64
CSIRO LAND and WATER Environmental Remote Sensing Group Arnold G. Dekker, Janet M. Anstee and Vittorio E. Brando CSIRO Land and Water, Canberra Technical Report 13/03, April 2003 Seagrass Change Assessment Using Satellite Data for Wallis Lake, NSW A consultancy report to the Great Lakes Council and Department of Land and Water Conservation

Seagrass Change Assessment Using Satellite Data for Wallis

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Seagrass Change Assessment Using Satellite Data for Wallis

C S I R O L A N D a nd WAT E R

Environmental Remote Sensing Group

Arnold G. Dekker, Janet M. Anstee and Vittorio E. Brando

CSIRO Land and Water, Canberra

Technical Report 13/03, April 2003

Seagrass Change Assessment Using Satellite

Data for Wallis Lake, NSW

A consultancy report to the Great Lakes Council and

Department of Land and Water Conservation

Page 2: Seagrass Change Assessment Using Satellite Data for Wallis

Environmental Remote Sensing Group

Arnold G. Dekker, Janet M. Anstee and Vittorio E. Brando

CSIRO Land and Water, Canberra

Technical Report 13/03, April 2003

Seagrass Change Assessment Using Satellite

Data for Wallis Lake, NSW

A consultancy report to the Great Lakes Council and

Department of Land and Water Conservation

Page 3: Seagrass Change Assessment Using Satellite Data for Wallis

Copyright © 2003 CSIRO Land and Water. To the extent permitted by law, all rights are reserved and no part of this publication covered by copyright may be reproduced or copied in any form or by any means except with the written permission of CSIRO Land and Water. Important Disclaimer To the extent permitted by law, CSIRO Land and Water (including its employees and consultants) excludes all liability to any person for any consequences, including but not limited to all losses, damages, costs, expenses and any other compensation, arising directly or indirectly from using this publication (in part or in whole) and any information or material contained in it. ISSN 1446-6163

Page 4: Seagrass Change Assessment Using Satellite Data for Wallis

CONTENTS

PART 1 - Seagrass Change Assessment Using Satellite

Data for Wallis Lake

1 INTRODUCTION............................................................................................3

1.1 Scope and aim............................................................................................................................................3

2 MEASUREMENT AND MODELLING...........................................................5

2.1 Introduction...............................................................................................................................................5

2.2 Optical modelling of the underwater light climate ..............................................................................6

2.3 Measurement of the above and underwater light field in Wallis Lake.............................................8

2.4 In-water optical modelling using HYDROLIGHT 4.1 .....................................................................11

2.5 Optical closure: modelled and in situ data..........................................................................................13

3 REMOTE SENSING: METHODS................................................................15

3.1 Introduction.............................................................................................................................................15

3.2 Introduction to Landsat data................................................................................................................15

3.3 Processing of Landsat at-sensor radiance to subsurface irradiance reflectance...........................16

3.4 Signal to noise ..........................................................................................................................................16

3.5 Supervised classification ........................................................................................................................18 Introduction ...............................................................................................................................................18 Method.......................................................................................................................................................19

4 RESULTS AND VALIDATION....................................................................20

4.1 Introduction.............................................................................................................................................20

4.2 Optical closure: modelled, in situ and remote sensing data..............................................................21

4.3 Classification results...............................................................................................................................22 The 12 September 2002 Landsat Image ..................................................................................................23 The 21 February 1995 Landsat Image .....................................................................................................25 The 30 March 1991 Landsat Image .........................................................................................................26 The 18 February 1988 Landsat Image .....................................................................................................27

4.4 Change detection between 18 February 1988 and 12 September 2002..........................................28

4.5 Validation field trip results....................................................................................................................31

4.6 Suspended sediment (tripton) concentrations ....................................................................................34

Page 5: Seagrass Change Assessment Using Satellite Data for Wallis

SEAGRASS CHANGE ASSESSMENT OF WALLIS LAKE CONSULTANCY REPORT

II

5 CONCLUSIONS AND RECOMMENDATIONS OF THE RESEARCH.....36

5.1 Results of this research...........................................................................................................................36 Optical modelling......................................................................................................................................36 Benthic substrate classification ................................................................................................................37

5.2 Future assessments – where to from here? .........................................................................................38

5.3 Transferability of methodology to management authorities............................................................40

5.4 Benefits of this research .........................................................................................................................40

PART 2 - Pathways for Implementation of Remote Sensing as an

Environmental Coastal Monitoring Tool

6 PATHWAYS FOR IMPLEMENTATION OF REMOTE SENSING AS AN ENVIRONMENTAL MONITORING TOOL FOR COASTAL LAKES........43

6.1 Scope and aim..........................................................................................................................................43

6.2 Methodology development requirements based on lake management authorities needs............44

6.3 Requirements for operationalisation of remote sensing....................................................................45

6.4 Pricing of remote sensor data and derived indicators and end-products.......................................46 Introduction ...............................................................................................................................................46 Raw remote sensing data acquisition costs..............................................................................................46 Notes for Table Interpretation ..................................................................................................................49

6.5 Concluding remarks...............................................................................................................................55

7 ACKNOWLEDGEMENTS...........................................................................57

8 REFERENCES.............................................................................................57

9 APPENDIX: DATA SUPPLIERS.................................................................58 Aerial Photography...................................................................................................................................58 Airborne Multi-spectral and Hyperspectral .............................................................................................58 Airborne LIDAR.......................................................................................................................................58 Satellite Multi and Hyper-spectral ...........................................................................................................58

Page 6: Seagrass Change Assessment Using Satellite Data for Wallis

SEAGRASS CHANGE ASSESSMENT OF WALLIS LAKE CONSULTANCY REPORT

III

Foreword

Graham Harris (CSIRO) suggested in his research overview report (Wallis Lake and its catchments: An overview and synthesis of existing data) within the Wallis Lake Catchment Management Plan (http://www.greatlakes.nsw.gov.au/Environ/wlcmp/wIndex.htm) that it was essential for the health of the Wallis Lake system that an ongoing monitoring program to be established. It is expected a monitoring program would provide important information to assess the success of the management plan and detect potential future problems. The Great Lakes Council (GLC) and the Department of Land and Water Conservation (DLWC) contributed seed funding to investigate the research and CSIRO Land and Water (CLW) contributed significant strategic research capacity to this study.

Page 7: Seagrass Change Assessment Using Satellite Data for Wallis

1

Part One

REPORT OF THE ASSESSMENT OF SEAGRASS CHANGE

USING SATELLITE DATA FOR WALLIS LAKE

Page 8: Seagrass Change Assessment Using Satellite Data for Wallis

2

Page 9: Seagrass Change Assessment Using Satellite Data for Wallis

3

1 Introduction

1.1 Scope and aim

This study investigated techniques using archived and current Landsat satellite data to characterise the benthic environment in Wallis lake: a shallow coastal lake (Figure 1.1). Mapping of substrate types and seagrass species has previously been undertaken successfully in other estuarine and coastal systems (Lee et al., 1999, Anstee et al., 2000 & Dekker et al., 2001) using sensors with higher spectral and spatial resolution (e.g. airborne hyperspectral scanners). In this project the potential use of archived Landsat data for this application was investigated. Landsat satellites are multispectral instruments that provide digital information over the earth’s surface (see section 3.2). Landsat imagery provides a source of archival data going back as far as 1984 and the French sensor SPOT has archival data from 1986. Much of the lessons learnt previously in developing the high spectral resolution techniques such as atmospheric correction and classification (Dekker et al., 2001) were used and applied to the lower spectral resolution Landsat data.

Seagrass has an important role in the maintenance of healthy estuary functioning. The health of seagrass is partly influenced by sedimentation and nutrient loading- two factors influenced by catchment processes and hence the health of the seagrass is a link to the broader catchment management. The mapping of seagrass distribution and change over time may provides estuary and catchment managers with information that can be used in the ongoing assessment of estuary health. In NSW only limited monitoring of seagrass distribution has been undertaken: in part due to costs and lack of suitable methodology. Historically, seagrass distribution assessment has been undertaken via aerial photography, which is labour intensive and somewhat subjective; making small changes over time ( the early warning signs of seagrass decline) harder to detect. The use of advanced satellite or airborne remote sensor technology for assessment of seagrass distribution provides an opportunity to undertake more cost effective, objective monitoring of seagrass health and condition, in support of catchment management activities. As more sensors become operational, the temporal coverage increases!

The aims of this project were:

• To produce maps of change of seagrass distribution from 1988 to 2002 in Wallis Lake.

• To assess the capacity of previous (1984 onwards) and existing moderate spatial resolution satellite data: Landsat 5 TM and Landsat 7 ETM+ (see section 3.2 for Landsat definitions) for the mapping of the ecosystems in Wallis Lake and it’s associated tributaries and estuaries.

• To collect in situ hyperspectral above and below water optical water column and substrate data. To apply CSIRO advanced processing methodologies for mapping of water quality indicators such as seagrass and macro-algae coverage and total suspended matter to the Landsat data.

• To assess the scope of relevant optical remote sensing techniques in detecting any of the following parameters (both detection and quantification): chlorophyll, cyanobacterial pigments (c-phycocyanin & c-phycoerythrin); coloured dissolved organic matter (CDOM); secchi depth, vertical attenuation coefficient of downwelling light (Kd); bathymetry, and macrophyte species such as seagrass and macroalgae.

Page 10: Seagrass Change Assessment Using Satellite Data for Wallis

4

• Figure 1.1: A Landsat TM False colour image of Wallis Lake with field sites from the 22-24 August 2001 campaign displayed in yellow

Part 1 of this report focuses on the methodology and results of the seagrass mapping project.

Section 2 discusses all the measurements undertaken, optical modelling and corrections used to create consistent comparable in situ measurements and modelled results.

Section 3 introduces the satellite data and the methodology employed to correct for the atmospheric and air/water interface effects then the classification technique used for the benthic substrate mapping on the Wallis Lake Landsat data is discussed.

Section 4 reports the results of the comparison of the in situ measurements and modelled results with the corrected satellite imagery. Classification results and change detection maps are displayed with the results of the field validation.

In Section 5 conclusions based on the results of the benthic substrate classifications are reported and recommendations are made for future monitoring and further research to provide enhanced information products.

Part 2 of this report focuses on pathways required for operational multi-temporal monitoring of coastal lakes. First, details on the methodology development and operational requirements of remotely sensing coastal lakes are presented, then options for improved data gathering using airborne and spaceborne sensors, including the recently launched Quickbird and SPOT5 satellites, are discussed.

Page 11: Seagrass Change Assessment Using Satellite Data for Wallis

5

2 Measurement and Modelling

2.1 Introduction

The radiance recorded by a remote sensing instrument contains a number of components when water masses are being imaged (Figure 2.1). Light from the sun and sky is partly reflected by the water surface, refracted at the water surface and some of this makes up part of the signal reaching the sensor. Light scattered in the atmosphere (path radiance) is also backscattered into the field of view of the sensor and makes up a large portion of the signal recorded, especially in the blue and green regions of the spectrum. Of the light that enters the water column, a proportion is backscattered by the water and suspended particles (including algae), a part is absorbed by the coloured dissolved organic matter (CDOM), the algae and the non-algal part of the total suspended matter called tripton, and some reaches the bottom where it is either absorbed or reflected back as a component of the upwelling radiation. This bottom reflected signal again goes through the various processes of attenuation in the water column, reflection and refraction at the water-to-air interface and atmosphere before it reaches the sensor.

• Figure 2.1 The pathways of light over and in a shallow water system

Since we intended to follow (as far as possible) a geophysical approach to the measurement and modelling of the image and in situ optical data, and to the inversion of this data to substrate type and bathymetry it was necessary to ensure that the properties of the underwater light field were internally consistent. In more practical terms a significant effort was spent on assuring that the Landsat radiance data, the in situ optical data, and the output from the optical models converged to the same reflectance (see sections 2.5 & 4.2) and vertical attenuation values. In the optical oceanographic community this is often referred to as “optical closure”.

The structure of this section is as follows: first an introduction to the principles of the underwater light field measurements and modelling; results of the in situ measurements undertaken, a discussion of the in-water optical modelling and closure of the modelled and in situ data.

Page 12: Seagrass Change Assessment Using Satellite Data for Wallis

6

2.2 Optical modelling of the underwater light climate

For measurement of environmentally relevant variables such as seagrass coverage or water column properties such as chlorophyll concentration from remote sensing, it is vital to understand: how light passes through the water column to the substrate, how it interacts with the substrate, how the resultant upward directed radiance is once more influenced by the water column before it leaves the water surface, and water surface and atmospheric effects.

The three essential optical properties for parametising an optical model are the substrate reflectance properties, the water leaving radiance signal and the vertical attenuation of both downwelling and upwelling light in the water column. This is the reason that vertical profiles of up- and downwelling light through the water column and the surface reflectances are measured in these studies. In August 2001, the RAMSES system (a submersible spectroradiometer) was used to collect substrate reflectance spectra in situ and to collect vertical profiles of up and downwelling light.

To better describe the rationale for the measurements, it is important to discuss some concepts of forward and inverse optical modelling of the remote sensing signal. Forward optical modelling is the modelling of the: i) substrate and water column optical characteristics in order to simulate subsurface irradiance reflectance, further forward modelling can take this reflectance and ii) pass it through the air/water interface and iii) propagate it through the atmosphere to simulate at remote sensor reflectance (it is also possible to replace the forward modelling of reflectance with the forward modelling of radiance). The inverse modelling refers to i) the inversion of the forward model in order to obtain subsurface irradiance reflectance data from the at sensor measured radiance and ii) the inversion of irradiance reflectance (just below the surface), R(0-) to water column concentrations of optically active substances or iii) substrate cover.

