• No results found

Diverse Uses of Hyperspectral Remote Sensing for Mapping and Monitoring of Wetlands

N/A
N/A
Protected

Academic year: 2021

Share "Diverse Uses of Hyperspectral Remote Sensing for Mapping and Monitoring of Wetlands"

Copied!
36
0
0

Loading.... (view fulltext now)

Full text

(1)

Shruti Khanna

Alexander Koltunov, Maria J. Santos, Erin L. Hestir,

Paul J. Haverkamp, Mui C. Lay & Susan L. Ustin

Center for Spatial Technologies and Remote Sensing (CSTARS)

Department of Land, Air & Water Resources

University of California, Davis

Remote Sensing Workshop for Water Resources Management

San Diego, California,

September 28 & 29, 2012

Diverse Uses of Hyperspectral

Remote Sensing for Mapping

and Monitoring of Wetlands

(2)

Track invasive species, monitor

spread and persistence

Map contamination after an

environmental disaster

Monitor stress and recovery in

wetland plant communities

Applications of Remote

Sensing

(3)

Track invasive species, monitor

spread and persistence

Map contamination after an

environmental disaster

Monitor stress and recovery in

wetland plant communities

Applications of Remote

Sensing

(4)

California

Area in detail below

Spatial

Large spatial extent (2600km

2

)

High spatial resolution data (3m)

Temporal

Multiple years (June 2004 - 2008)

Spectral

Spectral resolution (BW: 10 – 15 nm, 126 bands)

Spectral extent (400 to 2400 nm)

What is needed?

Multi-temporal images

Hyperspectral bands

Good spatial resolution

0 6 18 Km

The Sacramento –

San Joaquin Delta

(5)

Water hyacinth water absorptions

Water Hyacinth vs. Pennywort/Primrose

Water primrose water absorptions

R

ef

lec

tanc

e

* 10000

Wavelength in µm

(6)

Water hyacinth

Pennywort

Water primrose

2007 – NDII

Water hyacinth vs. Pennywort/primrose

Class

threshold

Value of NDII

No. of pi

x

el

s

of eac

h

s

pec

ies

(7)

spread

persistence

change in distribution

Track

(8)

June

2007

Increase in water hyacinth cover in Stone Lake

from 2004 to 2008

June

2004

June

2008

Water SAP Emergent Soil / Dry Veg Water Hyacinth Pennywort Water Primrose

(9)

Water

Submerged Aquatic Plants

Emergent (tule/cattail)

Soil / Dry Vegetation

Water Hyacinth

Pennywort

Water Primrose

Water hyacinth to

pennywort conversion in

Fourteen Mile slough

from 2004 to 2008

June

2007

June

2004

June

2008

(10)
(11)

Track invasive species, monitor

spread and persistence

Map contamination after an

environmental disaster

Monitor stress and recovery in

wetland plant communities

Applications of Remote

Sensing

(12)

The Macondo Oil Spill

Barataria

Bay

Terrebone Bay Chandeleur Islands Bird’s Foot

Biggest coastal oil spill in the US

One of the five biggest oil spills in

the world by volume

Map courtesy of National Geographic (surface oil) and

modified by Commission staff, NOAA/Coast Guard SCAT map (shoreline oiling)

(13)

Hydrocarbon

absorption

features

used to

detect oil

(Kokaly et al. 2012)

1720 nm

2300 nm

(14)

Oiled Shoreline in Barataria Bay

(Kokaly et al. 2012)

(15)

Track invasive species, monitor

spread and persistence

Map contamination after an

environmental disaster

Monitor stress and recovery in

wetland plant communities

Applications of Remote

Sensing

(16)

Erosion risk along

retreating vegetation edge

Sharp

transition

from green

to dead

(17)

NDVI

Response in

Oiled Areas

Oiled marsh

edge

(wide swath of dead and dying

vegetation)

Oil-free

marsh

edge

Black polygon

is the oiled

area

Oiled Pixels

(18)

mNDVI

ANIR

NDII

ADW at 940nm

Index trajectories w.r.t to distance to the

shore along

oiled

and oil-free shores

Plant stress due to oil contamination

(19)

0

10

20

30

40

50

0

2

6

10

14

18

26

%

s

oi

l/

N

P

V

pi

x

el

s

i

n 2010

%

r

ev

eget

at

ed

in 201

1

Average distance to shore (m)

2010

2011

Unrecovered

saltmarsh

Recovered

saltmarsh

Change detection to

determine recovery

after an year

(20)

Why remote sensing?

