2 Management Applicationsof Optical Remote Sensingin the Active River Channel

W. Andrew Marcus1, Mark A. Fonstad1 and Carl J. Legleiter2

1Department of Geography, University of Oregon, Eugene, OR, USA2Department of Geography, University of Wyoming, Laramie, WY, USA

2.1 Introduction

As a potential user of remote sensing for river monitoringand analysis, you may find yourself wondering ‘what canbe measured, mapped, and/or modeled with remote sens-ing?’; ‘will clients accept results based on remote sensing?’;and ‘why use remote sensing rather than classical fieldmethods?’ These are all critical questions, and answer-ing them correctly determines whether a remote sensingapproach will substantially benefit a project or detractfrom the project’s success.

This chapter addresses these questions in the context ofpassive optical imagery of the active river channel. ‘Passive’refers to the measurement of light occurring naturallyin the environment – reflected solar energy. This is incontrast to ‘active sensors’ such as radar and LiDAR,which emit a pulse of energy and record the return ofthat energy. Active sensors are explored in later chapters.‘Optical’ refers to the dominant wavelengths of light orig-inating from the sun: blue, green, red and near-infraredwavelengths. The blue, green and red wavelengths are‘visible light’ that we can see with our eyes. The near-infrared is invisible to the human eye, but is still amajor component of sunlight. Passive optical imageryis something we all have seen since childhood – mostphotos are passive optical imagery, as are panchro-matic, colour, and near-infrared air photos. This chapter

Fluvial Remote Sensing for Science and Management, First Edition. Edited by Patrice E. Carbonneau and Herve Piegay.© 2012 John Wiley & Sons, Ltd. Published 2012 by John Wiley & Sons, Ltd.

discusses imagery where the data are broken into narrow,discrete wavelengths of light, as with hyperspectral sen-sors. It also presents information on broad-spectrumimagery where many wavelengths are bundled together,as with panchromatic photography that encompasses allvisible wavelengths to form a single image representingoverall brightness.

The chapter also focuses on the ‘active channel’, whichis the low flow channel plus adjacent areas that are freeof vegetation and subject to scour or deposition undertypical hydrologic conditions. The active channel includesthe submerged channel, unvegetated mid-channel islands,chutes, and exposed bars. The active channel has beenthe focus of a great deal of remote sensing researchsince the mid-1990s, a period that coincides with theincreased availability of centimeter- to meter-resolutiondigital imagery. Prior to that time, research on remotesensing of rivers was generally limited to film-based aerialphotos or digital multispectral satellite imagery. Studiesof the active channel with digital multispectral data werelimited because the pixel size of satellite imagery was toolarge. A 30 or 80 m Landsat pixel from the 1970s or 1980s,for example, would cover the entire active channel pluslarge portions of the bank and floodplain in all but thelargest lowland rivers.

The vegetated floodplain is not included in our def-inition of the active channel and is not discussed in

19

20 Fluvial Remote Sensing for Science and Management

this chapter. The optical techniques for mapping ripar-ian vegetation on floodplains are, for the most part, thesame techniques used to map vegetation, regardless ofsetting. These methods are extensively discussed in manytext books, articles, and on-line sources and are alreadywidely used for resource management (for overviews seeDieck and Robinson, 2004; Jensen, 2007). In contrast,techniques specifically focused on the active channel aresufficiently new that they are not yet widely known to themanagement community.

Evaluating whether to use remote sensing requiresknowing which river features can be measured, mapped,and/or modeled in this manner. The first part of thischapter therefore reviews active channel features thatcan be mapped with passive optical imagery, the generalnature of the remote mapping techniques, and some ofthe limitations specific to each application. A subsequentsection summarises issues that are common to manyremote sensing applications (e.g., accuracy assessment).The chapter concludes with a discussion of factors toconsider when determining whether or not to use remotesensing rather than, or in addition to, other availabletechniques.

Finally, the focus throughout the chapter is on param-eters that can be monitored and mapped using remotesensing. These mapping applications may be an end-point in their own right, but are often just the startingpoint for management applications related to model-ing, planning, and active intervention in the stream. Webriefly mention some management applications such ashabitat assessment and flood planning, but for the mostpart we encourage the reader to think about, and imag-ine, the many uses to which the remote sensing-basedmeasurement and maps can be put.

2.2 What can be mapped with opticalimagery?

What can you measure and map with optical images? Atone level, the answer to this question is simple: anythingyou can see with the naked eye is potentially ‘mappable’with optical imagery. But the answer does not stop there;hyperspectral and multispectral imagery can detect fea-tures at resolutions and wavelengths not visible to thehuman eye. When looking at a clear-water stream, onetherefore can intuitively determine which features mightbe mappable with remote sensing (muddy streams wherethe bottom cannot be seen are generally poor candidatesfor remote sensing of features within the water column

Figure 2.1 Variations in depth (and other river features) can beeasily detected with the naked eye. In this example along theTrinity River, California, variations in depth are immediatelyapparent as variations in water darkness. Likewise, glides, rifflesand pools (i.e. biotypes) can be distinguished by variations indepth and surface turbulence, and variations in sediment sizeare apparent in the shallow water portions of the stream. Riverfeatures that are visible to the naked eye are features that, intheory, can be measured and mapped with opticalremote sensing.

or on the bed). In Figure 2.1, for example, one can gaugewhich areas are deeper by the darkness of the water.Likewise, the human eye can easily pick out variationsin sediment size, pieces of wood, algae on rock surfaces,surface turbulence, and so forth. All of these are there-fore good candidates for mapping via remote sensing.In addition, multi- or hyperspectral data might be ableto pick out more subtle variations in turbidity, depth,and turbulence.

Alas, life is rarely so simple. Our eyes and brains pro-cess a remarkable amount of information on the fly:brightness, colour, texture, shape, shadow, size, spatialcontext, rate of movement, and so on. In contrast, remotesensing algorithms typically use just one, or maybe twoof these factors at a given time. Often, therefore, a remotesensing-based, image processing approach misses the sub-tle identifying characteristics that are readily detected bythe human brain. What can be ‘seen’ with remote sensingtherefore often differs in subtle ways from what the eye-brain combination can detect. Yet sometimes the differentperspectives provided by remote sensing (especially withmultispectral and hyperspectral imagery) coupled withtrained users can detect more than the brain-eye combi-nation can see (e.g., Legleiter and Goodchild, 2005).

Much of the previous research on remote sensing ofrivers has focused on using imagery to map the same

2 Management Applications of Optical Remote Sensing in the Active River Channel 21

features that humans observe and map in the field. Thebulk of the following section summarises this kind ofwork. However, some of the more interesting advancesin remote sensing of rivers will probably result fromabandoning these anthropic constraints and letting theimages ‘speak for themselves;’ we briefly discuss someof these prospects in the conclusion of this chapter.Likewise, many of the most exciting advances in remotesensing of rivers are coming from devices that do notuse the visible and near-infrared wavelengths with whichpeople are most familiar. Radar, LiDAR, and thermalimagery and combinations of these with optical imageryare opening up a wide range of new remote sensingapplications for river management; the reader shouldbe sure to review those chapters to understand the fullspectrum of potential applications.

The following summary of optical image applicationshighlights recent studies that have achieved the highestaccuracies. Reviews by Mertes (2002) and Gilvear andBryant (2003) provide more information on the his-tory of different remote sensing sensors and applicationsin rivers, while Marcus and Fonstad (2008) and Feureret al. (2008) focus on reviews of optical imagery only.Marcus (2012) provides more detail on recent advancesin using passive and active instruments to measure thehydraulic environment of rivers. In the following discus-sion of potential applications we provide an overview ofhow various techniques work, identify major limitationsspecific to those applications, and briefly summarise theo-retical and practical limitations generic to all applications(e.g. image quality, resolution, shadow, logistics).

2.3 Flood extent and discharge

Documenting areas of flood inundation is critical foremergency response and floodplain planning. Mappingflood extents also is useful for identifying locations whereefforts to restore or maintain stream habitat and bio-diversity could be targeted. Finally, even when riversare not flooding, repeat mapping of inundated areaprovides insight into discharge variability over time aswell as an early warning of flood or drought hazardsin remote regions. The importance of remote sensingfor tracking flow variability and floods is particularlyimportant in areas without hydrologic monitoring sys-tems or where access to such data are limited (Brakenridgeet al., 2005).

Aerial photos have long been used to document floodextent, especially during the peak of major floods when

access to the river is limited and potentially dangerous.Some of the earliest applications of satellite imagerywere for measuring flood extents (Smith, 1997), butthe 17-day repeat cycle of the Landsat satellite did notallow monitoring of change during one flood event.Subsequently, the advent of satellites that covered thesame location on a daily basis enabled monitoring offloods with very coarse (1 km) spatial resolution data inlarge rivers (Barton and Bathos, 1989). More recently, 1to 5-m resolution satellite imagery provides a mappingtool in smaller streams. In contrast to satellites, however,aircraft can fly beneath cloud cover and capture floodevents that might be missed by satellite imagery.

