1. Data assimilation in land surface schemes Mathew Williams University of Edinburgh

2. Global forecast capability Key processes Challenges for the JULES team Initial conditions Parameters Errors EO & sensor networks Continual testing & development JULES

3. The philosophy of data assimilation Observations and models contain useful information about the target system But ALL observations and models are subject to error - and may be subject to bias Models and observations can be combined to optimise information

4. Improving estimates of land surface process MODELS OBSERVATIONS FUSION ANALYSIS MODELS + Capable of interpolation & forecasts ─ Subjective & inaccurate? OBSERVATIONS + Clear confidence limits ─ Incomplete, patchy ─ Net fluxes ANALYSIS + Complete + Clear confidence limits + Capable of forecasts + Data consistency

5. Approaches to data assimilation Sequential (predictor-corrector)

6. The Kalman Filter MODEL AtAt F t+1 F´ t+1 OPERATOR A t+1 D t+1 Assimilation Initial stateForecast Observations Predictions Analysis P Ensemble Kalman Filter Drivers

7. F t+1 AtAt AtAt The Ensemble Kalman Filter MODEL AtAt F t+1 OPERATOR F´ t+1 A t+1 D t+1 Assimilation Initial stateForecast Observations Predictions Analysis P Drivers

8. Observations – Ponderosa Pine, OR (Bev Law) Flux tower (2000-2) Sap flow Soil/stem/leaf respiration LAI, stem, root biomass Litter fall measurements

9. GPP C root C wood C foliage C litter C SOM/CWD RaRa AfAf ArAr AwAw LfLf LrLr LwLw RhRh D DALEC

10. Time (days since 1 Jan 2000) Williams et al (2005)

11. Time (days since 1 Jan 2000) Williams et al (2005)  = observation — = mean analysis | = SD of the analysis

12. Time (days since 1 Jan 2000) Williams et al (2005)

13. Time (days since 1 Jan 2000) Williams et al (2005) = observation — = mean analysis | = SD of the analysis

14. Data brings confidence Williams et al (2005)  = observation — = mean analysis | = SD of the analysis

15. Approaches to data assimilation Sequential (predictor-corrector) Inversion techniques – Monte Carlo – Adjoint

16. Monte Carlo Inversion MODEL F F´F´ OPERATOR Forecasts Observations Predictions P Drivers Initial J D Cost function

19. REgional Flux Estimation eXperiment (REFLEX) To compare the strengths and weaknesses of various MDF/DA techniques To quantify errors and biases introduced when extrapolating fluxes www.carbonfusion.org

20. REgional Flux Estimation eXperiment (REFLEX) FluxNet data MODIS MDF Full analysis Model parameters DALEC model Training Runs Deciduous forest sites Coniferous forest sites Assimilation Output

21. REgional Flux Estimation eXperiment (REFLEX) FluxNet data MODIS MDF Full analysis Model parameters DALEC model Testing site forecasts with limited EO data MDF MODIS Analysis FluxNet data testing Assimilation

22. Scaling

23. A multiscale approach www.abacus-ipy.org

24. Process models -upscaling EO data: -landcover -phenology Flux towers -processes -parameters Geostats -Spatial drivers -Uncertainty Tall tower /aircraft -Check on upscaling -Inversions

25. Linking fluxes to atmospheric [CO 2 ] Figure by Paul Parrish, data from J Moncrieff & J Grace

26. The Orbiting Carbon Observatory (OCO) Approach: Collect spatially resolved, high resolution spectroscopic observations of CO 2 and O 2 absorption in reflected sunlight Use these data to resolve spatial and temporal variations in the column averaged CO2 dry air mole fraction, X CO2 over the sunlit hemisphere Employ independent calibration and validation approaches to produce X CO2 estimates with random errors and biases no larger than 1 - 2 ppm (0.3 - 0.5%) on regional scales at monthly intervals OCO will acquire the space-based data needed to identify CO 2 sources and sinks and quantify their variability over the seasonal cycle Source: David Crisp, JPL

27. Spanning measurement scales OCO will make precise global measurements of X CO2 over the range of scales needed to monitor CO 2 fluxes on regional to continental scales. Spatial Scale (km) 1 2 3 4 5 6 CO 2 Error (ppm) 1 10 100 100010000 OCO Flask Site Aqua AIRS Aircraft 0 Flux Tower Globalview Network NOAA TOVS ENVISAT SCIAMACHY Tall tower Source: David Crisp, JPL

28. A strategy for JULES? LOCAL: DA for local parameter PDFs, process testing, C-water interactions, full state descriptions. FluxNet, IPSL, NCAR, ACCESS. REGIONAL: upscaling, coupling to/inverting atmospheric data/models. CarboEurope, ABACUS. GLOBAL: Global assimilation with optical, CO 2, water, temperature remote sensing, flasks. NCEO & CCDAS.

29. THANKS!

30. Tim Hill, PhD thesis Monte Carlo Inversion