Impacts of Snow Water Equivalent on Wildfire Events in the Sierra Nevada

EarthzineArticles, Climate, Original, Sections, Themed Articles, Wildfires Theme

Figure 1. The M261E ecological region in California. Our study area is further split into north and south sub-regions.

Figure 1. The M261E ecological region in California. Our study area is further split into north and south sub-regions.

Andrew Nguyen, San Jose State University;
Chase Mueller, University of Texas at San Antonio;
Roy Petrakis, University of Arizona;
Spencer Adkins, Brigham Young University
Olivia Kuss, Indiana University ‰ÛÒ Purdue University of Indianapolis;
Monica Kumaran, Harker High School;
Marc Meyer, USDA Forest Service Pacific Southwest Region;
Cindy Schmidt, Bay Area Environmental Research Institute

Abstract — High Sierra snow and ice provide the primary water supply for the Sierra Nevada ecosystem. Understanding how climate change affects high Sierra snowmelt and how these changes impact wildfire events is important for future forest management. Snow water equivalent (SWE) anomalies were averaged on a monthly basis and overall trends of snowpack availability and timing of snowmelt were examined throughout the Sierra Nevada from 2003-2012. Periods of decreased snowpack were examined alongside periods of decreased soil moisture (SM), increased soil temperature (ST), and increased wildfires. This project used NASA Earth observations such as the Moderate Resolution Imaging Spectroradiometer, onboard the Terra satellite, for snow cover and Landsat 5 Thematic Mapper for wildfire and vegetative analysis. We also used ancillary and modeled datasets such as temperature, precipitation, and water flow rate to provide a better understanding of the relationships among snowpack, soil properties, and wildfires. Computation of correlation coefficients for SWE to SM and ST to the number of fires indicated moderate to strong relationships. A Generalized Additive Model (GAM) was used to analyze the contribution of particular environmental variables to wildfire events and to create a wildfire risk map with an accuracy of 81 percent based on the most current fire occurrence data. This information enables fire managers to more accurately assess wildfire risk in the Sierra Nevada.

I. ENVIRONMENTAL IMPACTS OF WILDFIRE AND SNOWPACK

Wildfires cost California more than $800 million in annual property damage [1], [2] and have devastating long-term public health and economic consequences. Smoke and particulates from wildfires contribute to respiratory diseases and premature deaths, while the fires themselves change the hydrology and habitats of local ecosystems. Wildfires can also cause the loss and degradation of human capital and infrastructure. The burning of vegetation causes a release of stored carbon, exacerbating climate change [3]. Due to changes in climate, wildfires are predicted to become larger and more frequent over the next few decades [4].

Over the last decade, wildfires in the Sierra Nevada have increased substantially in area and frequency [5]. In addition, spring snowpack has declined, with further decreases projected for the mountainous western United States [6]. While research has focused on the connection of temperature and precipitation to fire activity, little is known about the effects of decreased snowpack in these areas. At the NASA Ames Research Center in Mountain View, California, our team analyzed the relationship between declining snowpack and increasing wildfires in the Sierra Nevada. We used three methods: 1) time-series correlation analysis among climatic and surface conditions (such as temperature and soil moisture), 2) case study fires and preceding climatic conditions to support the aforementioned correlations, and 3)åÊA Generalized Additive Model (GAM) to identify areas most susceptible to wildfire and to understand the conditions that contribute to wildfire. Our study focused on a large section of California called the M261E Sierra Nevada Section. This particular section is one of seven sections of the much larger province of Sierran steppe, mixed forest and coniferous forest labeled M261. An ecological section is a group of ecosystems that shares similar soil and landform characteristics. M261E consists mostly of steep mountains interrupted by deep valleys, and is characterized by a Mediterranean climate [7], [8]. To improve analysis, the M261E section was further divided by the study team into north and south sub-sections (Figure 1).

II. METHODS

Correlations and relationships between snow water equivalent and wildfire events

Figure 2. Methodology flowchart. Arrows indicate how specific datasets are connected and correlated for specific parts of the project. Box colors indicate how the data was used in this study.

Figure 2. Methodology flowchart. Arrows indicate how specific datasets are connected and correlated for specific parts of the project. Box colors indicate how the data was used in this study.

