1. Paul R. Moorcroft David Medvigy, Stephen Wofsy, J. William Munger, M. Dietze Harvard University Developing a predictive science of the biosphere

2. - we now have models that make predictions for the long-term responses of terrestrial ecosystems to climate change. - but are they predictive? carbon flux: land-air global mean temperature

3. - existing ‘big-leaf’ dynamic terrestrial biosphere models (DGVMs) are interesting, but largely unconstrained hypotheses for the effects of climate variability and change on terrestrial ecosystems. - models are fundamental to inference about the state of carbon cycle because the predictions of interest are at scales larger than those at which most measurements are made. atmospheric CO 2 meas. satellite observations (leaf phenology, soil moisture) Canopy CO 2 & H 2 O fluxes. forest inventories (vegetation dynamics) spatial scale 1m 2 1000km 2 100km 2 10km 2 1km 2 earth decades years months hours time scale - as a result, scaling is a key issue (Moorcroft 2006) Aircraft measurements of CO 2 & H 2 O fluxes

4. Ecosystem Demography Model (ED2) ha (~10 -2 km 2 ) (Moorcroft et al. 2001, Medvigy et al. 2006) of plant type i mortality growth water nitrogen carbon recruitment ~ 15 m leaf carbon fluxes evapo- transpir ation

5. carbon uptake (NEE tC ha -1 y -1 ) Harvard Forest LTER ecosystem measurements

6. - initialize with observed stand structure - model forced with climatology and radiation observed at Harvard Forest meteorological station. ED2 biosphere model Atmospheric Grid Cell ED-2 model fitting at Harvard Forest (42 o N, -72 o W) - 2 year model fit (1995 & 1996), in which model was constrained against: - hourly, monthly and yearly GPP and R total - hourly ET - above-ground growth & mortality of deciduous & coniferous trees

7. optimizedinitialobserved = optimization period Improved predictability at Harvard Forest: 10-yr simulations (1992-2001) Net Carbon Fluxes (NEP)

8. Improved predictability at Harvard Forest: 10-yr patterns of tree growth and mortality (1992-2001) observedinitialoptimized growth mortality = optimization period GPP respiration (r a + r h )

9. Improved predictability at Harvard Forest: 10-yr simulations (1992-2001) observedinitialoptimized = optimization period conifers hardwoods mortality growth

10. Vegetation model optimization: results model parameters are generally well- constrained: average coefficient of variation: 17% (= 95% confidence interval ) (-85, +160) Change in goodness of fit: 450 log-likelihood (  l) units (sig level:  l= 20)

11. Howland Forest Harvard Forest Howland Forest (45 o N, -68 o W) Howland forest Composition: growth observedinitialoptimized net carbon fluxes (NEP) (no changes in any of the model parameters)

12. Gross Primary Productivity (tC ha -1 mo -1 ) conifer basal area increment (tC ha -1 mo -1 ) hardwood basal area increment (tC ha -1 mo -1 ) Improved predictability at Howland Forest: 5-yr simulations (1996-2000) => model improvement is general, not site-specific

13. Regional Simulations - climate drivers : ECMWF reanalysis dataset - stand composition & harvesting rates: US Forest Service & Quebec forest inventory 1985 - 1995 - again, no change in any of the model parameters Harvard Forest

14. initial Regional decadal-scale dynamics of above-ground biomass growth (tC/ha/yr) observed optimized

15. Conclusions: Developing a predictive science of the biosphere structured biosphere models such as ED2 can be parameterized & tested against field measurements yielding a model with accurate: canopy-scale carbon & water fluxes tree-level growth & mortality dynamics (the processes that govern long- term vegetation change) capture observed regional scale variation in ecosystem dynamics without the need for site-specific parameters or tuning (scale accurately in space). capture short-term & long-term vegetation dynamics (scale accurately in time). Able to demonstrate that: shown that it is possible to develop terrestrial biosphere models that not only make predictions about the future of ecosystems but are also truly predictive.

16. optimization site Ameriflux site Future Directions: North American Carbon Plan (NACP): expanding to sub-continental scale.

17. Biosphere-atmosphere feedbacks Amazonia (Cox et al 2000) Amazonian deforestation predicted to change South American climate (Shukla et al 1990) Change in Annual Precipitation (mm) Santarem Flux tower (3 o S, -55 o W) Forest Inventory: Predicted collapse of the Amazon forests in response to rising CO 2

18. Collaborators: Steve Wofsy, Bill Munger, Roni Avissar, Bob Walko, D. Hollinger, Andrew Richardson Lab: Marco Albani, David Medvigy, Daniel Lipsitt, M. Dietze Acknowledgements References: Moorcroft et al. 2001. Ecological Monographs 74:557-586. Hurtt et al. 2002. PNAS 99:1389-1394. Albani & Moorcroft (2006) Global Change Biology 12:2370-2390 Moorcroft (2006) Trends in Ecology and Evolution 21:400-407 Medvigy et al. (2007) Global Change Biology (in review) Funding: National Science Foundation Department of Energy National Aeronautics and Space Administration

20. Soil decomposition model temperature sensitivity f(T) soil moisture sensitivity f(  ) relative decomposition rate  optimized initial 3-box biogeochemistry model (fast, structural & slow C pools)