Because water is generally visible to the human eyeon imagery, it is possible to trace or manually digitisethe wetted channel extent. Automated approaches formapping flow extent and discharge using satellite imageryare reviewed by Smith (1997). Mapping the extent of wateris typically done with the shortwave infrared bands. Wateris a strong absorber in these wavelengths and thereforevery dark on images compared to other features. It isour experience that automated approaches with meteror cm resolution imagery can confuse water with darkshadow and require manual editing to ensure reasonableresults. At coarser resolutions, however, the shadows aremixed with other features in the pixel (e.g., trees) andthe spectral signal is distinct from water. The task ofseparating water from other features is also simplifiedif one has change-over-time imagery. In this case, thedifference between dry and inundated conditions for agiven pixel is sufficiently large to enable change detectionalgorithms to accurately map surface inundation usinga variety of sensors, including the Landsat ThematicMapper (Kishi et al., 2001), AVHRR (Sheng et al., 2001),and the EO-1 hyperspectral sensor (Ip et al., 2006). Smith(1997, p. 1429) notes that there has been little changesince the 1970s in how inundated areas are mapped andthat the approach is ‘now considered operational.’

Discharge can be estimated directly from an inundatedarea if there is a nearby ground-based gauging stationand a correlation can be established between dischargeand flood extent. The highest accuracy discharge mea-surements using this approach are achieved if ‘gaugingreaches’ are identified where the wetted width varies dra-matically with changes in discharge (Brakenridge et al.,2005), a response that characterises many braided riversbut is more difficult to find in single thread channels. Toavoid confusion, it is worth noting that large changes inwidth in response to discharge variations is the oppositeof what is considered ideal for ground-based gauging

22 Fluvial Remote Sensing for Science and Management

stations, where it is preferable to have minimal variationin width in response to fluctuations in discharge. Space-based estimates of this sort are particularly valuable inareas where gauging stations are being discontinued. Cor-relations established in this manner can be extended toother rivers of similar morphology. Alternatively, floodvolume and depth can be derived by coupling floodinundation maps with digital elevation models, whichcan be coupled with hydrologic models to estimate dis-charge (Smith, 1997). In a related application, remotesensing-based measurements of flood extent can be usedto calibrate and validate numeric models of flooding(Bates, 2004).

Aerial flights and satellite imagery will doubtlesslyremain useful remote sensing tools for flow monitor-ing. Cloud and tree cover are major obstacles to theseapproaches, however, because the sensor cannot seethe water surface. Researchers and monitoring programstherefore are increasingly turning to active radar imagers,which penetrate forest and cloud cover and allow water tobe distinguished from other features as well as providingsurface elevation data for broad areas (Alsdorf et al., 2007;Schumann et al., 2009).

2.4 Water depth

The importance of water depth to monitoring, mappingand modeling river habitats has generated consider-able research interest in measuring depth from opticalimagery. As long as the water is clear enough to see to thebottom, there are three general approaches that providerelatively accurate depth estimates (Table 2.1).

The easiest depth mapping technique is the correla-tion approach, where the brightness of the image (i.e.,the pixel value) at a number of locations is correlatedto field measurements of depth at the same locations.The regression equation derived from this correlation isthen applied to the remainder of the image to estimatewater depths. There are a number of variations to thecorrelation approach. The highest accuracies (Table 2.1)are achieved by using multiple regressions where morethan one image band (e.g. red, green and blue bands)are correlated to depth (e.g., Gilvear et al., 2007; Lejotet al., 2007). Marcus et al. (2003) achieved high accura-cies using 128-band hyperspectral imagery that coveredthe visible and shortwave infrared wavelengths. They firstran a principal components analysis on the water portionof the image to remove noise and reduce the dimension-ality of the spectral signal, then ran a step-wise regression

for each biotype (riffle, pool, glide, etc.) to derive depthsthat were specific to different surface turbulence regimeswithin the stream. The major limitation specific to thecorrelation approaches is that water depths must bemeasured in the field at the same time as imagery is col-lected to avoid variations in discharge and channel shapethat could modify the relationship between pixel valuesand depth.

Physically-based models provide an alternative to sim-ple correlation approaches. These models are based onthe physics of how light moves through water, as is sum-marised in Chapter 3 and reviewed by Legleiter et al.(2004; 2009). From a management perspective, the mostreadily applicable of these models are those that avoidthe need for field teams to collect field data at the time-of-flight, which removes a major logistical constraint.Fonstad and Marcus (2005) developed a hydraulicallyassisted bathymetry (HAB) model that couples equationsdescribing light attenuation by the water column and thehydraulics of open channel flow with data on discharge,slope and channel width to map depths throughout astream (Figure 2.2). Their physical modeling techniquedoes not require field crews or data collection specific tothe project. The discharge data can come from nearbygauging stations, slopes can come from maps or othersensors, and width can be measured from the imagery.Moreover, the model is sufficiently simple that the math-ematical formulae used to compute depth estimates canbe implemented in a spreadsheet. Because the HAB tech-nique does not require field measurements, it can beused with historical imagery so long as discharge data areavailable for a nearby site.

Some researchers have expressed concern, however,that simple models like HAB do not consider the effects ofvariables like turbidity, substrate size and colour, surfaceturbulence, algae on rocks, and other factors that couldpotentially complicate depth mapping. In an extensivetheoretical and empirical experiment, however, Legleiteret al. (2009) demonstrated that these other factors canbe accounted for and accurate depth estimates achievedby using the natural log of an appropriate band ratio(Table 2.1). A band ratio is simply one band divided byanother, and typically the green band value is dividedby the red band value for the same pixel. Differences inpixel values due to sediment colour, sediment size, veg-etation, and other substrate characteristics are, in effect,normalized by band ratios, leaving the depth signal as theprimary factor driving variations in pixel values. Legleiteret al. (2009) also demonstrated that depth maps derivedfrom such ratios show remarkable resiliency across a range

2 Management Applications of Optical Remote Sensing in the Active River Channel 23

Tab

le2.

1A

ccu

raci

esfo

rde

pth

mea

sure

men

ts.A

ccu

racy

mea

sure

sre

pres

ent

the

hig

hes

tac

cura

cies

ach

ieve

du

sin

gop

tica

lim

ager

yto

map

bath

ymet

ry.D

etai

led

expl

anat

ion

sof

the

met

hod

sar

ein

the

cite

dre

fere

nce

s.U

pdat

edan

dm

odifi

edfr

omM

arcu

san

dFo

nst

ad(2

008)

.

An

alyt

ical

Loc

atio

nIm

ager

yA

ccu

racy

Au

thor

(s)

App

roac

hP

latf

orm

&Sp

atia

lSp

ectr

al#

ofB

ands

Met

ric

Opt

imal

Mon

itor

Res

olu

tion

Ran

geU

sed

Rep

orte

dV

alu

es(n

m)

Ach

ieve

d

Mu

ltip

lere

gres

sion

ofsp

ectr

a

Rh

one

Riv

er,

Fran

ceP

arag

lider

dron

e–

digi

tal

cam

era

5cm

,14

cmV

isib

le3

(r,g

,b)

r2,m

easu

red

vs.

obse

rved

0.53

alls

ubs

trat

es0.

81al

luvi

alsu

bstr

ate

0.90

vege

tate

dsu

bstr

ate

Lej

otet

al.,

2007

Mu

ltip

lere

gres

sion

ofpc

ban

ds

Lam

arR

iver

,W

Y,U

SAH

elic

opte

r–

PR

OB

E-1

1m

400–

2400

128

r2,m

easu

red

vs.

obse

rved

0.67

inh

igh

grad

ien

tri

fles

,0.

99in

glid

es

Mar

cus

etal

.,20

03

Ph

ysic

alm

odel

–fi

eld

test

ed

Bra

zos

Riv

er,

TX

,USA

Air

craf

t–

digi

tal

cam

era

20cm

Vis

ible

3(r

,g,b

)r2

,mea

sure

dvs

.ob

serv

ed0.

77Fo

nst

adan

dM

arcu

s,20

05P

hys

ical

mod

el–

fiel

dte

sted

Soda

Bu

tte

Cre

ek&

Lam

arR

iver

,W

Y,U

SA

Gro

un

d-ba

sed

hyp

ersp

ectr

alsp

ectr

o-ra

diom

eter

0.21

to1.

00m

400-

900

vari

ous

sen

sor

con

figu

ra-

tion

sev

alu

ated

r2,m

easu

red

vs.

obse

rved

0.79

to0.

