Drivers of burn severity in the northern Cascade Range, Washington, USA

Master's Thesis Abstract by C. Alina Cansler (2011)

Chapter 1: Remotely sensed indices of burn severity are now commonly used by researchers and fire managers to assess post-fire effects, but their relationship to field measurements of burn severity has not been evaluated in all ecosystems. In order to develop a geospatial atlas of burn severity for a 25 year period (1984-2008) in the northern Cascade Range, of Washington, USA, the accuracy of two remotely sensed indices of burn severity, the differenced Normalized Burn Ratio (dNBR) and the Relative differenced Normalized Burn Ratio (RdNBR), was assessed using field data from 639 plots located across four fires. In the process of comparing the two indexes, two methods of assessing field-based burn severity were compared: the Composite Burn Index (CBI) (Key and Benson 2006) and a newer version of the CBI that weighs severity scores by percentage cover (De Santis and Chuvieco 2009). The new weighted version of the CBI performed worse than the original unweighted version. Three methods of sub-sampling dNBR and RdNBR pixel values at field plot locations were compared, and bilinear interpolation, which uses the four nearest neighbors to determine pixel values, performed best. Four different statistical models of the relationship between field-based and remotely sensed burn severity were evaluated. The relative strengths of the dNBR and RdNBR indices were evaluated based on the R2 and the categorization accuracy of each model, and the best-fit model for each index was used to develop classification thresholds for categorical burn-severity images. The dNBR model had a higher R2 than the RdNBR model (R2 = 0.50 and R2 = 0.47, 1 respectively), but categorization based on the dNBR model (overall accuracy =59%, Kappa= 0.36) was lower than the RdNBR model (overall accuracy =59%, Kappa= 0.36). Both dNBR and RdNBR performed similarly and would be suitable for producing classified burn severity images in the northern Cascade Range, but RdNBR performed better and was used as the basis for categorical images of burn severity for all 125 fires in the northern Cascade Range burn-severity atlas. The slightly better performance of RdNBR falls in between the results of Miller and Thode (2007), in which RdNBR performed better than dNBR in the Sierra Nevada, and the results of Soverel et al. (2010), in which dNBR had higher classification accuracy than RdNBR in the Canadian boreal forest and the Canadian Rockies, possibly because variable pre-fire reflectance does not present as much of a challenge in the northern Cascade Range as in the Sierra Nevada.

Chapter 2: Quantifying attributes of mixed-severity fire regimes is challenging due to their high temporal and spatial variability. Traditionally, fire regimes have been described in terms of their predominant severity, frequency, and fire type. Increased availability of spatially explicit data and analysis tools and the use of natural fire regime characteristics as benchmarks for land management have led researchers and managers to take a more nuanced view of fire regimes. This view encompasses additional fire regime attributes, such as the seasonality, size, and spatial complexity of fires. This research quantifies patterns of severity and spatial pattern in the northern Cascade Range, WA, USA, compares variation in severity and spatial pattern within the larger study area, and examines how severity and spatial pattern vary with fire size. Geospatial fire occurrence records and a remotely sensed index of burn severity (Relative differenced Normalized Burn Ratio), were used to map fire severity for all fires greater than 10 ha (n = 125) that occurred during a 25 year period (1984-2008) in the northern Cascade Range, Washington, USA. Patterns of fire size, burn severity, patch size, the size hierarchy of the patch distribution, patch interior (area > 90 m from edge of patch), and spatial complexity were quantified for each fire. Results were summarized for four ecological subsections within the study area. Linear regression was used to test the relationship between fire size and burn severity, patch size, patch interior, and spatial complexity. ANCOVA was used to test if these relationships differed among subsections. The low elevation mixed-conifer forest in the eastern Cascades burned with less high severity, more moderate severity, smaller patches, and greater spatial complexity than the other 37 subsections. The north-east section of the study area, which has vegetation more similar to the northern Rocky Mountains, was dominated by large high-severity fires with very large patch sizes. Fires in the high-elevation forests along the Cascade Crest were smaller and were relatively evenly distributed between high severity, moderate severity, and unchanged areas. Burn severity, patch size, and the inequality of the patch size distribution (as measured by the Gini coefficient) increased with fire size. Severity, patch size, and patch interior increase more quickly with fire size in north-east section of the study area. This study found positive relationships between fire size and the proportion of area burned at high severity, and between fire size and the size and core area of high severity patches. These results suggest that under future climate fire may not only be larger in size but also may produce more high-severity fire with a different, more aggregated burn severity pattern. Results also indicate that the strength of the relationships between fire size and severity and fire size and spatial pattern are dependent on the local ecological context; differences in fire severity and spatial pattern within the northern Cascade Range reflect differences between each subsection’s vegetation, fuels, and topography. Future research focus on variation in burn severity, as opposed to fire size or area burned, will further elucidate the influence of fire on landscape spatial pattern.

Chapter 3: I examined the influence of annual climate and topographical complexity on the occurrence, size, severity, and within-fire severity pattern of fires in the northern Cascade Range of Washington, USA. Landsat Thematic Mapper (LTM) data were used to quantify fire severity for all fires greater than 10 ha (n = 125) that occurred during a 25-year period (1984-2008). Categorical burn-severity images were developed from an index of burn severity (Relative differenced Normalized Burn Ratio) derived from LTM data and parameterized with data from 639 field plots. My results show that the fire regime of the northern Cascade Range responds to annual climatic variation. Spring snowpack and summer temperature were negatively and positively correlated, respectively, with fire occurrence, and summer temperature was positively correlated with annual area burned, the proportion of landscape burned at high severity, and spatial aggregation of the high-severity class. Nevertheless, the within-fire severity mosaic reflects the underlying topographic complexity. Fires in areas with greater topographical complexity had increased spatial complexity of burn severity. Several recent studies in the western United States have documented a positive relationship between warm and dry climate and annual area burned. The relationship between climate drivers and fire-regime attributes identified in this study—a positive relationship between warm and dry conditions and the proportion of area burned at high severity and spatial aggregation of high-severity patches—adds nuance to the climate-area burned relationship documented in previous studies.