There are two forward modelling approaches to determine the relationship between the inherent and apparent optical properties of the water column and the water constituents: the analytical modelling and the radiative transfer modelling approach. Radiative transfer numerical models, such as HYDROLIGHT (Mobley, 1994; Mobley and Sundman, 2000a&b), compute radiance distributions and related quantities (irradiances, reflectances, diffuse attenuation coefficients) in the water column as a function of the water absorption and scattering properties, the sky and air-water interface conditions and the bottom boundary conditions. The analytical model is a simplification of the full radiative transfer equations, based on a set of given assumptions. Analytical models have the advantage that, due to their relative simplicity, they can be solved and inverted quickly. This is important in a remote sensing application where a model must be evaluated at every pixel of an image. Moreover analytical models can more easily be inverted.

Here it is necessary to introduce the concept of inherent and apparent optical properties. The inherent optical properties are those properties of the underwater light field that are not influenced by the amount or direction of light falling on a water body and transmitting through the water body. The apparent optical properties are more easily measurable with submersible spectroradiometers such as the RAMSES system. The apparent properties are dependent on the downwelling light stream amount and direction.

Inherent optical properties (IOP’s) are optical properties that are independent of the ambient light field (i.e., independent of changes in the angular distribution of radiant flux). These properties for light of a certain wavelength, are specified by the absorption coefficient a (m-1), the scattering coefficient b (m-1), and the volume scattering function β(Θ), which describes the angular distribution of scattered flux resulting from the primary scattering process. These definitions are based on the behaviour of a parallel beam of light incident upon a thin layer of a medium (Jerlov, 1976; Kirk, 1994). For remote sensing purposes, the backscattering coefficient bb (m

-1) is of more relevance, since this coefficient defines the amount of light scattered in a backward direction with respect to the incoming light (i.e. possibly towards the airborne or satellite sensor). The IOP’s of absorption and scattering are expressed in terms of absorption per metre: a (m-1) and scattering per metre: b (m-1). The values may be interpreted as follows: an absorption of 1.0m-1 indicates that an average photon entering the water column has a 63% chance of being absorbed in the first metre of water (the same relation applies to Kd and b). This may be calculated through the functional form e-az where a is the absorption coefficient (or Kd or b or bb) and z is the depth of the water column.

We refer to Aas (1987) for a complete derivation of the analytical model for the irradiance reflectance over optically deep waters. This model can act as a reference for understanding all

Page 13: Seagrass Change Assessment Using Satellite Data for Wallis

7

other models of this kind found in the literature. In terms of the backscattering and absorption coefficients the Aas (1987) analytical model for irradiance reflectance is written as:

b

b

du

ud

kba

brR

++=−

µµµ

)0( , (1)

where du

duud rrk

µµµµ

++

= , (2)

where ru and rd are the shape factors for up and downward scattering respectively as referred by

Aas (1987), µd is the average cosine for downwelling radiation and µu is the average cosine for upwelling radiation. This equation for R(0-) states that R(0-) is proportional to the backscattering divided by the sum of absorption and the second order backscattering. In more detail, the equation states that the irradiance reflectance is equal to a factor times the backscattering divided by the sum of absorption and the second order backscattering (whereby the second order backscattering is multiplied by a factor that accounts for up and downwelling shape factors and the average cosines for up and downwelling irradiance). The multiplication factor takes into account the downwelling shape factor and average cosines of the up and downwelling irradiances.

Although even this model contains approximations it may be expected to yield quite accurate results for turbid waters, as explained by (Aas, 1987). Various authors have developed analytical models for the subsurface reflectance, which can be related to the reflectance measured by remote sensing from (far) above the water surface. The subsurface irradiance reflectance plays an important intermediate role in many remote sensing applications on water quality.

Most studies neglect the variation in the shape factor and use empirical corrections based on the sun zenith angle instead of average cosines. For turbid waters the values for the average cosines for downwelling and upwelling light play a significant role (Dekker et al., 2001).

Optically shallow waters are waters where the substrate signal is detectable through the water column by a remote sensor. Examples are submerged macrophytes and sandy bottoms in lakes, seagrass and macro-algae fields in estuarine and coastal environments and coral reefs in tropical ocean environments. Maritorena et al., (1994) and Lee et al., (1999) describe the physics of an optical shallow water body (in equation 3) where part of the reflectance at the surface is composed of a bottom signal. In optically shallow waters, the subsurface upwelling irradiance, Eu(0-), is the sum of (i) upwelling irradiance originating within the water column (where none of the photons have interacted with the substrate), Eu(0-)C, and (ii) the upwelling irradiance reflected from the substrate (where each of the photons has interacted with the substrate) Eu(0-)B:

BuCuu EEE )0()0( −+−= (3)

Maritorena et al., (1994) provide equation 4 for the reflectance just below the surface of a homogeneous water body with a reflecting substrate,

)]exp()exp()[exp(),0( HRHAHKRHR CBd κκ −−−−+=− ∞∞ , (4)

where

R∞ = subsurface irradiance reflectance over a hypothetical optically deep water column; A= bottom albedo; H = water depth; κ B = vertical attenuation coefficient for diffuse upwelling light originating from the bottom; and κ C =vertical attenuation coefficient for diffuse upwelling light originating from each layer in the water column. If one is not able to separate the two upwelling light streams, assuming that κ B = κ C =κ , eqn 4 simplifies to eqn 5:

])(exp[)(),0( HKRARHR d κ+−−+=− ∞∞ (5)

Page 14: Seagrass Change Assessment Using Satellite Data for Wallis

8

2.3 Measurement of the above and underwater light field in Wallis Lake

Wallis Lake is large, shallow estuarine lake located about 360km north of Sydney in NSW. The catchment area is a significant resource for recreational activities and fishing. There have been significant environmental problems occurring within the catchment such as a hepatitis contamination of the oysters grown within the lake in the Christmas period of 1996-97. The impacts these problems demonstrated the need for an integrated management plan to maintain the health of the ecosystem. As part of a joint project between CSIRO and the Department of Land and Water Conservation (DLWC) and the Great Lakes Council (GLC), assessment of the lake’s benthic vegetation was undertaken on archival and current satellite imagery. With an understanding of the physical properties of the water and substrate species present, a baseline map could be produced and a map of the species coverage could be estimated from past imagery, in this case as far back as 1988.

A field campaign by CSIRO was conducted and the lake’s optical properties were measured and samples taken, at 7 locations within Wallis Lake and its tributaries in August 2001. This included profiles of downwelling irradiance and upwelling radiance within the water column, using the RAMSES spectroradiometer and using the HydroScat-6 backscatterometer for profiling measurements of backscattering (Figure 2.3) . In situ spectral reflectance measurements (Figure 2.5) of the benthic material were also collected using the RAMSES (Figure 2.4), while samples for spectrophotometric measurement of total absorption: the phytoplankton absorption, the tripton absorption and the CDOM absorption, were kept cool and dark until analysis in the laboratory at CSIRO Marine Research (Figure 2.2). For a detailed description of the spectrophotometric analysis methods see Clementson et al., (2001). Secchi depth, bottom depth, temperature and salinity measurements were undertaken at several sites, see Table 2.3, locations on Figure 1.1.

Site Name Site

Number

Secchi

Depth (m)

Bottom

Depth (m)

Water Temperature

(degrees Celcius)

Salinity

(µsiemens)

Wallamba River WL22_1 1.40 2.20 15.3 43.0

Pipers Ck Channel WL22_3 1.60 1.87 15.3 48.3

Pacific Palm WL23_5 - 1.40 12.3 44.4

Yahoo Island WL23_8 1.20 1.30 14.0 48.0

Coolongolook River WL23_10 1.20 2.30 14.0 46.0

Inlet WL24_1 2.00 3.00 14.9 50.9

Steps WL24_2 - 1.70 14.7 49.0

• Table 2.3: In situ field measurements taken in Wallis Lake during 22-24 August 2001. (‘-‘ indicates no measurement)

Page 15: Seagrass Change Assessment Using Satellite Data for Wallis

9

Absorption measured by Spectrophotometric Analysis from Wallis Lake samples collected 22-24th August 2001

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

400 450 500 550 600 650 700 750 800

Wavelength (nm)

Ab

sorp

tio

n c

oef

fici

ent

(m-1

)

a(total) a(detritus) a(phytoplankton) a(cdom)

• Figure 2.2 The laboratory measurements of absorption averaged over all sites. Note that the absorption by CDOM (Coloured dissolved organic matter) is the highest contributing factor to absorption in these lakes during this field campaign. Algal pigment absorption plays a significant role only in the wavelength range around 676 nm.

Backscattering measured by HydroScat-6 Wallis Lake 22-24 August 2001

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

400 450 500 550 600 650 700 750 800 850 900

Wavelength (nm)

Bac

ksat

teri

ng

co

effi

cien

t (m

-1)

wl22_1 wl22_3 wl23_8 wl23_10 wl24_1

• Figure 2.3 The in situ measurements of backscattering (see Figure 1.1 for location of sampling points) The results show a decrease in backscattering from the river water through the lake to the ocean waters at the lake entrance.

Page 16: Seagrass Change Assessment Using Satellite Data for Wallis

10

• Figure 2.4 The RAMSES field spectroradiometer consisting of an upwelling radiance sensor (the sensor pointing down) and a downwelling irradiance sensor (the sensor pointing upwards).

During the field campaign of 22-24 August 2001, information about the substrate and especially about seagrasses and macroalgae were collected at several location (see Figure 1.1 for locations and field sites) and measurements included:

• Spectral reflectances of key target materials taxomically identified by Wallis Lake local expert (Pia Laegdsgaard from the Coastal Ecology Group, DLWC) along specified transect(s).

• Survey points (differential global positioning system) of target materials and sampling points in conjunction with Wallis Lakes GLC staff. Additional D-GPS survey points were collected for the purpose of image geometric accuracy testing and resampling.

• Water sampling from selected sites for subsequent laboratory analysis by CSIRO Marine Research.

Field data and water laboratory analysis occurred (in Canberra at CSIRO Land and Water) for the development of a water radiative transfer solution applicable to the water quality and seagrass mapping with remote sensing sensors in Wallis Lake.

During all the measurements, geographical information was added by GPS (in WGS84). RAMSES spectral reflectance measurements (see Figure 2.4) were taken of the following samples:

• Gracilaria sp., red epiphytic algae, commonly found on Zostera or alone en masse

• Cystoseria trinodis, (Cockleweed) brown algae

• Sargassum sp., brown algae

• Chara sp., (Stonewort) green algae

• Chaetomorpha sp., filamentous green algae found in sheltered locations

• Posidonia australis, seagrass found in shallow clear waters

• Zostera capricornia, seagrass found in sheltered areas with sand or mud substrate

• Halophilia ovalis, seagrass found between other seagrass in sandy substrates

• Ruppia megacarpa, seagrass found in a range of habitats often mixed with Zostera.

Page 17: Seagrass Change Assessment Using Satellite Data for Wallis

11

Figure 2.5 contains a selection of averaged benthic reflectance spectra of species measured on 22-24 August 2001. The vegetation spectra are spectrally quite distinct, mainly due to a varying pigment composition as exemplified by the local spectral troughs in reflectance. This fact implies that hyperspectral remote sensing of Wallis Lake (either from aircraft or from satellites) will enable accurate discrimination and thus mapping of each of these individual species in those areas where the water column is sufficiently transparent to obtain a spectrum of the substrate species.

Benthic Reflectances Wallis Lake measured by RAMSES 22-24 August 2001

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

0.22

0.24

400 450 500 550 600 650 700

Wavelength (nm)

Ref

lect

ance

Gracilaria (red algae) avg Sargassum (brown algae) avg Cystoseira (brown algae) avgChara (green algae) avg Cheatomorpha (green algae) avg Posidonia (seagrass) avg Ruppia (seagrass) avg Zostera(seagrass) avg

• Figure 2.5 Benthic substrate reflectances measured on the 22nd-24th August 2001 using the

RAMSES field spectroradiometer. The wavelength range between 480 and 680nm allows maximal penetration of light into the water column and on to the substrate. The higher reflectance towards the NIR (700nm) is not relevant for remote sensing as the pure water absorption becomes significantly higher beyond 700nm.

2.4 In-water optical modelling using HYDROLIGHT 4.1

The radiative transfer model HYDROLIGHT will simulate spectra similar to those measured in the field and from the remote sensor if it is correctly parameterised. Once that match up level has been reached the HYDROLIGHT model makes it possible to run any type of simulation of reflectance at the surface (or in the water column) for substrates, water column properties, air/water interface effects and atmospheric effects. Thus HYDROLIGHT can be used as a tool to develop an algorithm to simulate bathymetry, benthic substrate and depth, turbidity, wind speed etc.

HYDROLIGHT 4.1, developed by Mobley (1994), is a radiative transfer model that computes radiance distribution and derived quantities for natural water bodies. The spectral radiance distribution L(z, θ, φ,λ) is computed as a function of depth (z) , direction (θ, φ), and wavelength (λ) within the water. Upwelling Radiance (Lu) is computed in all quarters above the sea surface. The water leaving radiance and reflected sky-radiance are computed separately, in order to isolate the water leaving radiance. Input of the model consists of the absorbing and scattering properties of the water body, the sea surface and of the bottom of the water column, and the sun and sky radiance incident on the sea surface. Outputs can be various irradiances, K -functions and reflectances (Mobley and Sundman 2000a; Mobley and Sundman 2000b).

Sun position is calculated from solar models incorporated in HYDROLIGHT using the following inputs: time (of RAMSES measurement), date, and geographical position. A semi-empirical RADTRAN atmospheric model implemented in HYDROLIGHT is used to model the atmosphere similar at the time of the in situ measurements.