98ev

alu

ated

acro

ssw

ide

ran

geof

con

diti

ons

Leg

leit

eret

al.,

2009

Ph

otog

ram

met

rypl

us

phys

ical

mod

el

Sou

thSa

skat

chew

anR

iver

,Can

ada

Air

craf

t–

film

-ba

sed

ordi

gita

lca

mer

a

4cm

Vis

ible

1M

ean

erro

r:St

dde

viat

ion

ofer

ror:

−0.0

25to

0.18

m±0

.18

to±0

.31

mL

ane

etal

.,20

10

24 Fluvial Remote Sensing for Science and Management

Figure 2.2 Depth map for the McKenzie River above Springfield, Oregon, developed using the true color air photo to the left and theHAB-2 technique (Fonstad and Marcus, 2005). Darker tones indicate deeper water, except where shadows obscure the water. Depthsin this reach vary from zero to approximately 1.2 m. Pixel resolution is ∼10 cm.

of spatial and spectral resolutions and environmentalconditions. One finding of particular note is that the bestdepth estimates usually occurred when one of the ratiobands was centered around 710 nm, due to the strongabsorption of near-infrared wavelengths by water, whichmakes the remotely sensed signal highly responsive tosubtle variations in depth.

Finally, in addition to correlation approaches and phys-ical models, classical photogramettry with stereo pairs canprovide high resolution, accurate results. Application ofthese techniques to determine water depth is compli-cated, however, by the need to know the height of thewater surface above the bed surface that is being mapped.Westaway et al. (2000, 2001) solved this by identifyingwater surface elevations at the channel edge and extend-ing these water surfaces to the remainder of the channel.More recently, Lane et al. (2010) coupled photogram-metric depth estimation techniques with optical physicsto achieve accurate depth measurements in areas cov-ered by only one photo (Table 2.1). To accomplish this,they developed depth-reflectance relations in areas ofstereo coverage and then applied these relationships toreflectance values in areas covered by only one photo.

Water clarity is the key limitation to all techniquesfor measuring water depth with optical imagery.The maximum depth that can be remotely measuredis the maximum water depth to which the light canpenetrate and return to the surface and be detectedby the sensor, which varies with water column optical

properties, wavelength, instrument sensitivity, andsubstrate composition (Legleiter et al., 2004).

2.5 Channel change

Managers and researchers have long used aerial photo-graphs to map channel boundaries, bars, floodplain cover,erosion and other features (Gilvear and Bryant, 2003).Channel change maps are important for documentingflood hazard, erosion hazard, and changes in habitatdiversity, as well as for understanding the causes of thosechanges. The concept behind using remote imagery tocreate two dimensional planimetric channel change mapsis simple: features visible on air photos can be tracedor digitised and transferred to map coordinates, and ifimages are available from different dates, the maps can beoverlain to document change over time.

A major component of channel change often relatesto the introduction of human features such as groynesand weirs. These features often stand out clearly on ver-tical images because of their shape and texture, whichmake them jump out to the naked eye, and because oftheir different composition relative to surrounding mate-rials, which enables their detection through automated,spectrally-based techniques. However, when covered byvegetation or buried in sediments, remote sensing maywell miss features of this nature.

2 Management Applications of Optical Remote Sensing in the Active River Channel 25

Since the 1980s the use of remotely sensed opticalimagery for channel change mapping has exploded,largely because of the widespread availability of Geo-graphic Information Systems (GIS). GIS made it mucheasier to digitise river features from aerial or satelliteimagery, georectify and reproject the features to a com-mon coordinate system, conduct a wide variety of spatialand change-over-time analyses, and produce maps fromthe results. The techniques for mapping river featureswith remote imagery are standard to any GIS or remotesensing software, and many college students now have theskills necessary to carry out the work.

The ease of mapping with GIS, however, has led to acertain degree of complacency regarding the results. Toooften change detection mapped from optical imagery isconsidered to be ‘real’, when in fact the change may bethe result of cumulative errors. Scale distortions in theoriginal imagery, poor selection of ground control pointsused to georeference the image, and the algorithms usedto transform the image to a specific coordinate systemall generate errors. These errors can be substantial andare compounded when attempting to detect change overtime (Hughes et al., 2006). Simply put, just because itlooks good does not mean that it is good. Analyses ofchannel change should include error assessment and useof stable control points.

Error of this nature can be significantly reduced usingorthorectification techniques, which correct for planformdistortion caused by topography and sensor geometry.The orthorectification techniques, however, are morecostly and complex. They require additional data layerson elevation, use more expensive software, need moretime to set up, and call for more specialised personnel.

In addition to GIS, an exciting advance in opticalremote sensing of rivers is the three dimensional map-ping of channels and channel change (this can also bedone with active sensors, as is discussed in elsewhere inthis volume). Previously, mapping elevation changes inchannels has been a challenge because it requires highaccuracy to capture channel bed elevation changes thatare often subtle (Brasington et al., 2003). It is also requiresmeasuring elevations of both the submerged and emer-gent parts of the channel throughout the entire activechannel.

Several approaches have been used to solve theseproblems with optical imagery. Westaway et al. (2000,2001) modified classical photogrammetric techniques toaccount for the effects of refraction. In addition, West-away et al. (2003) developed an alternative approachthat uses classical photogrammetry to derive topographic

information for exposed surfaces and a correlation-basedmethod of estimating water depth within the wettedchannel from the image data.

More recently, Lane et al. (2010) extended these tech-niques so that they can be used with historical imagery.In their procedure, overlapping images and photogram-metric principles are used to derive local elevations insideand outside the channel. A filter then identifies whichphotogrammetrically derived water elevations are likelyto be valid based on the certainty with which correspond-ing points can be identified on the two photos. The validwater depths are used to develop a relation between imagebrightness and depth that is applied to the rest of the chan-nel to map depths. These depths in turn can be subtractedfrom the elevation at the dry surface of the channel mar-gin to determine bed elevation. Lane et al. found that themethod provided accurate depth estimates and could beused to detect vertical changes of 0.40 m or more witha 67% confidence of change (Table 2.1). Although morecomplex than this brief description would suggest, theLane et al. methodology should ultimately be applicableanywhere: (1) tie points can be identified on historicalimagery that can be surveyed in the present day; (2) theimage scale is sufficiently fine to detect depth changes;and (3) the imagery of the submerged areas displays arange of brightness that can be correlated with depth.Marcus (2012) provides more in-depth summaries of thetechniques developed by Westaway et al. and Lane et al.

As with channel change in the horizontal dimension,it is critical to consider error when estimating changes inthe vertical dimension. For example, the 67% confidenceof correctly detecting a vertical change of 0.40 m in thesystem examined by Lane et al. (2010) is reassuring atone level, but also a sobering reminder that verticalchanges typical of many small streams will be missedor inaccurately portrayed using the photogrammetricapproach. Brasington et al. (2003) and Lane et al. (2003)provide guidelines on how to evaluate vertical error andits implications for detection of vertical change. Likewise,one can determine the potential range of depths that canbe mapped and the precision of bathymetric mappingtechniques by modeling the optical physics under varyingstream conditions (e.g. Legleiter et al., 2009).

2.6 Turbidity and suspended sediment

Turbidity, in its formal optical definition, refers to theamount of attenuation and back scattering of light due tosuspended solids and dissolved load (Davies-Colley and

26 Fluvial Remote Sensing for Science and Management

Smith, 2001). Turbidity in and of itself is an importantcomponent of water quality that controls light avail-ability for photosynthetic organisms. Turbidity is alsoan important control on other remote sensing applica-tions in rivers; for example, depth estimates from opticalimagery are far less likely to be reliable where the bottom isobscured (Figure 2.3). Finally, because turbidity is relatedto sediment concentrations, its measurement can provideinformation on timing of sediment movement, variabilityin sediment sources, and (when coupled with dischargedata) sediment load, although some approaches movedirectly from pixel values to suspended sediment concen-trations without reference to turbidity (e.g. Pavelsky andSmith, 2009).

There is an extensive literature describing the use ofoptical imagery to monitor turbidity and, to a lesserdegree, suspended sediment concentrations. Whensuspended sediments are remotely mapped, it is usuallyaccomplished via remote mapping of the turbidity.A correlation between turbidity and ground-based

Figure 2.3 Aerial image of the confluence of Soda Butte Creek(left) and the Lamar River (right), Wyoming, USA. Theturbidity in Soda Butte Creek prevents light penetration of thewater column, which in turn limits measurement of depths,substrates size, or other features below the water surface. Ifground measurements are available to calibrate imagereflectance to turbidity, however, the imagery can be used tomap turbidity and suspended sediment concentrations in thetwo streams.

measurements of suspended sediment concentrationsthen generates an estimate of suspended sedimentconcentrations. Because optical remote sensing imagesthe upper portion of the water column where suspendedsediment concentrations are usually lower, remotesensing approaches may underestimate depth-integratedsuspended sediment concentrations.

