 1 reading per hour Air temperature & relative humidity probe, 2m above ground: 30min averages recorded Wind direction and speed: 30min averages recorded Tipping rain gauge bucket: 30min totals recorded 4-way radiometer, incoming & outgoing shortwave & longwave radiation: 30min averages recorded
Below the Ground 30cm 60cm 90cm 8cm (Not to scale) CS615 Soil Moisture Probes: Measuring every 30 mins Soil heat flux plates: 30min averages recorded 5cm 10cm 2cm 20cm 50cm 100cm Soil temperature probes: Measuring every 30 mins
CSIRO Biosphere Model (CBM) / CABLE Long Wave Radiation Short Wave Radiation Precipitation LE H CO 2 G Snow (Not to scale) L1 L2 L5 L4 L3 L6 Wind Canopy model (Wang & Leuning, 1998): LE, H and CO 2 for a ‘sunlit’ and a ‘shaded’ leaf canopy; LE and H calculated from both vegetation and bare soil based on fraction of transmitted radiation through canopy. Six computational soil layers using the soil and snow scheme by Kowalczyk et al. (1994): Uniform properties for all layers; Individual volumetric moisture and temperature - moisture governed by Richard’s equation.
Ensemble Kalman Filter Time State Value EnKF Covariance
Ensemble Member Generation Perturbing meteorological variables: Random number generated at each time step in series, zero mean Random number generated once for each ensemble and applied to whole series, zero mean Turner et al., 2007. Remote Sensing of Environment, In Press.
Assimilation Over 1 Year Period (2005) LE+H assimilated on MODIS timescale – twice a day where SW radiation is >500Wm -2 (representing no cloud cover). Surface soil moisture on SMOS timescale – every 3 days.
Initial Conditions ♦ Observed Spin-up Using spin-up with best available parameters (1 January 2005)
Conclusions LE and H assimilation results are better than SM results for estimating LE and H, but slightly worse for soil moisture The land surface model used exhibits soil moisture and temperature biases when using standard parameters and forcing; this is likely to be typical of most NWP land models Temperature and moisture biases need to be accounted for using a bias-aware assimilation approach