Land Cover Mapping

1
Land Cover Mapping
John H. Lowry, Jr. R. Douglas Ramsey Ken
Boykin David Bradford Patrick Comer Sarah
Falzarano William Kepner Jessica Kirby Lisa
Langs Julie Prior-Magee Gerald Manis Lee
OBrien Keith Pohs Wendy Rieth Todd Sajwaj Scott
Schrader Kathryn A. Thomas Donald Schrupp Keith
Schulz Bruce Thompson Cynthia Wallace Cristian
Velasquez Eric Waller Brett Wolk
USU Logan
CDOW BLM NARSC Denver
EPA Las Vegas
CPFS Flagstaff
NMCFWRU Las Cruces
2
Gap Objectives

• Map the distributions of natural communities.
• Map predicted habitat of native animal species.
• Map the degree of management for biodiversity
  maintenance of land tracts and water bodies
  focusing on intent.
• Analyze the representation of biotic elements in
  the conservation network to identify gaps in
  long-term security.
• Provide this information to the public and those
  entities charged with land use research, policy,
  planning, and management.

3
Utah GAP Analysis (Edwards, Homer, Ramsey and
Falconer 1995)
4

• State-based vegetation classification systems
  (cover type legends)
• State-based mapping methods
• State-based mapping area

5
5 states map Salt Desert Shrub
State Land Cover Classes roughly equivalent to Salt Desert Shrub
Utah (1992) Salt Desert Shrub, Salt Desert Shrub-Mix
New Mexico (1996) Broadleaf Deciduous Desert Shrub
Nevada (1997) Salt Desert Shrub
Colorado (2000) Desert Shrub, Saltbrush Fans Flats
Arizona (1999) Shadscale-Mixed Grass-Mixed Scrub, Saltbush Scrub, Winterfat-Mixed Scrub
6
State-based land cover mapping efforts
1992 36 Cover Types
2000 52 Cover Types
1999 52 Cover Types
1996 42 Cover Types
Mixed Salt Desert Shrub
7
Objectives for Southwest Regional GAP

• Regionally standardized data and mapping methods
  
• Regionally consistent land cover legend
  (vegetation classification system)
• Eco-regional emphasis rather than state-based
  emphasis
• Improvements in vegetation map product

8
Mapping Scale Spatial Extent

• 528,000 square miles
• 340,000,000 acres
• 1.5 Billion pixels
• 1/5th of conterminous US
• 40 mapping zones
• Eco-regionally distinct
• Spectrally consistent
• Approximately state-based

9
Mapping Scale Spatial Resolution

• 5 state coverage 85 Landsat ETM 7 scenes
• 30 meter resolution pixels

10
Mapping Scale Thematic Resolution
NatureServe Ecological Systems
NVC Formation
NVC Alliance
NVC Association
NVC Class/Subclass
1,800 units
10 units
5,000 units
700 units
300 units
MRLC 2000 Proposal
Gap Analysis Program
National Park Mapping
(Natural/Semi-natural types)
(Slide Courtesy Pat Comer, Nature Serve)
11
National Vegetation Classification (NVC)
physiognomic levels Physiognomic Class
(woodland) Physiognomic Subclass (deciduous
woodland) Physiognomic Group (cold-deciduous
woodland) Subgroup (natural/semi-natural) F
ormation (temporarily flooded cold-dec. woodland)
Ecological System (Rocky Mt. Lower
Montane Riparian Woodland and Shrubland)
floristic levels Alliance (Populus
deltoides Temp. Flooded Woodland)
Association (Pop. deltoides / Distichlis spicata
Woodland)
12
Thematic Resolution Ecological Systems
Groups of plant communities and sparsely
vegetated habitats unified by similar ecological
processes, substrates, and/or environmental
gradients.
Inter-Mountain Basins Subalpine
Limber-Bristlecone Pine Woodland
Dominant Specie(s)
Lifeform
Region
Environmental Setting
13
Bioclimatic divisions differentiate Ecological
Systems
Colorado Plateau
Great Basin
Great Basin Pinyon-Juniper Woodland Ecological
System Pinus monophylla and Juniperus
osteosperma
Colorado Plateau Pinyon-Juniper
Woodland Ecological System Pinus edulis and/or
Juniperous osteosperma
14
Similar substrates group Alliances
Four-wing Saltbush Alliance
Shadscale Shrubland Alliance
Inter-Mountain Basins Mixed Salt Desert Scrub
Ecological System
15
Importance of Appropriate Mapping Scale
Spatial Extent (size of an area to be mapped)
Spatial Resolution (pixel size) Also
MMU (aggregation of pixels)
Thematic Resolution (level of thematic detail)
16
Methods

• Predictor Data
• Imagery Derived
• Ancillary Data

17
Regionally Standardized Predictor Data

• 85 Landsat 7 scenes
• Three seasons (spring, summer, fall)
• 1999-2002
• Region-wide 30 meter DEM
• Region-wide 10 class landform

18
Predictor Datasets Imagery Derived
July-Aug
Sept-Oct
ETM Bands 5, 4, 3
ETM Bands 5, 4, 3
19
Predictor Datasets Elevation
20
Summary of Predictor Datasets
Imagery Derived DEM Derived Other
NDVI or SAVI Elevation Geology
Brightness (T-cap) Aspect Soils
Wetness (T-cap) Slope Hydrology
Greenness (T-cap) Landform
Individual Bands

