optimizing remote sensing and gis tools for mapping and ... · optimizing remote sensing and gis...

Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/233157429

OptimizingRemoteSensingandGISToolsforMappingandManagingtheDistributionofanInvasiveMangrove(Rhizophora....

ArticleinMarineGeodesy·May2007

DOI:10.1080/01490410701296663

CITATIONS

13

READS

124

4authors,including:

Someoftheauthorsofthispublicationarealsoworkingontheserelatedprojects:

LocallyManagedMarineAreasNetworkViewproject

BrineDischargeFromDesalinationPlants–ImpactsonCoastalEcologyandCoastalCommunities

Viewproject

StacyJupiter

WildlifeConservationSociety

131PUBLICATIONS1,419CITATIONS

SEEPROFILE

DonaldCameronPotts

UniversityofCalifornia,SantaCruz

82PUBLICATIONS1,768CITATIONS

SEEPROFILE

AllcontentfollowingthispagewasuploadedbySusanA.Cochranon20October2015.

Theuserhasrequestedenhancementofthedownloadedfile.

https://www.researchgate.net/publication/233157429_Optimizing_Remote_Sensing_and_GIS_Tools_for_Mapping_and_Managing_the_Distribution_of_an_Invasive_Mangrove_Rhizophora_mangle_on_South_Molokai_Hawaii?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_2&_esc=publicationCoverPdfhttps://www.researchgate.net/project/Locally-Managed-Marine-Areas-Network?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_9&_esc=publicationCoverPdfhttps://www.researchgate.net/project/Brine-Discharge-From-Desalination-Plants-Impacts-on-Coastal-Ecology-and-Coastal-Communities?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_9&_esc=publicationCoverPdfhttps://www.researchgate.net/?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_1&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Stacy_Jupiter?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_4&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Stacy_Jupiter?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_5&_esc=publicationCoverPdfhttps://www.researchgate.net/institution/Wildlife_Conservation_Society?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_6&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Stacy_Jupiter?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_7&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Donald_Potts?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_4&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Donald_Potts?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_5&_esc=publicationCoverPdfhttps://www.researchgate.net/institution/University_of_California_Santa_Cruz?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_6&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Donald_Potts?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_7&_esc=publicationCoverPdfhttps://www.researchgate.net/profile/Susan_Cochran2?enrichId=rgreq-9c386a0cc72efc9df1a191950493eda2-XXX&enrichSource=Y292ZXJQYWdlOzIzMzE1NzQyOTtBUzoyODY2NTEwNDI1NDk3NzZAMTQ0NTM1NDM0Mjg3MQ%3D%3D&el=1_x_10&_esc=publicationCoverPdf

Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=umgd20

Download by: [USGS Libraries] Date: 20 October 2015, At: 08:16

Marine Geodesy

ISSN: 0149-0419 (Print) 1521-060X (Online) Journal homepage: http://www.tandfonline.com/loi/umgd20

Optimizing Remote Sensing and GIS Tools forMapping and Managing the Distribution of anInvasive Mangrove (Rhizophora mangle) on SouthMolokai, Hawaii

Mimi D'iorio , Stacy D. Jupiter , Susan A. Cochran & Donald C. Potts

To cite this article: Mimi D'iorio , Stacy D. Jupiter , Susan A. Cochran & Donald C. Potts (2007)Optimizing Remote Sensing and GIS Tools for Mapping and Managing the Distribution of anInvasive Mangrove (Rhizophora mangle) on South Molokai, Hawaii, Marine Geodesy, 30:1-2,125-144, DOI: 10.1080/01490410701296663

To link to this article: http://dx.doi.org/10.1080/01490410701296663

Published online: 15 May 2007.

Submit your article to this journal

Article views: 211

View related articles

Citing articles: 5 View citing articles

http://www.tandfonline.com/action/journalInformation?journalCode=umgd20http://www.tandfonline.com/loi/umgd20http://www.tandfonline.com/action/showCitFormats?doi=10.1080/01490410701296663http://dx.doi.org/10.1080/01490410701296663http://www.tandfonline.com/action/authorSubmission?journalCode=umgd20&page=instructionshttp://www.tandfonline.com/action/authorSubmission?journalCode=umgd20&page=instructionshttp://www.tandfonline.com/doi/mlt/10.1080/01490410701296663http://www.tandfonline.com/doi/mlt/10.1080/01490410701296663http://www.tandfonline.com/doi/citedby/10.1080/01490410701296663#tabModulehttp://www.tandfonline.com/doi/citedby/10.1080/01490410701296663#tabModule

Marine Geodesy, 30: 125–144, 2007Copyright © Taylor & Francis Group, LLCISSN: 0149-0419 print / 1521-060X onlineDOI: 10.1080/01490410701296663

Optimizing Remote Sensing and GIS Toolsfor Mapping and Managing the Distribution

of an Invasive Mangrove (Rhizophora mangle)on South Molokai, Hawaii

MIMI D’IORIO,1 STACY D. JUPITER,2 SUSAN A. COCHRAN,3

AND DONALD C. POTTS2

1Department of Earth Science, University of California, Santa Cruz,California, USA2Ecology and Evolutionary Biology Department, University of California,Santa Cruz, California, USA3United States Geological Survey, Pacific Science Center, Santa Cruz

In 1902, the Florida red mangrove, Rhizophora mangle L., was introduced to theisland of Molokai, Hawaii, and has since colonized nearly 25% of the south coastshoreline. By classifying three kinds of remote sensing imagery, we compared abilities todetect invasive mangrove distributions and to discriminate mangroves from surroundingterrestrial vegetation. Using three analytical techniques, we compared mangrovemapping accuracy for various sensor-technique combinations. ANOVA of accuracyassessments demonstrated significant differences among techniques, but no significantdifferences among the three sensors. We summarize advantages and disadvantagesof each sensor and technique for mapping mangrove distributions in tropical coastalenvironments.

Keywords AVIRIS, ASTER, aerial photography, habitat mapping, classificationaccuracy, alien species management, red mangrove

Introduction

Mangrove habitats are ecologically and geologically valuable coastal resources that harbora great diversity of organisms, filter dissolved and particulate materials crossing the land-seainterface, and stabilize and protect eroding coastlines (Lugo and Snedaker 1974). Becausethe condition of a mangrove ecosystem is intimately linked to the interaction of marine,terrestrial, and meteorological factors, mangrove systems respond rapidly to changingterrestrial, oceanographic, and climatic conditions over short time scales. Because they arerestricted to higher intertidal levels, mangroves are one of the first systems to experiencesea-level rise and are potentially sensitive indicators of coastal change induced by bothnatural and anthropogenic forces (Blasco et al. 1996; Ellison and Farnsworth 1997; Alongi2002). While mangrove forests can be seriously damaged by severe storms (Imbert et al.1996; Sherman et al. 2001), they also buffer inland areas against strong storm surges.

Received 14 March 2006; accepted 22 December 2006.Address correspondence to Mimi D’Iorio, Department of Earth Science, University of California,

1156 High St., Santa Cruz, CA 95064, USA. E-mail: [email protected]

125

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015

126 M. D’Iorio et al.

For example, coastal areas fringed by mangroves during the 2005 Asian tsunami wereless damaged than cleared regions (Danielsen et al. 2006). Therefore, the ability to mapmangrove ecosystems, evaluate their structure and condition, and detect changes rapidlyand affordably is important for assessing regional to global environmental change.

