Our results revealed an inverse relationship between snowmelt anomaly and total annual area burned in the Middle Rockies Ecoregion from 1984 to 2012. An earlier spring snowmelt was correlated with above-average annual area burned. These results support the findings of others (Medler et al. 2002, Westerling et al. 2006, Semmens and Ramage 2012). The implications of our findings are especially critical because the average annual area burned in the western US is projected to increase further, and years without fires are predicted to be increasingly rare (Westerling et al. 2011). Additionally, as the climate changes in the US Mountain West, snow is expected to melt progressively earlier in the season (Stewart 2009). Our findings suggest that snowmelt occurring earlier in the coming decades will exacerbate wildfire risk. If this relationship continues to persist, we can expect an increase of wildfire annual area burned in the twenty-first century.
The link between snowmelt and annual area burned is apparent for the large increase in fire activity from 2000 to 2012, which coincides with a decade of early snowmelt (Figure 2). Indeed, aside from 1988, all above-average fire years occur within the 2000 to 2012 subset (Figure 3). Examination of the scatterplot (Figure 3) reveals a clear pattern that the majority of above-average fire years follow an early snowmelt anomaly, similar to findings in Medler et al. (2002). We can see in the full period of record that there is only one point (1988) that experiences a higher than average fire season following a positive snowmelt anomaly (later than average snowmelt date). All other above-average fire years follow a negative snowmelt anomaly (earlier than normal melt). There are many years with a low total annual area burned, regardless of the snowmelt anomaly. However, the years with high fire totals almost exclusively follow early snowmelt anomalies. This indicates that late snowmelt anomaly has a suppressive effect on wildfires, whereas early snowmelt anomalies are one factor in many that contribute to large fire years.
Early snowmelt is one way to achieve low fuel moisture and extended fire seasons leading to increased annual area burned, yet it is certainly not the only mechanism. Climate indirectly affects wildfire occurrence, through control over fuel moisture (Gedalof et al. 2005, Miller et al. 2008, Littell et al. 2009, Fauria et al. 2011, Higuera et al. 2015). Dry fuel loads with adequate ignitions lead to wildfires. Fuel moisture is influenced by a variety of ecological and climatic processes including high summer temperatures, drought, and snowmelt timing. Of all the factors that lead to low fuel moisture, snowmelt timing is one of the earliest predictors, occurring several months before the fire season begins. Drought indexes, fuel moisture metrics, and spring-summer temperatures may be more highly correlated with annual area burned (Gedalof et al. 2005, Westerling et al. 2006, Miller et al. 2008, Spracklen et al. 2009, Higuera et al. 2015), but are temporally concurrent with wildfire season. Our model of classifying early snowmelt timing using Rutgers Global Snow Lab analysis of NOAA daily snow maps may provide months of foresight for wildland fire managers.
While climate variables such as PDSI and air temperature have a high explanatory power for wildfire (~50 %), snowmelt timing explains very little of the variability in annual area burned (~14 %). This is not a surprise, because snowmelt occurs months before the typical fire season, and there are many factors that can influence fire conditions in the interim. Nonetheless, fire managers should include snowmelt timing in any predictive analysis as snowmelt can directly influence fuel moisture levels, which serve as one factor in many that lead to large wildfire years.
We decided a priori that the year 1988 was an outlier as it clearly had the largest annual area burned and a slightly later than average snowmelt (Figure 2, Figure 3). The majority of the fires that make up the total acreage in our study area in 1988 occurred in the Greater Yellowstone Ecosystem, burning over 250 000 ha in the surrounding areas. The large 1988 fires can be explained by many factors aside from snowmelt timing. The primary consensus of why this fire complex became so large was the combination of unusual weather conditions characterized by prolonged summer drought and extreme wind coupled with the vast extent of continuous lodgepole pine (Pinus contorta Douglas ex Loudon) forest (Romme and Despain 1989, Turner et al. 1994, Higuera et al. 2010). Historical evidence suggests that similar large fire events have occurred in Yellowstone on the order of 150- to 300-year intervals, occurring circa 1240, 1540, and 1700 (Higuera et al. 2010). The extensive fires in 1988 were caused by interactions of climate and fuel conditions, and correspond to a natural historical MFRI of 150 yr to 300 yr (Romme and Despain 1989, Turner et al. 1994, Higuera et al. 2010).
Early snowmelt is correlated with above-average wildfire occurrence, but it is not the only factor. When looking at the complete data set, the Spearman’s rank correlation results suggest that late snowmelt’s suppression of the following fire season’s annual area burned is unlikely to be random. Omitting the year 1988 shows an even stronger inverse relationship for all but the most extreme fire year. Removing 1988 is justified because this well-studied complex was caused by extreme weather following snowmelt, and by omitting this data point we are better able to see the influence of periodic fires rather than widespread stand-replacing fires. Regardless of the cause of the 1988 fires, the fire regimes of this region are shifting due to earlier snowmelt and longer fire seasons, which is predicted to dramatically alter the ecosystem by the mid-twenty-first century (Westerling et al. 2011).
Further Investigations
Future research may seek to improve upon this analysis by using an updated MTBS dataset or higher-resolution imagery. The National Snow and Ice Data Center’s MOD10A2 data-set has recently become popular (Hall et al. 2006). However, with its period of record extending back only to 2000, analyzing these higher-resolution data at a regional scale over this short period may fail to identify the relationships at play. Indeed, analysis of the subset from 2000 to 2012 yields no statistically significant relationship between snowmelt anomaly and wildfire. Improvements upon the snowmelt timing logic may be helpful in investigating the influence of spring snow storms following a lengthy period of no snow.