To study the relationship between snowpack and wildfire, we examined many variables, including snow water equivalent (SWE), soil moisture (SM), soil temperature (ST), snow extent, ambient temperature, precipitation, and fire activity for October 2003‰ÛÒSeptember 2012 (Figure 2). Datasets for environmental variables were obtained through the NASA Earth Observing System’s Moderate Resolution Imaging Spectroradiometer (MODIS) instrument on the Terra satellite platform and Landsat 5 Thematic Mapper. Daily SWE data were acquired from the National Weather Service’s Snow Data Assimilation (SNODAS) program. We also obtained SM and ST data sets from the North American Land Data Assimilation Phase 2 åÊNOAH model. Datasets such as SWE, SM, and ST were modeled values validated with observational data (such as snow extent from the Terra MODIS sensor) prior to analysis. After calculating monthly averages for all datasets, SM and ST were correlated to ambient temperature and precipitation to understand the influence of climatic conditions on soil properties.

Prior to conducting time-series frequency analysis, each time series was converted into monthly cumulative departures (the residuals from the mean value of the series at each time step) to facilitate statistical comparisons between various data types. Frequency analysis of climatic and hydrologic time series was used to evaluate wildfire responses to environmental variability [9]. Correlation coefficients were generated for comparisons of SWE, SM, ST, and fire occurrence from October 2003‰ÛÒSeptember 2012 for the entire M261E region. SWE and SM were moderately positively correlated, with a correlation coefficient of 0.59 for the study area, which indicates that as SWE increases, SM will also increase, and vice versa. SWE was also moderately inversely correlated to ST, with a correlation coefficient of ‰ÛÒ0.43, indicating that as SWE decreases ST will increase, and vice versa. When comparing SM and ST to fire occurrence, similar patterns emerged. SM and fire occurrence were moderately inversely correlated, with a correlation coefficient of ‰ÛÒ0.56, indicating that as SM decreases, the likelihood of fire outbreak increases. Finally, a moderate correlation was observed between ST and fire outbreak, with a correlation coefficient of 0.61, indicating that as ST increases, the likelihood of fire outbreak also increases. These values indicate that from October 2003‰ÛÒSeptember 2012, SWE and fire activity were indirectly related through ST and SM.

Figure 3. Case study fires (red) within encompassing watersheds (blue). The BTU Lighting Complex Fire represents the north fire, the Piute Fire represents the south fire. These case studies helped show that SWE, ST, and SM are indicators of wildfire occurrence and can be applied to the study region

Figure 3. Case study fires (red) within encompassing watersheds (blue). The BTU Lighting Complex Fire represents the north fire, the Piute Fire represents the south fire. These case studies helped show that SWE, ST, and SM are indicators of wildfire occurrence and can be applied to the study region.

Case Study Comparisons

To supplement our first analysis, we conducted case studies focusing on specific fires from the north and south regions of M261E. Case study fires were chosen for the M261E sub-sections to establish variability in the north and south climatic regimes. These case studies were designed to isolate observed trends in data to specific wildfire events and to decouple trends from seasonal variation. For the north case study, we used the BTU Lightning Complex Fire. It ignited on August 8, 2008 and burned 21,080 hectares. This fire was located along the north side of the North Fork of the Feather River between the Rock Creek Reservoir and Lake Oroville (Figure 3). In the southern region, the Piute Fire was ignited on June 28, 2008 and burned 14,670 hectares. The fire was located south of Isabella Lake on Piute Peak (Figure 3). The criteria used to select the case study fires were: 1) the fire size must be more than 12,141 hectares and 2) the fires had to occur within the same year to allow for uniformity in seasonal climatic effects.

Monthly SWE, ST, and SM for the case study fire areas were averaged for the years leading up to the fire (October 2003‰ÛÒJune 2008). SWE was separately correlated with SM and ST using the same time series frequency analysis previously mentioned. In the northern case study, inverse correlations were observed for SWE and ST (‰ÛÒ0.47) similar to correlations for the entire region (‰ÛÒ0.42). Additionally, positive correlations were observed for SWE and SM (0.53) for both the case study area and the entire northern region. In the southern case study, inverse correlations were observed for SWE and ST (‰ÛÒ0.34) comparable to correlations for the entire south region (‰ÛÒ0.41). Similar positive correlations were also observed for SWE and SM for the south case study area (0.50) and the entire southern region (0.58). Similarities in the case study examples provided confidence that the variability in SWE and soil properties in relation to fire occurrences could be accurately applied to the entire M261E region.