Worldwide, most natural mangrove systems are declining from overharvesting, landreclamation, and aquaculture (Spalding et al. 1997; Alongi 2002), leading to large-scale ef-forts in many countries to restore degraded mangroves and preserve remaining natural man-grove systems. Hawaii, which lacked mangroves before European contact, provides an un-usual contrast because the success and rapid proliferation of introduced mangroves over thelast century has created other environmental concerns: habitat loss of four endangered water-birds, destruction of historic fishponds, and restructuring of the coastal wetland ecosystem(Allen 1998; Cox and Allen 1999). These impacts on native systems have encouraged localagencies to develop and implement alien species management strategies for mangroves.

Although the Hawaiian Islands are suitable climatic and geomorphic settings formangroves and propagules of some genera (e.g., Rhizophora) remain viable for severalmonths (Duke et al. 1998), biogeographic barriers to dispersal via the northern Pacificgyre prevented natural colonization prior to human introduction (Ricklefs and Latham1993). In 1902, the American Sugar Company deliberately imported the Florida redmangrove, Rhizophora mangle L., to Molokai to stabilize mudflats and protect the shorelinesand nearshore reef environments from increased agricultural runoff associated with thecultivation of the uplands for sugar cane (MacCaughy 1917). Moberley (1968) hypothesizedthat the later establishment of R. mangle in Kaneohe Bay, Oahu, over 50 km to the northwest,was the result of surface transport of pods driven by trade winds from Molokai to theadjacent island. R. mangle has now been positively identified on at least six of the eightmain Hawaiian Islands (Allen 1998).

The introduction of R. mangle substantially altered the dynamics and morphologyof Molokai’s southern coastal zone over the last century. Since 1902, R. mangle hasspread eastward along the Molokai coastline, leading to seaward progradation of theshoreline up to 400 m across the reef flat, filling of culturally important ancient fishpondsand encroachment onto private beachfronts (Wester 1981; D’Iorio 2003). Historical ratesof mangrove progradation (1915–2000) reflect increased sedimentation resulting fromchanging land use practices and are also correlated with watershed slope, coastal watershedarea, and average reef width (D’Iorio 2003). Mangroves may have improved water qualityalong some coastal reef flats by trapping terrigenous materials (Bigelow et al. 1989), butthey have also impeded drainage from narrow channels in urbanized areas, requiring costlyremoval (Wester 1981; Allen 1998; Chimner et al. 2006).

Other ecological consequences of mangrove introduction forecast by Egler (1947)have been equally dramatic. Contrary to early invasion theory which suggests that islandsare more susceptible to introductions because their biota are fragile and less competitive(Elton 1958), Simberloff (1995) proposed that the success of R. mangle in Hawaii is dueto the absence of native intertidal forest communities, rather than to superior competitiveabilities. The proliferation of R. mangle is hindering population recovery of four endangeredwaterbirds—the Hawaiian stilt (Himantopus mexicanus knudseni), Hawaiian duck (Anaswyvilliana), Hawaiian coot (Fulica alai), and Hawaiian moorhen (Gallinula chloropussandvicensis). Mangrove detritus, toxic to many detritivores, also may affect detritalfood webs in the sediment and pore waters (Alongi and Sasekumar 1992; Smith et al.2000). At the same time, the mangroves may be facilitating increases in some native andintroduced intertidal species by creating complex habitats within the network of prop rootsand increasing available energy supplies from accumulating leaf litter (Simberloff 1995).

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015

Mapping Invasive Mangroves on South Molokai 127

Remote sensing from spaceborne and airborne platforms has great potential forrapid assessment of rates of mangrove ecosystem change and for exploring impactsof introductions of alien mangrove species on natural systems, but all currentlyavailable remote sensing platforms require tradeoffs between cost, analysis time, level ofdiscrimination and accuracy. Despite extensive, labor-intensive post-processing demandsand the lack of global coverage (Lucas et al. 2002), aerial photography is still widelyused for mapping coastal environments, including mangrove systems (Saintilan and Wilton2001; Battiau-Queney et al. 2003; Higinbotham et al. 2004; Dahdouh-Guebas et al. 2004;Jupiter et al. 2006), but a growing body of work is using spaceborne remote sensingto investigate mangrove ecosystems worldwide. Multispectral satellite-based instrumentshave been used to define mangrove areas and assess baseline ecosystem conditions inmany places (Bina et al. 1978; Lorenzo et al. 1979; Dutrieux et al. 1990; Jensen et al.1991; Eong et al. 1992; Ramsey and Jensen 1996; Gao 1998; Ramirez-Garcia et al. 1998;Blasco and Aizpuru 2002; Haito et al. 2003; Jupiter et al. 2006), but analyses are oftenlimited by low spectral and spatial resolutions of the sensors (Green et al. 1998). Suchproblems may be overcome by the increased spectral resolution of hyperspectral sensors,such as the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), which offer manyadvantages for detailed wetland community mapping. For example, Hirano et al. (2003)used AVIRIS to identify mangrove plants to species level and to successfully map theinvasive lather leaf (Colubrina asiatica) in the Florida everglades. Yet widespread use ofhyperspectral technology is still limited by high acquisition costs, exceptionally large datafiles, long processing times, and difficulties in co-registering repeated overflights (Smithet al. 1998; Hirano et al. 2003).

We used the reef- and mangrove-fringed southern coastline of Molokai as a case study tocompare three kinds of remote sensing imagery, differing in spectral and spatial resolutions,and three analytical approaches for mapping invasive mangrove distributions. The imagetypes were color aerial photography, multispectral satellite Advanced Spaceborne ThermalEmissions and Reflectance Radiometer (ASTER), and airborne hyperspectral AVIRISdata. We analyzed each three ways to evaluate the ability to discriminate mangrove fromnonmangrove habitat. By integrating in situ field data within a geographical informationsystem (GIS), we assessed the accuracy of each combination of imaging sensor andanalytical approach and determined which provides the most effective and accurateclassifications.

This study had three goals:

1. To validate the use of three widely used remote sensing technologies, integratedwith field data and accuracy assessment, as a tool for mapping mangrove cover andtracking ecosystem responses to invasive species.

2. to assess the accuracy of three standard processing techniques used to analyze andclassify remote sensing data of varying spatial and spectral resolutions, and

3. to discuss the application of GIS and remote sensing for coastal management inHawaii.

Materials and Methods

Study Area

Molokai (21◦N, 157◦W) is the fifth largest island in the Hawaiian chain, covering about676 square kilometers and rising 1515 m above sea level (Figure 1). The south coast has

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Figure 1. Shaded relief map of the island of Molokai with all locations mentioned in the text. Upperleft inset shows Molokai’s location within the Hawaiian Archipelago. Lower inset shows the Pala’austudy region in an aerial photograph taken 24 April 2000: types of surface cover are indicated bysymbols, the Pakanaka fishpond is outlined (black dotted line) and the seaward extent of mangrovesin 2000 is shown (dotted white line).

the largest living fringing reef in the Hawaiian Islands, a nearly continuous, 2 km wide,carbonate structure that extends for roughly 50 km.