The vast majority of literature on turbidity andsediment concentrations focuses on oceans, lakesand estuaries. The research that does exist for riversfocuses almost entirely on large systems like the Amazon(Mertes et al., 1993), Yellow (Aranuvachapun andWalling, 1988) or Yangtze Rivers (Liu et al., 2008), whereturbidity, suspended sediment concentrations, and sur-face characteristics of the streams are relatively constantover long reaches. In contrast, a large host of factorsin smaller systems alter the relation between turbidity,suspended sediment concentrations, and reflected lightover short distances. Surface turbulence, water depth,substrate colour and size, aquatic vegetation, sun anglerelative to the water, and local variations in sedimentcolour all can alter the signal received by the sensorindependent of any actual changes in turbidity. Differentinputs and variable timing of flow from tributaries alsocan alter the correlations between turbidity and sedimentconcentrations. In short, it remains difficult to accuratelymap turbidity and suspended sediment concentrationsin small and medium streams with remote sensing.

The majority of approaches for estimating turbidity orsediment concentrations use field measurements to estab-lish empirical relations between image values, turbidity,and suspended sediment. Mertes et al. (2002) provide anoverview of image processing techniques for mapping sus-pended sediment concentrations. Presently, there are nowidely accepted standard approaches for setting up theseempirical relationships. Most researchers use some com-bination of red and/or near infrared bands to map turbid-ity, although green bands can be of use, especially whenchlorophyll concentrations strongly influence turbidity.The relationships set up between sediment concentra-tions or turbidity and spectral signatures include spectralmixture analysis (Mertes, 1993), simple logarithmic rela-tions (Aranuvachapun and Walling, 1988), normaliseddifference relations (Han et al., 2006), simple regression(Wang et al., 2009), multiple regression (Shibayama et al.,2007), General Additive Models (Bustamante et al., 2009)and neural networks (Teodoro et al., 2008). As of now,researchers seem inclined to choose the method thatgenerates the best fit to the field data without havingstrong theoretical arguments for choosing one method

2 Management Applications of Optical Remote Sensing in the Active River Channel 27

over another. Moreover, simple methods often seem togenerate results that are nearly as accurate as more com-plex approaches. Bustamante et al. (2009), for example,found that the simple Water Turbidty Index of Yagamataet al. (1988) yielded an r2 value of 0.75 for measured ver-sus modeled turbidity values, while a far more complexGeneral Additive Model yielded an r2 of 0.79. Similarly,Wang et al. (2009) found that one variable regressionusing band 4 (near infrared) of Landsat provided excel-lent estimates of suspended sediment concentrations inthe Yangtze River (Table 2.2).

The wide range of techniques for estimating turbiditywith optical imagery indicates the need for a more gen-eral theoretical basis for turbidity measurement. Dekkeret al. (1997) reviewed the optical theory that underliesturbidity mapping by remote sensing and identified sev-eral issues of particular importance. Light interacts withturbid waters in a non-linear manner, making it hard todevelop empirical relations that can be transferred fromone system to another. Sun angle relative to the surfaceand sensor can also strongly alter turbidity estimates. Themany factors controlling turbidity and its reflected signalhave hindered development of physical models that canbe applied without local calibration data. Mertes et al.(1993) developed an approach based on spectral mixtureanalysis that holds promise, but even their techniquerequired calibration to laboratory data. For the fore-seeable future, managers therefore will have to rely onestablishing empirical relations between local turbidityand the sensor signal. Empirical approaches of this naturecan produce accurate results, but are difficult to imple-ment in remote rivers or during flood conditions, whichare often the periods of greatest interest to managers.

2.7 Bed sediment

Mapping bed sediment size is important for documentingin-stream habitat for fish, macroinvertebrates and otherorganisms, for characterising flow resistance for hydraulicand flood inundation models, and for modeling sedimenttransport and channel stability. There is a substantialbody of literature on ground-based optical measurementof sediment size where the pixel resolution is far smallerthan the sediment size. In this case the individual particlescan be seen with the naked eye on the imagery, so thefocus shifts to automating the procedure in order todelineate particle boundaries and measure particle axes(e.g. Raschke and Hryciw, 1997; Graham et al., 2005).More recently, terrestrial laser scanning (an active sensing

system) has been used to map exposed sediment sizes atreach scales (Hodge et al., 2009). These approaches toground-based mapping of sediment size are useful at thescale of an individual plot, bar, or reach but are notfeasible over longer lengths of stream where thousandsto millions of photos or ground-based surveys might berequired to cover the entire area.

It is only recently that airborne optical imagery has beenavailable at sufficiently fine spatial resolutions to measuresediment size over long lengths of stream. The coarserresolution of these photos compared to ground images,however, limits what they can detect.

Rather than trying to measure individual grains as isdone with ground photos, approaches for measuring sed-iment size with aerial imagery use the image semivariance,a statistical technique that characterises the ‘graininess’or texture of an image. The concept is simple. Areas withlarger sediment sizes have more shadows cast by the largeclasts and therefore have a more heterogeneous texture.In contrast, surfaces with much finer sediments have lessof a size difference between clasts, have less shadowingeffect, and are more homogenous in appearance. Thesevariations in image texture within a given window (e.g.35 × 35 pixels) can be measured by a two-dimensionalvariogram (among other techniques), with higher valuesindicating more brightness variation from one pixel tothe next. Carbonneau et al. (2004, 2005) discovered thatthe values from a two dimensional variogram are linearlyrelated to grain size for both dry and submerged sedi-ments. A linear regression between field measurementsof the median particle size, D50, and the two dimensionalsemi-variance of the image for the same locations wasused to develop equations that could then be applied tothe remainder of the image to accurately estimate D50

(Table 2.2). This technique can also be used to map otherpercentiles of the sediment grain size distribution (D16,D84, etc.) so long as: (a) the window size is large enough toget a stable semivariance signal; (b) the sediment patchesare relatively uniform at the scale of the window; and (c)the grain size fraction (e.g. D50) is larger than the imageresolution (Carbonneau et al., 2005). Using this tech-nique with 3 cm resolution imagery, Carbonneau et al.were able to continuously map sediment size along 80 kmof the St. Marguerite River in Quebec.

Assuming that the sediment can be seen through thewater, image resolution is the biggest limitation on mea-suring sediment size. The general rule of thumb is thatthe smallest size of sediment that can be mapped withblack and white or three band colour imagery is equal tothe pixel resolution of the image (Carbonneau, 2005).

28 Fluvial Remote Sensing for Science and Management

Tab

le2.

2A

ccu

raci

esof

sedi

men

tan

dtu

rbid

ity

mea

sure

men

ts. A

ccu

racy

mea

sure

sre

pres

ent

the

hig

hes

tac

cura

cies

ach

ieve

du

sin

gop

tica

l im

ager

yto

map

bath

ymet

ry.

Det

aile

dex

plan

atio

ns

ofth

em

eth

ods

are

inth

eci

ted

refe

ren

ces.

Upd

ated

and

mod

ified

from

Mar

cus

and

Fon

stad

(200

8).

Feat

ure

An

alyt

ical

Loca

tion

Imag

ery

Acc

ura

cyA

uth

or(s

)

App

roac

hP

latf

orm

&Sp

atia

lSp

ectr

al#

ofB

ands

Met

ric

Opt

imal

Mon

itor

Res

olu

tion

Ran

geU

sed

Rep

orte

dV

alu

esA

chie

ved

(nm

)

Tu

rbid

ity

Sim

ple

regr

essi

onof

spec

tra

Tam

paB

ayes

tuar

y,FL

,U

SA

Sate

llite

(MO

DIS

)25

0m

620–

670

1r2

,mea

sure

dvs

.ob

serv

ed0.

73C

hen

etal

.20

07

Gen

eral

ized

Add

itiv

eM

odel

Gu

adal

quiv

irR

iver

,Spa

inLa

nds

at-5

TM

Lan

dsat

-7E

TM

+30

m45

0–

2350

eval

u-

ated

vari

esw

ith

sett

ing

r2,m

easu

red

vs.

obse

rved

0.79

Bu

stam

ante

etal

.,20

09

Susp

end

edse

dim

ent

Reg

ress

ion

Yan

gtze

Riv

er,

Ch

ina

Lan

dsat

-7E

TM

+30

m77

0–86

01

(ban

d4)

r2,m

easu

red

vs.

obse

rved

0.86

to0.

87W

ang

etal

.,20

09

Bed m

ater

ial

grai

nsi

ze

Spat

ial

vari

ogra

mSa

inte

Mar

guer

ite

Riv

er,Q

ueb

ec,

Can

ada

Hel

icop

ter

–di

gita

lcam

era

3cm

Vis

ible

3(r

,g,b

)r2

,mea

sure

dvs

.ob

serv

ed0.

