GEOG 566






         Advanced spatial statistics and GIScience

April 6, 2017

An investigation into the correlation between snowpack and post-wildfire forest greening.

Filed under: 2017,Final Project,My Spatial Problem 2017 @ 10:14 am

Research Question

My spatial question looked into the relationship between winter snowpack and re-vegetation following a severe wildfire. I used variograms and binned scatter plots to characterize this relationship in a visual manner.

Datasets

To estimate snowpack, I used our research group’s new snow cover frequency (SCF) product, which uses satellite reflectance data from the Moderate Resolution Imaging Spectroradiometer (MODIS) to measure the percentage of snow-covered days over a user-defined span of time. MODIS imagery is acquired daily and is available from the year 2000 to the present. The spatial resolution of MODIS imagery is 500 meters. This project considered October 1 to May 1 to be the relevant timeframe for each year included in the analysis. So in the example below, the highest pixel value for SCF (0.58) corresponds to the interpretation that 58% of the valid days from October 1 to May 1 were snow-covered.

MODIS reflectance data will also be used to estimate forest greening. The Enhanced Vegetation Index is calculated using red, infrared, and blue wavelength bands to estimate canopy greenness, a quality which depends on leaf area, chlorophyll content, and canopy structure. EVI images are readily available in the form of 16-day composites collected at a resolution of 500 m. For each summer season (June 1 – September 30), EVI data will be condensed across bands to create a maximum summer EVI image which will be utilized in the spatio-temporal analysis.

Wildfire burn perimeter and severity data will be obtained through the Monitoring Trends in Burn Severity (MTBS) project. Analysis will be constrained to wildfires within the CRB occurring over forested land cover with large areas of high burn severity. This data is available back to 1984, but will only be considered post-2000 due to the limiting data availability of MODIS imagery.

To look at soil texture, I will be using the State Soil Geographic dataset (STATSGO), which includes soil polygon shapefiles along with soil texture information that I have joined to the spatial layer. This layer was created by generalizing more detailed soil survey maps. When these detailed maps were not available, the survey used data on geology, topography, vegetation and climate to predict the probable classification.
 

Hypothesis

The primary relationship driving this study is that between winter snowpack and the following summer’s revegetation following a severe wildfire. My hypothesis was that snowpack is most strongly correlated with forest greening directly following a severe wildfire, and that the strength of this correlation decreases with time following the wildfire. Additionally, I expected to find the snowpack-greening relationship to be strongest where soils are coarse and skeletal with low water-holding capacity.

Approach

My research question utilized two approaches. First, I conducted a series of variograms and cross-variograms with the general goal of understanding the patchiness of my data. I initially created variograms of both SCF and EVI raster layers for the pre- and post-fire year. Overall I found that the patchiness of both the snow and vegetation data remained consistent across the pre- and post-fire years. I then created cross-variograms between SCF and EVI to compare spatial autocorrelation between my snow cover data and vegetation data to see how these variables change together across space within the perimeter of the Spur Peak wildfire, and to see how this relationship changes temporally.

Second, and more in line with my thesis research, I shifted my attention to the Pot Peak wildfire, where I had already been analyzing the post-wildfire snow-revegetation relationship prior to this class using a different approach (Figure 1). Specifically, I plotted EVI against SCF with EVI as the dependent variable. To standardize the changes in snow and vegetation, I’ve plotted the change in EVI against the change in SCF by subtracting each post-fire year condition from the pre-fire year conditions.

To elucidate the potential role that burn severity and soil type have on this relationship, I’ve binned the data in two ways. To monitor potential effects of different soil textures, I delineated and separated the SCF/EVI raster data by soil polygon boundary. To visualize the effect that burn severity might have on this relationship, I used burn severity thresholds recommended by Miller and Thode (2007) to colorize by highly burned, moderately burned and control (low/no burn) pixels. Finally, for every year and for each class of burn severity, I conducted a regression analysis including a regression line and the corresponding R-squared values and p-values.

Figure 1: Pot Peak wildfire burn severity map. Colored polygons represent different soil types, brighter pixels indicate higher burn severity.

Results, Part 1: Cross variograms

A couple of apparent trends are evident in the time-series of cross-variograms shown below (Figures 2a-d). First, although each year shows two clear bumps in the cross-variogram, the lag distance of those bumps shift significantly from year to year. My guess would be that the initial shift from pre-fire to the first year post-fire was primarily due to the variation in burn severity across the landscape. The shifts in ensuing years may be due to varying rates of vegetation recovery or perhaps due to changes in snowpack distribution caused by the wildfire. It’s also worth noting that the valley between these two peaks remains fairly consistent at a lag distance of 5000 meters. In tandem, these trends suggest that there may be two patch sizes for this wildfire (of shifting size), but that the between patch distance remains fairly consistent at about 5000 meters. While I learned quite a bit regarding the usage of variograms and how to interpret them, these findings ultimately didn’t shed much light on the spatial question that I was asking, so I decided to change routes into a completely different toolbox.

Figure 2a: Pre-fire cross variogram between SCF and EVI                                     Figure 2b: First year post-fire cross variogram.

  

                        Figure 2c: Second year post-fire cross variogram                                                 Figure 2d: Third year post-fire cross variogram

 

Results, Part 2: Regression scatter plots

For the second part of my project, I switched gears completely and brought in some of the work I had been doing outside of class on the Pot Peak wildfire in central Washington. I’ve included the scatter plots for all 3 soil types below (each group of 6 plots is a single soil polygon), but only conducted a regression analysis on the Wedge-Fernow soil type (Figure 3a), characterized by a primarily ashy and pumiceous soil texture.

