Effects of Vegetation Change on the Home Range of Black Bears (Ursus Americanus) with Emphasis
Effects of vegetation change on the home range of black bears (Ursus americanus) with emphasis on clear cutting
Aurora Hagan, Department of Fisheries and Wildlife, Undergraduate Program, University of Minnesota, Saint Paul, MN 55108, USA
Jaime Nielsen, Department of Natural Resources Science and Management, Graduate Program, University of Minnesota, Saint Paul, MN 55108, USA
Krista Trenda, Department of Fisheries and Wildlife, Undergraduate Program, University of Minnesota, Saint Paul, MN 55108, USA
Images of June and Lily from the North American Bear Center
Introduction
Black bears (Ursus americanus) do not migrate, but they have large territories that range from 10-50 sq. kilometers for females and 100-140 sq. kilometers for males. The bears cover this area in order to find food for themselves and for their young. Past studies have shown that these movement patterns are largely based on a food gradient for both the location of and the availability of food. The territorial extent will become smaller as food supplies decrease and the bears have been found to travel outside of their territory during times of adequate nutrient supply (The Friends of Algonquin Park). This suggests that black bears are not simply wandering around within a set territory, but are actively making decisions based on the environmental conditions.
It is theorized that physical disturbances to the deciduous and coniferous forests in which they live may also have a significant impact on movement patterns of the bears and the extent of their territory. Logging is one such disturbance and is common in the areas in which black bears are found.
Objectives
The purpose of the project is to analyze GPS data points collected on adult female bears’ home ranges and analyzing them in contrast to vegetative changes over seasons. We will do an analysis of these home ranges, and determine if certain areas are preferred due to changes in vegetation in Northern Minnesota. Using images from different seasons and multiple years over a four year time-span can show an overall change between years and changes in vegetation. If there is a correlation between vegetation use, mainly clear cuts, and black bear movement, then it can be determined to what degree these areas are used by bears. Future work could then be conducted to further assess reasons for preferences to different treatment sites.
Materials
The study area is in St. Louis County, MN near Lake Vermilion. As requested by the bear researchers, the exact location cannot be revealed.This is a natural area with coniferous and deciduous tree species and contains several lakes and small wetlands. Using ERDAS measuring tool and creating polygons around the clipped images the approximate size of the clipped image is 630 square kilometers and the Minimum Convex Polygon (MCP) of the bears’ ranges is approximately 64.28 square kilometers. This is taken using the measuring tool on ERDAS and creating polygons.
This project was performed using 2 types of computer programs, ArcGIS 2010 and ERDAS 2010/2011.ArcGIS was used in processing collected GPS points while ERDAS was used to process and analyze all remote sensing imagery. Sets of GPS points for two bears were provided by researchers from the North American Bear Center, which show the movements of each bear at varying intervals from 2008 through 2010. Data points were collected during the months from spring to fall at approximate two hour intervals.The time frames per bear were different; Lily was tracked for 5 months in 2009 and2010, while June was tracked for 5 months in 2009 and2010. The GPS points for 2008 were removed from this project as they contained only one month of sporadic data.
Landsat 4-7 Combined (TM) images were acquired from the USGS Global Visualization Viewer (GloVis). Images were chosen based on the availability of Level 1 data, time of year, and were selected to have minimal cloud cover over the study area.The Geo-TIFFLevel 1 data contains 7 bands (Blue, Green, Red, NIR, mid-NIR) which can be combined into one pyramid layer and analyzed by ERDAS. A range of images were chosen to align with the GPS points (appendix A).
An unsupervised classification was run on each image to classify cover type. A post classification procedure was performed to produce three separate thematic change maps.The vegetation maps were combined with relating GPS points to determine if there were correlations between cover type and bear location, while the change maps were used to determine if clear cutting has an impact on the overall bear range.
To assess the level of accuracy in classification per image, a supervised accuracy assessment was performed in ERDAS.Each classified image was compared to a relating infrared and/or panchromatic image of the area by process of a stratified random technique to assign points.