Measures direct biophysical processes

e.g. loss of chlorophyll and other pigments, water stress

Provides Geospatial Information

Provides wall-to-wall coverage

Samples difficult to reach locations

Multi-temporal data allows

Monitoring change in distribution, spread and persistence

Monitoring stress and recovery

(21)

LDCM

(Landsat Data Continuity Mission and

HyspIRI

(Hyperspectral InfraRed Imager):

Future of Satellite Remote Sensing

Spectral extent

UV to Thermal

380 to 2500 nm 420 to 2450 nm

Bandwidth

broadband (variable) 10 nm

6.5 to 10 nm

Number of bands 11 bands

210 bands

244 bands

Spatial resolution 15m (pan), 30m

60m

30m

Temporal resolution 16 days revisit

19 days revisit

on demand

LDCM

HyspIRI

Specifications:

(22)

HyspIRI,

2020

Detect pigment loss and

water stress

Map communities and

functional types

Reduce uncertainties in

ecosystem composition

feedbacks

Improve inputs to climate

models

Applications

LDCM,

Feb 2013

Map current and historical extent of

invasive species (depends on patch

size, e.g. tamarisk, ice plant)

Detect links between climate,

weather and invasive species

distributions

Detect land cover change over time

(Four decades of data)

Help in climate change forecasting

(23)

Acknowledgements

Field crews of CSTARS, CDBW, CDFA and USGS, Baton Rouge, LA for collection of field data

Mui Lay, Jaylee Tuil, Sinzee Tran & Shawn Kefauver for data preparation

George Scheer for IT support

Marcia Carlock at CDBW for funding image acquisition over the Delta

Dr. Diane Wickland at NASA for providing access to the AVIRIS database

Dr. Elizabeth Blood at NSF for funding the Gulf Oil Spill project

Dr. Mike Whiting and Dr. Jose Zarate for collecting calibration data in the field

References

Khanna, S., M. J. Santos, S. L. Ustin, and P. J. Haverkamp. 2011. An integrated approach to a biophysiologically based classification of floating aquatic macrophytes. International Journal of Remote Sensing, 32:1067-1094.

Khanna S, Santos MJ and Ustin SL (2010c) Concurrent effects of natural disturbance and control on water hyacinth distribution in the Sacramento-San Joaquin Delta, Dissertation: Chapter 2, PhD Ecology, University of California, Davis, California Santos, M. J., S. Khanna, E. L. Hestir, M. E. Andrew, S. Rajapakse, J. A. Greenberg, L. W. J. Anderson, and S. L. Ustin, (2009).

Use of Hyperspectral Remote Sensing to Evaluate Efficacy of Aquatic Plant Management in the Sacramento-San Joaquin River Delta, California. Invasive Plant Science and Management, 2: 216 – 229.

Graham, B., W. K. Reilly, F. Beinecke, D. F. Boesch, T. D. Garcia, C. A. Murray, and F. Ulmer. 2011. Deep Water: The Gulf Oil Disaster and Future of Offshore Drilling. National Commission on the BP Deepwater Horizon Oil Spill and Offshore Drilling. Kokaly, R. F., B. Couvillion, J. Holloway, S. Ustin, S. Peterson, S. Khanna, S. Piazza, and G. Aiken. 2011. Spectroscopic remote

sensing of the distribution and persistence of oil from the Deepwater Horizon spill in Barataria Bay marshes. Remote Sensing of Environment, accepted.

(24)

Water

Trees

Grasslands

Shrubs

Soil/Rock

Color Infrared Landsat data over South Lake Tahoe Basin

Support Vector Machines Classification

Landsat land cover classification

Climate Change

fore-casting at regional

level requires

land-cover and surface

properties derived

from Landsat Data

(25)

Water vs. Submerged Aquatic Plants

R

ef

lec

tanc

e

* 10000

Wavelength in µm

(26)

Summary of Results

Accuracies: 62% - 100%

Kappas: 0.6 - 0.9

Useful inputs

VIS, NIR & SWIR reflectance

Narrow band indices

Water indices & continuum removals

Water hyacinth classification method

Hybrid decision trees

Biophysiologically based, with statistical support and expert

knowledge

(27)