We concentrated on a 6 km2 area in the coastal Pala’au region where Rhizophoramangle saplings were originally planted in 1902 (Allen 1998) (Figure 1). In this region,rainfall averages 70 cm yr−1, falling mainly in the winter during brief, severe Kona stormsthat cause sheet wash and gullying. Climate changes, variations in land-use practices,and overgrazing of uplands by feral and domestic animals have led to vegetation loss,accelerated erosion, and extensive coastal sedimentation (Allen 1998). The coastal plainlandward of the mangrove fringe is an extensive mud flat composed predominantly offine-grained, lateritic mud and clay. This mud flat is hypersaline and rarely inundated bytides. It is occupied by dense Batis maritima (pickleweed) populations, sparse Pluchea

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Table 1Sensor Specifications for Aerial Camera (1:10000 scale); ASTER (AdvancedSpaceborne Thermal Emissions and Reflcetion Radiometer) and, AVIRIS(Advanced Visible/Infrared Imaging Spectrometer); VNIR = Visible—Near

Infrared; SWIR = Short Wave Infrared; TIR = Thermal InfraredAerial Camera∗ ASTER AVIRIS∗∗

Platform Airborne Spaceborne AirborneSwath Width Variable 60 km 10.5 kmSpatial Resolution Variable VNIR: 15 m 17 m

SWIR: 30 mTIR: 90 m

Spectral Range 400–700 nm 520–12000 nm 400–2450 nmNumber of Bunds 1 VNIR: 4 bands 224

SWIR: 5 bandsTIR: 5 bands

Band Width 300 nm 60–700 nm 10 nm

∗Non-digital, using film emulsion technology.∗∗High altitude AVIRIS flown at 20 km.

indica (Indian fleabane) patches, salty mud flats, and infrequent groves of Prosopis pallida(kiawe) (Figure 1).

Remote Sensing Data Sources

All images used in this study were acquired during 2000, with specifications for each datatype listed in Table 1.

Aerial photography. Color aerial photographs were collected for the U.S. GeologicalSurvey by Air Survey Hawaii in April 2000 at a scale of 1:10,000. We scanned the imagediapositives at 1200 dpi using a high-resolution photogrammetric scanner to preservehigh spatial resolution (0.25 m) and to maintain geometric relationships during digitalconversion. Digital stereo images (overlapping 60%) were orthorectified using surveyedground control points (±10 cm) and then mosaicked into a rectified orthophotomosaiccovering approximately 20 km2 of the south coast of Molokai. The root mean squarederror (RMSE) values for the rectification, coupled with the ground resolution and groundcontrol accuracy, provide an average error of ±0.5 m horizontally and ±1.0 m verticallyfor placement of every image pixel. The rectified orthophotomosaic served as the gridtemplate for construction of a reference point matrix for accuracy assessment. The digitalorthophotomosiac was resampled to 1 m and to 20 m (using nearest neighbor transformation)for comparing results from varying spatial resolutions.

ASTER multi-spectral imagery. ASTER images were collected by NASA’s EarthObservation Systems’ Terra satellite in November 2000. The data were acquired as ASTERLevel 1B Calibrated Radiance at the Sensor, and delivered as radiometrically calibrated andgeometrically corrected radiances (Wm−2 sr−1µm−2) that had been processed to removeeffects of rotation of the earth and the variable position of the spacecraft. The radiance valueswere converted to apparent surface reflectance using the atmospheric correction algorithm,ACORN v. 3.12 (Analytical Imaging and Geophysics 2001), which applies Modtran 4

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


radiative transfer modeling to quantify and remove influences of aerosol scattering andatmospheric gases. Because ASTER spatial resolution varies with the spectral region, theVNIR bands were resampled from 15 to 30 m before combining with the SWIR bands forprocessing.

AVIRIS hyperspectral imagery. AVIRIS hyperspectral imagery was collected byNASA in April 2000 from a modified ER-2 aircraft flying at approximately 20,000 m,yielding 17 × 17 m ground pixels along a ∼10.5 km wide swath flown from east towest along the south coast. The AVIRIS data were converted to apparent reflectance usingthe ATREM atmospheric correction (ATmospheric REMoval Program) (Gao et al. 1993).ATREM uses an atmospheric scattering model adapted from MODTRAN 5S radiativetransfer code to convert raw radiance values at the sensor to actual surface reflectance.

Pre-Processing

Since the inland limit of the Pala’au mangrove forest was not completely encompassed bythe ASTER imagery and clouds covered approximately 15% of the region, all datasets werespatially subset to the same 6 km2 region of Pala’au that was visible in the noncloudedparts of the ASTER image. Because ATREM correction applied to AVIRIS data tends toovercompensate in water absorption regions of the spectrum, wavebands in these regions(0.94, 1.13, 1.37, 1.9 µm) were masked out. Wavebands at CO2 absorption features (1.6and 2.0 µm) were also masked out. All clouds and marine and near-shore reef environmentswere masked from all datasets by applying band thresholding techniques to a MinimumNoise Fraction transformation (MNF). The MNF transformation is a two-step process thatdecorrelates noise from the data, rescales the data with unit variance in every band, and isfollowed by a standard principle components analysis (Lee et al. 1990; Hirano et al. 2003).Band thresholding of primary MNF bands can help identify similar materials, which canbe excluded or masked from subsequent analyses.

Classification Techniques

We processed all three datasets with ENVI version 3.5 (Research Systems Inc., Boulder,CO., USA) and PCI Geomatics’ APEX photogrammetric software (for orthorectificationof aerial photographs). The following image analysis approaches were then applied to eachimage: (I) visual interpretation; (II) ISODATA unsupervised classification; and (III) SpectralAngle Mapping supervised classification. Each technique was applied to the same Pala’auregion subset from color aerial photographs, combined VNIR-SWIR ASTER imagery,and the AVIRIS imagery. The accuracy of each mangrove classification was quantifiedindividually with a contingency-based error matrix calculated between classified pixels andreference data.

(I) Visual interpretation. The images were enhanced to heighten tonal contrasts andfacilitate mangrove canopy discrimination using an equalization stretch combined with a3 × 3 median filter. Mangrove habitats were delineated by manually digitizing the outeredge of the canopy based on visual contrasts and referenced field data. To assess observerbiases, the images were visually interpreted independently by two individuals: one familiarwith the site (I) and one who had not been to the site (I∗).

(II) Unsupervised classification. An Iterative Self-Organizing Data Analysis Tech-nique (ISODATA) unsupervised classification was performed on each dataset. This iterativeclustering routine is based on minimum spectral distances between each pixel and a

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


candidate cluster (ERDAS 1999). For all classifications, we specified 100 iterations, aconvergence threshold of 99%, 25 maximum classes, and a minimum of 10 pixels perclass. Mangrove classes were distinguished based on mean spectral reflectance curves andspectral features from field validation sites.

(III) Supervised classification. Using field data to guide the selection of homogeneousmangrove regions on each dataset, we applied ENVI’s Spectral Angle Mapper (SAM)supervised classification, which treats each spectrum as a multi-directional vector andcalculates the n-dimensional angle between training spectra and the image spectrum ofeach pixel (see Fig. 9 in Hirano et al. 2003). We delineated regions of interest (ROIs) inknown zones of mangroves and then used their mean spectra as input training classes todrive the SAM. Low SAM values reflect small spectral angles between training and imagespectra, indicating high spectral correlation to the training endmember.

Georectification

After processing, all data were resampled to 5 m pixel sizes using the nearest neighbormethod and geocorrected to the aerial orthophotomosaic with a second-order polynomialwarping transformation. Fifty points were selected from the base orthophotomosaic andtied to points on the warp images. The RMSE values for registration of the ASTER andAVIRIS images were 1.8 pixels (9 m) and 1.3 pixels (6.5 m), respectively.

