| Synthesizing information for the environmental sciences |
| Potential from emerging tools | |
| Jim Clark | |
| Nicholas School of the Environment | |
| Department of Biology | |
| Department of Statistical Science |
| Example: the climate change puzzle |
| Uncertain climate predictions | |
| Complicated patterns in ecological data |
| Models of current climate |
| Precipitation (control runs) | ||
| Convective storms | ||
| Multiple moisture sources (Gulf, Atlantic) | ||
| Location and strength of: | ||
| Bermuda High | ||
| Nocturnal jet | ||
| Hurricane moisture in summer | ||
| Models of future climate |
| 2 x CO2 Precip | |
| Dry winter, summer |
| Models of future climate |
| hot |
| Models of response |
| SE aridity and savanna expansion |
| Models of response |
| Climate change impacts on species | ||
| Based on climate envelopes | ||
| Correlations with long-term averages | ||
| Plants respond to weather | ||
| Future climates will be novel | ||
| Ignore subgrid processes | ||
| Source of immigrants? |
| Source of immigrants? |
| Biodiversity responds to spatial heterogeneity |
| Subgrid processes: | ||
| Growth and survival respond to gaps | ||
| Seed has limited production and dispersal |
| Varies by species |
| What do ÔprocessÕ data say? |
| Residents vs potential invaders |
| Could coastal plain spp invade the Piedmont? | |
| Could Piedmont spp invade the S. Appalachians? | |
| Relative demographic performance of residents vs potential invaders? |
| Seedling germination |
| Seedling growth rates |
| Seedling survival |
| Process data climate correlations |
| No advantage to potential invaders for any demographic rate in any local habitat | |
| Complex | |
| WhatÕs missing? |
| Challenges |
| How will each species respond to changing climate, in the context of environmental variation? | |
| Differential responses with size and age | |
| Causal factors multiscale, interact | |
| Causal factors and responses partially observed. |
| Outline |
| Prediction framework | |
| Assimilating models and data |
| Outline |
| Prediction framework | |
| Assimilating models and data |
| Prediction framework |
| X,y data | |
| M model | |
| qM parameters | |
| XÕ scenario | |
| yÕ prediction | |
| Outline |
| Prediction framework | |
| Assimilating models and data |
| Our approach |
| Long-term demographic monitoring (1991-) | ||
| Experiments | ||
| Individuals to small plots to landscapes | ||
| Intervention designs | ||
| Manipulate environment | ||
| Introduce potential immigrants | ||
| Detailed environmental data | ||
| Hierarchical modeling to | ||
| Integrate all sources of information | ||
| Accommodate the unknowns | ||
| Long term, heterogeneous data |
| Mapped stands | ||
| All life history stages | ||
| Seed rain | ||
| Seed banks | ||
| Seedlings | ||
| Saplings | ||
| Mature trees | ||
| Interventions | ||
| Canopy gaps | ||
| Nutrient additions | ||
| Herbivore exclosures | ||
| Fire | ||
| Environmental monitoring | ||
| Canopy photos | ||
| Soil moisture | ||
| Temperature | ||
| Wireless sensor networks | ||
| Remote sensing | ||
| Heterogeneous data |
| Simulator for analysis/prediction |
| Missing the connections |
| Hierarchical Bayes model |
| Light availability: ground to remote data |
| Ground measurements |
| Remote sensing |
| Exposure to sky |
| Solar geometry models |
| Estimates combine three data types |
| What makes this possible? |
| Inference |
| State space model: Fixed and random individual effects |
| Inference |
| Maturation status |
| Inference |
| Seed prediction from estimated fecundity |
| Inference |
| Seed prediction from estimated fecundity |
| Estimates of hidden states |
| Sources of variability/uncertainty in fecundity |
| Inference |
| Semiparametric mortality |
| Mortality risk with size, growth rate |
| Allocation: among individuals |
| Many estimates |
| 27 parameters for growth, fecundity, and associated process and observation errors | |
| T fixed year effects | |
| 40 mortality risk estimates (34 growth rate classes, 6 diameter classes) | |
| 2n random effects (growth and fecundity) | |
| 5nT latent states (canopy area, diameter, fecundity, maturation status, mortality risk) | |
| n > 103 trees | |
| T = 16 years (1992 – 2007) | |
| Slide 45 |
| Slide 46 |
| Prediction framework |
| X,y data | |
| M model | |
| qM parameters | |
| XÕ scenario | |
| yÕ prediction | |
| The implications: Scalable Landscape Inference & Prediction (SLIP) |
| Prediction that incorporates sources and scales for knowns and unknowns | |
| How to be efficient? |
| SLIP Light Model |
| Light computation explicit each m2 | |
| Individual trees | |
| Graphics hardware based algorithm |
| Detail where it matters |
| SLIP Dispersal Algorithm |
| Forward simulation: individuals respond to landscape influences |
| Competitive exclusion |
| with random effects |
| The advantages of multilevel modeling |
| Fully exploit data through coherent assimilation | |
| Synthesis of all information | |
| Predictions follow directly from estimates |
| The challenges |
| Modeling: When is complexity warranted? | ||
| Abundant, heterogeneous data | ||
| Products important | ||
| Repeated application, e.g, spp, sites, intervals | ||
| Ex: software for broad use, dissemination; NWP | ||
| Doing the computation | ||
| Efficient algorithms, data structures | ||
| Collaborators |
| Lab participants: | ||
| Brian Beckage (Univ Vermont) | ||
| Jim Clark | ||
| Mike Dietze | ||
| Janneke HilleRisLambers (Univ Washington) | ||
| Ines Ibanez | ||
| Shannon LaDeau (Smithsonian Inst) | ||
| Allen McBride | ||
| Jason McLachlan (Notre Dame Univ) | ||
| Jackie Mohan (Harvard Univ) | ||
| Mike Wolosin | ||
| Pete Wykoff (Univ Minnesota) | ||