To assess runoff variability and to validate SWE values within the watersheds encompassing the case study fires, we included data from two river gage downstream of the fires. A comparison of SWE to river gauge data for the north and south watersheds yielded a moderately positive correlation of 0.55 for both areas, further validating the SNODAS model data. These case studies supported the linkage of SWE with SM and ST, indicating that ST and SM were important in determining wildfire conditions.

Modeling wildfire risk using a generalized additive model (GAM)

Fig_4

Figure 4: The fire risk map produced for our study area. This map is based on climatic and surface conditions for 2010‰ÛÒ2012 WY.

GAM is an ecosystem forecasting model that relies on absence and presence points of an independent variable, such as fire, to determine its relationship with selected dependent variables, such as various climate and surface conditions [10]. For this study, we obtained absence points by analyzing a classified land cover image from the National Land Cover Database. We generated points over areas that had a low probability of burning (open water, perennial ice/snow or barren land). We also generated presence points using fire polygons from the CAL FIRE Fire and Resource Assessment Program for the applicable water years. Dependent variables that are capable of creating suitable conditions for wildfire include SWE, SM, ST, maximum temperature, and precipitation.

The model was used to evaluate the influence of each these variables to wildfire occurrence. Based on the model results, SWE, SM, and ST are the three primary indicators of wildfire risk.

Additionally, the model generated a percentage-based fire risk map to identify areas prone to wildfire, given the climatic conditions and soil properties. In this study, risk was defined as the probability of wildfire occurring based on the area’s climatic and surface conditions. Risk maps for water years (WY) 2007‰ÛÒ2009 and 2010‰ÛÒ2012 (Figure 4) were created. Using the ‰ÛÏPredict GAM from Table‰Û tool, 80.9 percent accuracy was observed for predicted fire risk for the WY 2007‰ÛÒ2009 risk map. This accuracy was derived using fire presence and absence points from WY 2010‰ÛÒ2012 as inputs into the tool (Figure 5). This was a high level of accuracy given the inherent variability in the ecosystem, including differing climate conditions throughout M261E. It was interesting to note that this model was originally developed to track suitable living conditions for marine life on coastal regions. It has also been applied to determine habitat suitability for invasive plant species and algal blooms along the California coast. To our knowledge, the model has not been used to evaluate conditions for something as dynamic as wildfire.

Figure 5. Risk map reflecting conditions for 2007‰ÛÒ2009 WY. Wildfire occurrences (black) for the 2010‰ÛÒ2012 WY were overlaid to delineate areas of risk predicted by the model. The model's predictions resulted in 80.9 percent accuracy.

Figure 5. Risk map reflecting conditions for 2007‰ÛÒ2009 WY. Wildfire occurrences (black) for the 2010‰ÛÒ2012 WY were overlaid to delineate areas of risk predicted by the model. The model’s predictions resulted in 80.9 percent accuracy.

III. CONCLUSIONS

The primary focus of this project was to determine how fire activity is affected by climatic (SWE, precipitation, and ambient temperature) and soil (SM and ST) variables within the Sierra Nevada region known as M261E. Establishing relationships among SWE, SM, ST, and wildfire events helped create a link between fire activity and environmental variability. Our observations indicate a strong relationship between SWE and fire activity. The BTU Lightning Complex and Piute case study fires enabled further validation and analysis of the modeled data sets and variables in the years leading up to their respective fires. This supports the connection between wildfire occurrence and SWE, SM and ST. GAM can also be used to develop a wildfire risk assessment while defining the most influential variables for a given time period and study area.

