using spectral data to discriminate land cover types
TRANSCRIPT
Using spectral data to discriminate land cover types
Multispectral and Hyperspectral Imaging for Land Cover and Habitat
Mapping
VegetationAll of the plant life in a particular region (e.g., the vegetation
of Wyoming) or period (e.g., Pleistocene vegetation)Examples: Lodgepole pine forest; sagebrush steppe; mixed grass
prairie
Land CoverEverything that occupies the land surface including human
disturbance and unvegetated (barren) areasExamples: Lodgepole pine forest, urban, unvegetated sand dunes
HabitatThe area that an organism occupies and the type of
environment an organism or population of organisms need to survive.Examples: Lodgepole pine forest; sagebrush within 5 km of water
To review…
Vegetation – lodgepole pine trees + understory species together comprise a type of vegetation
Sagebrush and associated plants are another
Land cover includes vegetation or bare dunes or other surface types
Habitat – may include access to water
Habitat – can require particular spatial arrangement of patches
General procedures are the same (see air photo lecture)
Satellite based mapping is usually digital, but can be manual like with air photos
Satellite mapping usually relies more on spectral information than does mapping from aerial photos
Satellite based mapping is often for larger areas than aerial photo mapping and at coarser spatial resolution
Air Photos vs. Multi/Hyperspectral Mapping
Large area coverageFrequent and regular return timesRadiometric and geometric consistencyHigher spectral resolutionMany custom digital tools available for
analysis
Other advantages of satellite data for land cover mapping
Satellite Mapping Disadvantages/Challenges
Resolution trade-offs (e.g., lower spatial resolution) though this is changing
Spectral confusion sometimes does not allow us to distinguish cover types that users need to map
Return time sometimes too slow (Landsat 16 days)
Sensor failures out of our controlLess control over mission parameters
Spectral Confusion
Spectral confusion in overlap zone even though peaks are distinct. Sometimes there is LOTS of overlap
The biggest advantage of satellite data is the spectral information they containTo distinguish cover types with satellites they
must be spectrally different from one another
If cover types are spectrally ambiguous (spectral confusion), we must include other data to accurately distinguish them (e.g., environmental data)
So how do we map different classes of vegetation, land cover, and habitat using satellites?
How does one group image pixels into predetermined thematic classes when they may or may not be spectrally distinct from other thematic classes?Requires thoughtful choice of classesRequires knowledge of the spectral properties of
classes including spectral variabilityRequires knowledge of sensor characteristicsRequires knowledge of spatial-environmental
relationships of typesRequires a great deal of patience to sort out all
of the above and apply the knowledge logically
The classification problem:
Image classification questionsWhat factors (e.g., reflectance, environmental
relationships, context, history, etc.) make one feature of interest different from another?
Can you capture the differences accurately with obtainable spatially distributed data?
How can you best exploit the differences between features statistically or otherwise?
What is Image Classification?Image classification includes all of the steps
necessary to group pixels into thematic classes – self similar groups representing some common feature
Classification tries to use all necessary and available information to distinguish classes
Note that “image classification” is the process of making a map. A “classification scheme” is a list of types in the map legend.
Digitally creating groups of pixels that (hopefully) correspond to similar cover types on the ground
Two main strategies:Unsupervised classification – group pixels
using the pixel values to determine natural spectral groups in the data
Supervised classification – group unknown pixels with pixels representing known types from the ground by measuring their similarity to the known pixels.
Per pixel classification of digital imagery
Use many bands simultaneously (multivariate) to create a map of classes
Classification
Land cover/habitat mapping in Wyoming using satellites
Landsat Image – Pinedale 2005
First satellite based statewide mapping in WyomingEarly map by Fred Porter and others at the
Wyoming State Geological SurveyBased on MSS dataProduct was a paper map of the stateCourse resolution of types Availability?
Gap Analysis Land Cover Statewide coverageCompleted in early 1990s using 1989 imageryHand digitized from Landsat TM imagesCoarse MMU (1 km2)More detailed list of cover types (41) than
earlier statewide mapDigital data still available (WyGISC) and still
used, though recently replaced with a new regional Gap Analysis
1st Wyoming Gap Land Cover Map
Wyoming Maps in National ProgramsNational Land Cover Database (NLCD 2001,
2006, and 2011 complete and downloadable (www.mrlc.gov)
Landfire (completed 2001, revised 2010, and updated regularly)
Gap Analysis (ReGap). Complete for Wyoming as part of the Northwest Regional Gap Analysis.
NLCD 2011All of Wyoming
Broad cover types
NLCD 2006 for Wyoming (from MRLC Viewer)
NLCD 2006 Laramie
Landfire (2012) for Wyoming
Covers entire U.S. (See www.landfire.gov)
Wyoming ReGap Data
Wyoming Mapping in State AgenciesRecent mapping projects with goal of
improving resolution of statewide map productsSpatial resolution generally at full TM/ETM+
resolution (30 m pixels) but sometimes aggregated to slightly larger MMUs (2 acre)
Resolution of cover types with more detail than GAP and other programs
Local field input to improve accuracy
Primary PlayersWyoming Game and Fish Dept.Wyoming State BLM OfficeU.S. Forest Service (National Forests only)WyGISCOther agencies and private contractors
Biggest challenge in Wyoming is matching what we can do over large areas with satellites to what is needed by land managers on the groundResolution mismatchesSpectral confusionAccuracy problems
Land managers also need land cover structural information (cover density, height of vegetation, etc.)
Habitat managers need to integrate temporal differences in habitat – e.g., crucial winter range vs. summer range, etc.
Wyoming Challenges
Combining spectral mapping with environmental modeling can help us solve spectral confusion
Future challenge will be to understand the influence of land use and disturbance history on the distribution of land cover.
Addressing challenges