21
Training Sites

• Sample Variability of the landscape
• Ecological
• Spectral
• Topographic gradients
• In order to train the predictor data

22
Training Samples Strategy

• Ground-based opportunistic field sampling
• Sufficient data to assign an Alliance label to
  each site

23
Geographical Stratification
24
GPS CENTER POINT
WITH THE ENTRY OF GPS COORDINATES, THE VIEW ZOOMS
TO THE GPS LOCATION AND PLACES AN OPEN CIRCLE
CENTERED ON THE GPS POINT
25
(No Transcript)
26
A minimum of two digital photos are taken of each
site
Photos are named using the field site ID format
with a numeric modifier to identify photo 1, 2,
etc.
UT081300GM03-1.jpg
UT081300GM03-2.jpg
27
Training Data Sources
3000 air photo interpretation sites US Forest
Service
28
Training Data Sources
3000 air photo interpretation sites US Forest
Service
5200 Sites from other organizations (USGS
Landfire BLM)
29
Training Data Sources
3000 air photo interpretation sites US Forest
Service
5200 sites from other organizations (USGS
Landfire BLM)
7800 field work RSGIS Lab in collaboration
with BLM UDWR
16000 total sample sites
30
Regional Total 93,000
31
Classification Method Classification Trees

• Data-mining software for decision-making and
  exploratory data analysis
• Identifies complex relationships between multiple
  independent variables to predict a single
  categorical class
• Predictor variables may be categorical or
  continuous
• Recursively splits the predictor data to create
  prediction rules or a decision tree.

32
Mining the Predictor Layers
33
(No Transcript)
34
Simplified Example Splits on 2 variables
35
Simplified Example Tree output for 2 variables
36
Comparison with 1995 Utah GAP Vegetation Map

• Utahs Great Basin Eco-Region ( 17.5 M acres,
  300 x 120 Miles)
• Approximately 5 mosaicked Landsat 7 scenes
• 3000 sample sites (1700 USU/1300 other sources)

37
Comparison for Great Basin Eco-Region (Partial
List)
1995 Utah GAP Cover Types 2003 SWReGAP Cover Types (Preliminary)
Agriculture Cultivated Crops/Irrigated Agriculture
  Pasture/Hay/Non-irrigated Agiculture
Barren Playa
  Cliff and Canyon
  Active and Stabilized Dunes
  Volcanic Rock and Cinderlands
Desert Grassland Semi-Desert Grassland
Grassland Invasive Perennial Grassland
  Invasive Annual Grassland
  Invasive Annual Forbland
Greasewood Greasewood Flats Complex
Pinyon-Juniper Pinyon-Juniper Woodland
Pinyon  
Juniper  
Sagebrush Big Sagebrush Shrubland
  Xeric Mixed Sagebrush Shrubland
Sagebrush/Perennial Grass Montane Sagebrush Sagebrush Steppe
  Semi-Desert Shrub Steppe
Salt Desert Scrub Mixed Salt Desert Scrub
Urban Low Density Developed
  Medium-High Density Developed
Wetland Arid West Emergent Marsh
38
1995 GAP Vegetation Map
2003 GAP Veg. Map (preliminary)
39
Park Valley Example
40
1995 GAP 30 M
2003 GAP 30 M
1995 GAP Pub.1KM
41
Clarkston Example
Clarkston
42
1995 GAP 30 M
2003 GAP 30 M
1995 GAP Pub.1KM
Clarkston
Clarkston
Clarkston
43
Edge-matching between three mapping areas
44
(No Transcript)
45
125 Land cover classes 109 Ecological
Systems 1-acre min. mapping unit
46
(No Transcript)
47
Validation, Accuracy and Appropriate Uses

• Accuracy Assessment when map is completed
• Internal Validation concurrently with mapping
  effort
• Large landscape monitoring and planningscales of
  1 100 k 1 250k

48
Assessing Map Quality

• Consider our approach Internal Validationhold
  out with replacement
• Offers quantitative measure of model predictive
  ability for most modeled classes
• With replacement assumes final map is better
  than the assessment indicates
• Performed on each mapping zone independently
• Objective to produce the best map possible with
  available training data

49
Accuracy Assessment with 20 withheld data
Southern Wasatch Range
50
Overall Validation (sum of the matrices)

• Combined error matrices for all mapping zones
• Ignored classes with lt 20 validation samples or
  unevenly validated in the region
• Total samples 17,030
• Overall agreement 61
• Kappa statistic 0.60
• Representing
• 85 of 125 classes
• 91 of land area

51
Focus on Ecological Similarity
Objective
Re-examine error matrix within the context of
ecological similarity between classes
52
Validation Results 3 Similarity Classes
Original Matrix
53
Data Availability Distribution

• Data by
• Region, State, Ecoregion
• User defined box
• ArcInfo Imagine
• Database of legend descriptions with photos
• Databases of field samples with photos
• Map quality assessment
• Error matrices
• Qualitative assessments
• Detailed methods description documents
• Predictors used
• Samples used
• Etc.

• http//earth.gis.usu.edu/landcover.html

More Information