Page 18: Seagrass Change Assessment Using Satellite Data for Wallis

12

To achieve optical closure between HYDROLIGHT and the in situ data, we needed to run various simulations experimenting with different parameterisations. The model used for the HYDROLIGHT simulations is the “ABCASE2” (a standard HYDROLIGHT model). This is a generic four-component IOP model with flexibility in how the user can define component optical properties. Where the concentrations and IOPs of the four components (pure water, chlorophyll, CDOM and mineral particles) are entered in the model. The total absorption, aTOT in this respect can be represented by:

TSSCDOMCHL acacacaa TSSCDOMCHLWATERTOT*** ⋅+⋅+⋅+=

(6)

where

aWATER = absorption due to water

cCHL = concentration of chlorophyll

a*CHL = specific absorption of chlorophyll

cCDOM = concentration of CDOM

a*CDOM = specific absorption of CDOM

cTSS = concentration of TSS (minerals)

a*TSS = specific absorption of TSS (minerals)

The parameterisation of HYDROLIGHT can be described as:

• Component 1: pure water. The “pure water” absorption values of Pope and Fry (1997) are used as the HYDROLIGHT default. It is one of the latest and most accurate available values.

• Component 2: Chlorophyll. Assumed is a concentration profile of chlorophyll constant with depth. The entire scattering is attributed to component 4, so the scattering of Chlorophyll is assumed to be 0. This is possible as the total backscattering estimated in situ with Hydroscat-6 is attributed to the total suspended matter (that contains algal biomass).

• Component 3: CDOM (coloured dissolved organic matter). Log transformed (laboratory measured) CDOM spectral data, gave an absorption of 0.026 m-1 at 500nm. The reference wavelength of 500 nm is associated with an exponential decay function with a slope of 0.014. The accuracy of this regression is minimal 97 %.

• Component 4: total suspended matter (TSS). The concentration profile is also assumed to be constant with depth. The a * (minerals) is used to specify the absorption of the TSS.

• There was assumed to be no internal source, caused by bioluminescence, nor inelastic scat-tering caused by (Chlorophyll and CDOM) fluorescence or Raman scattering. By using the laboratory absorption data and the Hydroscat backscattering data some of these effects are implicit within the parameterisation. Furthermore these effects are expected to be so small that it is more important to get optical closure without these effects. A possible sophistication could be running the model with these parameterisations. Incorporating these internal sources and elastic scattering sources has to be done with care and each of the parameterisations has to be valid for the waters under study. That went beyond the scope of this research.

• The wavelength selection is defined with corresponding Landsat bands wavelength input files.

• A semi-empirical sky model and a wind speed of 5 m s-1 define the air-water surface boundary conditions.

• A solar zenith angle of 44-70 degrees and a 0-50 % cloud cover defined the sky conditions ranges for the 4 Landsat images processed..

• The atmospheric parameters were obtained by modelled radiosonde data for each month over a 25 year period from 1957-1975 for Williamtown (45 kilometres southwest of Wallis Lake).

• The direct (solar) and diffuse (sky) components of the downwelling sky irradiance are directly calculated from the RADTRAN model in HYDROLIGHT.

Page 19: Seagrass Change Assessment Using Satellite Data for Wallis

13

• A file with reflectance defines the bottom reflectance or boundary spectral condition. Reflectance data obtained by the RAMSES measurements are used to specify the different reflectance files.

Part of the HYDROLIGHT output is the simulation of R(z-), the irradiance reflectance at different depths in the water column over a substrate at z = max depth. In the case of z = max depth, the output is the substrate reflectance; in the case of z = 0 the output is the subsurface irradiance reflectance.

2.5 Optical closure: modelled and in situ data

The input parameters resulted in a reasonable optical closure between HYDROLIGHT and the in situ RAMSES data. Figures 2.6 & 2.7 show the result of the optical closure (at field sites WL22-1 & WL22-3) between RAMSES in situ measured spectra of Lu/Ed and the HYDROLIGHT simulated Lu/Ed. Once this stage has been reached, all permutations of concentrations can now be simulated as long as the parameterisation of the model with specific inherent optical properties is correct. At this stage of “knowledge” it becomes possible to simulate thousands of possible permutations of combinations of concentrations, substrates and substrate depths with which it is possible to train neural networks, test analytical inversion algorithms etc.

Page 20: Seagrass Change Assessment Using Satellite Data for Wallis

14

Hydrolight Simulation R(z) over Zostera substrate compared with in situ RAMSES spectra at site WL22-1

in Wallis Lake 22 August 2001

0

0.005

0.01

0.015

0.02

0.025

0.03

400 450 500 550 600 650 700 750

Wavelength (nm)

Lu

/Ed

In Situ: z=0.2m In Situ: z=0.9m Hydrolight: z=0m Hydrolight: z=1m

• Figure 2.6: Reflectance spectra corresponding to field site WL22-1, located on the Wallamba

River at depths of 0.2m and 0.9m (z= depth). Simulated reflectance from HYDROLIGHT (light to dark blue). In the ideal situation the HYDROLIGHT spectra (light blue) matches the in situ RAMSES spectra.

Hydrolight Simulation R(z) over Zostera substrate compared with in situ RAMSES spectra at site WL22-3

in Wallis Lake 22 August 2001

0

0.005

0.01

0.015

0.02

0.025

0.03

400 450 500 550 600 650 700 750

Wavelength (nm)

Lu

/Ed

In Situ: z=0.4m In Situ: z=0.7m Hydrolight: z=0m Hydrolight: z=1m

• Figure 2.7: Reflectance spectra corresponding to field site WL22-3, located on Pipers Creek channel. Simulated reflectance from HYDROLIGHT (light to dark blue) and reflectance from RAMSES (light to dark green). In the ideal situation the HYDROLIGHT spectra (light blue) matches the RAMSES spectra (light green).

Page 21: Seagrass Change Assessment Using Satellite Data for Wallis

15

3 Remote Sensing: Methods

3.1 Introduction

Satellite imagery of Wallis Lake spanning 14 years was used to monitor the change in seagrass communities. The Landsat archives were browsed in order to select cloud free imagery of consistent level of quality (see Table 3.1). The quality of the imagery was assessed on the absence of sun glint, wind-induced waves as well as the absence of river outflow induced turbidity affecting the seagrass “visibility” through the water column.

Date sensor quality Notes 12th September 2002 L7 ETM high little sun glint or wind waves 21st February 1995 L5 TM reasonable little sun glint or wind waves sensor striping

visible, particularly in the rivers. 30th March 1991 L5 TM reasonable little sun glint or wind waves turbid river water. 18th February 1988 L5 TM high little sun glint or wind waves.

• Table 3.1: Landsat imagery time series

3.2 Introduction to Landsat data

The Landsat program has been run since 1973 providing calibrated medium spatial resolution data (80 m resolution until 1984, 30 m afterwards, see Table 3.2) to a varied user community including the academic and scientific communities, local and federal governments as well as commercial operators. Landsat has provided multi-temporal coverage meeting some of the needs of science, education, businesses and governments for information and changing aspects of the Earth’s surface.

The Landsat Program’s aim is to provide global repetitive acquisition of medium resolution multispectral data of the Earth's surface. Landsat provides a source of global, calibrated, medium spatial resolution measurements that can be used for multi-temporal assessment or change detection. Landsat has acquired a long term record of the Earth's continental surfaces as seen from space, and as an archival environmental record it is unmatched in quality, detail, coverage, and value.

The platforms that carry Landsat have other remote sensor systems and data relay systems along with attitude-control and orbit-adjust subsystems, power supply, receivers for ground station commands and transmitters to send the data to ground receiving stations.

System Launch (end of service)

Instrument/s

Resolution (meters)

Landsat 1 23 July 1972 (6 January 1978)

RBV, MSS

80, 80

Landsat 2 22 January 1975 (25 February 1982)

RBV, MSS

80, 80

Landsat 3 5 March 1978 (31 March 1983)

RBV, MSS

30, 80

Landsat 4 16 July 1982 (failed August 1993)

MSS, TM

80, 30

Landsat 5 1 March 1984 (current data no longer available)

MSS, TM

80, 30

Landsat 6 Failed on Launch ETM 15 (panchromatic), 30(multispectral)

Landsat 7 15 April 99 (still operational)

ETM+ 15 (panchromatic), 30(multispectral)

• Table 3.2: Landsat Family (RBV = Return Beam Vidicon (camera), MSS = Multispectral Scanner System, TM= Thematic Mapper, ETM = Enhanced Thematic Mapper, ETM+ = Enhanced Thematic Mapper Plus)

Page 22: Seagrass Change Assessment Using Satellite Data for Wallis

16

The Landsat 7 instrument was launched on April 15, 1999 and is currently operated by the US Geological Survey (USGS). The Enhanced Thematic Mapper Plus (ETM+) carried on Landsat 7 is an eight-band multispectral scanning radiometer with bands in the visible, near-infrared, short-wave, and thermal infrared (see Table 3.3). The pixel resolution is 30 metres in the visible & near infrared bands, 15 meters in the panchromatic and 60 meters in the thermal band. The coverage is continuous (constantly collecting data along the orbit) with a 16-day repeat cycle, and a standard scene size of about 170 x 183 kilometres. Scenes can be delivered in many sizes. The Landsat data pricing has become increasingly cheaper with time. A standard 170 x 183 kilometres scene costs about $AUD1200 at the moment (it used to be $AUD8000). The expectation is that pricing will decrease in the future.

Band Range Landsat 5 TM

(micrometers)

Landsat 7 ETM+

(micrometers)

1 0.45-0.52 0.45 - 0.52

2 0.52-0.60 0.53 - 0.61

3 0.63-0.69 0.63 - 0.69

4 0.76-0.90 0.78 - 0.90

5 1.55-1.75 1.55 - 1.75

6 10.40-12.50 10.40 - 12.50

7 2.08-2.35 2.09 - 2.35

8 - 0.52 - 0.90

• Table 3.3 Band designations of Landsat 5 and 7 sensors

3.3 Processing of Landsat at-sensor radiance to subsurface irradiance reflectance

The first step of the Landsat data processing involves the correction of at-sensor measured upwelling radiance from Wallis lake for atmospheric effects. Although the physics of atmospheric correction of remote sensing data over waters is essentially the same as for terrestrial targets, there are a few practical differences that need to be addressed. For any water body it is the signal coming from within the water body that is desired. On land it is the surface reflected signal that is of interest. For water bodies the surface reflected signal is considered as noise, and is composed of the reflected component of diffuse skylight and of the direct sunlight reflected from the water surface. Water bodies in general reflect (as subsurface irradiance reflectance) in the range of 1% to 15% of downwelling irradiance. The majority of waters reflect between 2% and 6% of downwelling irradiance. Thus to obtain for example, 40 levels of irradiance reflectance in the range of 2 to 6% reflectance we need a minimal accuracy of atmospheric correction to 0.1% reflectance (Dekker et al., 2001).

A model-based solution to this problem has been applied in this study using the “coastal Waters and Ocean MODTRAN-4 Based ATmospheric correction” (“c-WOMBAT-c”) procedure implement-ed in IDL/ENVI® (see Brando & Dekker, 2002 for details). MODTRAN is a radiative transfer model developed by Geophysics Division of the Air Force Phillips Laboratory. C-WOMBAT-c converts radiometrically calibrated imagery to apparent reflectance (for terrestrial targets) and subsurface reflectance (for water targets) images.

3.4 Signal to noise

In order to understand the detection limits of an environmental variable with a remote sensor it is necessary to know or estimate the Signal to Noise Ratio (SNR) as it exists for each image. One could refer to the instrument provider’s SNR, however these are invariably determined under laboratory conditions, often involving a bright lamp and an integrating sphere. In actual remote sensing environments there are the sources of noise in the image data such as atmospheric

Page 23: Seagrass Change Assessment Using Satellite Data for Wallis

17

variability, the air water interface with swell, wave and wavelet induced reflections and refraction of diffuse skylight and direct sunlight.

To estimate the sensitivity of the images from the two Landsat sensors the environmental noise equivalent radiance difference (NE∆LE) and the environmental noise equivalent reflectance difference (NE∆R(0-)E) need to be determined. These are image based measures and are dependent on instrument SNR together with environmental influences such as atmospheric and air water interface variability.

The NE∆LE is calculated from the at-sensor radiance image according to Dekker & Peters (1993):

NE∆LE = σ(L)

where σ(L) is the standard deviation in each band over an as homogeneous as possible area of optically deep water within the image, the size of the uniform area can be determined by increasing the number of pixels step by step (1, 3x3, 5x5, 7x7, etc) until σ(L) reaches a first asymptotic limit. Care has to be taken that whilst increasing the size of the uniform area no actual water body heterogeneities are included in the sampled area. The NE∆R(0-)E is calculated in the same manner using the atmospherically and air-water interface corrected R(0-) image.

These calculations were done on a few Landsat images using the ocean as a homogeneous surface for a training target; the results for the 2002 and 1988 images are presented in Figure 3.1 & 3.2. These results indicate that Landsat 5 is less sensitive than Landsat 7. In terms of reflectance Landsat 5 can resolve about 0.5 to 1 % reflectance differences. It needs to be realised that as Landsat only has 256 digital levels of radiance detection available the theoretical limit is about 256:1.

Based on a comprehensive discussion of spectral resolution, spatial band location and sensor sensitivity it is unlikely that a Landsat-type sensor can accurately map chlorophyll, cyanobacterial pigments, secchi depth, Kd, CDOM and bathymetry based on the optical modelling approach as described in Dekker et al., (2001).

Estimate of Environmental Noise for Landsat7 ETM+ Data from 12 September 2002

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

400 450 500 550 600 650 700

Wavelength (nm)

NE

∆LE (

W m

-2 s

r-1 µ

m-1

)

0.0%

0.2%

0.4%

0.6%

0.8%

1.0%

1.2%

1.4%

1.6%

NE

∆R(0

-)E

NEDLE Dark NEDLE Light NEDR(0-) Dark NEDR(0-) LightNE∆LE Dark NE∆LE Light NE∆R(0-)E LightNE∆R(0-)E Dark

• Figure 3.1: The environmental noise equivalent radiance difference (NE∆LE) [left axis] and environmental noise equivalent R(0-) difference (NE∆R(0-)E) [right axis] for Landsat7 ETM+ over ocean water. The values were calculated on dark and light stripes that were visible in the data. Note: NE∆(R0-) is not a percentage error, but a difference expressed in percentages.

