GIS Modeling Approach to Forest Habitat Mapping
GIS Modeling Approach to Forest Habitat Mapping
I. History/Background
II. Methodology
III. Results
IV. Future Considerations
What is a Forest Habitat Type?
• Alexander Von Humboldt
• Braun-Blanquet system in Europe
• Minnesota DNR Native Plant Communities
• Dr John Kotar at University of Wisconsin Madison
John Kotar’s Forest Habitat Type
Classification for North Central Minnesota
• Simplified classification for forest managers
• Based on late-successional plant communities
• Aids in management decisions – Ecologically Based
Forest Management
Goals and Objectives
• Creation of an automated method for producing ecosystem
(habitat type) maps
• Demonstrate the use of Visual Basic and ArcGIS as a tool
for habitat type mapping
• Quantify the accuracy of two different methods of habitat
type mapping: the raster and the vector method
Why Create a Forest Habitat Model?
• 17,000 total cover type polygons
• 3-5 days for each township, with over 30 townships total
Study Area
Forest Habitat Type Model Requirements
• Soils – SSURGO
• Cover Type
• Habitat type sample points – stratified systematic
random sample based on cover type
• Digital Elevation Model – Raster
• First Law of Geography
Vector Habitat Typing Model
• Pivot tables showing probability of each habitat type
occurring on which soil type and cover type
*Courtesy of Cheryl Adams Blandin Forestry
Cover Type by Habitat Type Pivot Table
Habitat Type AAbAa AbArAo AbArV AbArV-Ly AbFnAu AbFnThAn
Red Pine 2 0 0 0 0 0
Jack Pine 0 1 0 0 0 0
Balsam Fir 5 9 9 14 10 0
White Spruce 25 3 2 3 10 0
Upland Black Spruce 12 6 1 2 6 0
Black Spruce 0 3 4 2 1 2
Tamarack 0 2 0 0 0 1
N White Cedar 0 0 1 1 3 12
N Hardwoods 52 20 1 1 10 0
Cover Type
Lowland Hdwds 0 3 0 0 40 35
Balm O' Gilead 3 2 3 0 23 2
Aspen 194 262 41 38 177 2
Birch 10 26 5 3 8 0
Lowland Grass 0 0 0 0 0 0
Lowland Brush 0 0 0 0 0 1
Marsh 0 0 0 0 1 0
Stagnant Spruce 0 0 0 0 0 0
Grand Total 303 337 67 64 289 55
• Use the top two
Vector Model
Four Step Process
1. Join the habitat type points layer with the cover
type layer
2. Assign the “easy” ones
3. Add based on adjacency, soil type, and cover
type
4. Add based on only soil type, and cover type
Step One – Spatial Join
Step Two – Add Known Types
• Add all of the Marsh, Muskeg, Lowland Grass, Lowland
Brush, etc.