80fo

rD

50C

arbo

nn

eau

etal

.,20

04

Spat

ial

vari

ogra

mSa

inte

Mar

guer

ite

Riv

er,Q

ueb

ec,

Can

ada

Gan

try

–di

gita

lca

mer

a0.

3m

mV

isib

le1

(gra

ysc

ale)

r2,m

easu

red

vs.

obse

rved

0.93

for

dist

ingu

ish

ing

san

dfr

omla

rger

clas

ts

Car

bon

nea

uet

al.,

2006

Lin

ear

un

mix

ing

Rib

ble

Est

uar

y,L

anca

shir

e,U

KD

aeda

lus

1268

Air

born

eT

hem

atic

Map

per

1.75

mV

isib

leto

Shor

t-w

ave

IR

10r2

,mea

sure

dvs

.ob

serv

ed0.

79fo

r%

clay

0.60

to0.

84fo

r%

san

dB

est

indr

ier

con

diti

ons

Rai

ney

etal

.,20

03

2 Management Applications of Optical Remote Sensing in the Active River Channel 29

For example, an image with 10 cm resolution can be usedto map sediments of 10 cm or larger. In turn, this meansthe image cannot be used to accurately characterise thefull distribution of sediment sizes (unless the finest sed-iments are 10 cm in size), because the smaller sedimentsare not being ‘seen’ by the sensor. Managers applyingthe technique with panchromatic or colour imagery willalmost certainly need to charter special flights to collectimagery at the resolution of the sediment size of interest.For example, Carbonneau et al. (2004, 2005) chartered ahelicopter to fly at 155 m above the river to acquire 3 cmresolution imagery.

If multispectral or hyperspectral imagery are available itmay be possible to map features smaller than the nominalimage pixel size. Rainey et al. (2003), for example, usedspectral mixture analysis to map fine sediment sizes inan estuary. This type of ‘pixel unmixing’ is accomplishedbased on knowledge of the spectral characteristics of thedifferent pure end members present within a pixel. Forexample, it is easy to estimate how much black andwhite paint are mixed together to create a grey colourin a single pixel; the darker the grey, the larger theproportion of black paint. Rainey et al. (2003) used theoptical theory that small spaces between fine grains actas blackbody cavities. The black body behaviour fills inthe spectra for each pixel to a degree proportional tothe grain size, enabling estimates of the proportion ofsand and mud, even in pixels that were 1.75 m in size(Table 2.2). Unfortunately, the optical theory only workswell in relatively dry sediments. As of yet, there are notechniques for measuring sediment smaller than the pixelsize in submerged sediments. This means that issues suchas sand embededness cannot at present be characterizedwith airborne imagery.

2.8 Biotypes (in-stream habitat units)

Biotypes, also called ‘in-stream habitats’, ‘micro-habitats’or ‘morphologic units’, refer to features such as riffles,pools, glides, and exposed bars. From an optical perspec-tive, these features vary in a number of important waysthat can be captured by remote sensing. Optical variationsassociated with depth, surface turbulence, substrate size,and vegetation associated with the units enable differenti-ation of features that are essentially composed of the samematerial – water. Likewise, the differences in compositionbetween exposed bars, water and vegetation make it easyto manually map bars if the image resolution is appro-priate. Automated mapping of bars with classification

algorithms is also relatively straightforward, provided thefeatures are not in deep shadow.

Biotypes are receiving increasing attention from streammanagers. In Europe, biotype mapping is a mechanism fordefining biodiversity in streams under the Water Frame-work Directive (Dodkins et al., 2005). In the UnitedStates, many agencies use in-stream habitats to char-acterise stream health. A number of schemes exist fordefining the different kinds of biotypes, leading to confu-sion around definitions; one person’s glide can be anotherperson’s run. A good summary of the different biotypeclassification schemes and their overlap is provided inTable 1 of Milan et al. (2010).

Despite their growing importance, there is relativelylittle research on mapping biotypes with remotely sensedoptical imagery. Researchers have achieved good resultsmapping in-stream habitats using supervised classifica-tion, a classical technique that is included on all remotesensing software. To apply this technique, the user firstidentifies a number of pixels, called training sites, on theimage that are characteristic of each class of interest (e.g.,riffles, pools, glides). The algorithm then maps other pix-els on the image that have spectral signals similar to thetraining sites. Remotely sensed biotype maps can be moreprecise than ground-based maps. Ground surveyors willoften lump many small features into a larger adjacentfeatures (e.g., lumping low velocity stream margins intothe riffle that dominates that stream reach). If the pixelsize is small enough, the imagery will differentiate thesemany smaller extent variations.

Legleiter et al. (2002) and Marcus (2002) found thatmapping accuracy is sensitive to the number of spectralbands used for classification. Simply put, more bandsare better. Hyperspectral imagery (discussed in anotherchapter in this volume) thus provides better results thanmultispectral imagery (Marcus, 2002), and imagery thatincludes short wave infrared is better than imagery thatonly spans the visible wavelengths (Legleiter, 2003). Thespatial resolution of imagery relative to the size of thechannel being mapped is also a crucial consideration.Marcus et al. (2003) documented higher mapping accu-racies with 1-m imagery in a 5th order stream than ina 4th order stream, which in turn had higher accuraciesthan in a 3rd order stream (Table 2.3). The drop inaccuracy results from the ‘mixed pixel’ problem, whichoccurs when one pixel encompasses multiple features. Inthe stream context, a 1-m pixel in a third order streamis more likely to include portions of two units (e.g., aglide and pool) than in the 5th order stream, where thebiotypes are much larger.

30 Fluvial Remote Sensing for Science and Management

Tab

le2.

3A

ccu

raci

esfo

rbi

otyp

e,al

gae

and

woo

dm

appi

ng.

Acc

ura

cym

easu

res

repr

esen

tth

eh

igh

est

accu

raci

esac

hie

ved

usi

ng

opti

cal i

mag

ery

tom

apba

thym

etry

.

Det

aile

dex

plan

atio

ns

ofth

em

eth

ods

are

inth

eci

ted

refe

ren

ces.

Upd

ated

and

mod

ified

from

Mar

cus

and

Fon

stad

(200

8).

Feat

ure

An

alyt

ical

Loca

tion

Imag

ery

Acc

ura

cyA

uth

or(s

)

App

roac

hP

latf

orm

&Sp

atia

lSp

ectr

al#

ofB

ands

Met

ric

Opt

imal

Mon

itor

Res

olu

tion

Ran

geU

sed

Rep

orte

dV

alu

esA

chie

ved

(nm

)(n

m)

Bio

typ

esP

rin

cipa

lcom

pon

ent

redu

ctio

nfo

llow

edby

supe

rvis

edm

axim

um

likel

ihoo

dcl

assi

fica

tion

Lam

arR

iver

,W

Y,U

SAH

elic

opte

r–

PR

OB

E-1

140

0– 2400

128

Pro

duce

r’s

accu

racy

86%

Mar

cus

etal

.,20

03

Pri

nci

palc

ompo

nen

tre

duct

ion

follo

wed

byco

-kri

gin

g

Lam

arR

iver

,W

Y,U

SAH

elic

opte

r–

PR

OB

E-1

140

0to

2400

128

Pro

duce

r’s

accu

racy

94%

Goo

vaer

ts,

2002

Alg

aeM

atch

edfi

lter

Cac

he

Cre

ek,

WY

,USA

hel

icop

ter

140

0to

2400

128

Use

r’s

accu

racy

75%

–in

ten

tion

ally

over

esti

mat

ed#

alga

esi

tes

tom

ake

sure

non

ew

ere

mis

sed

Mar

cus

etal

.,20

01

Reg

ress

ion

Swan

Riv

er,

Au

stra

liaC

ASI

(spe

ctra

lm

ode

wit

h28

8ba

nds

)on

airc

raft

Not re

port

ed42

3– 946

1at 75

0n

mr2

,mea

sure

dvs

.ob

serv

ed

tota

lcel

ls0.

94di

nofl

agel

late

s,0.

44ch

loro

phyl

l-a,

0.77

chlo

roph

yta

0.70

diat

oms

0.59

chry

toph

ta0.

95

Hic

ket

al.,

1998

Woo

dT

hre

shol

dfi

lter

onm

ult

iple

pcba

nds

Lam

arR

iver

,W

Y,U

SAH

elic

opte

r-P

RO

BE

-11

400

to24

0012

8P

rodu

cer’

sac

cura

cy85

%M

arcu

set

al.,

2003

Th

resh

old

filt

eron

1pc

ban

dU

nu

kR

iver

,A

K,U

SAA

ircr

aft

–di

gita

lca

mer

a0.