Page 24: Seagrass Change Assessment Using Satellite Data for Wallis

18

Estimate of Environmental Noise for Landsat5 TM Data from 18 February 1988

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

400 450 500 550 600 650 700

Wavelength (nm)

NE

∆LE (

W m

-2 s

r-1 µ

m-1

)

0.0%

0.2%

0.4%

0.6%

0.8%

1.0%

1.2%

1.4%

1.6%

NE

∆R(0

-)E

NEDR(0-)E AREA 1 NEDR(0-)E AREA 2 NEDRE Dark NEDRE LightNE∆LE Dark NE∆LE Light NE∆R(0-)E Area 1 NE∆R(0-)E Area 2

Figure 3.2: The environmental noise equivalent radiance difference (NE∆LE) [left axis] and environmental noise equivalent R(0-) difference (NE∆R(0-)E) [right axis] for Landsat5 TM over ocean water. The values were calculated on dark and light stripes that were visible in the data. Note: NE∆(R0-) is not a percentage error, but a difference expressed in percentages.

3.5 Supervised classification

Introduction

With limited overall understanding of the substrates that included field data for only three days (not coinciding with any of the satellite images), we had to make a decision on the most appropriate technique to trial for the purpose of this research.

A fully parameterized system is where all permutations of substrate reflectance are known as well as scene-derived knowledge of the optical properties of the water column are available. In this system, it is conceivable to invert (given sufficient spectral bands with water column penetration capacity and spectrally variable substrates) the relationship between R(0-) and the desired variables that need to be estimated. This level of sophistication for a water body such as Wallis Lake is yet to be realized. Therefore it was decided to apply the ENVI-based Maximum Likelihood supervised classification technique after some experimentation with other classifying techniques, such as Spectral Analysis Method (SAM) or isoclassification.

Maximum Likelihood is based on statistics (mean; variance/covariance), where a Bayesian Probability Function is calculated from the selected regions of interest (ROIs). Each pixel is then placed into the class to which it most likely belongs. This was done with the Wallis Lake Landsat data, using the three visible bands. The near infrared band was judged to be overly influenced by sensor striping over water caused by the very low reflectance values, often less than 1%. The classification results seem to identify known patterns of benthic vegetation and have been subjected to DLWC and GLC verification as reported in Section 4.3.

Page 25: Seagrass Change Assessment Using Satellite Data for Wallis

19

Method

The goal of this classification method is to be as objective as possible to make it repeatable for other data sets for multi-temporal comparison. Regions of interest (ROIs) were selected based on a pseudo true colour combination using the first three Landsat bands centred at 485nm, 560 nm and at 660nm. The ROIs have been selected according to homogeneous texture on the RGB over the entire image. The ROIs selection criteria included: targets that are reasonably homogeneous, pixels within the ROI generally having a low within class low variance, selected in a range of water depths and coincided with field data or past study sites. These ROI files have been saved and may be used on future imagery thus enabling objective repetition of this methodology! It should be noted that the selection of ROIs is most effective when the image analyst has some prior knowledge or field experience of the region.

The aim of the classification was to separate the benthic material within the lake into sets of spectral classes that represent the patterns and texture of the ecosystem. These classes, or its attributes, are created by the classifier which have been trained with the spectral classes (ROIs) collected from the imagery. The Maximum Likelihood classification was run with ENVI 3.5 image processing software using the spectra as input and with a probability threshold value of 0.05. Pixels with probabilities less than this value for all classes were unclassified.

The spectra represented most of the pixels in the image and there were some classes that could be identified using the field data, however some appeared to be mixtures of the field data or quite different altogether. These unidentified classes were labelled unknown and validation field trips in October and November 2002 focused on the identification of these classes.

The aim was to group classes that were spectrally similar and to link them to an environmentally relevant label. Firstly, the location (collected by a Global Positioning System, GPS) and associated substrate information was used to label known classes; substrate maps and the Field Guide publication (Laegdsgaard, 2001) were used to identify only the major groupings of substrates as they contains little geographic detail; the existing CSIRO spectral library as well as the Maximum Likelihood classification scheme was used. In most of the regions 6-11 substrate classes, 3 water types (river, turbid river & lake) and 1 sand class could describe the spectral variation available in the Landsat data. Post-classification, the spectral classes were labelled using the parameterized HYDROLIGHT model.

Page 26: Seagrass Change Assessment Using Satellite Data for Wallis

20

4 Results and Validation

4.1 Introduction

Wallis Lake was selected as a representative coastal lake to ascertain the potential of Landsat data for mapping changes in benthic substrates. Wallis Lake estuary consists of lakes and rivers with interconnecting channels (Figure 4.1) .Based on past Wallis Lake studies (West et al., 1985) it appeared that the lake has a complex substrate environment. Unvegetated substrates vary from mud to sand to coarse sand with shell fragments. Vegetated substrates vary from very lightly vegetated with Halophilia to densely vegetated areas of mixtures of Zostera, Posidonia and various macro-algae (e.g. at Pipers Creek and Pacific Palms) to monospecies fields of Zostera or Posidonia at various levels of density. Thus, Wallis Lake was a challenging target but, if successful, a representative target of other coastal lakes.

Four Landsat images from 1988 to 2002 (spanning 14 years) were selected for atmospheric correction and classification (see Table 3.1). The quality of the images affected the quality of the resultant classifications. Regions with visible striping impacted on the classification by mislabelling classes. High quality images (such as the 12 September 2002) had little evidence of this in the classification.

• Figure 4.1: Wallis Lake topographic map. Derived from NATMAP – Bulahdelah sheet no. 9333,

scale 1:100000, 1983.

Page 27: Seagrass Change Assessment Using Satellite Data for Wallis

21

4.2 Optical closure: modelled, in situ and remote sensing data

With atmospheric (iterative) parameterisation for the September 2002 Landsat image we achieved a reasonable closure between the Landsat atmospherically and air-water interface corrected data to R(0-) and the in situ measured R(0-) as shown in Figure 4.2. Figure 4.2 shows that closure is not perfect with only band 560nm within the signal to noise range for Landsat. The Landsat spectra for site WL22-3 has a similar shape and amplitude to that of the RAMSES reflectance (measured at 0.2 and 0.9 m depths), even though there are 13 months intervening between the field campaign and the satellite overpass. As the Landsat images were not coincident with the field measurements, it is important to emphasise that we were still able to reach a measure of optical closure with non coincidental data.

Once this stage of optical closure has been reached the Landsat imagery can be used as an in situ measured reflectance at each pixel. All permutations of concentrations can now be simulated as long as the parameterisation of the model with specific inherent optical properties is correct. At this stage of “knowledge” it becomes possible to simulate thousands of possible permutations of combinations of concentrations, substrates and substrate depths with which it is possible to train neural networks, test analytical inversion algorithms etc. In addition the stringent requirement for simultaneous field data and remote sensor overpass data acquisition is no longer necessary! After a well parameterized atmospheric model (c-WOMBAT-c) is applied to Landsat data (coincident with the field measurements), optical closure can be obtained between the HYDROLIGHT and the in situ field measurements or remote sensing data. In fact a well parameterised atmospheric and in-water model reduces the necessity for fieldwork considerably in future, as the application for Wallis Lake becomes mature.

It is recommended to focus on the optical closure measurement and modelling in future research in Wallis Lake.

Hydrolight Simulation R(z) over Zostera substrate compared with in situ RAMSES spectra

and the Landsat spectra for site WL22-3

0

0.005

0.01

0.015

0.02

0.025

0.03

400 450 500 550 600 650 700 750

Wavelength (nm)

Lu

/Ed

In Situ: z=0.4m In Situ: z=0.7m Hydrolight: z=0m Hydrolight: z=1m Landsat WL22-3

• Figure 4.2: Reflectance spectra corresponding to field site WL22-3, located on Pipers Creek channel. Simulated reflectance from HYDROLIGHT (light to dark blue)Landsat (red) and reflectance from RAMSES (light to dark green). In the ideal situation the HYDROLIGHT spectra (light blue) and the RAMSES spectra (light green) matches the Landsat spectra (red).

Page 28: Seagrass Change Assessment Using Satellite Data for Wallis

22

4.3 Classification results

Classes were labelled according to the criteria described in the method section, and generally fell into sensible groupings (with knowledge of the lake ecosystem) with help from our existing field knowledge and past studies. In some images some pixels occurred where most benthic spectral information was lost due to sun glint, turbidity or wind waves and these pixels often formed their own classes across depth and substrate boundaries. After analysis, they were either discarded from the classification, or labelled to the closest associated spectral class.

In the end we had four main sources of information used in the classification and labelling:

• past substrate maps and studies, including Laegdsgaard (2001) & West et al., (1985),

• the field data collected during the 22-24 August field trip,

• HYDROLIGHT results using the above field data as discussed in Section 2.4,

• an existing spectral library on substrate vegetation at CSIRO.

The data from the 22-24 August 2001 field survey were most useful as they were located by GPS.

Validation field trips were undertaken in October and November 2002 and are discussed in Section 4.5.

Page 29: Seagrass Change Assessment Using Satellite Data for Wallis

23

The 12 September 2002 Landsat Image

The Landsat image of 12 September 2002 was assessed as being high quality, with little sun glint or wind induced waves that may reduce the depth of penetration of light into the Lake. This image was acquired closest to a field validation campaign and was thus processed first. The ROIs coincided with as many field positions as possible, where measurements of the substrate occurred. In the classification (Figure 4.3) unidentified classes were targeted in the October and November 2002 field work carried out by GLC and DWLC, and subsequently labelled as Chara and Macroalage. One class that was most closely spectrally related to Posidonia in the southern region of the lake has been labelled as unknown, as there is no evidence that Posidonia occurs in this area. There appeared to be less Zostera than in past images, with replacements by a macroalgae class. This macroalgae class may be a heavy infestation of epiphytes on the Zostera or may have replaced the Zostera.

• Figure 4.3: Classification of the Landsat 7 ETM+ data from 12 September 2002. This image was

closest to a field validation campaign.

Page 30: Seagrass Change Assessment Using Satellite Data for Wallis

24

A subset of the 12 September 2002 classification was defined from the Bridge at the mouth of the Lake, west to Wallis Island then south to Green Point (locations found in Figure 4.1), to calculate seagrass coverage, as shown in Figure 4.4. This area was chosen to be a representative area for testing future multi-temporal analyses.

• Figure 4.4: Seagrass classes of the Landsat 7 ETM+ classification data from 12 September 2002

within the region from the Bridge at the mouth of the Lake, west to Wallis Island then south to Green Point. This area was defined using an ArcView shape file (walarea.shp) as defined by DLWC (Graham Carter, per. comm).

The area covered by seagrass species are displayed in Table 4.1 as total area covered and as a percentage of the total classified region (water covered) of 16.27 km2. The three major seagrass classes, are displayed in Table 4.1 by combining all subgroups: Zostera/Macroalgae, Sparse Zostera and Sand/Zostera are combined into just one class – Zostera; Ruppia, Halophila/Ruppia/Sand and Sparse Halophila/Ruppia are combined into the class – Ruppia & Halophila.

Species Area km2 Area % Zostera 1.71 10.53 Posidonia 2.03 12.49 Ruppia & Halophila 3.15 19.37

• Table 4.1: Area of seagrasses as defined in the 12 September 2002 Classification within the boundary specified in Figure 4.4

Within the Zostera class, the four sub-groups (Figure 4.4) are displayed within Table 4.2 as total area in km2 and as a percentage of the total Zostera in the region which was 1.71 km2 (from Table 4.1).

Sub-Group Area km2 Area % Zostera 0.20 11.61 Sand/Sparse Zostera 1.01 59.12 Zostera/Macroalgae 0.00 0.26 Sparse Zostera 0.50 29.01

• Table 4.2: Area of Zostera as defined in the 12 September 2002 Classification within the boundary specified in Figure 4.4

Page 31: Seagrass Change Assessment Using Satellite Data for Wallis

25

The 21 February 1995 Landsat Image

The Landsat image of 21 February 1995 was assessed as being of reasonable quality, with little sun glint or wind waves but with sensor stripping visible, particularly in the rivers. The water in the lake was more turbid than for the 1988 (Figure 4.7) or 2002 (Figure 4.3) images. The images of less than high quality tend to map Posidonia where it does not occur (in the southern basin). Higher spectral and spatial remote sensing data will probably prevent this from occurring. In the 1995 classification (Figure 4.5), there appeared to be less macroalgae than in the 2002 image although Posidonia was classified in the southern part of the Lake where it is unlikely to occur – mapping incorrectly may be due to reduced image quality. Some Posidonia has been found (West et al., 1985) on the mid western edge of the Lake between Coomba and Little Flat Point, which has been mapped as Posidonia in Figure 4.5. Large homogeneous beds of Zostera appear in the mid and southern parts of the Lake with consistent beds of Ruppia and the Ruppia/Halophilia mixed class.

• Figure 4.5: Classification of the Landsat 5 TM data of 21 February 1995. The image quality was medium with a higher turbidity(or less clarity) in the lake. These classification results are less reliable than for the 1988 and 2002 images.

Page 32: Seagrass Change Assessment Using Satellite Data for Wallis

26

The 30 March 1991 Landsat Image

The Landsat image of 30 March 1991 was assessed as being of reasonable quality, with little sun glint or wind waves but turbid river water. As in the 1995 image there appears to be some sensor striping and again Posidonia was classified in the southern part of the Lake where it is unlikely to occur– mapping incorrectly may be due to reduced image quality. Large homogeneous beds of Zostera appear in the mid and southern parts of the Lake and consistent beds of Ruppia and the Ruppia/Halophilia mix (Figure 4.6). Beds of Posidonia mapped out in the north and around Wallis Island, which are consistent with the Laegdsgaard (2001) field guide and maps of West et al., (1985).