As the climate warms, increasing amounts of Sierra Nevada precipitation will turn from snow to rain, effectively lengthening and intensifying the fire season. Our data, along with future vegetative and data trend analysis of the study region, may improve the predictive risk assessments of wildfire events. This will also enable forest managers to improve monitoring and fire abatement strategies, such as prescribed burning to decrease large fire outbreaks like the recent Rim Fire near Yosemite National Park. Additionally, forest managers and hydrologists can monitor watersheds currently experiencing a decline in spring snowpack. To prepare for future wildfires, remotely sensed and in situ data will be useful in determining areas at risk and to assist in forest management decisions.

IV. ACKNOWLEDGEMENTS

The team would like to acknowledge the NASA DEVELOP Program for its support in this project.

Andrew NguyenåÊis a graduate of California State University, East Bay. He has earned a Bachelor of Science in geography. His current position as a project coordinator at NASA’s DEVELOP Program, in California, entails conducting multiple Earth science-based research projects. Using satellite imagery, these projects spread across a wide range of scientific disciplines from forest ecology to wetland restoration.åÊ Prior to the DEVELOP Program, he served in the U.S. Marine Corps. as a map analyst, an infantry squad leader, and regimental security leader . He is currently a second-year graduate student in California San Jose State University’s geography program.

REFERENCES
[1] A. Westerling, and B. Bryant, ‰ÛÏChange and wildfires around California: Fire modeling and loss modeling,‰Û Sacramento, CA: California Climate Change Center, 2006. [Online]. Available:åÊ www.energy.ca.gov/2005publications/CEC-500-2005-190/CEC-500-2005-190-SF.PDF. [Accessed: 1 August 2013].

[2] J. Wu, A. Winer, and R. Delfino, ‰ÛÏExposure assessment of particulate matter air pollution before, during, and after the 2003 Southern California wildfires,‰Û Atmospheric Environment, vol. 40, no. 18, pp. 3333-3348, January 2006.

[3] B. Bond-Lamberty, S.D. Peckham, D.E. Ahl, and S.T. Gower, ‰ÛÏFire as the dominant driver of central Canadian boreal forest carbon balance,‰Û Nature, vol. 450, pp. 89-92. November 2007. [Online]. Available: http://www.nature.com/nature/journal/v450/n7166/pdf/nature06272.pdf. [Accessed: 1 August 2013].

[4] D. V. Spracklen, et. al., ‰Impacts of climate change from 2000 to 2050 on wildfire activity and carbonaceous aerosol concentrations in the western United States,‰Û J. Geophys. Res., vol. 114, October 2009. [Online]. Available: http://ulmo.ucmerced.edu/pdffiles/09JGR_Spracklenetal.pdf. [Accessed: 10 July 2013].

[5] A. L. Westerling and B. Bryant, ‰ÛÏWarming and earlier spring increase western U.S. forest wildfire activity,‰Û Sci., vol. 313, no. 5789, pp. 940-943, July 2006.

[6] M. C. Mastin, K. J. Chase, and R. W. Dudley, ‰ÛÏChanges in spring snowpack for selected basins in the United States for different climate-change scenarios,‰Û Earth Interactions, vol. 15, no. 23, pp. 1-18, July 2011.

[7] J. E. Keeley, ‰ÛÏFire and invasive species in Mediterranean-climate ecosystems of California,‰Û presented at the Proc. of the Invasive Species Workshop: the Role of Fire in the Control and Spread of Invasive Species. Fire Conf. 2000, the 1st Nat. Congr. on Fire Ecology, Prevention, and Manage., 2002, Nov. 27-Dec. 1, 2000, Tall Timbers Research Station, Tallahassee, FL.

[8] A. Pupacko,‰ÛÏVariations in northern Sierra Nevada streamflow: Implications of climate change,‰Û J. Amer. Water Resources Assoc., vol. 29, no. 2, pp. 283‰ÛÒ290, 1993.

[9] R.T. Hanson, M.W. Newhouse, M.D. Dettinger, ‰ÛÏA methodology to assess relations between climatic variability and variations in hydrologic time series in the southwestern United States,‰Û Journal of Hydrology., 287. 252-269. 2004.

[10] R. L. Young, J. Weinberg, V. Vieira, A. Ozonoff, and T.F. Webster, ‰ÛÏGeneralized additive models and inflated type I error rates of smoother significance tests,‰Û Computational Stat. & Data Anal., vol. 55, no. 1, pp. 366-374, 2011.