45V

isib

le3

(r,g

,b)

Ove

rall

use

r’s

accu

racy

89%

Smik

rud

and

Pra

kash

,20

06

2 Management Applications of Optical Remote Sensing in the Active River Channel 31

Several researchers have developed alternativeapproaches to supervised classification. Maruca andJacquez (2002) achieved good results using a spatiallyagglomerative cluster technique, and Goovaerts (2002)achieved high accuracies using a geostatistical co-krigingapproach (Table 2.3). Legleiter and Goodchild (2005)examined the potential of a ‘fuzzy’ approach to streamclassification that provides a more realistic representationof the gradual transitions between similar habitatunits. Their work revealed far more complexity inthe stream patterns than is captured in the simpleeither/or dichotomy of biotype classification. None ofthese approaches, however, are presently available inexisting software packages. Special programming skillsare therefore needed to implement them, making themless accessible to the management community.

Regardless of the approach, no physical models yet existthat allow remote mapping of biotypes in the absence offield data. Because the spatial pattern of biotypes canchange with variations in discharge, personnel must be inthe field at or near the time of image acquisition to mapbiotypes. The field maps are then used to select pixelsto train the classification algorithms. Given that fieldmapping of biotypes is necessary regardless of mappingapproach, using remote sensing only makes sense if thespatial extent of mapping far exceeds what a survey teamcan readily achieve while in the field.

2.9 Wood

Wood in rivers plays a major role in forming andaltering stream habitats in many streams. Wood can forcesediment deposition and transport, alter in-stream habi-tats and stream morphology, provide organic debris formacroinvertebrates, and create shelter for fish. Emplac-ing wood is a major tool used in stream restoration.Promoting wood accumulation is often a central goalof riparian management strategies (e.g. Fox and Bolton,2007; Jochem et al., 2007).

In theory, remote detection of wood in river channelsor on exposed bars should be relatively straightforward.At its simplest level, if sufficiently fine resolution imageryis available, one can see the wood on the images and mapit manually. Even from an automated perspective, detec-tion of wood should be relatively simple. Mapping depth,turbidity, channel change, and biotypes all require detect-ing variations, often subtle, in one feature type (water).

In contrast, mapping wood, at least against a backgroundof water or sediment, requires detecting the differencebetween features with distinctly different compositions.

Automated mapping of wood in and along rivers can, infact, be relatively simple if one has hyperspectral imageryor fine resolution, high quality colour imagery Marcuset al. (2003) and Smikrud and Prakash (2006) both usedsimilar approaches to detect wood with relatively highaccuracies (Table 2.3). Both sets of researchers first cal-culated principal components from their hyperspectralimagery to isolate the spectral signatures of different fea-tures. They then applied a matched filtering techniqueto the principal component images to detect wood. Thematched filter operates by using some wood pixels withinthe image to train an algorithm that then finds similarpixels elsewhere in the image. A major advantage of thematched filtering technique is that there is no need forfield teams if some wood can be seen clearly on theimagery. Principal component transformations are avail-able in all remote sensing software and matched filteringis included in an increasing number of these packages.

If hyperspectral imagery is available, the matched filterwill almost certainly detect wood in areas where it cannotbe seen with the naked eye on the same image. In this case,the algorithm is ‘unmixing’ the pixel to detect locationswhere wood makes up only a portion of the pixel. Thiscan lead to confusion on the part of users, who maybelieve the classification is showing wood where it doesnot exist. In addition, finding pieces of wood smaller thana pixel may cause problems if the user is seeking to mapwood of a given size or larger. For example, many fluvialwood studies only seek to document wood that is 1 or 2 min length and at least 10 cm in diameter – large enoughto affect the flow. Hyperspectral imagery may detectmuch smaller wood, making it difficult to determinewhich pieces are of sufficient size to alter flow dynamics.Pixel unmixing approaches or shape detection algorithmsmight provide a solution to this issue, but no research hasyet been attempted along these lines.

2.10 Submerged aquatic vegetation(SAV) and algae

SAV and algae are important to ecosystem health, provid-ing food and cover for a wide range of species, removingtoxins from water and sediments, and stabilising streambeds. On the other hand, an over abundance of SAV

32 Fluvial Remote Sensing for Science and Management

or the presence of certain species can be damaging toecosystem health, water quality, and human structures.Problems associated with SAV and algal blooms rangefrom eutrophication to acute toxicity associated withblue green algae to fouling of engineering works. Remotemapping of SAV and algae for management applicationsshould thus seek to provide more than a simple percentcover map; ideally, it would also provide information onspecies composition, biomass, and physiology.

Almost all SAV research has focused on lakes, deltas,wetlands, estuaries and coastal waters. As with turbidity,where research also has centered on large water bodies,focusing on these environments has not significantlyadvanced remote sensing of SAV in small stream systems,where local-scale variations can confound the spectralsignal. Work from large water bodies thus provides aguide to mapping SAV and algae, but should not betransferred directly to smaller streams without additionalstudy and validation.

Silva et al. (2008) review remote sensing and plantphysiology issues to consider in mapping SAV. A centralissue in mapping SAV is the relatively strong absorption ofoptical wavelengths by water. An identical plant thereforelooks different to the sensor when it is emergent, justbeneath the surface, or more deeply submerged. Similarly,changes in plant structure, age, and reflectance confusethe identification of plants and complicate estimationof biomass.

Researchers thus use field measurements that incor-porate plant- and location-specific variations to developregressions that use individual bands, band ratios, orprincipal components to predict biomass (Silva et al.,2008). Models of this sort have yielded R2 values of 0.79(Armstrong et al., 1993) to 0.85 (Zhang, 1998) for com-parisons of measured and estimated SAV biomass in largerelatively stationary water bodies. Regression-based esti-mates reach a plateau, however, beyond which biomasscontinues to increase without a corresponding changein the spectral reflectance – the signal becomes saturated(Figure 2.4). In addition to biomass, remote sensing hasbeen used to map SAV community type, chlorophyllconcentration (Penuelas et al. 1993), photosynthetic effi-ciency (Penuelas et al. 1997, 1993), and foliar chemicalcomposition (LaCapra et al. 1996).

Hyperspectral data are useful for separating SAV andalgal chlorophyll signals (Williams et al. 2003) and iden-tifying invasive species (Underwood et al., 2006). Eventhe additional spectral information, however, does notentirely overcome the complex signals generated byvariable turbidity, water depths, and plant physiology(Hestic et al., 2008). Regardless of sensor type or plat-form, the variability in results among research projectsand the potential complexities in mapping SAV indicatethat – from a management perspective – this applica-tion is still in a developmental rather than an opera-tional phase.

Figure 2.4 Submerged aquatic vegetation (SAV), Browney Brook, County Durham England, and algae along a side channel of theTummel River, Scotland. The distinct spectra of chlorophyll relative to water and substrate enables mapping of general locations ofSAV and algae, although separation of species and mapping of parameters such as biomass is more problematic.

2 Management Applications of Optical Remote Sensing in the Active River Channel 33

As with SAV, the literature on algae focuses on largewater bodies. Mapping of marine algae is so well estab-lished that satellites have been launched largely to monitorchlorophyll concentrations and map phytoplankton (e.g.,the Coastal Zone Color Scanner in 1978). High spatial res-olution imagery can provide good measurements of algalblooms over time in lake settings (Hunter et al., 2008).Hoogenboom et al. (1997) and Quibell (1991) describesome of the basic optics to consider in mapping algaewith remote imagery.

Research on algae mapping in rivers suggests signif-icant potential for this application. Hick et al. (1998)found high correlations between the 750 nm band and anumber of algal parameters (Table 2.3) and concludedthat only three to four bands were needed to map algalblooms over large areal extents, provided that in-watercalibration data are available. Marcus et al. (2001) usedhyperspectral imagery and pixels from an algae filled poolto train a matched filter to find similar sites throughouta backcountry region. In a subsequent survey of thestream, 75% of the sites they classified as algae had algae(Table 2.3) and led to the discovery of four previouslyunknown amphibian sites. This application demonstratedthe value of remote sensing for exploratory purposes ininaccessible regions.

2.11 Evolving applications

The applications discussed above are ones for which therehas already been a substantial body of work. Severalevolving lines of research are also worth tracking depend-ing on management needs. Forecasting and detection ofice breakup on large rivers is receiving increased atten-tion (Pavelsky et al., 2004; Morse and Hicks, 2005, Kaaband Prowse, 2011). This application has the potential tobecome a major tool in inaccessible sub-polar and polarrivers, just as detection of floods has become one of theestablished applications of remote sensing. There is alsoan emerging effort to use remote sensing measurementsfrom streams to map derived variables such as streampower (Jordan and Fonstad, 2005; Carbonneau et al.,2011), although most of these require merging active andpassive sensors.

