Mapping nonnative plants using hyperspectral imagery

Download Report

Transcript Mapping nonnative plants using hyperspectral imagery

Mapping nonnative plants using
hyper spectral imagery
Emma Underwood
Susan Ustin
Deanne Dipietro
Outline
 Introduction
 Methodology
 Results
 Discussion
 Conclusions
Introduction
 Nonnative plants : are plants that have been
introduced in to the environment intentionally or
accidentally by human activity.
 Threaten - global diversity
- ecosystem functioning
 Pimentel et al (2000) estimated that 50,000
invasive species have been introduced into the U.S.
causing economic losses of $137 billion per year;
approximately $35 billion annual cost for plant
invasions alone.
 The global extent and rapid increase in invasive
species is recognized as a primary cause of global
biodiversity loss.
 Nowhere is the ecological threat more clearly seen
than in California where more than 1025 plant
species have been added to the flora (Rejmánek &
Randall, 1994).
Ice plant(Carpobrotus edulis)
 Common ice plant
(Carpobrotus edulis), is an
exotic plant species that
invades coastal plant
communities from the
North Coast of California
to Mexico.
 CalEPPC (1999) rated C.
edulis as an A-1 species
(The Most Invasive Wild
land Pest Plant:
Widespread).
Jubata grass (Cortaderia jubata)
 Poses a significant
threat to
mediterranen
ecosystems
 prolific wind dispersed
seeds
 tolerance of a broad
range of habitats and
 its competitiveness for
light,moisture,and
nutrients(Cowen,1976).
Key requirement
Delineation the spatial extent
Severity or intensity of invasion
Provides
 baseline for monitoring future expansion
 assists in identifying targets for control
activities
Remote sensing-invasive plants mapping
 Offers significant opportunities for providing
timely information on invasions of nonnative
species into native habitats.


Larger spatial area
Short period of time
 Researchers exploit unique phenological, spectral,
or structural characteristics of the nonnative
species in the image to distinguish them from the
mosaic of species around them.
 Two ways


High spatial ,but low spectral resolution-black &white or color infrared aerial
photographs
Digital images with greater spectral resolution although coarser spatial
resolution.
 Aerial photography - inexpensive,
- large amount of archival data
- very fine spatial resolution
But
-
relies on nonnative plant
possessing visually detectable
unique characteristics
- Extensive manual labor
- Can collect data over a
relatively small spatial area
 Digital multispectral imagery- automated image
processing
- large spatial
coverage
 Successful applications of AVHRR, TM and ADAR
However, invasive species populations
Can be detected only after they become dense and
widespread.
AVIRIS (Airborne Visible/infrared imaging
spectrometer)
 Increased spatial




resolution
Fine spectral
resolution
224 spectral bands
400-2500nm at 10nm
resolution
20 m spatial resolution
when flying at 11km.
Limitations of using
traditionally available
wavebands
- with color infrared
green vegetation is
observed as shades of
red.
Objective
 To investigate the use of AVIRIS imagery to
detect the invasive species ice plant (Carpobrotus
edulis) and Jubata grass(Cortaderia jubata).
 To compare three techniques for processing the
imagery : minimum noise fraction(MNF),Continuum
removal ,and band ratio indices.
Study site
 VAFB is used primarily for
developing and testing
missiles and satellite
launches for the
Department of Defense
and NASA.
 836 Vascular plantsquarter of them are
invasive species.
 C.edulis and C.jubata have
invaded two native
community types :coastal
dune scrub community and
maritime chaparral.
The focus of this research is the
encroachment of ice plant and jubata
grass in to these native communities
and specifically on the ability of
AVIRIS to identify pixels of different
densities of these species.
Field work at VAFB
5 community types
Coastal Scrub-iceplant
Intact Coastal Scrub
Chaparral-iceplant
Intact chaparral
Pampas-chaparral
Methodology
Pre-processing
- Atmospheric correction
- Masks( NDVI > 0.2)
Processing
- Minimum Noise Fraction
- Continuum Removal
- Band Ratios
Compared processing techniques
- Ability to delineate iceplant extent
- Detect iceplant density
- Assess ease of processing and repeatability
Image Classification
 Supervised Classification : requires the user to
decide which classes exist in the image, and then
These samples known as training areas and are
then input into a classification program, which
produces a classified image.
 Type of supervised Classification:Maximum
likelihood classifier
Maximum likelihood classification calculates the
probability that a given pixel belongs to a specific
class.
 Unsupervised Classification: when features are
separated solely on their spectral properties .
Processing Techniques
 Minimum Noise fraction(MNF) : Often used
method for reducing redundancy of information
between bands
 reduce and compress the data
 increase speed of processing
Continuum removal: is a procedure that facilitates
the distinction of similar absorption bands in hyper
spectral curves.
Band ratio indices
 Emphasize the biochemical and the
biophysical properties of the vegetation
contained in physiologically important
bands.
Definitions
 Confusion matrix : is a square array of numbers
set out in rows and columns which express the
number of sample units(i.e.,pixels,clusters of
pixels etc..) assigned to a particular category
relative to the actual category as verified on the
ground.
 Overall accuracy-the total number of correctly
classified samples( i.e,the major diagonal) divided
by the total number of sample units in the entire
matrix.
 Producer’s accuracy :is the number of correctly
classified samples of a particular category divided
by the total number of reference samples from
that category.
 User’s accuracy : is the number of correctly
classified samples of a particular category divided
by the total number of samples being classified as
that category.
Results
Comparison of 3 processing techniques
Results
 Overall accuracy-MNF performed best
 User’s accuracy-is important
User’s accuracy-MNF lowest
 MNF is able to produce a more accurate map of
multiple vegetation classes
 Continuum removal and band ratio techniques are
suited for detecting species with distinct
characteristics.

Discussion
 Evaluated three approaches- accuracy
- logistics of processing
- ease of interpretation
 Confusion matrices demonstrated that MNF
performed best for classifying all five vegetation
communities.
– However,in terms of classifying one of the target
species,ice plant,the continuum removal and the
band ratio methods performed better.
Key findings..
MNF
 Worked well identifying different densities of ice plant
X Difficult to interpret
Band Ratio
 Intuitive
Continuum Removal
 Good results for presence / absence of ice plant
 Easy to use
X Speckled result, accuracy not improved when lumped
Conclusions
The benefit of this research has been to
contribute to the knowledge base of land
managers at VAFB
 by providing improved information on the
spatial extent and the density of the ice plant
and jubata grass, which will lead to better
protection of the native biodiversity.