The most noticeable result is the apparent strong relationship that exists in the first year following the wildfire for both the moderate and high burn pixels. The R-squared values were 0.753 and 0.657 for the moderately and highly burned pixels, respectively. The p-values for these pixels were also quite low. Therefore, in this first year post-fire, we see a significant relationship suggesting that for an increase in snow cover frequency, a resulting increase in greenness (EVI) is likely to be observed. Interestingly, we see a complete dissolve of this relationship in the second and third years following the wildfire. The R-squared values drop to near zero and the p-values increase significantly. The control plots show no significant correlation in any of the post-fire years.

For the other two soil type polygons (Figures 3b and 3c) no significant trend exists in any years following the wildfire. The only apparent instance where the moderate and high burn pixels have noticeably differently relationships between snow cover and greenness following the wildfire is within the Saska-Ramparter soil type (Figure 2c). However, looking back at the entire burn zone (Figure 1), we can see that there is a clear separation of a highly burned patch of land to the north and a moderately burned patch of land to the south. There is likely some topographic factor, perhaps vegetation type or elevation, that is driving this difference as opposed to the scale of burn severity.

Figure 3a: Scatter plots and regression analysis for the Wedge-Fernow soil type polygon. Red points correspond to highly burned vegetation, yellow points correspond to moderately burned vegetation, and blue points correspond to unburned or very-low burned vegetation. Included are the R-squared and associated p-values for both the moderately and highly burned areas.

           

            Figure 3b: Scatter plots for the McCree-Ardenmont soil type                                 Figure 3c: Scatter plots for the Saska-Ramparter soil type

 

Significance

Snow accumulation has already been shown to influence peak summer forest greenness, especially at moderate elevations (Trujillo et al. 2012). The post-wildfire relationship between snowpack and forest revegetation is critical to understand as current trends of increasing temperatures, more ephemeral snowpack and intensifying wildfire activity are all forecasted to continue. Consequently, an expanding area of western mountain regions are becoming vulnerable to disturbance, revegetation and successional growth.

Forest managers and watershed managers may find the analysis of this research useful in preparing for future climate regimes. As wildfires continue to become more prevalent, having a comprehensive understanding of the ecological impacts of such disturbances will be critical for effective management of post-wildfire landscapes.

The ecological implications of this research are multi-faceted, especially regarding the changing climate affecting western U.S. mountains. Forests in the CRB are significant contributors to carbon sequestration, as western forests are responsible for 20-40% of total carbon sequestration in the contiguous U.S. (Schimel et al. 2002; Pacala et al. 2001). Depending on how successful forests recover following wildfires, western forests’ role as carbon sinks versus carbon sources may become more uncertain in future climate scenarios.

Software learning

ArcGIS: Coming into this class, I was already very comfortable with ArcGIS tools and capabilities. I did learn about a couple of new tools, including adding XY coordinates to a point shapefile to allow for smoother spatial analysis in R.

R: Before this class, I was only familiar with R in the context of a past statistics course using tabular data. Through this research question, notably the variogram analysis, I became more comfortable reading in raster layers to R, searching for appropriate R packages, and ultimately analyzing spatial data.

MATLAB: For the second half of my project, I primarily worked in MATLAB in creating the scatter plots. Through this process I learned a great deal regarding general plotting syntax, data access from a csv or text file, and categorical data binning.

Statistical learning

While I was already familiar with the creation and interpretation of scatter plots and regression statistics, the concept of semivariograms and cross-variograms was completely new to me. Working together with other students and sharing our thoughts regarding variogram and cross-variogram interpretation proved to be extremely helpful in understanding what the variograms were telling me. Throughout the course of this project, I even observed a couple of seminar talks in which variograms were referenced, so I was thankful to be able to quickly grasp the implications of the variograms that were being displayed.

Response to exercise/tutorial comments

Following Exercise 3, it was suggested that perhaps I should change research methods by stepping away from variograms and instead involve the research that I was already conducting outside of class. This was the primary reason for switching analysis approaches for my final project. Following my Tutorial 2 presentation, we discussed the limitations of my current method, especially the fact that there are so many other influential topographic factors that play into the revegetation of the burn zone. Moving forward, I’m planning on comparing the 5 years before the wildfire with the 5 years after the wildfire over the exact same extent to account for this issue.

References

Miller, J. D. & Thode, A. E. Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sensing of Environment 109, 66–80 (2006).
Pacala, S. W., Hurtt, G. C., Baker, D. & Peylin, P. Consistent land- and atmosphere-based U.S. Carbon Sink Estimates. Science 292, 2316–2320 (2001).
Schimel, D. et al. Carbon sequestration studied in western U.S. mountains. Eos Trans. AGU 83, 445–449 (2002).
Trujillo, E., Molotch, N. P., Goulden, M. L., Kelly, A. E. & Bales, R. C. Elevation-dependent influence of snow accumulation on forest greening. Nature Geosci 5, 705–709 (2012).

 

Print Friendly, PDF & Email


No Comments

No comments yet.

RSS feed for comments on this post.

Sorry, the comment form is closed at this time.

© 2019 GEOG 566   Powered by WordPress MU    Hosted by blogs.oregonstate.edu