• Figure 4.6: Classification of the Landsat 5 TM data from 30 March 1991. The image quality was medium with a higher turbidity in the lake. These classification results are less reliable than for the 1988 and 2002 images.

Page 33: Seagrass Change Assessment Using Satellite Data for Wallis

27

The 18 February 1988 Landsat Image

This image was assessed as high quality, with little sun glint or wind waves. Beds of Posidonia mapped out in the north and around Wallis Island, which are consistent with the past surveys and maps of this date. Large areas of Zostera covered the central and southern basins of the Lake forming homogeneous beds adjacent to Ruppia and Ruppia/Halophilia beds in the Lake’s central east basin (Figure 4.7). An unknown class (labelled ‘unknown’) was not identified from the field spectra but is likely to be similar to the Zostera/Macroalgae class in the September 2002 image or Zostera with a mud or silt substrate. Further field measurements may lead to the identification of this unknown class at a later stage.

• Figure 4.7: Classification of the Landsat 5 TM data from 18 February 1988. This image and the

2002 image are of the highest quality and the most suitable for change detection.

Page 34: Seagrass Change Assessment Using Satellite Data for Wallis

28

4.4 Change detection between 18 February 1988 and 12 September 2002

To identify regions of change from the 1988 to 2002 high quality images, everything except the class of interest (Zostera, Posidonia & the combined Ruppia/Halophila class) was masked out in the 1988 and the 2002 images. The difference of the 2 images for each class can be seen in the Figure 4.8 where red indicates loss, blue indicates no change and green indicates addition. White pixels are classified as another class not included as a seagrass species. Single pixel ‘noise’ can be seen in these images. Information in single pixels is less accurate than information in adjoining groups of pixels.

To illustrate change and the vagaries of change over time, a change detection map for Zostera was produced for the 1988-1991 and 1991-1995 images (Figure 4.9). The 1988-1991 image (on the left) clearly indicates Zostera loss from the central eastern basin. Zostera loss or change is not uncommon and could be the result of seasonal or environmental conditions, although both images were acquired approximately the same season (February 1988 and March 1991). Zostera had regrown in the 1991-1995 image at the scar feature identified in the 1988-1991 change detection map.

Page 35: Seagrass Change Assessment Using Satellite Data for Wallis

29

• Figure 4.8 (from left to right) Zostera, Posidonia & Ruppia/Halophila change from 1988-2002, with red = loss, green = gain and blue = no change. White pixels within the lake indicate a class not identified as seagrass.

Page 36: Seagrass Change Assessment Using Satellite Data for Wallis

30

• Figure 4.9: Left - Zostera change from 1988-1991, Right – Zostera change from 1991-1995, with red = loss, green = gain and blue = no change. The highlighted region shows a

scar where Zostera was lost in 1991 but had regrown by 1995.

Page 37: Seagrass Change Assessment Using Satellite Data for Wallis

31

The seagrass trend in Wallis Lake seems to be an overall decline in probably just one seagrass species – Zostera. Posidonia, Ruppia and Halophila seem to be stable with no gross changes in the 14 year period from 1988 until 2002. However, the Zostera has undergone significant change and adaptation. From quite early in the time series (between 1988 and 1991) some Zostera beds were reduced in size, in particular the Zostera bed around Coomba Bay (localities are identified in Figure 4.1) and a bed south of Pelican Islandwhere a bright feature was identified, like a sandy scar in the middle of a Zostera meadow. The former never regrew as Zostera but the latter area was regrown as Zostera as shown in the 1995 image.

The overall change in Zostera has been heaviest in the middle and southern parts of the lake, in particular Coomba Bay, Little Flat Point, between Brushy and Deepwater Points, and south of Pelican Island (see Figure 4.9). Field surveys and guides (eg Laegdsgaard, 2001) have identified new Posidonia beds along sections including Little Flat Point. There appears to be some Posidonia ‘gain’ (that is, addition) along this region together with some small additions along the south west side of the central basin.

There are two small areas of Zostera gain between Pacific Palm and Booti Island and Little Snake and Pelican Island. The latter region seems to be at the expense of the Ruppia/Halophilia seagrass, which otherwise has remained reasonably stable.

The change in Zostera distribution from 1988 to 2002 may be exaggerated as the classification has identified several macroalgae classes due to a heavy infestation of epiphytes or replacement of Zostera entirely in some places. If the Zostera still exists under this macroalgae class then it needs to be identified and labelled accordingly. It would probably need a higher spectral resolution sensor to resolve epiphytes on seagrasses. A more focussed field campaign and spectral analysis of the southern basin of the lake would clarify this issue.

4.5 Validation field trip results

Two validation field trips were conducted by DWLC and GLC in October and November 2002. The results and comparison with the September 2002 Landsat based classification are tabulated below. The field validation was undertaken with specific intent to check the results of the classification, in particular, classes not aligned with the field spectral measurements were investigated. It is important to note that these field campaigns were designed to check regions in the validation that were uncertain.

A difficulty in interpreting the results of this comparison is that both the field GPS and the Landsat data have unclear spatial accuracies. It is estimated that the Landsat imagery is accurate to about 1.5 pixel, equivalent to 45 m. The GPS used on the 10 October was different from the GPS used on the 1 November. Personal communication by Graham Carter mentioned that the GPS used on 1 November did not have the required accuracy (to adequately pin point field information with image data) or a lesser accuracy than the one used on 10 October 2002, thus the table below should be interpreted with caution. In addition, we are comparing field data from 10 October 2002 (Table 4.3) and 1 November 2002 (Table 4.4) with Landsat data of 12 September 2002.

Page 38: Seagrass Change Assessment Using Satellite Data for Wallis

32

WGS84 Northings

WGS84 Eastings

Field Result 10 Oct 2002 based on point measurements

Landsat TM 7 September 2002 Classification Result

Match (0=no, 0.5= next pixel, 1=yes)

452376 6437449 Healthy Posidonia 160mm, to the west, Gracilaria and Cystoseira

Ruppia & Halophila (inshore from a posidonia bed) 0

453560 6431744 Green Island Posidonia Posidonia 1 453890 6434384 Posidonia Posidonia 1 453860 6433184 Posidonia Posidonia 1 453904 6434426 600mm healthy posidonia Posidonia 1 453800 6431744 Halophila and Ruppia Ruppia & Halophila 1 453860 6428744 Halophila and Ruppia Ruppia & Halophila 1 453710 6432764 Zostera Zostera 1 454040 6429314 Zostera Sparse Ruppia/Halophilia

(on edge of Zostera bed 0.5 454400 6430274 Zostera and Algae? Zostera & Macroalgae 1 453080 6430184 Chara? Chara 1 453200 6430454 Chara? Chara 1 452660 6429014 Zostera Sparse Zostera 0.5 452840 6428804 Sparse Halophila And

Ruppia Posidonia/Sparse Ruppia/Halophilia 0.5

454910 6423134 Macroalgae Zostera/Macroalgae 0 454160 6423794 Chara? Zostera/Macroalgae 1 451910 6423224 Macroalgae Zostera/Macroalgae 0.5 453260 6423704 Zostera and Posidonia Zostera & Macroalgae 0.5 454190 6421844 Sparse Zostera unclassified - on land? 452180 6425324 Posidonia Posidonia 1 Correctly Matched

76%

• Table 4.3: 10 October 2002 Field Trip Results (provided by G. Tuckerman - GLC) versus the classification results for the least understood areas. The results for other areas are significantly better, but not quantified. These results represent a ‘worst case’.

Page 39: Seagrass Change Assessment Using Satellite Data for Wallis

33

WGS84 Eastings

WGS84 Northings

Field Result 1 November 2002 Landsat TM 7 12 September 2002

Classification

Match (0=no, 0.5= next

pixel, 1=yes)

453635 6427827

Isolated Posidonia Patches Surrounded by semi bare sand located in central eastern section of lake. 10m x10m -central point Ruppia 0

453775 6428419

Isolated Posidonia Patches Surrounded by semi bare sand located in central eastern section of lake. 20m x 20m central point Ruppia & Halophila 0

455410 6422660

Chara (stonewort) silt covered -relative fine uniform layer - with chara forming continuous grass like bed. Points define polygon

Riverwater (next to Macroalgae & Chara) 0.5

455432 6422677 Riverwater (next to Macroalgae & Chara) 0.5

455462 6422709 Posidonia 0 455436 6422721 Zostera 0 455421 6422726 Zostera 0

454888 6423454

Uniform bed of "clean" (unsilted, no algae) Chara with numerous large shells > 20mm x 40mm water depth 1.5 m Macroalgae 1

454887 6423456 Macroalgae 1 454887 6423458 Macroalgae 1

454907 6423578 Zostera (on edge of Macroalgae bed) 0.5

454884 6423591 Zostera/Macroalgae (on edge of Chara bed) 0.5

455192 6423569 Green Chara bed interspersed with 30-50% Gracilaria Chara 1

455211 6423559

Sparse Zostera (on edge of Macroalgae bed) 0.5

455192 6423568 Chara 1

455192 6423606

Sparse Zostera (on edge of Macroalgae bed) 0.5

455157 6423594 Zostera (on edge of Macroalgae bed) 0.5

451933 6426629 Thick stemmed Zostera 1.5 m depth Zostera 1

451243 6426959 dead thick Zostera Chara (next to Zostera) 0.5

454903 6423449 Chara lawn with numerous shells Macroalgae 1

454543 6422749 chara lawn -no shells 1.3m depth

Posidonia surrounded by Macroalgae & Chara 0.5

455233 6423599 Gracilaria Unclassified 454663 6422879 Ruppia and Chara algae covered Chara 1

452105 6422190 Heavy algae covered Chara 60%/ Ruppia Zostera (next to Macroalgae & Chara) 0.5

452683 6421889 Gracilaria and mud Ruppia (next to Macroalgae & Chara) 0.5

452713 6421979 fine silt Macroalgae 0

Correctly Matched 54%

• Table 4.4: 1 November 2002 Field Trip Results (provided by G. Carter - DLWC) versus the classification results for the least understood areas. The results for other areas are significantly better, but not quantified. These results represent a ‘worst case’.

Table 4.4 shows that isolated and sparse Posidonia is mapped in the classification as Ruppia, which is logical as we have no sparse Posidonia signal in our spectral library, and Ruppia maps out the closest in the Maximum Likelihood classification. There is a tendency for Chara to be mapped

Page 40: Seagrass Change Assessment Using Satellite Data for Wallis

34

as Macroalgae which indicates the spectral library needs some refinement or a higher spectral resolution sensor should be used.

A general recommendation is to develop an integrated methodology for validating remotely sensed images with field data. This would then take into account the specific two dimensional information that is contained within 1 pixel, for example, a 30 by 30m area in the case of Landsat.

4.6 Suspended sediment (tripton) concentrations

A component of this study included the application of advanced processing methodologies for mapping water quality indicators such as total suspended matter, Secchi depth and Kd using Landsat data.

There are two modelling approaches to determine the relationship between the inherent and apparent optical properties of the water column and the water constituents: the analytical modelling and the radiative transfer modelling approach (see Section 2.2).

One method for retrieval of concentrations is a direct inversion of an analytical model using a linear matrix inversion method (MIM). The MIM has been applied to retrieve optical water quality variables from airborne hyperspectral imagery (Hoogenboom et al., 1998; Lee et al., 2001) and to satellite hyperspectral imagery (Brando & Dekker, 2002). Brando & Dekker (2002) retrieved concentrations of chlorophyll, CDOM and tripton (the non-phytoplanktonic part of the total suspended solids), inverting three spectral bands of Hyperion satellite hyperspectral reflectance imagery of Deception Bay, Queensland. Using this methodology images of CDOM, CHL and tripton concentrations in the Bay were produced. Due to the low sensitivity of the Landsat sensor and the low concentration of chlorophyll found in Wallis Lake only tripton concentrations were retrieved.

The September 2002 Landsat image was masked to exclude depths of less than 2.5 meters, where visibility of the substrate would be likely. The MIM must be applied to regions where the subsurface irradiance reflectance signal is over an optically deep water column (that is, where water depth is greater than the secchi depth). During the August 2001 field trip secchi depths of up to 2.0m (Table 2.3) were measured, therefore, with this information and analysis of the 12 September 2002 image, a cut-off of 2.5m was estimated as being the maximum secchi depth for that image. The MIM was applied on the image at greater than 2.5m and Figure 4.10 displays the result. The concentrations retrieved are within the range measured during the August field trip (Table 4.5), for example, the WL24-1 field site visited in August 2001 had a tripton concentration of 12.5 mg L-1 and the MIM result for the September imagery retrieved 11.4 mg L-1.

The map of tripton concentrations, Figure 4.10 (displayed in 30 by 30 meter pixels) achieves a spatial density never to be obtained operationally by in situ measurements.

In the optically deep areas of the lake given in Figure 4.10, it may be possible to also map secchi depth and Kd, provided that tripton is the main optical component. More focused field and laboratory research would explicitly address the issue of Landsat mapping secchi depth and Kd in the Wallis Lake environment.

Page 41: Seagrass Change Assessment Using Satellite Data for Wallis

35

• Figure 4.10: Tripton concentration for the Landsat 7 image from 12 September 2002 for water depths greater than 2.5 meters. The field site positions from the 22-24 August 2001 campaign are displayed above and the results of the TSS measurements during this time are listed in Table 4.5.

Sample Site description Tripton (mg L-1) WL22-1 Wallamba River 15.4 Wl22-3 Pipers Creek Channel 16.6 WL23-5 Pacific Palms 13.0 WL23-8 Yahoo Island 20.5 WL23-10 Coolongolook River 17.2 WL24-1 Tidal Inlet 12.5 WL24-2 Steps 11.5

• Table 4.5: Tripton concentration measured in Wallis Lake during 22-24 August 2001, as located on Figure 4.10.

