Description of the research question I am exploring.

The broad question I am exploring is, “How will climate change affect fire regimes in the Pacific Northwest in the 21^st century?” or stated as an overarching hypothesis:

Over the 21^st century, projected changes in climate will cause changes in fire regimes in the Pacific Northwest by influencing vegetation quantity, composition, and fuel conditions.

I am exploring this question in the context of model vegetation and fire results from the MC2 dynamic global vegetation model (DGVM). MC2 is a gridded, process model with modules for biogeochemistry, fire, and biogeography. Inputs consist of climate and soil data. Outputs from the model include vegetation type, carbon fluxes and pools, hydrologic data, and values related to fire, including carbon consumed by fire and fraction of grid cell burned.

MC2’s current fire module computes fuel conditions within each grid cell. Fire occurrence is modeled when conditions exceed a set fuel condition threshold. An ignition source is always assumed to be present. This threshold-and-assumed-ignition algorithm has the potential to underestimate fire occurrence in areas that rarely or never meet the fuel condition threshold and to overestimate fire occurrence in areas that frequently exceed the fuel condition threshold. I am currently implementing a stochastic ignitions algorithm that allows the user to set an overall daily ignition probability and applies a Chapman Richards function to a fuel condition measure to determine probability of an ignition spreading into a fire.

I will be running the model with historical climate (1895 to 2010) and future climate (2011 to 2100) to produce potential vegetation results (i.e. land use not taken into consideration). Historical data are downscaled from PRISM data, and future data are downscaled from output data produced by the CCSM4 Climate model using the CMIP 5 representative concentration pathway (RCP) 8.5. The model will be run at a 2.5 arc minute resolution (approximately 4km x 4km cell size).

I will compare the output from the 20^th century to that of the 21^st century and characterize differences in fire regime spatially and temporally. This will be the first run of the MC2 with the new stochastic ignitions algorithm.

(I have added several references below related to what is discussed here.)

The dataset I will be analyzing

The dataset I will be analyzing will come from MC2 model runs described above. The extent of the dataset is from 42° to 49° latitude and from -124.75° to -111° longitude (from the southeast corner of Idaho west to the US coast and north to the Canadian border), comprising 169 x 331 spatial grid cells of size 2.5 x 2.5 arc minutes. Outputs are on an annual basis from 1895 through 2100. Water and barren land are mapped out of the dataset.

Outputs include variables for various carbon pools, fluxes, vegetation characteristics, and fire characteristics. Those I will be analyzing include carbon consumed by fire and fraction of cell burned. I will be summarizing the data over the time dimension to compute mean time between fires (essentially fire return interval, but over a shorter time period than might be appropriate for calculating a true fire return interval).

Hypotheses

• Vegetation, elevation, and climate will cause fire regimes to cluster spatially through influences on fuel quantity, composition, and condition.
• Projected increased temperature and change in precipitation patterns will cause fire to be more frequent and/or more severe through influences on fuel quantity, composition, and condition.
• Shifting climate characteristics will cause regions with similar fire regimes to shift in location due to changing fuel quantity, composition, and conditions.

Kinds of analyses

The first analysis I will do is a cluster analysis using mean time between fires, carbon consumed by fire, and fraction of cell burned. I will first summarize data over six time periods to produce six datasets: four 50-year periods (1901-1950, 1951-2000, 2001-2050, and 2051-2100), and two 100-year periods (1901-2000 and 2001-2100). Then I will run a cluster analysis (type to be determined) on each dataset.

Using two or more of the resulting clustered datasets I will explore the differences among clusters within each dataset and between datasets (likely using Euclidian distance between clusters).

I will map clustering results back onto the landscape in order to explore spatial patterns within each dataset and differences in spatial patterns between datasets. I will also compare the spatial pattern of clustering results to the spatial extents of EPA Level III ecoregions to see how well or poorly they align.

If time permits, I will do further analyses to characterize the relationship between vegetation type distribution, climate factors, and fire regime clusters.

Expected outcomes

I expect that cells with the same statistical cluster will be concentrated geographically, that for historical data, these concentrations will align closely with EPA Level III ecoregions, that cluster characteristics will be different between time periods, and that geographical groupings of clusters will shift generally northward and towards higher elevation somewhat between historical and future time periods.

From previous runs of the MC2 and preliminary observations of results from the runs for this project, I know that dominant vegetation type shifts from conifer to mixed forests west of the crest of the Cascade Mountains. Within this region, I expect a large shift in fire regime, with carbon consumed falling and mean time between fires decreasing over much of this region. In other regions, I expect general decreases in the mean time between fires due to warmer temperatures and locally drier summers. I also expect carbon consumed to generally remain constant or locally increase due to more favorable growing conditions.

Importance to science and resource management

Studies using DGVMs commonly produce map and graphic results showing extent and intensity of change over uni- or bidimensional spatiotemporal domains. This approach will provide more quantifiable differences using a multidimensional analysis. The ability to characterize fire regimes this way will allow for better model parameterization and validation, which in turn may lead to greater confidence in model projections.

Model results will provide projected changes across an ecologically and economically important region. Results will help resource managers understand and plan for potential change.

Level of experience

• Arc: Medium, a little rusty
• ModelBuilder and Python: Expert, especially with python.
• R: Medium, a little rusty

References

The Beginner’s Guide to Representative Concentration Pathways: http://www.skepticalscience.com/rcp.php

Bachelet, D., Ferschweiler, K., Sheehan, T.J., Sleeter, B., Zhu, Z., 2015. Projected carbon stocks in the conterminous US with land use and variable fire regimes. Global Change Biol., http://dx.doi.org/10.1111/gcb.13048

Daly, C., Halbleib, M., Smith, J.I., Gibson, W.P., Doggett, M.K., Taylor, G.H., et al., 2008. Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol. 28 (15), 2031–2064, http://dx.doi.org/10.1002/joc.1688.