Moving beyond technique development, researchersare now beginning to apply the remote sensing methodsdiscussed above to better understand rivers, although thiswork is just now beginning to appear in the publishedliterature. Marcus and Fonstad (2008, 2010), for example,used continuous data on depths derived from remote

sensing to show that classical downstream hydraulicgeometry relations do not provide accurate predictions ofdepth. This finding has significant implications for streamrestoration, which frequently uses hydraulic geometry toestimate target depths (and other parameters) for natu-ralising streams. Ongoing work by the authors and othersis examining the potential to use parameters derivedfrom remote sensing to document habitat impacts oflow-head dams, model water quality, and target riverreaches for restoration. The list of potential applicationsis expanding rapidly, especially if one includes work thatis examining fusion of optical data with LiDAR or radar,which can enable extraction of other hydraulic vari-ables such as water surface slopes, velocity, and Froudenumbers. Carbonneau et al. (2011), for example, usedtopographic information derived from radar imagery toobtain water surface slopes and used optical imageryto measure wetted widths and water depths, then com-bined these data to derive stream power and velocitiesevery meter along 16 km length of a stream. Because theymapped water depth and substrate size for every squaremeter along the stream, they were also able to derivecontinuous metrics of spawning habitat suitability for theentire stream.

2.12 Management considerationscommon to river applications

Our discussion so far has focused on reasons for usingremote sensing, potential applications in river settings,and some of the potential complications specific to indi-vidual applications (e.g., problems associated with depthmeasurement). But there are also issues common toalmost any effort to use remotely sensed data to obtainriver information on turbidity, SAV, or whatever. Theseissues are discussed in detail in Marcus and Fonstad(2008). Rather than repeat their work, we summarisetheir discussion in Table 2.4 and focus here on issuesof particular relevance to stream managers. In doing so,we assume that the manager is working with a remotesensing professional who has addressed issues of dataacquisition, quality and analysis. In this circumstance, themanager will be given remote sensing results, probably inthe form of maps and, preferably, accuracy assessments.The manager is then faced with interpreting those results,presenting them to other users, and determining whetherand to what extent this information can be incorporatedinto management plans.

34 Fluvial Remote Sensing for Science and Management

Table 2.4 Common issues in river remote sensing. The issues highlighted below commonly arise in the context of using optical

remote sensing to map river variables. The table does not summarize issues that are common to remote sensing across all

applications, such as atmospheric and geometric corrections, standard image processing issues, or large data storage requirements.

Category Common Issues in River Applications Potential Solutions

Logistics: see Aspinall et al.,2002; Marcus andFonstad, 2008

Timing of flights: Aircraft-based imagerycannot be acquired when variable needs tobe measured

– acquire your own aircraft– use satellite imagery– use existing imagery for historical analysis

Timing of field data: Field data collected underconditions different than those under whichthe image data were acquired

– schedule field data collection during periodsof relatively stable stream conditions (e.g.low flow period)

Location: Field sites inaccessible or toodangerous to collect ground validation data

– substitute data from similar settings– collect data at less dangerous time (e.g. use

post-flood indicators of depths)– model data to simulate plausible values

Expertise: Requires personnel with significantexpertise in remote sensing

– contract with existing experts– use existing web-based data sets (e.g. for

flooding)– seek out automated approaches

Optical environment: seeLegleiter et al., 2009;Marcus and Fonstad, 2008

Obstructed view of river: from above – use low elevation platforms (e.g. hand heldballoon, drones)

– use active device (LiDAR, radar)

Turbidity: blocks light penetration of watercolumn

– use ground-based measurements (e.g. totalstation surveys, sonar, ground-based radar)

Shadows: create different radiance values overidentical features

– mask out shadowed areas– develop algorithms for shadow removal

Sun-target-sensor geometry: obscures features(e.g. reflections, sun glint)

– plan image acquisition to avoid unfavorableconditions

– use algorithms to normalize lightingconditions

Local variations: turbulence, substrate color,SAV, etc. generate different reflectances for agiven measure (e.g. depth)

– use band ratios to normalize for variations– stratify image to measure variable separately

in each category

Imagery & ground data: seeAspinall et al., 2002;Legleiter et al., 2002, 2009;Marcus et al., 2003

Location precision: must be high to match smallfeatures on imagery and in streams

– map directly to the imagery– set up benchmark/targets to tightly

co-registered imagery to ground surveys

Spatial resolution: may not be sufficient todetect small stream features

– acquire imagery with high spatial resolutionfor smaller streams

– use pixel unmixing for features with distinctspectra (e.g. wood)

Spectral coverage: Narrow spectral resolutionover a broad spectral range sometimesneeded to separate features (e.g. biotypes,SAV) that have similar compositions

– use hyperspectral imagery– use mapping algorithms that do not rely to

such a large extent on between-spectravariations (e.g. semi-variograms, kriging)

2 Management Applications of Optical Remote Sensing in the Active River Channel 35

2.13 Accuracy

Remote sensing results are almost always less than 100%accurate when compared to ground data. In fact, thelarge majority of the highest accuracies achieved withremote sensing of rivers range between about 75 and90% (Tables 2.1, 2.2 and 2.3). Accuracies are limited toless than 100% due to the previously discussed obsta-cles that are specific to individual applications, as wellas a number of issues common to most optical remotesensing projects. These ‘generic’ issues are summarisedin Table 2.4 and discussed in more detail by Aspinallet al. (2002), Legleiter et al. (2002, 2009) and Marcus andFonstad (2008). Figure 2.5 shows how the optical envi-ronment can vary over short distances with viewing angleand location, highlighting some of the issues identified inTable 2.4.

In addition, certain river settings do not work wellfor optical remote sensing. In particular, high energy andsmall headwater streams tend to have the view of the watercolumn obscured by turbulent flow, in-channel featureslike boulders and wood, and overhanging vegetation. Inaddition, the small size of these streams means that veryfine resolution imagery is necessary to capture the finescale of variations in the stream (e.g., the rapid transitionfrom a step to a pool in a step-pool system).

Do these levels of uncertainty mean that remote sens-ing results are too fraught with error to be useful for

management applications? The simple answer is ‘no’ – forseveral reasons.

First, the metrics classically used to assess ‘accuracy’of remote sensing data might not be entirely appropriate(Marcus et al., 2003, Legleiter et al., 2011). Standard meth-ods of characterising accuracy in remote sensing assumethe ground data are correct. Differences between ground-based and remote sensing results thus are assumed torepresent error in the remote sensing. But what if theremote sensing data is actually more reliable and infor-mative than the ground data? Marcus (2002) and Legleiteret al. (2002), for example, argued that their remote sens-ing maps of biotypes were more accurate than their fielddata. This was because surveyors on the ground combinedlarge sections of river into one unit (e.g., a riffle), even if‘mini-glides’ were present within the riffle. In contrast, thehigh spatial resolution imagery would also map most ofthat same unit as a riffle, but also map some pixels as glidesdepending on local variations in surface turbulence anddepth. In this case the remote sensing imagery is probablymore precise in its mapping of fine resolution features.The determination as to whether the remote sensing mapis more or less accurate depends on whether you are adetail-oriented ‘splitter’ or a ‘lumper’ focused on the bigpicture; others may be more comfortable representingthis kind of natural variability using fuzzy approaches,as described by Legleiter and Goodchild (2005). Similararguments can be made regarding remote sensing mapsof wood and bed sediment size.

(a) (b)

Figure 2.5 Photos from a bridge over the Garry River below Killecrankie, Scotland, demonstrating how the optical environment canchange dramatically over short distances (Table 2.4). (a) Looking upstream, portions of the river are obscured by trees, shadows alterthe lighting in some areas, and reflections obscure features. (b) Looking downstream from the same bridge at approximately the samephoto scale, the different viewing angle enables the camera to readily captures variations in depth and substrate color and size.

36 Fluvial Remote Sensing for Science and Management

Second, sometimes ground based measurements aremore precise, but the broad areal coverage provided bythe imagery makes remotely sensed maps more globallyaccurate. Marcus and Fonstad (2008) make this case withthe example of field surveys of cross sections, which areprecise and very accurate at that one location. Yet thosecross sections are poor predictors of depth a short distanceaway from the survey location. The +15 cm precision ofthe remote sensing depths at all locations in the streamthus provides better global bathymetric accuracy than thehigh local accuracy of the cross section survey. Similararguments can be made for SAV, turbidity, and algae, allof which require relatively time consuming local surveysthat are representative of values at or near the sample site.

Third and finally, remote sensing maps are clearlythe most accurate alternative when they are the onlyalternative. This occurs when historic data are neededor where sites are inaccessible or too dangerous forfield surveys. Reconstruction of historic channel change,mapping of algae in roadless areas, and mapping of floodsand ice breakup all represent cases discussed above whereremote sensing might provide the only viable mappingalternative.