Page 42: Seagrass Change Assessment Using Satellite Data for Wallis

36

5 Conclusions and Recommendations of the Research

There are two levels of conclusions. One level concerns the conclusions from this research with respect to the potential of satellite remote sensing for mapping substrates and determining change over time, which will be addressed in this section. The other level of conclusions and recommend-ations concern the question of what would be the necessary level of investment to make any of these methodologies operational for shallow water mapping purposes, which will be addressed in Part 2 of this report.

5.1 Results of this research

As we intended to follow as much as possible a geophysical approach to measurements and modelling of the image and in situ optical data and the inversion of this data to substrate type it was necessary to:

• process the remotely sensed imagery to subsurface irradiance reflectance,

• process the in situ measurements of the water column and the substrates,

• parameterise the radiative transfer and analytical models to enable simulation of remote sensing type spectra from a shallow lake environment, with variables of the light field that are consistent with each other.

In more practical terms, a significant effort was spent on assuring that the satellite radiance data, the in situ optical data, and the output from the optical models converged to the same reflectance (see Figures 2.9 & 2.10) and vertical attenuation values. In the optical oceanographic community this is often referred to as “optical closure”.

Optical modelling

A reasonable closure (within signal to noise capabilities) was achieved between the Landsat atmospherically and air-water interface corrected data to R(0-) and the in situ measured R(0-), especially at Site WL22-3. The HYDROLIGHT 4.1 model was correctly parameterised and was used as a tool to simulate the effects of bathymetry, benthic substrate and depth, turbidity, wind speed on the reflected signal from the lake.

For the other locations HYDROLIGHT could approach only the shape of the Landsat image spectra, so the type of substrate or mix of substrates at a particular site influences the magnitude of simulated reflectance (this does require further research to resolve). Once this stage was reached the Landsat imagery was used to estimate an in situ measured reflectance at each pixel at site WL22-3. Furthermore, all permutations of concentrations can now be simulated as long as the parameterisation of the model with specific inherent optical properties is correct.

At this stage of “knowledge” it becomes possible to simulate thousands of possible permutations of combinations of concentrations, substrates and substrate depths with which it is possible to train neural networks, test analytical inversion algorithms etc. As much time was spent achieving this closure, it was not possible to carry out a sensitivity analysis using the fully parameterised HYDROLIGHT model. It is recommended to perform this type of research to understand the effects of changes in any optical condition (turbidity, concentration, substrate cover density) on the R(0-) and on the products that are derived from R(0-).

Page 43: Seagrass Change Assessment Using Satellite Data for Wallis

37

Benthic substrate classification

From our field measurements and from the Landsat imagery it appeared that Wallis Lake has a complex substrate environment. Unvegetated substrates vary from muddy to silty to sandy to coarse sand. Vegetated substrates vary from very lightly vegetated with Halophila to densely vegetated areas of mixtures of Zostera, Posidonia and various algae (e.g. at Pipers Creek and Pacific Palms) to monospecies fields of Zostera or Posidonia at various levels of density.

The Maximum Likelihood Classifier was selected as the most appropriate tool for this classification, after limited success using various supervised techniques (such as Spectral Angle Mapper [SAM]) and unsupervised techniques (such as isoclassification). Maximum Likelihood separated the substrate into classes that seemed comparable with the Wallis Lake Study vegetation map produced by West et al., (1985).

In order to reduce the dimensionality of the data to meaningful groupings/communities, use was made of as much information as was available concerning the actual substrates - the four sources of information were:

• Published classification of the substrates and benthic communities in the region based on field transects in and Laegdsgaard (2001), Webb et al., (1999) and West et al., (1985)

• the field data collected on 22-24 August 2001

• the results of the HYDROLIGHT modelling,

• an existing spectral library on substrate vegetation at CSIRO.

Quality of the Landsat image affected the quality of the resultant classification. Regions with visible striping impacted on the classification by causing mislabelling of classes whereas high quality images (such as the 12 September 2002 image) had little evidence of this in the classification.

From the change detection classifications from 1988 until 2002, the seagrass Zostera seems to be in decline. Posidonia, Ruppia and Halophila seem to be stable with no gross changes in the 14 year period from 1988 until 2002. However, the Zostera has undergone significant change and adaptation. From quite early in the time series (between 1988 and 1991) there appeared to be a reduction in Zostera beds, with some areas not returning by 2002. The overall change in Zostera has been heaviest in the middle and southern parts of the lake, in particular Coomba Bay, Little Flat Point, between Brushy and Deepwater Points, and south of Pelican Island (see Figure 4.8). This loss of Zostera from 1988 to 2002 may be exaggerated as the classification has identified several macroalgae classes due to a heavy infestation of epiphytes or replacement of Zostera entirely in some places. If the Zostera still exists under this macroalgae class then it needs to be relabelled. Hyperspectral remote sensing could resolve this issue more clearly.

This project has compiled satellite imagery spanning 14 years over Wallis Lake and has monitored the change in seagrass communities. Over this large area remote sensing has significant advantages over other more traditional techniques. After classification and validation of the data sets, remote sensing becomes even more cost-effective, particularly with regards to multi-temporal change detection assessment. It is also a very effective method for monitoring the effect of management controls applied in that period.

Remote sensing methodologies based on digital data and using methodologies that are well documented have the additional advantage of being objective and repeatable. This project has been based on archived data available from the Landsat sensors, but data with higher spatial and spectral resolution and better signal to noise ratio is, and will be, available. These improvements will result in higher quality, more accurate data, however, it is unclear whether mapping exact change is really environmentally relevant. These issues need discussion with GLC and DLWC before future recommendations be made.

The final results of Landsat classification has surpassed most expectations where the original mission was to separate seagrass as a whole from macroalgae and bare substrate. The patterns of the seagrasses spatial distribution together with maps of Tripton are valuable for the management of Wallis Lake as they achieve a spatial density not possible with traditional in situ methodologies.

Page 44: Seagrass Change Assessment Using Satellite Data for Wallis

38

5.2 Future assessments – where to from here?

CSIRO recommends that several key locations be identified for multi-temporal assessment. These locations should be used as indicator sites in the long term monitoring of the Lake. The sites should be chosen with regard to varied usage and species. As Posidonia is less variable over the seasons than Zostera, the monitoring should occur in similar seasons and climatic conditions (that is, similar rainfall). A general recommendation is to develop an integrated methodology for validating remotely sensed images with field data. This would take into account the specific two dimensional information that is contained within 1 pixel, for example, a 30 by 30m area in the case of Landsat images.

An analysis of hyperspectral data would provide higher separability and spatial and spectral variation than Landsat imagery. Although there would be a significant increase in the cost of data acquisition, hyperspectral data provide many more opportunities than multispectral imagery. Hyperspectral data have been used to successfully map rock platform vegetation, mangroves, salt flats and water quality parameters such as total suspended sediment (TSS), chlorophyll and carbon dissolved organic matter (CDOM) concentrations. The results of the benthic vegetation spectra in Figure 2.5 are an illustration that most species will be able to be discriminated using hyperspectral remote sensing data.

As the DLWC and the GLC are also concerned with the whole catchment area, it would be even more worthwhile and cost-effective to look at the joint needs for remote sensing data for the lake and the catchment. An airborne hyperspectral survey or a Quickbird image coverage of the Lake plus relevant surrounding catchment could a valuable management resource for current and future management.

After consultation (with CSIRO, GLC & DLWC), a Quickbird image was acquired on 24 January 2003, covering most of the Lake. Quickbird imagery has 0.64m resolution in black & white and 2.4m pixels in the first four Landsat bands. The cost of acquisition was $4450 after a promotional discount of 40%. Figure 5.1 illustrates the coverage and resolution provided by this dataset. The Quickbird colour data in Figure 5.1 has been enhanced to emphasise the water features. Fine details can be seen and identification of small patches of seagrass species can be matched clearly with field data. As Quickbird data has approximately the same spectral characteristics as those used in the Landsat classification of this project, it has the potential to provide greatly increased spatial accuracy. As the spatial resolution is much finer than Landsat, there would be less pixel mixing (the averaging of all the spectra within the pixel dimensions), and more pure pixels, which also improves the capacity of the classifier to distinguish species. Moreover, Quickbird data allows additional analysis of texture and pattern!

Page 45: Seagrass Change Assessment Using Satellite Data for Wallis

39

• Figure 5.1: Quickbird satellite images acquired over Wallis Lake, 24 January 2003.

Page 46: Seagrass Change Assessment Using Satellite Data for Wallis

40

5.3 Transferability of methodology to management authorities

In a subsequent study the gaps in knowledge could be filled in after which fieldwork and methodology development would not require significant attention. What would require attention is the development of a protocol, defining the methodology together with the development of faster processing methodologies. Initially these developments would still need to be done at CSIRO, although (depending on the level of sophistication of the LMA requirements) it is conceivable to initiate a training of highly skilled GIS or laboratory staff at the LMA to carry out more of the processing. Current developments in satellite remote sensing of oceans could be emulated such as: the generation of a large number of lookup tables for atmospheric correction and air-water interface correction; and the generation of spectral libraries that simulate most seagrass, macro-algae species and substrate types with varying depths and types of water column. Once such data are available, fast processing methodologies such as neural networks can be automated to process the satellite remote sensing data. This could require a period of 2 to 5 years to accomplish (depending on the level of involvement by all the involve agencies and developers). Once accomplished it could be easily applied to all NSW coastal lakes.

5.4 Benefits of this research

For large area coverage and for remote areas remote sensing has significant advantages over other more traditional techniques. Fieldwork can be done in accessible parts of a further inaccessible area. With the information derived from the fieldwork, image classification methods (based on supervised/unsupervised methods or on geophysical inversion) can be applied to the entire image(s). This has the benefit of being applied on inaccessible regions of the site and on a different date. Moreover, after initial (and perhaps secondary validation) fieldwork the requirement for fieldwork drastically diminishes. Thus, remote sensing becomes more cost-effective as it is applied more often, especially in multi-temporal change detection mode such as in an auditing process that needs to be carried out once every year, 2 years or more, in order to detect the effect of management controls applied in that period.

Remote sensing methodologies based on digital data and using methodologies that are well documented have the additional advantage of being objective and repeatable. This would be difficult to achieve if, for example, field-teams had to do similar analyses over extensive areas. To summarise, the benefits of remote sensing are:

• Production of accurate spatially comprehensive benthic substrate maps • Large area coverage without associated extra costs • Mapping inaccessible regions • More cost effective than fieldwork to survey the entire region • Multi-temporal analysis, ie archival data available and comparable with future data sets • Repeatable and objective – not operator dependent • Rapid assessment • Methodology applicable to data sets of improved spectral and spatial resolution • Easy assessment of environmental change processes enabling adaptive management of the lake system (and surroundings, if included) • In the future, increased spatial, spectral and temporal resolution will enable remote sensing data and information to be the tool of choice for monitoring the environmental state and change detection

Page 47: Seagrass Change Assessment Using Satellite Data for Wallis

Part 2

PATHWAYS FOR

IMPLEMENTATION OF REMOTE SENSING AS AN

ENVIRONMENTAL MONITORING TOOL FOR

COASTAL LAKES

Page 48: Seagrass Change Assessment Using Satellite Data for Wallis
Page 49: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

43

6 Pathways for Implementation of Remote Sensing as an Environmental Monitoring Tool for Coastal Lakes

6.1 Scope and aim

The results of the project “Seagrass Change Assessment using Satellite Data for Wallis Lake” (see Part 1) clearly indicate that it is feasible to detect changes in benthic vegetation type and cover using the Landsat satellite data. Landsat data were chosen for the fact that they are available for 1984 until now at 30 m pixel resolution (with a standard image size of 180 by 180 km). Extensive archives exist of Australia resulting in data costs that are low per unit area.

The attractiveness of the methodology is that it is objective and can be repeatedly applied. The data source is objective, unlike many human interpreter based methodologies. The data per square kilometre is cheap and all encompassing: no extrapolations are necessary from point measurement to the entire area.

For an organisation like DLWC the attractiveness of investing in remote sensing lies in the fact that as it is applied to larger area or more lakes the amount of work decreases per lake and with each next application.

Once a complete spectral library of substrates and water types are available (for example - NSW coastal lakes), one methodology can map all lakes. This is in contrast with the traditional methods that require a large staff presence in the field for each mapping exercise. In addition the data collected (as it is at high spectral resolution) can be easily adapted to any current or future optical remote sensing system. Thus, the field and laboratory analyses have permanent value.

More and more sensors are becoming available leading to cheaper data pricing. Each time more of this type of work is done in Wallis Lake, the classifications can improve and be retrospectively applied to the remote sensing data archive. The collection of a spectral library of substrates, seagrasses and macro-algae in Wallis lake (or similar lakes) will lead to a situation where it will not be necessary to perform in situ field campaigns to be able to analyse the remote sensing data.

It becomes important to follow up this scientific and exploratory work to determine how other similar satellite or airborne digital remote sensing can be used operationally by lake management authorities. This requires an analysis of the lake water management authorities needs for detection and monitoring of environmentally relevant changes to the lake ecosystem.

We need to explore the current and near future capabilities of the remote sensors from satellites and from aircraft for addressing these needs. A very important step is to present scenario’s for operational use of remote sensing derived spatial and temporal information. Aspects need to be discussed such as accuracy and reliability, scale, costs, transfer of methodology etc.

Part 2 of this report will end with recommendations in the short term (one year) and in the mid-to-longer term for implementation of remote sensing as an environmental modelling tool for coastal lakes.