Accuracy Assessment

Green and Mumby (2000) recommended gathering field data from at least 50 sites perhabitat class for a rigorous accuracy assessment of multispectral classifications, althoughdata from high-resolution aerial photos may be substituted if access is difficult. Becauseof limited access to mangrove canopies, we were only able to collect 42 field validationpoints within the mangrove forests and along the fringes. At each location, an area of 20 m2

was scored for dominant ground cover type: mangrove, nonmangrove, or mixed. We thenviewed the aerial photographs in stereo to select an additional 165 points (for a total of 207)in the mangrove class. The nonmangrove reference class consisted of 150 points collectedin the field, using a handheld GPS (


Figure 2. Distribution of mangroves detected by three analytical techniques (rows) for three kinds ofimagery (columns). Techniques are: I = visual interpretation with site familiarity; II = unsupervisedISODATA classification; III = supervised SAM classification. Nominal pixel sizes for each imagerare given in parentheses. Dashed lines on ASTER images enclose regions obscured by cloud.

This statistic reflects errors of commission which result when a pixel is committed to anincorrect class.

Tau coefficients indicate how many more pixels were correctly classified than expectedby chance (Ma and Redmond 1995) and are used to assist consideration of accuracy byproviding a readily interpretable coefficient with an approximately normal distribution thatcan be used in statistical comparisons. Tau values were arcsine square root transformedand analyzed in a two-way Model I ANOVA without replication to assess relativecontributions to overall variance from sensor and classification technique. To determinedifferences in significance between techniques, we analyzed the transformed Tau values ina one-way Model I ANOVA and examined multiple comparisons using Tukey’s post-hoctest.

Results

Analyses of the same mangrove area, derived from three kinds of imagery, are summarizedin Figure 2. Tau coefficients varied among sensors and analytic techniques, with a two-wayANOVA indicating that analytical processing technique contributed significantly to thevariance in Tau coefficients (P < 0.01) and accounted for 69% of the overall variation(Figure 3, Table 2a). Sensor type explained only 13% of the overall variation and did

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Figure 3. Mean tau coefficients for processing techniques: (I) visual interpretation with site famil-iarity; (I∗) visual interpretation without site familiarity; (II) unsupervised ISODATA classification;and (III) supervised SAM classification. Techniques (I, I∗) labeled a are significantly different fromtechnique (III) labeled b. Error bars are standard error for each technique.

not contribute significantly to the overall variance (Table 2a). Tau coefficients showedsignificant differences between visual interpretation (I and I∗) and supervised classificationtechniques (III), while no significant differences existed between unsupervised ISODATA(II) and any of the other analytical techniques (Figure 3, Table 2b). Across all sensors,the visual interpretation technique had the highest mean overall accuracy, followed byISODATA unsupervised classifications and SAM classifications (Table 3).

Table 2aSummary of a two-way Model I ANOVA conducted on the Taucoefficients (Figure 3) for the overall accuracies of mangrove

classification

Source of variation df SS MS Fs

Sensor 3 260.6 86.9 2.29 NSTechnique 3 1328.2 442.7 11.65∗∗

Error 9 342.1 38

Total 15 1930.9

NS = not significant, ∗∗ = P


Table 2bPairwise comparisons from Tukey’s post-hoc tests of one-way Model I ANOVA of Tau

coefficients for accuracies of mangrove mapping techniques

VisualInterpretation

(I)

VisualInterpretation

(I∗)UnsupervisedISODATA (II)

SupervisedSAM (III)

Visual Intepretation (I) 1.000Visual Intepretation (I∗) 0.862 1.000Unsupervised ISODATA (II) 0.128 0.400 1.000Supervised SAM (III) 0.002∗∗ 0.009∗∗ 0.140 1.000

∗∗ = P < .01.

(I- I∗) Visual Interpretation

The visual interpretation technique consistently had higher overall accuracies than the othertechniques for all three kinds of imagery (Table 3). The highest Tau values were for visualinterpretation of color aerial photography using both the 1 m and 20 m datasets; accuracieswere not significantly different when mangrove areas were delineated by individualsfamiliar (I) and unfamiliar (I∗) with the fieldsite (Figure 3, Table 2b). The overall accuracy ofvisual classifications ranged from 83.3% to 98.0%. Producer accuracies for all sensors werelower than user accuracies due to conservative digitizing of mangroves at ecotone borders.

(II) ISODATA

The unsupervised ISODATA classifications consistently had intermediate accuracies, lowerthan the visual ones but always more accurate than the supervised SAM classifications.ISODATA overall accuracies ranged from 79.6% to 89.6% (Table 3). The ISODATAalgorithm failed to include all morphological varieties of R. mangle within the ASTER andAVIRIS classifications (errors of omission) and consistently excluded regions containingdwarf mangroves along the landward fringe, leading to lower producer accuracies(67.5–85.5%). User accuracies (95.3–96.3%) were also slightly lower for ISODATAclassifications, probably because of misclassification of some upland pixels as mangroves(errors of commission) (Table 3).

(III) Spectral Angle Mapping

Because the training end member spectra used to generate the SAM classifications wereaverages of image-derived spectral signatures from known pure mangrove stands, thesupervised SAM classifications were generally more conservation than other techniques,and also had the lowest average overall accuracies (65.5–72.2%) (Table 3). Mean Tau valuesfor SAM classifications were significantly lower than for visual interpretation techniques(I and I∗) (Table 2b). Errors of omission reduced the producer accuracies for the mangroveclass across sensor types to 44.4–52.2% (Table 3). The final SAM classifications did notcontain the full morphological array or all geographic locations of mangroves within thezone and were heavily biased towards regions where the training class spectra were selected(i.e., dense mangrove cover). User accuracies of SAM classifications were lowest for aerialphotographs, which had the greatest number of upland pixels misclassified as mangroves.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015

Tabl

e3

Acc

urac

yas

sess

men

tsfo

rea

chim

ager

and

anal

ytic

tech

niqu

e.A

naly

tical

tech

niqu

es(I

–III

)ar

ede

fined

inTa

ble

2b.N

omin

alpi

xels

izes

are

give

nin

pare

nthe

ses.

Acc

urac

ym

eans

for

each

sens

orar

eal

sosh

own

with

corr

espo

ndin

g95

%co

nfide

nce

inte

rval

s(%

)

Imag

erTe

chni

que

Ove

rall

Acc

urac

y(%

)95

%C

.I.(

%)

Use

r’s

Acc

urac

y(%

)95

%C

.I.(

%)

Prod

ucer

’sA

ccur

acy

(%)

95%

C.I

.(%

)

AIR

PHO

TO

(1m

)I

98.6

96.7

–99.

599

.597

.0–>

99.9

98.1

95.0

–99.

4I∗

95.5

92.8

–97.

310

0.0

97.6

–100

.092

.387

.7–9

5.3

II81

.577

.1–8

5.2

96.1

91.5

–98.

471

.064

.5–7

6.8

III

70.3

65.4

–74.

895

.589

.6–9

8.3

51.2

44.4

–57.

9A

IRPH

OT

O(2

0m

)I

98.0

95.9

–99.

110

0.0

97.7

–100

.096

.693

.1–9

8.5

I∗93

.890

.8–9

5.9

100.

097

.6–1

00.0

89.4

84.4

–92.

9II

89.6

86.0

–92.

496

.292

.2–9

8.3

85.5

80.0

–89.

7II

I65

.560

.5–7

0.3

92.0

84.8

–96.

144

.437

.8–5

1.3

AST

ER

(30

m)

I84

.179

.9–8

7.6

86.5

80.9

–90.

785

.279

.5–8

9.5

I∗90

.586

.9–9

3.2

98.8

95.5

–>99

.984

.278

.4–8

8.7

II86

.782

.7–8

9.9

96.3

92.0

–98.

579

.673

.4–8

4.7

III

70.5

65.5

–75.