Change detections were done to analyze changes in vegetation and see if there was a correlation between selected years. This would help show if there had been clear-cutting done in the bears’ home ranges.
Procedures
The original GPS coordinates were given in Lat/Long. To be able to use the GPS points in ERDAS they first had to be imported into ArcMap and converted from Lat/Long coordinates to UTM 1984 WGS 15N coordinate system in order to match satellite imagery. To see the bears’ movements in different months and years a separate shape file was created per bear per month in order to relate them more directly to the time frames chosen for comparison. This was done for 2009 and 2010 GPS points.
The satellite imagery was processed and analysed in ERDAS 2010/2011. Using images that were downloaded from USGS, the images were clipped to focus on the study area and these clipped images were then used for all later use. All images underwent the same processes during this project.
Before the images were clipped, the 7 bands were combined, per image, to produce a stacked pyramid layer usable for analysis by ERDAS. This was done by selecting the raw TIF files contained in the Level 1 download and running them through a layer stacking process.
The clipped version of the stacked pyramid layers were then processed through an unsupervised classification to assign cover types. The classification was set up to produce 16-20 classes, run 15 iterations, with a convergence threshold of 0.95. The maps for October 2008 and September 2011 were more broadly classified and contain five total classes, while all other maps contain eight to nine classes. Each was then recoded to merge all like classes. The final six vegetation maps were combined with the relating GPS.
A supervised accuracy assessment was performed to access the level of accuracy in classification per image. Each classified image was compared to a relating infrared and/or panchromatic image of the area by process of a stratified random technique to assign points. The stratified random accuracy assessment was set to select 100 random points with a minimum of 10 points per class.
The last part of the project was to produce maps which show change in vegetation over a time period. This was done by comparing a map of an earlier time period to one of a later period. The compared map sets were: October 2008 to September 2011, July 2006 to July 2010, and April 2010 to August 2010.
Results
July 2006 and 2010 maps that were classified to a broader scale to show vegetation changes over a period of time (See Figure 10, App. B). The overall accuracy assessment of July 2006 is 74.07% and July 2010 is 72.22%. October 2008 to September 2011 show generalized change detection with an accuracy assessment of72.2%. The accuracy assessment of April 2010 is 73.74% and is 78.79% for August 2010. These assessments reflect decent accuracies of vegetation classifications. However, one of the potential reasons for the assessments not being as accurate as they could be is because of cloud cover and shadows in the images affecting classifications and changes over time. Figures 1 and 2 are zoomed in to capture the focused areas to visually see correlation between vegetation changes and bear movement. The maps showing broad classification of cover type for the full extent of the study area can be seen in Appendix B. A trend toward bare ground and low growth areas can be seen in both maps.
Fig. 1: October 2008 Vegetation Map
Blue: water, dark green: coniferous, light green: deciduous, yellow: low growth, tan: bare ground/roads, black points represent June, red points represent Lily
Fig. 2: September 2011 Vegetation Map
Blue: water, dark green: coniferous, light green: deciduous, yellow: low growth, tan: bare grounds/roads, black points represent June, red points represent Lily
April to August 2010 maps changes in cover types to help see general trends during the seasonal movement patterns of the bears and also to represent a broad scale change in vegetation during the monitoring time period. The red areas represent changes from forested areas to bare ground or low vegetation. The orange areas represent changes from bare soil or low vegetation to forested areas. These changes may or may not be consistent with what changes actually occur.
Fig. 3: Change detection map from April to August 2010
Blue: water, black: unchanged vegetation, red: from forest to bare soil/roads/low vegetation, orange: from bare soil/roads/low vegetation to forest, purple points represent Lily, white points represent June
As seen in all maps in all date ranges, the movement pattern of the bears tends to be within forested areas and around bare ground and low growth areas. Using the Google Earth function through ERDAS provides a better visualization to further classify vegetation. Information provided designates a large percentage of those bare ground and low growth areas to be clear cuts at varying stages. Past research has shown that black bears will use these low growth areas as a food source and value from having management strategies that allow for these areas to be in close proximity to the forested areas which provide shelter and safety (Zager et al., 1983).