Page 50: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

44

6.2 Methodology development requirements based on lake management authorities needs

The requirements of lake management authorities are 1. to understand the changes from the past to the present as an indicator of system variability due

to natural and anthropogenic influences 2. to understand the effect of current management practices on the lake ecosystem 3. to predict the effects of current and future management on the lake ecosystem 4. to be enabled to deal with calamities (both natural and anthropogenic) In general there appear to be 3 types of uses of remote sensing for water management authorities: 1. real time or near real time use for emergency management 2. project based use 3. monitoring and change detection Lake management authorities (LMA) want to assess environmentally relevant changes in their lake ecosystems. If this needs to be addressed by remote sensing a dialogue is necessary between the LMA and the remote sensing methodology developers to establish what would be the most suitable and feasible indicators to be detected and monitored by remote sensing. In Part One of this report we presented results of our research work. It represents a subset of the most suitable historical satellite data. A more extensive archival research is possible if deemed necessary by the LMA. In this previous research we attempted to determine the maximum extent to which seagrass, macro-algae and substrate type could be discriminated without limiting the number of species to be studied. Thus, a crucial recommendation for further work is for the LMA authorities to decide on the relevant species and substrate cover types to address their environmental assessment and change detection needs.

We will consider the two uses of remote sensing that do not have considerable infrastructural requirements (that is, we are excluding the near real time requirement): project-based and monitoring type applications. Project-based remote sensing can often be more expensive than monitoring-based remote sensing as there are few repetitive costs and often detailed information is required. For monitoring purposes (depending on the frequency: that is, once every three months or once every three years) remote sensing costs play a more important role.

There are a range of scale issues for which the LMA need to determine their requirements.

• Temporal scale requirements are based on the changes and the speed of change in indicators that is anticipated or deemed necessary to determine. Filtering out the effects of seasonal changes on more overriding trends of change are factors that may need to be included.

• Spatial scale required is also of importance (and an important driver of costs). The higher the spatial resolution (that is, the smaller the pixels) the higher the data acquisition and processing costs become. Spatial scale is linked to accuracy too. It is probably safe to say that one should not rely on the change of one pixel in the image as being relevant: it could be a geolocation mismatch of anything up to one pixel (quite a high standard of geocorrection). What is relevant however is the total area coverage of a group of pixels that classify a certain type of substrate. For example an increase of Posidonia from one to two pixels in an area may seem like a doubling but is possibly caused by the patch of Posidonia being fully imaged in one pixel in the first image but being split over two pixels in the second image. However a patch of Zostera of 80 pixels increasing to 90 pixels is relevant as pixels that go from a coverage of one to being split over two pixels are compensated by pixels going from half cover to no cover.

• Spectral scale mainly relates to how many species and substrate types (mud, silt, sand) are needed to be imaged and mapped. The higher the spectral resolution the higher the discrimination of species becomes. It might become feasible with high spectral resolution remote sensing data to determine overgrowth of algae on seagrass leaves.

• If water column information is also required (eg chlorophyll, CDOM etc) then more spectral resolution is required.

Page 51: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

45

If price was no concern most LMA would opt for high spatial resolution airborne hyperspectral data as it addresses all the above mentioned issues at the highest level.

6.3 Requirements for operationalisation of remote sensing

If we presume the LMA define their requirements with respect to indicators, frequency, scales and costs, it becomes possible to establish pathways for more operational implementation of remote sensing.

In the initial stage (reported in Part One) fieldwork is performed, laboratory analyses are done, remote sensing image processing methods are developed (including atmospheric correction and air-water interface correction) and tested. Several incremental loops are required to finetune the results to the best possible given the initial inputs. Often it is found that the fieldwork was not performed in optimal locations (concerning the capture of most present species and types of substrate cover) as the full spatial analysis from the satellite imagery was not yet available.

In a subsequent study the gaps in knowledge could be filled in after which fieldworks and methodology development would not require attention. What would require attention is the development of a protocoled methodology and the development of faster suitable processing methodologies. Initially these developments would still need to be done at CSIRO, although (depending on the level of sophistication of the LMA requirements) it is conceivable to initiate a training of ‘highly’ skilled GIS or laboratory staff at the LMA to carry out more of the processing.

Current developments in satellite remote sensing of oceans could be emulated such as: the generation of a large number of lookup tables for atmospheric correction and air-water interface correction, and; the generation of spectral libraries that simulate most seagrass and macro-algae species as well as substrate types with varying depths and types of water column.

Once such data are available, fast processing methodologies such as neural networks can be trained to process the satellite remote sensing data automatically. This could require a period of two to five years to accomplish (depending on the level of involvement by all the involve agencies and developers). Once accomplished it could be easily applied to all NSW coastal lakes.

It is highly recommended to carry out at least one high spatial and spectral resolution remote sensing analysis for the lake in order to determine a baseline (time zero) that can act as an environmental information system for the next few years. This is probably best carried out by an airborne remote sensing system such as CASI or HYMAP. Although it is not yet established what high spatial resolution satellite sensors (such as QuickBird, IKONOS and SPOT5) can accomplish.

Subsequently based on LMA requirements, high spatial resolution satellite data or medium resolution data could be used for more frequent low cost change detection.

Page 52: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

46

6.4 Pricing of remote sensor data and derived indicators and end-products

Introduction

Costs of implementing remote sensing derived information within a management organisation can be split into four components:

1. Raw remote sensing data acquisition costs and necessary field and laboratory

observations 2. Preprocessing and processing of remote sensing data to variables such as seagrass

species 3. IT infrastructure that enables fast, reliable delivery of relevant data from source to end-user 4. Integration of spatially explicit remote sensing derived relevant environmental information

into knowledge system of end-user management authority.

Raw remote sensing data acquisition costs

Raw remote sensing data costs are often seen as a bottleneck for applications of remote sensing for management agencies involved in inland, coastal or marine research. There are several reasons why this is, perhaps, the wrong way of regarding these costs.

The issue is that remote sensing is often competing with established ways of monitoring. These established ways of monitoring are often much more expensive and much less spatially encompassing than remote sensing data. For example sending out a boat and crew to perform chlorophyll measurements costs (if all costs are included) approximately $5k to $20k per day. For that 6 to 15 point samples are taken and deemed to be representative of the site.

Remote sensing (depending on the sensor) will give for the same amount of money (and far less if used correctly) fully spatially co-registered information on chlorophyll, total suspended matter, seagrass and macro-algae distribution.

On the other hand LMA’s need to maintain their existing information gathering infrastructure till remote sensing has proven its reliability. The consequence is that remote sensing derived information is initially seen as an extra cost and then has to compete against all other innovations clamouring for attention by LMA’s.

Some general rules exist for the costs of different types of remote sensing data acquisition. Satellite data are comparatively low cost per area unit for fine to coarse spatial resolution data. When considering time series, it is important to note that archived satellite data is often discounted for historical data over a particular area.

Satellite data processing costs are often low per unit area (as compared to airborne remote sensing techniques) as there is generally lower data volume per area. The new generation of high spatial resolution, mid-spectral resolution sensors such as Quickbird and IKONOS are more intermediate in costs between airborne hyperspectral and satellite multispectral data. Satellite data taken at regular intervals and at the same time of day, is less flexible then airborne data.

For airborne remote sensing, aircraft and pilot costs are about the same, regardless of the sensor on-board. Advanced airborne sensor data (e. g. hyperspectral) is becoming cost-competitive with air photography. Airborne data processing, analysis and ‘value-adding’ varies depending on levels of processing, atmospheric correction, data volume and level of spatial accuracy (3 metre RMS, or better). Hyperspectral scanner data have been shown to be more cost-effective in some applications compared with a more traditional survey involving standard field sampling and air photo interpretation (Mumby et al., 1999). As hyperspectral data can deliver many more indicators the cost per indicator per unit area is an interesting parameter (see figure 5.1). Also, once available an aircraft based sensor has maximum flexibility in targeting, multiple images, stereo imaging etc. On a more general note it is advisable to seek all spatial information requirements within organizations and between organizations dealing with spatial data as shared data-acquisition for multiple purposes can significantly decrease the cost per indicator per area.

Page 53: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

47

Price per indicator for a 2 m resolution hyperspectral airborne campaign

0

50

100

150

200

250

300

350

400

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

km2

Pri

ce p

er in

dic

ato

r (i

n t

ho

usa

nd

s A

UD

)

2 indicators 4 indicators 6 indicators 8 indicators 10 indicators 12 indicators 14 indicators

16 indicators 18 indicators 20 indicators

• Figure 5.1: Cost per indicator for a hyperspectral airborne remote sensing application.

Approximate Costs of Airborne Imaging Spectrometryper unit area per resolution of pixels

50

100

150

200

250

300

350

400

450

100 200 300 400 500 600 700 800 900 1000

Area surveyed km2

cost

s (t

ho

usa

nd

s o

f A

UD

)

1 m 2 m 4 m 10 m

• Figure 5.2: Cost scenarios for airborne spectrometry

Page 54: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

48

Costs of Quickbird 2.4 m satellite data(note that polygons may be acquired to optimize coverage)

0

10

20

30

40

50

60

70

0 100 200 300 400 500 600 700 800 900 1000

Area surveyed km2

cost

s (t

hou

san

ds

of

AU

D)

2.4m

• Figure 5.3: Cost scenarios for Quickbird satellite data.

Page 55: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

49

Table 5.1 outlines the contemporary sources of image and surface data that are deemed most applicable for water quality monitoring and assessment. Costs are in Australian dollars (unless specified) and are valid at the time of writing this report. In the interest of brevity this table contains abbreviations that represent various data source and types.

Notes for Table Interpretation

Data Type Code

Airborne

Photography A-Ph

Video A-Vd

Laser Depth Sounder A-Lds

Scanner Multi-spectral A-Ms

Scanner Hyper-spectral A-Hs

Radar A-Rd

Laser A-Ls

Satellite

Multi spectral (Fine resolution) S-MsF

Multi spectral (Medium resolution) S-MsM

Hyper spectral S-Hs

Radar S-R

Page 56: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

50

Spatial (pixel) resolution

Extremely fine <5m EF

Fine 5-20m F

Medium 20-250m M

Coarse 250 - >1000m C

Spectral Rake is a mode of collection where a limited number of points are collected across the scene.

Spectral resolution

Low panchromatic or analogue images L

Medium multi spectral, 3-20 spectral discrete bands Ms

High hyper spectral, contiguous spectral bands Hs

(typically between 20 and 300 bands)

Note:

User dependent = sensor deployment should occur according to the users specifications (date, time & location).

Weather dependent = sensor deployment should occur under optimal weather conditions

Page 57: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

51

Sensor Spectral Resolution Spatial Resolution Temporal Resolution Raw Data acquisition Cost (Indicative) AAAiiirrrbbbooorrrnnneee(((AAA---PPPhhh)))

Photography Panchromatic, colour, and colour infra-red

EF to F User and weather dependent $90 per frame

AAAiiirrrbbbooorrrnnneee mmmuuullltttiii---ssspppeeeccctttrrraaalll (((AAA---MMMsss)))

Digital Video Spectral filters EF to F User and weather dependent ~$300 per km2

Daedalus 1268 12bands 7 VIS/NIR, 3 SWIR and 2 TIR EF to F User and weather dependent

$2,700 per hour (2m geocoded product)

ADAR Spectral filters EF to F User and weather dependent Price on application AAAiiirrrbbbooorrrnnneee HHHyyypppeeerrr---ssspppeeeccctttrrraaalll (((AAA---HHHsss)))

CASI-spatial mode

Up to 20 bands VIS-NIR

EF to F User and weather dependent $200-400 per km2

CASI-spectral mode

Up to 256 bands VIS-NIR Spectral rake* User and weather dependent $200-400 per km2

HYMAP 126 bands VIS-SWIR

EF to F User and weather dependent $100-400 per km2 ($40,000 per day)

AAAiiirrrbbbooorrrnnneee LLLaaassseeerrrsss (((AAA---LLL)))

ALTM 3025 25,000 pulses/per/second EF to F User dependent ~$4,000 p/h

LADS 3.24x106 soundings per hour EF to F User dependent Price on application

SSSaaattteeelllllliiittteee MMMuuullltttiii---ssspppeeeccctttrrraaalll (((SSS---MMMsssMMM))) --- MMMeeedddiiiuuummm rrreeesssooollluuutttiiiooonnn

Landsat TM 7+

8 bands (1 Pan, 4VNIR, 2SWIR, 1TIR)

30m multispectral (15m panchromatic)

16 days Superscene (240km North-South*250km E-W) $3240 Equivalent to $0.06 per km2 Subsets available at reduced cost eg $1500 185x185km

SPOT 4 4 bands 20m multispectral (10m pan)

26 days Superscene (140km North-South*260km E-W) $3500 Equivalent to $0.10 per km2 Subsets available at reduced cost

SPOT 5 5 Bands (3VNIR, 1SWIR+1 Pan)

EF (2.5m panchromatic) F (10m VNIR)

26 days

Full Scene (60kmx60km) $5000 standard 10m colour archive data equivalent to $1.40 per km2 $4360 for 169km2 2.5m colour archive data ($26 per km2) to $19000 orthorectified 2.5m colour ($530 per km2)

ASTER 14 (3VNIR, 3SWIR, 5TIR) 15m VIS 90m TIR 250m

16 days Need to request acquisition Free for validation, then $200 per scene

Page 58: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

52

MODIS 36 500m land 1000m water

Twice daily, or every other day Free

SSSaaattteeelllllliiittteee MMMuuullltttiii---ssspppeeeccctttrrraaalll (((SSS---MMMsssFFF))) --- FFFiiinnneee RRReeesssooollluuutttiiiooonnn

Ikonos 4 VNIR 1m panchromatic 4m multi-spectral

3 days Pointable sensor

From $12 per km2 pan archive data $37 per km2 for 4m multi-spectral $50 per km2 for 1m pan & 4m multi-spectral bundle Up to $290 per km2 -precision 1m pan & 4m multi-spec bundle Minimum order: archive data = 49km2, otherwise = 100km2

Quickbird 4 VNIR 0.61m panchromatic 2.5m multi-spectral

1-3.5 days depending on the latitude Pointable sensor

$43 per km2 for 2.4m multi-spectral $51 per km2 for 1m pan & 2.4m multi-spectral bundle or 4-band pan sharpened data $290 per km2 -precision 1m pan & 4m multi-spec bundle Minimum order: archive data = 25km2, otherwise = 64km2

SSSaaattteeelllllliiittteee HHHyyypppeeerrr---SSSpppeeeccctttrrraaalll (((SSS---HHHsss)))

Hyperion 224 bands 30m 16 days (near TM) Pointable sensor

$3400 for 315km2 (7.5x42km strip) ($11/km2) $5100 for 1390km2 (7.5x185km strip) ($3.70/km2)

ALI

10 bands (1 Pan 6 VNIR 3 SWIR)

15m panchromatic 30m multi-spectral

16 days (near TM) Pointable sensor

$3400 for 1554km2 (37x42km strip) ($2.20/km2) $5100 for 6845km2 (7.5x185km strip) ($0.75/km2) If bundled with hyperion data ALI is only an additional $1,400 (for the 42 or 185km strip!)