2007 – mNDVI

Submerged Aquatic Plants vs. Water

Submerged

Plants

Water

Class

(28)
(29)

Spatial

Resolution

Spectral

Bands

SAP Users

Accuracy

SAP Producers

Accuracy

Overall

Accuracy

Overall

Kappa

3m

6 bands

73.6

53.1

81.1

0.67

125 bands

81.7

51.7

82.2

0.69

15m

6 bands

79.6

53.1

82.4

0.69

125 bands

81.7

51.7

82.2

0.69

30m

6 bands

59.8

36.4

72.8

0.51

125 bands

40.3

20.5

62.7

0.31

Simulating Submerged Aquatic Plants

Classification at Landsat Spectral and

Spatial Scales

(30)

Classification using C5.0

Ensemble boolean decision trees

C5.0 maximizes Information Gain Ratio to split the

training set into two homogeneous groups

It uses the ratio to choose a split that minimizes

branches

This curbs overfitting the data – a common problem with

automated decision trees

Winnows attributes using only important predictors

Ensemble trees boost accuracy by learning from

(31)

Hybrid Strategy

Hybrid approach

Biophysiological basis of classification

ANOVA to decide useful variables

Histograms to decide thresholds

Expert knowledge used in building tree

(32)

Essentials for successful classification

Using multiple inputs in addition to image bands

Different techniques can gather distinct information about target

classes

Using methods that can accommodate

Multi-modal distributions of target classes

Multiple data types with different ranges and magnitudes

Be robust and automated when dealing with huge datasets over

vast spatial extent

Collecting enough training and validation data

Matching target class complexity with required data

Study system complexity

(33)

The primary gateway for

biological invasions in the

Western United States

The Sacramento – San Joaquin Delta

California

(34)

The classification problem

Multiple Inputs

Reflectance bands from the imagery

LiDAR derived tree height

Spectral Indices

Broad band e.g. NDVI & narrow band indices e.g. CAI

Indices measuring specific properties: water,

chlorophyll, cellulose

Multiple Techniques

Continuum removals over water absorption features

Linear Spectral Unmixing (LSU)

(35)

ANIR – example of an angle index

0

2000

4000

6000

8000

0.4

0.8

1.2

1.6

2

Wavelength in Micrometers

R

ef

lec

tanc

e*

10,

00

0

Dry soil

Moist soil

Dry vegetation

Green vegetation

+

NIR SWIR1 RED

ANIR

a

b

c

Phase transition

from green

vegetation to

dry vegetation

(36)

100 200 300 400 500 600 700 800 0 10 20 30 40 50 A v er ag e w at er ab so rp ti o n d ep th at 1. 2 µm

Distance to nearest oiled pixel (m) 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 0 10 20 30 40 50 A v er ag e A N IR

Distance to nearest oiled pixel (m)

Index trajectory in Sep 2010 Index trajectory in Aug 2011

Index trajectories w.r.t

to distance from the

nearest oiled pixel

0.15 0.20 0.25 0.30 0.35 0.40 0.45 0 10 20 30 40 50 A v er ag e m NDV I

Distance to nearest oiled pixel (m)

Sep 2010 Aug 2011 Sep 2010 Aug 2011 Sep 2010 Aug 2011 Seasonal differences

Recovery from oil contamination

References

Related documents

(Adjei, et. 2014) In chapter one, we proposed a greedy search algorithm for searching data fitting models with equally accurate predictive power as the original skill graph but

This work assumes that the answers to these and other questions are implicitly contained in the ARR’s data and a process mining based approach could be utilized to discover

This epithelium is composed mainly by four cell types: (1) Basal Cells (BCs) present on the basal lamina of airway epithelium from the upper airways until the respiratory

Next, we used pathway analysis to identify major path- ways and systems (Table 2) that may be affected by the observed olanzapine-induced methylation changes in the cadherin

Initially, most students will need full-time special aural/oral program for hearing impaired children, with emphasis on all language skills, concept development, auditory

A type system for a Haskell-like programming language that supports context-dependent overloading and follow the Hindley-Milner approach of providing context- free type

As § 42(1) IFA 1993 was applicable to foreign vehicles only, an Austrian corporation investing according to the principles of risk spreading could not be qualified as an investment

Cheyney(Cheyney , 2008) To examine the processes and motivations involved when women in the U.S. choose to have a home birth. Modified grounded theory. Phase 1: Pacific