1. Ecological Context of Climate Impacts on Fire: Wildland Fire Area Burned in the Western U.S. 1916-2003 Jeremy Littell with Don McKenzie and Dave Peterson

2. “Actually, the Forest Service where I work has sort of changed its mind about forest fires, and so I don’t say, ‘Only you can prevent forest fires’ anymore.” “We’re more of a faith-based Forest Service now. So we feel that if there is a forest fire, maybe there was supposed to be one. Maybe we should pray for it to stop. It’s not up to us to play fire fighter.” -Smokey the Bear, Prairie home Companion, 16 April 2005 95-97% of the area burned is burned by 3-5% of the fires, right?

3. Context: 20 th Century Area Burned Wildland-fire area burned: policy and resource management metric WFAB ecologically imperfect Are there useful, ecologically interesting relationships between climate and this data? Data compiled from multiple sources (NIFMID)

4. 20 th C. Fire and Climate in the West McKenzie et al. 2004. Conservation Biology Mechanism  response Westerling et al. 2003. BAMS. Statistical prediction

6. Scaling Responses and Drivers: Compromise Scale of Fire Data: –Annual State WFAB –Monthly 1° x 1° WFAB –To annual eco-regions Scale of Climate Data –Monthly HCN –Match to eco-regions –Seasonal for several climate divisions:

7. Objective: Ecologically-specific, Meso-scale Fire Climatology Goal: Develop ecosystem-specific diagnostic climate-fire relationships for western U.S. Strategy: Reconstruct wildland fire area burned for as many western U.S. eco-provinces as possible. Method: Evaluate climate-fire relationships for reconstructed area-burned by eco-province.

8. Eco-provinces

9. Gridded Eco-provinces

10. Reconstructing Eco-province Area Burned from State and Grid Data Scale fire datasets to compatible spatial and temporal resolution. Logarithmic regression models, training period is 1980-2000: Log (Cascade Mixed (gridcells))= log (WA + OR) Use modeled relationship to hind-cast Cascade Mixed for full 1916-2003 dataset.

13. Seasonal Climate: Aggregated HCN Climate Divisions Temperature, Precipitation, PDSI –Annual –Winter (ONDJFM) –“Growing Season”: (AMJJAS) –Summer: (JJAS) PCA Aggregation

14. 1916-2003 Eco-province Area Burned vs. Climate Log~Log linear approximation

15. Log Models Produce Unacceptable Errors

16. 20 th C. WFAB is Gamma Distributed

17. State Data is Gamma Distributed

18. Aggregated Eco-provinces Retain Similar Distribution

19. Reconstructing Eco-province Area Burned from State and Grid Data Scale fire datasets to compatible spatial and temporal resolution. Non-linear, gamma-specific regression models, training period is 1980-2000: Cascade Mixed (gridcells)= WA + OR, specify log link and σ 2 ~ μ 2 Use modeled relationship to hind-cast Cascade Mixed for full 1916-2003 dataset.

22. Reconstructed Distribution

23. Reconstruction Diagnostics

24. Observed vs. Predicted WFAB

25. * * 1-(residual deviance/null deviance)

26. 1980-2000 Reconstructions vs. Climate Reconstructed Forest-Dominated Ecoprovince Fire Time Series vs. climate variables Antecedent precipitation seems to be more important than temperature in forested systems during the model calibration period.

27. 1916-2003 Reconstructions vs. Climate Reconstructed Forest-Dominated Ecoprovince Fire Time Series vs. climate variables

28. 1980-2000 Reconstructions vs. Climate Reconstructed Grassland/Shrubland/Desert Dominated Eco-province Fire Time Series vs. climate variables As antecedent favor vegetation growth, fire area burned increases. This is consistent with fuel or fuel AND climate limited systems.

29. Reconstructed Grassland/Shrubland/Desert Dominated Eco-province Fire Time Series vs. climate variables 1916-2003 Reconstructions vs. Climate

30. Caveats Is the state-level data correctly normalized for reporting area in the early 20 th century? Is the proportion of area burned attributable to climate (vs. land use, vegetation dynamics, or fire exclusion) reasonably stationary?

31. Summary Eco-province wildland fire area burned is strongly influenced by climate between 1980 and 2000. Reconstructions of 1916-2003 eco-province area burned indicate that similar but weaker relationships occur for the entire period. Climate has exerted a detectable influence on 20 th century wildland fire regardless of other influences.

32. Conclusion Climate  fire impacts are ecologically- dependent, and vary because of combined influences of seasonality, ecosystem vegetation, and climate regime. Meso-scale, mechanism  response approaches are useful in evaluating these relationships.

33. Prospectus Cross-validation to improve confidence about uncertainty. Investigation of ocean/atmosphere controls on climate “supply” in extreme fire years. Attribution of unexplained residual deviance in eco-province/climate models Future extrapolation given future climate scenarios and constraints of eco-provinces

34. Acknowledgements Tony Westerling and Tom Swetnam