110

0.0

95.3

–100

.048

.041

.1–5

4.9

AV

TR

IS(2

0m

))I

89.8

86.2

–92.

698

.895

.6–>

99.9

83.3

77.5

–87.

8I∗

83.3

79.0

–86.

898

.794

.9–9

9.9

71.9

65.4

–77.

7II

79.6

75.1

–83.

595

.891

.0–9

8.3

67.5

60.8

–73.

6II

I72

.267

.3–7

6.7

99.1

94.4

–>99

.952

.245

.4–5

9.0

135

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


In practice, the user accuracies may be even lower than those reported because, throughchance, our reference data did not include many points in the uplands that were commonlymisclassified as mangroves by all three techniques.

Discussion

Optimization of Remote Sensing Tools

Previous comparisons of remote sensing tools for mangrove mapping emphasized variousgoals, including the importance of sensor selection for discriminating mangrove fromnonmangrove vegetation (Green et al. 1998), identifying distinct mangrove assemblages inmultispecies communities (Sulong et al. 2002; Wang et al. 2004), and detecting mangrovesalong a narrow fringe (Manson et al. 2001). In this study of a monospecific mangrove system,sensor choice had only minor impacts on the outcome of the accuracy assessments acrossall techniques. The result is consistent with the structure of the Rhizophora mangle zone onMolokai’s south coast: it is broad, homogeneous, and sufficiently different architecturally toallow spectral discrimination from adjacent vegetation, such as the low lying Batis maritimaand shrubby Pluchea indica. Therefore, accurate results may be achieved effectively usingcoarser spatial resolutions (15–30 m), and decisions about the sensor of choice should focuson other characteristics, such as cost, processing time, repeatability, and management goals.

Although the high spatial resolution of aerial photography is invaluable for discriminat-ing substrates along narrow ecotones (Manson et al. 2001), acquisitions can be expensive,particularly for large-scale coverage, and often require intensive pre- and post-processing(e.g., mosaicking, georectification) compared to other sensors. The total cost of an aerialphoto survey is determined by many factors, including the age, scale, and image qualityaffect both the initial price and the cost of processing. Historical aerial photographs canbe acquired at a low price (US$5–50 per image) but are prone to warping and requiresignificantly more processing effort compared to the more expensive ($5,000–8,000 perflight) orthorectified aerial products currently available. The relatively small footprint ofaerial photographs (usually


heavy metals), soil nutrient levels, water availability, rapid temperature flux, and abnormalconcentrations of atmospheric carbon dioxide, all of which can modify spectral reflectancesof plants (Woolley 1971; Filella and Penuelas 1994; Adams et al. 1999; Martini and Silver2002; Thelen et al. 2004).

Figure 4. Image-derived spectral signatures from AVIRIS, ASTER and air photos for R. manglegrowing in five different cross-shore environmental settings. Bottom right inset shows a schematicdiagram of a hypothetical N-S cross section through the Pala’au mangrove habitat depictingthese various environmental settings, as well as the shallow geologic structure, surface cover andmorphological zonation of R. mangle.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


When effects of environmental patchiness and/or resource availability are taken intoaccount, intraspecific spectral variation often becomes greater than interspecific differences(Price 1994; Cochrane 2000), which can further confound classifications. Indeed, wefound that mapping accuracies of both the ASTER and AVIRIS imagery were lowerin the supervised SAM and unsupervised ISODATA classifications due to higher errorsof omission because the algorithms failed to recognize all morphological varieties ofR. mangle, which result from physiological responses to environmental gradients across thecoastal interface (e.g., salinity, wave action, inundation, sedimentation). Dwarf mangroves,bordering the coastal interface (seaward) and the margin of the evaporate mudflat(landward), are shorter and exhibit a denser network of prop roots than the tall coastal andinland dense canopy trees (Figure 4). Protected within the rock walls of coastal fishponds,mangrove propagules and young seedlings are abundant, while larger, more mature treesbuffer the interface, growing within crevices in the fishponds’ seaward facing wall. Thearchitectural and physiological properties of dwarf mangroves on both the seaward andlandward fringes produced pixels with reduced reflectance in near infrared wavelengths andincreased reflectance at chlorophyll absorption wavelengths (0.45–0.52 µm and 0.63–0.69µm), due to lower leaf densities and chlorophyll concentrations compared to trees withinthe tall, dense, interior canopy (Figure 4).

Many of these issues can be addressed using very high spatial resolution multi- (e.g.,Quickbird, IKONOS, CASI) and hyperspectral sensors (e.g., HyMap, HyperSpecTIR).While Hirano et al. (2003) could only map a class of R. mangle in Florida to 40% accuracywith AVIRIS, Green et al. (1998) mapped four mangrove classes to 86% accuracy in themultispecies Turks and Caicos system, using higher spatial resolution CASI (8 bands, 1 mpixels) imagery. High spatial resolution coupled with high spectral resolution would likelyreduce omission errors observed from both ASTER and AVIRIS data due to the sensors’inability to isolate mangrove spectra from mixed pixels on the border with nonmangrovehabitats, which frequently occurred in sparsely settled dwarf mangrove zones along thesouth shore of Molokai.

When total costs of image acquisition and labor required for processing (for bothnew and archival data) are balanced against classification accuracy for each sensor-analysis combination, visual interpretation and unsupervised ISODATA classificationsfrom multispectral ASTER data produced the most favorable cost:benefit ratio for mappingmangroves on south Molokai. ASTER imagery is available in multiple formats and withvarying levels of post-processing at less than $US100 per scene. Level 1B data areradiometrically and geometrically calibrated radiance at the sensor for the 14 band spectralimage and require little to no post-processing aside from atmospheric and occasionallygeospatial corrections. Also, unlike the aerial photographs or AVIRIS data, the entireMolokai study area was captured in one ASTER scene, requiring analysis of a singleimage and eliminating the processing workload associated with multiple image mosaickingand illumination adjustments. ASTER imagery is easily accessed online, provides nearlycontinuous global coverage, and is suitable for change detection studies with its revisittime of 16 days. Although cloud cover can be a problem and georeferencing may needto be improved with ground control, ASTER has already been used to produce robustclassifications of global mangrove distributions (Haito et al. 2003). Crosstalk (when opticalsignals from one band leak into another band) has recently been identified as a potentialproblem in the SWIR region (Iwasaki and Tonooka 2005), but free crosstalk correctionsoftware can be downloaded from the ASTER Science Project website.

Our results should not be extrapolated directly to other ecosystems; we emphasizethat both the sensor and processing requirements will change based on the size of the

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


region of interest and the spatial and temporal scales of questions being asked (Mumbyet al. 1999; Phinn et al. 2000). For instance, in multispecies mangrove systems, theadvantages of high spatial resolution, hyperspectral imagery for accurately discriminatingspecies, and species assemblages may necessitate investing in higher cost imagery. Inaddition, while there were no significant differences between visual classifications ofMolokai mangroves by familiar and unfamiliar users, site familiarity is likely to bemore important in more complex habitats. Wherever possible, data from different sourcesshould be integrated to improve mapping resolution and accuracy (Held et al. 2003). Forexample, most multi- and hyperspectral data cannot be used to create stereo views in whichmangrove vegetation can be classified based on height (Hirano et al. 2003), but this canbe addressed by integrating the imagery with a digital elevation model (Phinn et al. 2000).Combinations of optical (e.g., photography, Landsat TM, CASI) and microwave (e.g.,RADARSAT-1, synthetic aperature radar (SAR)) data can also yield higher classificationaccuracies than optical data alone and increase the ability to discriminate groups withinmangrove assemblages (Ramsey and Jensen 1996; Souza, Filho and Paradella 2002;Held et al. 2003).

