GEOG 566






         Advanced spatial statistics and GIScience

May 21, 2017

Tutorial 2: Calculate Disturbance Severity Patch Sizes with SDMtools in R

Filed under: Tutorial 2 2017 @ 4:29 pm

Overview

Often patch dynamics are examined under the lens of single disturbance events, but contemporary disturbance regimes are demonstrating that large-scale disturbances can overlap in time and space.  The spatial configuration of patches from an initial disturbance may influences the spatial configuration of patches of a sequential disturbance.  The broader research question is focused on determining if there is a relationship between patches from an initial disturbance of beetle outbreak and a sequential disturbance of wildfire. Thus far, I have conducted analysis to determine if there is any inherent patchiness for the burn severity raster and the beetle-outbreak raster.  Both have indicated some level of patchiness. I wanted to go back and characterize the patches for each disturbance to generate summary statistics about patches for each disturbance.  The Entiako burn perimeter delineated our landscape for both disturbance events. The Entiako fire burned in 2012 in central interior British Columbia. I asked the following questions:

What is the variability in patch size for burn severity classes and beetle severity class?

Methods

The burn severity and beetle severity raster generated from Landsat imagery using dNBR and NDMI pixel indexing approaches respectively. Initially, I planned to conduct the analysis in FRAGSTATS that assesses landscape spatial patterns.  The raster seemed unreadable in the commonly used GEOTIFF format.  I ended up conducting the analysis in R version 3.3.2 (Team 2016) with the SDMtools Package (VanDerWal 2015).   The SDMtools package is based on the work by McGarigal et al. (2012) FRAGSTATS software package and calculates many of the same metrics. The following spatial equation was considered as a guide yi = f(yi-h,i+h).

In order to calculate patch metrics, I converted the continuous raster into a classified raster.  The burn severity classification was based on break points in the dNBR values defined by Reilly et al. (2017), and the beetle severity classification was based on cross-validation with orthorectified aerial photographs.  Burn classes included a low (plus unburned), moderate, and high, and beetle class included none, low-moderate, and moderate-high. In the SDMtools package, I calculate patch metrics for each disturbance severity class, burn severity class, Figure 1 and beetle severity class Figure 2.  The following PDF includes step-by-step instructions with r-code for generating patch metrics:

ATaluccigeog566_exercise3_r_19May2017

 

Figure 1: Burn Severity Classes for Patch Metric calculations.

 

Figure 2: Beetle Severity Classes for Patch Metric Calculations

Results

 

In the SDMtools package, I ran two analyses, one on the whole landscape and one on each burn severity class. For the first analysis, I found that high burn severity was the smallest component for this fire event only accounting for about 15 percent of the landscape (see Table 1). This was interesting, because lodgepole pine is often characterized by ‘stand replacing’ high severity fire, however, moderate burn severity can also account for ‘stand replacing’ fire in lodgepole pine dominated landscapes, which was about 45 percent of the landscape (see Table 1). The landscape was dominated by low-moderate and moderate-high beetle severity with only a small portion unaffected by beetle outbreak (see Table 2).

 

Table 1: Burn severity class summary for patches, area, and proportion of landscape

Burn Severity Class Number of Patches Total Area in Hectares Proportion of Landscape
Low 382 2767.14 0.394
Moderate 448 3179.79 0.452
High 422 1082.70 0.154

 

Table 2: Beetle severity class summary for patches, area, and proportion of landscape

Beetle Severity Class Number of Patches Total Area in Hectares Proportion of Landscape
None 985 780.93 0.111
Low-Moderate 656 2968.56 0.423
Moderate-High 709 3273.30 0.466

 

 

The second analysis examined the patch metrics for each disturbance severity class. Table 3 and 4 includes patch area metrics by burn severity and beetle severity respectively. While the output includes many other variables, I wanted to look at the variability in patch size for disturbance type and class. The high burn severity maximum and mean patch size are the most interesting part of Table 3, because they are quite different from the low and moderate burn severity.  Further statistical analysis is needed to determine if they numbers are statistically different.  The summary of patch size metrics for both disturbance types indicate that there are many small patches for all class types.

 

Table 3: Burn severity classes patch size summary in hectares

Burn Class Number of patches Minimum Patch size Maximum Patch size Mean Patch Size Median Patch size SD
Low 382 0.09 1449.45 7.24 0.18 77.7
Moderate 448 0.09 1461.69 7.10 0.18 75.3
High 422 0.09 445.95 2.57 0.18 22.5

 

Table 4: Beetle severity classes patch size summary in hectares

Burn Class Number of patches Minimum Patch size Maximum Patch size Mean Patch Size Median Patch size SD
None 985 0.09 77.58 0.79 0.18 4.04
Low-Moderate 656 0.09 2380.86 4.53 0.18 93.07
Moderate-High 709 0.09 1461.60 4.62 0.18 57.68

Critique of the method – what was useful, what was not?

This analysis attempt was much more complicated than anticipate. The complications mostly came from the raster format.  The raster was formatted as floating point, which was not compatible with the stand alone Fragstats Software nor the patch grid extension in Arc GIS. The patch grid extension for Arc GIS is based on Fragstats and generates similar patch metrics (Rempel et al. 2012). I did not realize the issue was related to the  floating point raster until after I figured out how to run the SDMtools package in R (VanDerWal 2015). I did trouble shoot and figure out the raster issue, which lead to converting the raster from floating point to integer in R.  Additionally, a major challenge was figuring out how the patches are defined in SDMtools package. Fragstats and Patch Grid both offer an option to define the cell neighborhood rule as 4 or 8.  This option of defining the neighborhood cell rule was unclear to me in the SDMtools package. Based on a brief comparison of data outputs from Fragstats and SDMtools, I believe that patch identification tool in SDMtools classifies patches base on the 8 cell neighborhood rule. I think that calculating patch statistics in R is less cumbersome and allows for better documentation of methods. While the results of this analysis are interesting to consider, they do not address the overlap in disturbance events, which could be quite interesting.

 

 

References

McGarigal, K., A. Cushman, and E. Ene. 2012. FRAGSTATS v4: Spatial patterns analysis program for categorical and continuous maps. Univeristy of Massachusetts, Amherst, Massachusetts.

Reilly, M. J., C. J. Dunn, G. W. Meigs, T. A. Spies, R. E. Kennedy, J. D. Bailey, and K. Briggs. 2017. Contemporary patterns of fire extent and severity in forests of the Pacific Northwest, USA (1985-2010). Ecosphere 8:e01695.

Rempel, R. S., D. Kaukinen, and A. P. Carr. 2012. Patch Analyst and Patch Grid. Ontario Ministry of Natural Resources. Center for Northern Forest Ecosystem Research, Thunder Bay, Ontario.

Team, R. C. 2016. R: A language and environment for statisitical computing. R Foundation for Statistical Computing, Vienna, Austria.

VanDerWal, J. 2015. Species Distribution Modelling Tools: Tools for processing data associated with species distribution modeling exercises.

 

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