Discussion
Some problems that were encountered were with the GPS points, Landsat imagery, NAIP imagery, and ERDAS. The problems with the GPS points in 2008 were limited points over a two month period. October2010 had limited points for June and at inconsistent collection intervals.Landsat imagery can only use Level 1 data. Theformat we were able to download was GeoTIFF which provides all seven bands, and the other images are in jpeg format. The jpeg format does not have a projection and cannot be used in ERDAS with the GPS points. The only downloadable images for 2009 had clouds, so were unusable. NAIP imagery is great imagery, but the files are very large and could download, but when unzipping the files
The images are projected in NAD 83 and images from USGS are projected in WGS 84. For images from July 2006 and July 2010 were classified differently and whenrecorded ERDAS grouped some of the classes together which caused anunusable change detection map.
For planning on future projects look at the GPS data points first to help decide time ranges of imagery needed. Only download GloVis Level 1, unless the user has knowledge to use jpeg images. With limited images available in GloVis, it would be useful to know how to correct cloud cover pixels.For better classification and accuracy assessment of images the NAIP would be useful. To understand better if black bears prefer clear-cut areas better images would be needed, and data on areas that have multiple stages of clear-cut growth.
While it is known that black bears spend a lot of their time in forests, it would be interesting to find out just how much time they spend in clear-cuts and new growth areas. In order to assess time spent in certain vegetation, however, more consistent collection times of GPS points would be needed instead of the inconsistent times that we were given. While we did not have time to look at the times associated with GPS points and locations in and/or around these areas, it would be interesting if future work could be done to answer these questions. The maps and information we were able to create could be useful for future studies and management of black bears and habitats.
Acknowledgements
We would like to thank Dr. Lynn Rogers and Sue Mansfield for providing us with the GPS coordinates for June and Lily. Thank you, also, to our instructors, Joe and Lian, for the help provided during this project.
References
The Friends of Algonquin Park. "The Science Behind Algonquin's Animals - Research Projects - Black Bear." The Science Behind Algonquin's Animals - New Server. Web. 12 Oct. 2011.
Zager, P., C. Jonkel, and J. Habeck. 1983. Logging and Wildfire Influence on Grizzly Bear Habitat in Northwestern Montana. Bears: Their Biology and Management. 5:124-132.
Appendices
Appendix A:
- 2006/, July 11; TM Sensor; L4-7 Combined scene
- 2008, Oct. 04; TM Sensor; L4-7 Combined scene
- 2010, April 17; TM Sensor; L4-7 Combined scene
- 2010, July 06; TM Sensor; L4-7 Combined scene
- 2010, Aug. 23; TM Sensor; L4-7 Combined scene
- 2011, Sept. 19; ETM+ SLC-off; L4-7 Combined scene
Appendix B:
Fig. 4: July 2006 Vegetation Map
Fig. 5: October 2008 Vegetation Map
Fig. 6: April 2010 Vegetation Map
Fig. 7: July 2010 Vegetation Map
Fig. 8: August 2010 Vegetation Map
Fig. 9: September 2011 Vegetation Map
Fig. 10: October 2008-September 2011 Change Detection
Red depicts changes in vegetation