Implications for Mangroves in Hawaii

Management plans for many invasive species recommend complete removal of thenonnative species, but total eradication of Hawaiian mangroves would have complexconsequences.

1. Mangrove removal might assist population recovery of the four endangered,endemic waterbirds; none of these species use mangrove habitat for foraging ornesting, and the trees provide shelter for some of their mostly nonnative predators(Adams et al. 1999).

2. Removing mangroves from areas of restricted flow, such as fishponds and anchialinepools, might improve local water quality by increasing water flow and reducing leaflitter accumulation and breakdown, which is decreasing available dissolved oxygenlevels (Cox and Jokiel 1996).

3. Conversely, the sediment trapping ability of the mangroves may be improving waterquality on adjacent coral reefs. On Molokai, Bigelow et al. (1989) noted reducedturbidity near high mangrove basal areas. This is consistent with data from Australiaand China where mangrove forests are net sinks of terrestrially derived sedimentsand organic matter (Alongi and McKinnon 2005; Alongi et al. 2005).

4. Although the presence of mangroves may have increased populations of oppor-tunistic exotics (e.g., the Samoan crab, Scylla serrata), it may have simultaneouslyincreased species richness of native of soft-sediment dwellers (Demopoulos andSmith 2001).

5. In other parts of the world, mangroves also serve as important fish nursery habitats(Nagelkerken et al. 2000; Mumby et al. 2004; Islam and Wahab 2005), thoughspecific links between mangrove area and fishery yields have yet to be determinedfor Hawaii.

6. Cost of complete mangrove removal would be extremely high (>$100,000/hawith machinery, >$350,000/ha manually; Allen 1998), and there are potentiallyimportant ecological and economic motivations for preserving certain regions.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


One management practice, currently used in Hawaii, is targeted removal of mangrovesfrom selected archaeological and important bird habitat sites. For example, mangroveswere successfully controlled in the early 1990s from Kaloko and Aimakapa fishponds inthe Kaloko-Honokohau National Historic Park on the main island (Pratt 1998). Both sitesare culturally significant, and Aimakapa is also one of the two most important breedingsites for the Hawaiian coot and Hawaiian stilt (Pratt 1998). Mangrove control was alsoinitiated in the mid-1990s in the Nuupia Ponds Wildlife Management Area (NPWMA)within the Kaneohe Bay Marine Corps Base on Oahu, but due to the high reproductive rateand probable lack of propagule predators, ongoing maintenance control is required (Coxand Allen 1999).

Because mangrove control is costly and not necessarily immediately effective, remotesensing tools have the potential to enable managers to prioritize sites for restoration and tomonitor responses to management efforts. The rate of unmanaged mangrove encroachmentinto fishponds, bird habitats, or other sites of cultural significance can be quantified at abroad resolution from satellite imagery and at a finer resolution, over a longer temporalrange, from aerial photography. The rates of encroachment and proximity to other mangrovepatches, which may reseed cleared regions, may then be considered with site value and costswhen assessing the urgency for restoration. Combining this information with thematic layersderived from remote-sensing analyses in a geographic information system can increase thepower of decision making. Within a GIS, spatial statistics can then be performed to derive“hot spots” for targeted mangrove removal or those areas which should be considered formangrove protection to enhance the ecosystem functions of sediment trapping and fisherieshabitat.

Conclusion

The 1902 introduction of the red mangrove to Molokai presents a unique researchopportunity to explore the applications of remote-sensing technology for monitoringinvasive species in tropical coastal ecosystems. Now established on at least six of the maineight islands, the increasing distribution of Rhizophora mangle has encouraged review ofthe state’s invasive species laws and associated resource management policies, but currentlythe fate of R. mangle in Hawaii is unknown.

Remote sensing provides various options for continuous monitoring of mangrovehabitat over time. Determining the most cost-effective sensor-technique combination foraccurate classification of mangrove habitat is essential for building a reliable speciesdistribution baseline that can be used for designing invasive species policy and for directingcoastal resource management strategies. In the present case, unsupervised classificationsfrom ASTER multispectral imagery were the most cost-effective options for efficiently andaccurately mapping the monospecific mangrove distribution on South Molokai. However,the performance of classification techniques and from different data sources will varydepending on the complexity of the mangrove systems.

In Hawaii, monitoring the coastal change associated with mangroves is essential tounderstanding how the natural coastal ecosystem reacts to invasive species introductions andadapts overall to changing climatic regimes. Accurate and timely information regardingits changing distribution is critical to the management of mangrove forests throughoutthe Hawaiian Islands. The findings of this remote-sensing research provide a strongfoundation for future mangrove monitoring, encroachment management, and integratedcoastal resource planning in Hawaii.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Acknowledgements

This study was conducted as part of the Coral Reef Project of the USGS Coastal andMarine Geology (CMG) Program, and we thank all members of this research team for theirsupport and encouragement. We also thank the NASA JPL for funding and acquiring theAVIRIS 2000 Hawaii High Altitude data. We thank Dr. Michael E. Field (USGS—CMG)for initiating and guiding this study, Dr. Gary Griggs and Dr. Eli Silver for valuableediting insights and suggestions, and Jim Maragos, Bruce Richmond, Cheryl Hapke, AnnGibbs, Josh Logan, and Curt Storlazzi of the USGS Pacific Science Center for invaluableassistance with image processing and GIS applications. Last, we thank other members ofthe Hyperspectral Imaging Project (HIP) and Coastal Imaging Lab at UCSC for ongoingacademic advice and scientific collaboration.

References

Adams, M. L., W. D. Philpot, and W. A. Norvell. 1999. Yellowness index: an application of spectralsecond derivatives to estimate chlorosis of leaves in stressed vegetation. International Journalof Remote Sensing 20:3663–3675.

Agresti, A., and B. A. Coull. 1998. Approximate is better than “exact” for interval estimation ofbinomial proportions. The American Statistician 52:119–126.

Analytical Imaging and Geophysics, LLC. 2001. ACORN user’s guide, stand-alone version. ACORNversion 3.12, January 2001. Analytical Imaging and Geophysics, LLC, Boulder, Colo.

Allen, J. A. 1998. Mangroves as alien species: the case of Hawaii. Global Ecology and BiogeographyLetters 7:61–71.

Alongi, D. M., and A. Sasekumar. 1992. Benthic communities. In Tropical Mangrove Ecosystems,A. I. Robertson and D. M. Alongi (eds.). Washington, DC: American Geophysical Union,pp. 137–171.

Alongi, D. M. 2002. Present state and future of the world’s mangrove forests. EnvironmentalConservation 29:331–349.

Alongi, D. M., and A. D. McKinnon. 2005. The cycling and fate of terrestrially-derived sedimentsand nutrients in the coastal zone of the Great Barrier Reef shelf. Marine Pollution Bulletin51:239–252.

Alongi, D. M., J. Pfitzner, L. A. Trott, F. Tirendi, P. Dixon, and D. W. Klumpp. 2005. Rapid sedimentaccumulation and microbial mineralization in forests of the mangrove Kandelia candel in theJiulongjiang Estuary, China. Estuarine, Coastal and Shelf Science 63:605–618.

Battiau-Queney, Y., J. F. Billet, S. Chaverot, and P. Lanoy-Ratel. 2003. Recent shoreline mobility andgeomorphologic evolution of macrotidal sandy beaches in the north of France. Marine Geology194:31–45.

