The ability of high resolution aerial imagery to determine shrub location and size in creosote flats of the Sonoran and Mojave Deserts

Background/Question/Methods

The vegetation of the Sonoran and Mojave deserts is characterized by shrublands dominated by Larrea tridentata(creosote bush) and other smaller co-dominant species.  These shrubs create fertility islands that harbor native and invasive annual plants, providing considerable amounts of fuel for fires in the dry season.  The spatial patterns of shrubs and fertility islands they create can have major effects on spread of wildfire.

We assessed the ability to infer information about shrub distribution from orthorectified aerial imagery (at 24-50cm resolution).  In both deserts we studied regions of approximately 1 hectare, including areas affected by wildfires in 2005.  We used ImageJ software to determine the location and size of shrubs visible in the aerial image and compared these data to data gathered in the field using a GPS device (accuracy 30-50cm) and hand measurements of canopy size. We assessed the ability of the aerial image to accurately determine shrub size using linear regression, and the spatial pattern of the shrub population using the pair correlation function and the mark correlation function.  We also determined if the images could be used to determine the locations of burned shrubs and to distinguish Larrea from Ambrosiain the Mojave.

Results/Conclusions

In the Sonoran, we were able to use the 24cm imagery to determine the shrub size in the field (Area_Field=0.84 + 1.06*Area_Image, R²=0.67), and locate shrubs over 0.04 m2 in area within 1m of the GPS locations.  In the burned area, many shrubs that we classified as dead or as a stump were discernible on the 24cm imagery.

In the Mojave, where Larrea and Ambrosia are co-dominant, we used the mean pixel intensity to distinguish between species. The minimum shrub visible at 50cm resolution is 0.24 m2, which excluded about 1% of the Larrea population and 22% of the Ambrosia population. These images were able to estimate shrub size, but with more bias than in the Sonoran (Area_Field=0.58 + 0.48*Area_Image, R²=0.74). 

The choice of threshold pixel intensity value to segment shrub objects from the background introduced a tradeoff between accuracy for count and location-based statistics (number of shrubs and univariate point pattern analysis) and accuracy for area-based statistics (area estimation and marked point pattern analysis). Use of aerial imagery to obtain information on shrub spatial patterns can reveal how wildfires affect the geometry and spatial pattern of perennial shrubs and the resulting distribution of fuel for wildfires.