Step Three – Adjacency
• Query soil type layer to select one particular soil
type
• Query habitat type layer to select the polygons
that have already been habitat typed
• Select all the polygons that touch the edge of the
already habitat typed polygons
• From that group select all the polygons with a
certain cover type
• From that group select all the polygons that have
their centers in the selected soil types
• Assign the appropriate habitat type
Step Three – Adjacency
Step Four – Only Soil and Cover Type
• Query the soil type layer to select a certain soil
type
• Query the habitat type layer to find all of the
polygons that have their centers within the certain
soil type
• Select FROM THE SELECTED the polygons that
have the cover type being looked for
• DESELECT all of the polygons that have already
been habitat typed
• Add the habitat type
Step Four – Only Soil and Cover Type
Sample and Verification Points
• Three sampling schemes: 30, 40, and 50%
stratified random sample based on PLS section
Percentage # of Points for Model Creation # of Points for Verification
30 61 138
40 80 119
50 109 90
Results
Fifty Percent Sampling Rate
• Classification based on 109 samples
• Overall assigned habitat types percentage: 81%
• 26 samples classified correctly, 64 classified
incorrectly, 90 total samples
• 26/90 = 29% correct, 64/90 = 71% incorrect, Khat =
0.1676
Results
Results
Forty Percent Sampling Rate
• Classification based on 79 samples
• Overall assigned habitat types percentage: 79%
• 31 samples classified correctly, 88 classified
incorrectly, 119 total samples
• 31/119 = 26% correct, 88/119 = 74% incorrect,
Khat = 0.1485
Results
Results
Thirty Percent Sampling Rate
• Classification based on 61 samples
• Overall assigned habitat types percentage: 79%
• 35 samples classified correctly, 84 classified
incorrectly, 138 total samples
• 35/138 = 25% correct, 84/138 = 75% incorrect,
Khat = 0.1362
Results
Accuracy Measure by Sampling Percent
Accuracy Measures 50% 40% 30%
Percentage of Polygons Assigned a Habitat Type 81% 79% 79%
Overall Accuracy 29% 26% 26%
Number of Plots Classified Correctly 26 31 35
Number of Plots Classified Incorrectly 63 87 101
Percentage Classified Correctly 29% 26% 26%
Percentage Classified Incorrectly 71% 73% 73%
Khat 0.1676 0.1484 0.1362
Percentage Omission 71% 74% 74%
Percentage Commission 6% 5% 5%
Raster Model
Raster Modeling Technique
1. Convert cover type and soil type to raster
2. Reclassify the cover type layer to discard all of
the areas that won’t be considered for forest
management
3. Add together the cover type and soil type.
Separate out cover type/soil type combinations
that could be classified into more than one
habitat type – reclassify the slope to a 1-10
scale
4. Add together slope and cover type/soil type
combinations – reclass to meaningful scale
5. Add all of the multi habitat type combinations
with the single habitat type codes
6. Reclassify to meaningful scale and apply
majority filter
Step One – Cover Type Reclassify
Possible Soil Type Cover Type Combinations
Possible Values V1 V2 V3 V4 V5 V6 V7 V8 V9 V10
AAbAa 46 47 66 67 71 72 73 74 76 77
AbArAo 66 68
AbArV
AbArV-Ly 26 62
AbFnAu
AbFnThAn 29 35 46 52
AbFnThAn-Moss 29 35 46 52
AbPiG
AbPiPl 11 12
AbPiV 30 66
AbThLe 27 28 29 33 34 35
AbThSp 33 35
ATiCa 38 46 58 66 65 73 76 68
ATiPo 46 47 66 67 73 74
FnLa 45 49 52 56
FnTiAt 46 66
PmCh 27 28 33 34
PmPl 27 28 33 34
Step Two – Reclass Slope and
Add to Individual Cover/Soil Combinations
Step Two
Step 3 – Add Together All Output from Step 2
Along With Output from Step 3
Step 5 – Reclass to Meaningful Output and
Apply Majority Filter
Raster Results
Accuracy Measures 50% 40%
Unfiltered 4 Filter 8 Filter Unfiltered 4 Filter 8 Filter
Overall Accuracy 21% 21% 21% 18% 19% 20%
Number of Plots Classified Correctly 15 15 15 18 18 19
Number of Plots Classified Incorrectly 56 56 56 79 78 77
Percentage Classifed Correctly 21% 21% 21% 18% 19% 20%
Percentage Classifed Incorrectly 79% 79% 79% 92% 81% 80%
Khat 0.07856 0.07792 0.07792 0.0753 0.07521 0.0874
Percentage Omission 79% 79% 79% 81% 81% 80%
Percentage Comission 7% 7% 7% 7% 7% 7%
Accuracy Measures 30% All
Unfiltered 4 Filter 8 Filter Unfiltered 4 Filter 8 Filter
Overall Accuracy 23% 23% 24% 22% 23% 23%
Number of Plots Classified Correctly 25 25 26 36 36 37
Number of Plots Classified Incorrectly 83 82 81 125 124 123
Percentage Classifed Correctly 23% 23% 24% 22% 23% 23%
Percentage Classifed Incorrectly 77% 77% 76% 78% 77% 77%
Khat 0.11632 0.11677 0.12754 0.10144 0.10149 0.10853
Percentage Omission 77% 77% 76% 78% 78% 77%
Percentage Comission 6% 6% 6% 6% 6% 6%
Comparison of the Accuracy Measures Between
the Raster and Vector Model
Accuracy Measures 50% 40% 30%
Vector Raster Vector Raster Vector Raster
Overall Accuracy 29% 17% 26% 16% 25% 19%
Number of Plots Classified Correctly 26 15 31 19 35 26
• Excludes verification points that fell on 'NoData'
pixel values
• Dark areas (Not the Same) are most commonly lowland areas that
were removed from the raster model
Future Considerations
• Increased sampling points
• Vector model order dependency
• Geographic extent of study area
– Soil types 619 and 622B missing
– Occur with: AbArV, AbArV-Ly, AbFnAu, AbPiV,
AbPiG, and FnTiAt
• Polygons with multiple sample points
Thanks and Questions?
Thank you to Cheryl Adams of UPM Blandin Forestry
for all of the data used in this study. I would also like
to thank Misha Blinnikov, Jorge Arriagada, and Benjamin
Richason III for all of their guidance, suggestions, and
motivation.
Jesse Adams
JSA GIServices
5406 Oneida Street
Duluth, MN 55804
Phone: 218.525.1172
Fax: 218.525.1382
Mobile: 608.215.8464
Email: jesse@jsagiservices.com
Website: www.jsagiservices.com