Bigelow, K. A., K. Alspach, R. Lohle, T. McDonough, P. Ravetto, C. Rosenfeld, G. Stender, and C.Wong, 1989. Assessment of mangrove ecosystem of west Moloka’i, Hawaii, with additional sitesurveys of Moauni Beach Park and Ualapu’e Fishpond. Marine Option Program, University ofHawaii, Honolulu, HI.

Bina, R. T., B. R. de Jesus, Jr., and E. Lorenzo, 1978. Mangrove inventory of the Philippines usingLandsat multi-spectral data and image 100 system. In Proceedings of the Twelfth InternationalSymposium on Remote Sensing of Environment, pp. 2343–2359. Environmental ResearchInstitute of Michigan.

Blasco, F., and M. Aizpuru. 2002. Mangroves along the coastal stretch of the Bay of Bengal: Presentstatus. Indian Journal of Marine Sciences 31:9–20.

Blasco, F., P. Saenger, and E. Janodet. 1996. Mangroves as indicators of coastal change. Catena27:167–178.

Chimner, R. A., B. Fry, M. Y. Kaneshiro, and N. Cormier. 2006. Current extent and historicalexpansion of introduced mangroves on O’ahu, Hawai’i. Pacific Science 60(3):377–383.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Cochrane, M. A. 2000. Using vegetation reflectance variability for species level classification ofhyperspectral data. International Journal of Remote Sensing 21:2075–2087.

Congalton, R. 1991. A review of assessing the accuracy of remotely sensed data. Remote Sensing ofEnvironment 37:35–46.

Cox, E. F., and J. A. Allen. 1999. Stand structure and productivity of the introduced Rhizophoramangle in Hawaii. Estuaries 22:276–284.

Cox, E. F., and P. L. Jokiel. 1996. An environmental study of Nuupia Ponds Wildlife ManagementArea, Marine Corps Base Hawaii, Kaneohe Bay. Final report.

Dahdouh-Guebas, F., I. Van Pottelbergh, J. G. Kairo, S. Cannicci, and N. Koedam. 2004. Human-impacted mangroves in Gazi (Kenya): predicting future vegetation based on retrospective remotesensing, social surveys, and tree distribution. Marine Ecology Progress Series 272:77–92.

Dahdouh-Guebas, F., A. Verheyden, W. De Genst, S. Hettiarachchi, and N. Koedam. 2000. Fourdecadal vegetation dynamics in Sri Lankan mangroves as detected from sequential aerialphotography: a case study in Galle. Bulletin of Marine Science 67:741–759.

Danielsen, F., M. K. Sorensen, M. F. Olwig, V. Selvam, F. Parish, N. D. Burgess, T. Hiraishi, V. M.Karunagaran, M. S. Rasmussen, L. B. Hansen, A. Quarto, and N. Suryadiputra. 2005. The Asiantsunami: a protective role for coastal vegetation. Science 310:643.

D’Iorio, M. M. 2003. Mangroves and shoreline change on Molokai, Hawaii: assessing the role ofintroduced Rhizophora mangle in sediment dynamics and coastal change using remote sensingand GIS. Ph.D. Dissertation. University of California, Santa Cruz, CA.

Duke, N. C., M. C. Ball, and J. C. Ellison. 1998. Factors influencing biodiversity and distributionalgradients in mangroves. Global Ecology and Biogeography Letters 7:27–47.

Dutrieux, E., J. Denis, and J. Populus. 1990. Application of SPOT data to a base-line ecological studyof the Mahakam delta mangroves (East Kalimantan, Indonesia). Oceanologica Acta 13:317–326.

Egler, F. E. 1947. Arid southeast Oahu vegetation, Hawaii. Ecological Monographs 17:383–435.Ellison, A. M., and E. J. Farnsworth. 1997. Simulated sea level change alters anatomy, physiology,

growth, and reproduction of red mangrove (Rhizophora mangle L.). Oecologia 112:435–446.Elton, C. S. 1958. The ecology of invasions by animals and plants. Methuen, London.Elvidge, D. E. 1990. Visible and near infrared reflectance characteristics of dry plant materials.

Remote Sensing of Environment 11:1775–1795.Eong, O. J., G. W. Khoon, W. Y. Ping, and W. H. Kheung, 1992. Identification of mangrove vegetation

zones using MicroBRIAN and Landsat imagery. In Proceedings of the Third ASEAN Scienceand Technology Week Conference, pp. 383–389. University of Singapore.

ERDAS. 1999. ERDAS Field Guide, Fifth Edition. ERDAS, Atlanta, GA.Filella, I., and J. Penuelas. 1994. The red edge position and shape as indicators of plant chlorophyll

content, biomass, and hydric status. International Journal of Remote Sensing 15:1459–1470.

Gao, B. C., K. B. Heidebrecht, and A. F. H. Goetz. 1993. Derivation of scaled surface reflectancesfrom AVIRIS data. Remote Sensing of Environment 44:145–163.

Gao, J. 1998. A hybrid method toward accurate mapping of mangroves in a marginal habitat fromSPOT multispectral data. International Journal of Remote Sensing 19:1887–1899.

Green, E., C. Clark, P. Mumby, A. Edwards, and A. Ellis. 1998. Remote sensing techniques formapping mangroves. International Journal of Remote Sensing 19:935–956.

Green, E. and P. Mumby, 2000. Mapping mangroves. Remote Sensing Handbook for Tropical CoastalManagement. A. J. Edwards. Paris, UNESCO: 183–198.

Haito, H., M. F. Bellan, A. Al-Habshi, M. Aizpuru, and F. Blasco. 2003. Mangrove research andcoastal ecosystem studies with SPOT-4 HRVIR and TERRA ASTER in the Arabian Gulf.International Journal of Remote Sensing 24:4073–4092.

Held, A., C. Ticehurst, L. Lymburner, and N. Williams. 2003. High resolution mapping of tropicalmangrove ecosystems using hyperspectral and radar remote sensing. International Journal ofRemote Sensing 24:2739–2759.

Higinbotham, C. B., M. Alber, and A. G. Chalmers. 2004. Analysis of tidal marsh vegetation patternsin two Georgia estuaries using aerial photography and GIS. Estuaries 27:670–683.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Hirano, A., M. Madden, and R. Welch. 2003. Hyperspectral image data for mapping wetlandvegetation. Wetlands 23:426–448.

Imbert, D., P. Labbe, and A. Rousteau. 1996. Hurricane damage and forest structure in Guadaloupe,French West Indies. Journal of Tropical Ecology 12(5):663–680.

Islam, M. S., and M. A. Wahab. 2005. A review on the present status and management of mangrovewetland habitat resources in Bangladesh with emphasis on mangrove fisheries and aquaculture.Hydrobiologia 542:165–190.

Iwasaki, A., and H. Tonooka. 2005. Validation of a crosstalk correction algorithm for ASTER/SWIR.IEEE Transactions on Geoscience and Remote Sensing 43:2747–2751.

Jensen, J. R., E. Ramsey, B. A. Davis, and C. W. Thoemke. 1991. The measurement ofmangrove characteristics in southwest Florida using SPOT multispectral data. GeocartographyInternational 2:13–21.

Jupiter, S. D., D. C. Potts, S. R. Phinn, and N. C. Duke, 2006. Natural and anthropogenic changes tomangrove distributions in the Pioneer River Estuary (QLD, Australia). Wetlands Ecology andManagement: DOI 10.1007/s11273-006-9011-9.