• Table 5.1 : Current Remote Sensing Data Sources (December 2002) , modified from Lymburner (2001)

Page 59: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

53

In order to present some examples two sensor system image data types were chosen as examples. These two types of data have unique capabilities and cover many indicators.

• An airborne hyperspectral data collection (the example is for a 2 m resolution dataset, similar calculations can be done for any spatial resolution between 0.4 and 10 m)

• A Quickbird satellite data collection (2.4 m resolution multispectral bands similar to Landsat)

Airborne hyper-spectral data such as CASI (Compact Airborne Imaging Spectrometer) and HYMAP are high spatial resolution, flexible and are capable of delivering the most indicators at the highest level of confidence due to their high spectral and radiometric resolution. Satellite multispectral data (such as Quickbird and Ikonos) are very high spatial resolution (0.6 m black & white; 2.4 m multi-spectral) and are attractive as their cost per square kilometre are intermediate between high resolution airborne and medium resolution satellite and has similar capacity of indicator discrimination as the Landsat TM system.

The costs for acquisition of airborne imaging spectrometry are relatively high (Figure 5.2). A 2000 km2 area would cost approximately 670K AUD at 1 metre resolution but only 170K AUD at 10 metre resolution. The spatial resolution at which the data is flown has a significant impact on the cost. For a CASI the width of the imaged area ranges from 500 metres wide at 1 metre resolution to 5000 metres wide at 10 metre resolution. The amount of flying is therefore drastically reduced when the required spatial resolution becomes coarser. The overall costs level increase with increasing area is less than linear as economy of scale issues start playing a role. In the case of Wallis Lake key sites identified as being of ecological or environmental importance could be flown at high resolution, and the rest mapped using coarser imagery, which would substantially decrease costs. If a HYMAP or a CASI would be flown to only image the lake areas where substrate vegetation is likely to be visible the costs could go down to 40K AUD .

The costs of acquisition of Quickbird satellite data are lower than for the airborne imaging spectrometry example (Figure 5.3). For a 1000 km2 area the costs are 65K AUD. Quickbird allows end-users to select any shape of image as long as the polygon sides have a minimal length of 5 km (that is 25km2 is the minimum for archived data). However Quickbird data is multi-spectral only and can thus not resolve as many indicators as airborne imaging spectrometry.

Figure 5.4 shows a subset of the Landsat scene of Wallis Lake (near Yahoo Island) with an example of Quickbird data with the box denoting the location in the Landsat image. On Quickbird imagery (Figure 5.5) pattern and texture of the seagrasses becomes visible, something not possible with Landsat data resolution. Moreover, a 1.5 pixel accuracy in Landsat data equivalent to a 45 m variability, whereas Quickbird, with a similarly defined accuracy, gives a 3.6 m variability.

The acquisition costs for the raw data collection pricing for relevant sensors for a region the size of Wallis Lake (approximately 85 km2) are listed in Table 5.2.

Page 60: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

54

• Figure 5.4: The Landsat September 2002 image with the white box indicating the Quickbird extent in Figure 5.5

• Figure 5.5: A subset of the Quickbird image of Wallis Lake showing spatial details unavailable in Landsat

Page 61: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

55

Sensor Spectral Resolution Spatial Resolution

Cost $AUD

Landsat TM 7+

8 bands (4VNIR, 2SWIR + 1TIR)

M 30m (15m panchromatic)

$560 for 25km2 ($22.40/km2) $1500 185km2 ($8.11/km2)

ALI

10 bands (1 Pan 6 VNIR 3 SWIR)

M 30m (10m panchromatic)

$3400 for 1554km2 (37x42km strip $2.20/km2) $5100 for 6845km2 (7.5x185km strip $0.75/km2) If bundled with Hyperion data ALI is only an Additional $1,400 (for the 42 or 185km strip!)

Spot 5 1 Pan 3VNIR 1 SWIR

EF to F 2.5 or 5m Panchromatic 10m VNIR 20m SWIR

$1900 for 400km2 (Standard programmed level) 10m colour ($4.70/km2) $4360 for 169km2 (Precision 2B) 2.5m colour archive data ($26/km2) Up to $6060 for 169km2 (orthorectified programmed level 3) 2.5m colour ($36/km2)

Quickbird 4VNIR

EF to F 0.61m panchromatic 2.5m multi-spectral

$3660 for 2.4m multi-spectral ($43 per km2) $4400 for bundled 1m pan & 2.4m multi- spectral data or 4-band pan sharpened data ($51 per km2) Promotional prices often up to 40% off.

Ikonos 4VNIR

EF to F 1m panchromatic 4m multi-spectral

$3700 for 100km2 for 4m multi-spectral ($37/km2) $5000 for 100km2 for bundled pan & 4m multi-spectral data ($50/km2)

Hyperion 224 bands M 30m

$3400 for 315km2 (7.5x42km strip) ($11/km2)

CASI Up to 256 bands VIS-NIR

EF to F $17000-34000 ($200-400/km2)

HYMAP 126 bands VIS-SWIR EF to F

$8500-34000 ($100-400/km2)

• Table 5.2 The raw data collection pricing for relevant sensors for Wallis Lake seagrass mapping.- assuming an areal coverage of 85 km2 , the approximate size of Wallis Lake. (Currency conversions used: 1€=$AUD1.837, $US1=$AU1.696)

6.5 Concluding remarks

The main conclusion is that remote sensing is most cost-effective if it is applied over larger areas. There is some indication that high resolution remote sensing data gathering methodology may have higher overall costs but is not much different from cheaper methods of remote sensing if it is calculated in price per application/indicator per area mode. However, the main driver behind decisions of what sensor to use should be the LMA requirements for spatially comprehensive data and the use they would make of it.

For large area coverage and for remote areas remote sensing has significant advantages over other more traditional techniques (fieldwork can be done in accessible parts of a further inaccessible area). With the methodologies derived from the fieldwork, image classification

Page 62: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

56

methods (based on supervised/unsupervised methods or on geophysical inversion) can be applied to the entire image(s), also in inaccessible parts. Moreover, after initial (and perhaps secondary validation) fieldwork the requirement for fieldwork drastically diminishes. Thus, remote sensing becomes more cost-effective as it is applied more often, especially in multi-temporal change detection mode such as in an auditing process that needs to be carried out once every year, 2 years or more, in order to detect the effect of management controls applied in that period.

Remote sensing methodologies based on digital data and using methodologies that are well documented have the additional advantage of being objective and repeatable. This would be difficult to achieve if for example, field-teams had to do similar analyses over extensive areas.

Page 63: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

57

7 Acknowledgements

Thanks to Graham Harris (CSIRO) for initiating this study, Graham Carter (DLWC) for his useful comments and additions to this report (including the substrate photos). Gerard Tuckerman (GLC) & Pia Laegdsgaard (DLWC) provided field data and expert local knowledge. We also appreciated the help of Nicole Pinnel, Liis Sipelglass, Cesar Urrutia and Guy Byrne during this project. Peter Fearns (CSIRO Marine Research) provided valuable comments and suggestions in the review of this document.

8 References

Aas, E. (1987). Two-stream irradiance model for deep waters: Appl.Opt., v. 26, p. 2095-2101.

Anstee, J.M, Dekker, A.G., Byrne, G.T., Daniel, P., Held, A. & J. Miller (2000). Use of hyperspectral imaging for benthic species mapping in South Australian coastal waters. Presented at 10th Australian Remote Sensing and Photogrammetry Conference, Adelaide, Australia, The Remote Sensing & Photogrammetry Association of Australia.

Ball-AIMS (2000). Ball AIMS Data collection and processing report, Sydney Harbour Trial for DSTO Maritime Operations Division, Canberra, ACT, Australia, p. 55.

Brando, V. E. and A.G. Dekker (2002). (submitted). "Satellite hyperspectral remote sensing for estimating estuarine and coastal water quality," IEEE - Transactions on Geoscience and Remote Sensing.

Clementson, L. A., Parslow, J.S., Turnbull, A.R., McKenzie, D.C. and C.A. Rathbone (2001). The optical properties of waters in the Australasian sector of the Southern Ocean: Journal of Geophysical Research, v. 106, p. 31611-31626.

Dekker, A. G., Brando, V. E. , Anstee, J. M. , Pinnel, N. , Kutser, T., Hoogenboom, H. J., Pasterkamp, R., Peters, S. W. M., Vos, R. J., Olbert C. and T. J. Malthus (2001). Imaging spectrometry of water, In: Imaging Spectrometry: Basic principles and prospective applications: Remote Sensing and Digital Image Processing, v. IV: Dordrecht, Kluwer Academic Publishers, p. 307 - 359.

Dekker, A. G., and S. W. M. Peters (1993). The use of the Thematic Mapper for the analysis of eutrophic lakes: A case study in The Netherlands: International Journal of Remote Sensing, v. 14, p. 799-822.

Hoogenboom, H.J., Dekker, A.G. and J.F. De Haan (1998). "Retrieval of chlorophyll and suspended matter in inland waters from CASI data by matrix inversion," Canadian Journal of Remote Sensing, vol. 24, pp. 144-152.

Jerlov, N.G. (1976). Marine Optics. Elsevier, Amsterdam, The Netherlands.

Kirk, J.T.O. (1994). Light and photosynthesis in aquatic ecosystems. Cambridge University Press, 401p.

Laegdsgaard, P. (2001) A field guide for the identification and monitoring of the Seagrasses and Macroalgae in Wallis Lake, Land and Water Conservation, Centre for Natural Resources, NSW Government.

Lee, Z., Carder, K. L.., Mobley, C. D., Steward, R. G. and J. F. Patch (1999). Hyperspectral remote sensing for shallow waters: 2. deriving bottom depths and water properties by optimization: Applied Optics, v. 38, p. 3831-3843.

Lee, Z., Kendall, L. C, Chen, R. F. and T. G. Peacock (2001). (in press) Properties of the water column and bottom derived from Airborne Visible Imaging Spectrometer (AVIRIS) data: (JGR).

Lymburner, L. (2001). Estimating riparian vegetation functions in the Nogoa catchment of the Fitzroy river basin using remote sensing and spatial analysis. Confirmation Report.

Page 64: Seagrass Change Assessment Using Satellite Data for Wallis

PATHWAYS FOR COASTAL LAKE MONITORING CONSULTANCY REPORT

58

Department of Civil and Environmental Engineering University of Melbourne and CRC Catchment Hydrology CSIRO Land and Water.

Maritorena, S., Morel, A.. and B. Gentili (1994). Diffuse reflectance of oceanic shallow waters: influence of water depth and bottom albedo: Limnol.Oceanogr., v. 39, p. 1689-1703.

Mobley, C. D. (1994). Light and water; Radiative transfer in natural waters: London, Academic Press, 592 p.

Mobley, C. D., and L. K. Sundman.(2000a). HYDROLIGHT 4.1 Users' Guide, WA, Sequoia Scientific, Inc, p. 85.

Mobley, C. D., and L. K. Sundman (2000b), HYDROLIGHT 4.1 technical documentation, WA, Sequoia Scientific, Inc, p. 76.

Mumby, P. J. and A. R. Harborne (1999). Development of a systematic classification scheme of marine habitats to facilitate regional management and mapping of Caribbean coral reefs. Biological Conservation 88: 155-163.

Pope, R. M., and E. S. Fry (1997). Absorption spectrum (380-700 nm) of pure water. II. Integrating cavity measurements: Appl.Opt., v. 36, p. 8710-8723.

Wallis Lake Catchment Management Plan – Volume 1 – State of the Catchment Report http://www.greatlakes.nsw.gov.au/Environ/wlcmp/wIndex.htm

Webb, McKeown & Associates. (1999). Wallis Lake Estuary Processes Study. Forster-Tuncurry: Great Lakes Council; unpublished consultants report: p.1-129.

West, R.J., Thorogood, C., Walford, T. and R.J. Williams (1985). An estuarine inventory for New South Wales, Australia. Department of Agriculture, NSW.

9 Appendix: Data Suppliers

Aerial Photography

NSW Dept. Land and Property Inf. www.lpi.nsw.gov.au

Airborne Multi-spectral and Hyperspectral

Digital Video - Specterra Systems www.specterra.com.au Daedalus 1268 - Air Target Services www.airtargets.com.au CASI - Ball-AIMS (currently under review – contact SKM

for further details) HYMAP - Hyvista Corporation www.hyvista.com.au

Airborne LIDAR

LIDAR - AAM Geoscan http://www.aamsurveys.com.au/

Satellite Multi and Hyper-spectral

Landsat ETM - ACRES www.auslig.gov.au SPOT - ACRES www.raytheon.com.au ASTER - NASA http://asterweb.jpl.nasa.gov/ QuickBird – Digital Globe http://www.digitalglobe.com/ IKONOS - Space Imaging http://www.spaceimaging.com/ MODIS - NASA http://modarch.gsfc.nasa.gov/ Hyperion - NASA http://eo1.gsfc.nasa.gov/Technology/Hyperion.html