Fig. 11: Failed Change Detection Map of July 2001-July 2010
Appendix C:
Accuracy Assessments for all Images
July 2006
ACCURACY TOTALS
------
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy Accuracy
------
Unclassified 0 0 0 ------
water 11 12 11 100.00% 91.67%
wetland 2 1 1 50.00% 100.00%
coniferous15 13 10 66.67% 76.92%
mixed forest 8 12 7 87.50% 58.33%
deciduous11 12 9 81.82% 75.00%
coniferous/bare 11 10 9 81.82% 90.00%
low growth 9 10 5 55.56% 50.00%
low growth 14 11 8 57.14% 72.73%
mixed forest 0 0 0 ------
deciduous0 0 0 ------
coniferous 0 0 0 ------
bare ground/road 0 0 0 ------
low growth 0 0 0 ------
deciduous0 0 0 ------
mixed forest 0 0 0 ------
urban 0 0 0 ------
Totals 81 81 60
Overall Classification Accuracy = 74.07%
October 2008
ACCURACY TOTALS
------
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy Accuracy
------
Unclassified 0 0 0 ------
Water/Wetlands 20 18 16 80.00% 88.89%
Coniferous 23 23 18 78.26% 78.26%
Low Growth 12 16 8 66.67% 50.00%
Deciduous 19 17 13 68.42% 76.47%
Bare Ground/Road16 16 10 62.50% 62.50%
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Coniferous 0 0 0 ------
Low Growth 0 0 0 ------
Low Growth 0 0 0 ------
Deciduous 0 0 0 ------
Bare Ground/Road0 0 0 ------
Deciduous 0 0 0 ------
Deciduous 0 0 0 ------
Bare Ground/Road0 0 0 ------
Bare Ground/Road0 0 0 ------
Totals 90 90 65
Overall Classification Accuracy = 72.22%
April 2010
ACCURACY TOTALS
------
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy Accuracy
------
Unclassified 0 0 0 ------
Water 16 16 16 100.00% 100.00%
Wetland, Cloud 1 2 1 100.00% 50.00%
Wetland 18 15 11 61.11% 73.33%
Low Vegetation 6 4 2 33.33% 50.00%
Coniferous 17 21 14 82.35% 66.67%
Deciduous 15 16 12 80.00% 75.00%
Mixed, Low Vege 7 6 4 57.14% 66.67%
Bare Soil, Road 17 15 11 64.71% 73.33%
Bare Soil, Road 2 4 2 100.00% 50.00%
Coniferous 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Mixed Forest, L 0 0 0 ------
Deciduous 0 0 0 ------
Deciduous 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Totals 99 99 73
Overall Classification Accuracy = 73.74%
July 2010
ACCURACY TOTALS
------
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy Accuracy
------
Unclassified 0 0 0 ------
Water 12 12 11 91.67% 91.67%
shadows/wetland 10 10 9 90.00% 90.00%
coniferous 15 13 8 53.33% 61.54%
Mixed forest 10 11 5 50.00% 45.45%
coniferous 13 13 10 76.92% 76.92%
coniferous/bare 8 10 4 50.00% 40.00%
urban 9 11 9 100.00% 81.82%
low growth 13 10 9 69.23% 90.00%
coniferous 0 0 0 ------
bare ground/road0 0 0 ------
roads/ low grow 0 0 0 ------
deciduous0 0 0 ------
deciduous0 0 0 ------
low growth 0 0 0 ------
clouds 0 0 0 ------
clouds 0 0 0 ------
Totals 90 90 65
Overall Classification Accuracy = 72.22%
August 2010
ACCURACY TOTALS
------
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy Accuracy
------
Unclassified 0 0 0 ------
Water 15 16 15 100.00% 93.75%
Wetland, Cloud 1 1 0 0.00% 0.00%
Wetland 14 12 8 57.14% 66.67%
Low Vegetation 2 2 1 50.00% 50.00%
Coniferous 24 26 19 79.17% 73.08%
Deciduous 22 20 18 81.82% 90.00%
Mixed Forest, L 9 9 6 66.67% 66.67%
Bare Soil, Road 10 11 9 90.00% 81.82%
Bare Soil, Road 2 2 2 100.00% 100.00%
Coniferous 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Mixed Forest, L 0 0 0 ------
Deciduous 0 0 0 ------
Deciduous 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Bare Soil, Road 0 0 0 ------
Totals 99 99 78
Overall Classification Accuracy = 78.79%