Kokaly, R. F., R. N. Clark, and K. E. Livo. 1998. Mapping the biology and mineralogy ofYellowstone National Park using imaging spectroscopy. In Summaries of the Seventh Annual JPLEarth Science Workshop, R. O. Green (ed.). Pasadena, CA: NASA Jet Propulsion Laboratory,pp. 245–254.

Lee, J. B., A. S. Woodyatt, and M. Berman. 1990. Enhancement of high spectral resolutionremote sensing data by a noisy-adjusted principal components transform. IEEE Transactions onGeoscience and Remote Sensing 28:295–304.

Lorenzo, E. N., B. R. de Jesus, Jr., and R. B. Jara, 1979. Assessment of mangrove forest deteriorationin Zamboanga Peninsula, Philippines using Landsat MSS data. In Proceedings of the ThirteenthInternational Symposium on Remote Sensing of Environment, pp.1737–1745. EnvironmentalResearch Institute of Michigan.

Lucas, R. M., J. C. Ellison, A. Mitchell, B. Donelly, M. Finlayson, and A. K. Milne. 2002. Useof stereo aerial photography for quantifying changes in the extent and height of mangroves intropical Australia. Wetlands Ecology and Management 10:161–175.

Lugo, A. E., and S. C. Snedaker. 1974. The ecology of mangroves. Annual Review of Ecology andSystematics 5:39–64.

Ma, Z., and R. L. Redmond. 1995. Tau coefficients for accuracy assessment of classification of remotesensing data. Photogrammetric Engineering and Remote Sensing 61:435–439.

MacCaughy, V. 1917. The mangrove in the Hawaiian Islands. Hawaiian Forester and Agriculturalist14:361–365.

Manson, F. J., N. R. Lonergan, J. M. McLeod, and R. A. Kenyon. 2001. Assessing techniques forestimating the extent of mangroves: topographic maps, aerial photographs and Landsat TMimages. Marine and Freshwater Research 52:787–792.

Martini, B. A., and E. A. Silver. 2002. The evolution and present state of tree-kills on MammothMt., CA: tracking volcanogenic CO2 and its lethal effects. Summaries of the Eleventh AnnualJPL Airborne Earth Science Workshop. R. O. Green (ed.). NASA Jet Propulsion Laboratory,Pasadena, CA.

Moberley, R. Jr. 1968. Loss of Hawaiian littoral sand. Journal of Sedimentary Petrology 38:17–34.Mumby, P. J., A. J. Edwards, J. E. Arias-Gonzalez, K. C. Lindeman, P. G. Blackwell, A. Gall, M. I.

Gorczynska, A. R. Harborne, C. L. Pescod, H. Renken, C. C. C. Wabnitz, and G. Llewellyn.2004. Mangroves enhance the biomass of coral reef fish communities in the Caribbean. Nature427:533–536.

Mumby, P. J., E. P. Green, A. J. Edwards, and C. D. Clark. 1999. The cost-effectiveness of remotesensing for tropical coastal resources assessment and management. Journal of EnvironmentalManagement 55:157–166.

Nagelkerken I., G. Van Der Velde, M. W. Gorissen, G. T. Meijer, T. van’t Hof, and C. denHartog. 2000.Importance of mangroves, seagrass beds and the shallow coral reefs as a nursery for importantcoral reef fishes, using a visual census technique.Estuarine, Coastal and Shelf Science, 51:31–44.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015


Phinn, S. R., C. Menges, G. J. E. Hill, and M. Stanford. 2000. Optimizing remotely sensed solutionsfor monitoring, modeling, and managing coastal environments. Remote Sensing of Environment73:117–132.

Pratt, L. W. 1998. Vegetation management strategies for three national historical parks on Hawai’iIsland. The Pacific Cooperative Studies Unit and the Hawaii-Pacific Islands Ecosystems StudiesUnit Network Technical Report No. 121 University of Hawaii, Honolulu.

Price, J. C. 1994. How unique are spectral signatures? Remote Sensing of Environment. 49:181–186.Ramirez-Garcia, P., J. Lopez-Blanco, and D. Ocana. 1998. Mangrove vegetation assessment in the

Santiago River Mouth, Mexico, by means of supervised classification using Landsat TM imagery.Forest Ecology and Management 205:217–229.

Ramsey, E. W. III, and J. R. Jensen. 1996. Remote sensing of mangrove wetlands: relating canopyspectra to site-specific data. Photogrammetric Engineering & Remote Sensing 62:939–948.

Rasolofoharinoro, M., F. Blasco, M. F. Bellan, M. Aizpuru, T. Gauquelin, and J. Denis. 1998. Aremote sensing based methodology for mangrove studies in Madagascar. International Journalof Remote Sensing 19:1873–1886.

Ricklefs, R. E. and R. E. Latham. 1993. Global patterns of diversity in mangrove floras. In Speciesdiversity in ecological communities: historical and geographical perspectives, R. E. Ricklefsand D. Schluter (eds.). Chicago, IL: University of Chicago Press, pp. 215–229.

Saintilan, N., and K. Wilton. 2001. Changes in the distribution of mangroves and saltmarshes in JervisBay, Australia. Wetlands Ecology and Management 9:409–420.

Sherman, R. E., T. J. Fahey, and P. Martinez. 2001. Hurricane impacts on a mangrove forest in theDominican Republic: damage patterns and early recovery. Biotropica 33(3):393–408.

Simberloff, D. 1995. Why do introduced species appear to devastate islands more than mainlandareas?. Pacific Science 49:87–97.

Smith, C. R., M. C. Austen, G. Boucher, C. Heip, P. A. Hutchings, G. M. King, I. Koike, P. J.D. Lambshead, and P. Snelgrove. 2000. Global change and biodiversity linkages across thesediment-water interface. Bioscience 50(12):1108–1120.

Smith, G. M., T. Spencer, A. L. Murray, and J. R. French. 1998. Assessing seasonal vegetation changein coastal wetlands with airborne remote sensing: an outline methodology. Mangroves and SaltMarshes 2:15–28.

Souza Filho, P. W. M., and W. S. Paradella. 2002. Recognition of the main geobotanical features alongthe Braganca mangrove coast (Brazilian Amazon Region) from Landsat TM and RADARSAT-1data. Wetlands Ecology and Management 10:123–132.

Spalding, M., F. Blasco, and C. Field. 1997. World Mangrove Atlas. The International Society forMangrove Ecosystems, Okinawa, Japan.

Sulong, I., H. Mohd-Lokman, K. Mohd-Tarmizi, and A. Ismail. 2002. Mangrove mapping usingLandsat imagery and aerial photographs: Kemaman district, Terengganu, Malaysia. Environment,Development and Sustainability 4:135–152.

Thelen, K. D., A. N. Kravchenko, and C. D. Lee. 2004. Use of optical remote sensing for detectingherbicide injury in soybean. Weed Technology 18:292–297.

Wang, L., W. P. Sousa, P. Gong, and G. S. Biging. 2004. Comparison of IKONOS and Quickbirdimages for mapping mangrove species on the Caribbean coast of Panama. Remote Sensing ofEnvironment 91:432–440.

Wester, L. 1981. Introduction and spread of mangroves in the Hawaiian Islands. Association of PacificCoast Geographers Yearbook 43:125–137.

Woolley, J. T. 1971. Reflectance and transmittance of light by leaves. Plant Physiology 47:656–662.

Dow

nloa

ded

by [

USG

S L

ibra

ries

] at

08:

16 2

0 O

ctob

er 2

015

View publication statsView publication stats
https://www.researchgate.net/publication/233157429

optimizing remote sensing and gis tools for mapping and ... · optimizing remote sensing and gis...

Documents