None of the points above are intended to imply thatremote sensing is always the most accurate or the preferredalternative. Rather, our point is that accuracy measuresmust be considered in the context of the managementneeds for local precision, global accuracy, and the rangeof alternatives available for gathering data.

2.14 Ethical considerations

Remote sensing of rivers can raise ethical issues nottypically encountered with classical ground-based data.Some of these issues are addressed in guidelines regard-ing professional conduct (ASPRS, 2007), data sharing asit relates to national security (Federal Geographic DataCommittee, 2005), and the use of remote sensing to bene-fit people of all backgrounds and economic levels (UnitedNations General Assembly, 1986). Many countries alsohave a history of legal precedent regarding privacy and theuse of remote imagery. We do not attempt to provide anin-depth discussion of these issues, but rather, raise someconcerns that are particularly relevant to river managers.

The use of remote sensing raises social equity issues,potentially providing users who have access to remotesensing technology an advantage in accessing the riverresource. This situation is nothing new; timber and min-eral companies, for example, have used remote sensing

for decades to acquire information they do not sharewith other parties. However, because rivers are often apublicly owned and managed resource and are connectedto up- and down-stream users, the issue of privileging aset of technologically savvy users over other residents ofthe basin is particularly problematic with rivers. If a smallnumber of users have information on the best fish habitat,gravel supplies, reaches with high quality water, or otherfactors, they can potentially exploit the river resource totheir own benefit, but not necessarily to the benefit ofothers living within the same watershed. Whether this isviewed as good or bad depends on the perspective of theindividuals and society.

One solution to the concern regarding private exploita-tion is to make all remote sensing results publicly available,an approach advocated by the United Nations (UnitedNations General Assembly, 1986). Yet allowing this degreeof access raises concerns regarding intellectual propertyand environmental protection. Particularly with high spa-tial resolution imagery, the maps may reveal the locationsof sensitive resources (e.g., habitat for an endangeredspecies), potentially jeopardising the very resource oneis trying to protect (Marcus and Fonstad, 2008). Man-agers may therefore want to consider aggregating the datato a coarser spatial resolution, as is done with the U.S.Census, where results are reported at the block level butnot the household level. Alternatively, one could makethe locations blurry or indistinct, obscuring the mapcoordinates to prevent resource exploitation. But theseapproaches invite the same criticism leveled before, wherea privileged elite have access to information that otherscannot share. Competing social goals thus leave managershaving to grapple with such issues on a case-by-case basis.

As with any technology, the potential for use andmisuse of river remote sensing is broad. The importanceof water to human society and the connectivity of riversmake the potential benefits and costs all the greater inrivers. It will largely be up to river managers, who have thebroadest access to information about rivers and their userpopulations, to contemplate and develop uses for remotesensing that most benefit rivers and the communities thatdepend upon them.

2.15 Why use optical remote sensing?

The reasons for using remote sensing for river manage-ment are similar to those reasons for considering remotesensing in any setting. Among the most important con-siderations are which features can be remotely mapped

2 Management Applications of Optical Remote Sensing in the Active River Channel 37

(discussed in the first part of the chapter), image avail-ability and cost; the spatial extent, coverage, scale andresolution of imagery; the need to collect repeat mea-surements; logistical constraints; the availability of digitaldata; the availability of multispectral data; and accuracy.In rivers, the potential to protect personnel from danger-ous field situations can be a major consideration as well.The tradeoffs among these various factors need to be care-fully attended to before a project commences: managerswho assume remote sensing will meet their mapping andmonitoring needs may be disappointed and upset.

From a management perspective, logistics and costconsiderations often drive the decision as to whether touse remote sensing. In turn, the spatial extent of thestudy and the number of measurements needed over timeare typically primary factors driving logistics and cost.Assuming that the variable in question can be mappedwith remote sensing, the general rule is that the larger thespatial coverage and the greater the number of measure-ments needed over time, the more attractive the remotesensing option becomes. For example, consider the com-mon scenario where information on channel depths isneeded for flood inundation modeling. If the model-ing involves only one relatively short river reach at onepoint in time, surveyors can often measure the neces-sary cross sections with relative ease in just one day. Incontrast, use of remote sensing would require acquisitionof remote imagery, specialised software, and expert per-sonnel, making this alternative cost prohibitive. However,if the intent is to collect cross sections throughout theentire river, then the cost of remote sensing becomesincreasingly reasonable relative to mobilising field crewsfor prolonged periods of time. Even in the case of onetime, local data collection, the remote sensing approachcould be cost effective if existing imagery were availablefor the appropriate date and trained personnel were athand. The one time application of remote sensing makeseven more sense in large or fast-flowing rivers where fielddata collection could place personnel at risk.

Moreover, the preceding example undersells the poten-tial utility of remote sensing. Field measurements typicallyprovide point or transect data. In contrast, remote sens-ing imagery provides continuous measurements of theentire channel over long distances, assuming the chan-nel can be seen by the sensor. In the context of floodinundation analysis, remote sensing does more than justsurvey isolated cross-sections as a field crew would; italso provides continuous depth measurements for theentire channel at the resolution of the imagery. Thisenables more detailed monitoring of depth changes and

more sophisticated modeling, which in turns enhancesthe potential to develop more accurate models of floodinundation.

Furthermore, unlike the field-based cross sections, theimagery is more versatile in its applications. Once theimagery has been acquired, depending on its spectralcoverage and spatial resolution, it can potentially be usedto map features such as riparian cover, macrophytes,sediment size, wood, or biotypes (glides, riffles, etc.). Thedigital image format also provides a permanent recordthat is useful for validation by independent parties andallows for the possibility of developing historic changemaps for variables that cannot be detected using presentapproaches. If this sounds farfetched, consider that priorto the year 2000, the only variables that remote sensingimagery was used for were flood extent, depth, turbidity,and riffle/pool delineation; this list that is now muchlonger (Marcus and Fonstad, 2008).

Yet many of the positive attributes listed above arehypothetical. Most river managers do not know if they willneed continuous coverage, monitoring of other variables,or more sophisticated modeling at a later date. What theydo know is that they need a specific measurement now.Furthermore, techniques for remote sensing of rivers arestill being developed; there are no standard approachesthat have been endorsed by regulatory agencies or incor-porated into readily available software packages. Finally,and perhaps most importantly, our personal experiencesuggests that projects requiring new image data mustincorporate significant flexibility. In general, one wishesto acquire imagery for the right place at the right time withthe right specifications (resolution, signal to noise ratio,etc.). We have considered ourselves lucky on occasionswhere we have achieved two of these three objectives.Because the costs are realised in the present, applicationof remote sensing for river management, especially if newdata must be acquired, can be risky business.

Remote sensing becomes a reasonable alternative toground-based field surveys when it can monitor or mapthe variables of interest and when it can do so on a costeffective or safer basis than ground-based techniques. Thecost-benefit ratio varies with the factors discussed aboveand with the risk averseness of the management agency.What is certain is that remote sensing will become anever more viable option in the future. Increasingly, rivermanagement goals extend beyond the active channel toinclude the floodplain and nearby landscapes, so thatmultiple users contribute to the purchase of image data.Moreover, the cost of imagery has generally decreasedover the past decade, the availability of software and

38 Fluvial Remote Sensing for Science and Management

personnel who can implement remote sensing algorithmsis growing, and methods are being more widely testedand accepted. Present trends suggest that the utility andcost effectiveness of remote sensing will only improve inthe coming years.

References

Alsdorf, D., Bates, P.D., Melack, J., Wilson, M., and Dunne, T.2007. Spatial and temporal complexity of the Amazon floodmeasured from space. Geophysical Research Letters 34(8):Article number L08402.

American Society for Photogrammetry and Remote Sensing(ASPRS). 2007. Code of Ethics. http://www.asprs.org/membership/certification/appendix_a.html. Accessed 30November 2007.

Aranuvachapun, S. and Walling, D.E 1988. Landsat-MSS radi-ance as a measure of suspended sediment in the Lower YellowRiver (Hwang Ho). Remote Sensing of Environment 25(2):145–165.

Armstrong, R.A 1993. Remote sensing of submerged vegeta-tion canopies for biomass estimation. International Journal ofRemote Sensing 14(3): 621–627.

Aspinall, R.J., Marcus, W.A., and Boardman, J.W. 2002. Consid-erations in collecting, processing, and analyzing high spatialresolution, hyperspectral data for environmental investiga-tions. Journal of Geographical Systems 4(1): 15–29.

Barton, I.J. and Bathols, J.M. 1989. Monitoring floods withAVHRR. Remote Sensing of Environment 30(1): 89–94.

Bates, P. 2004. Remote sensing and flood inundation modeling.Hydrologic Processes 18: 2593–2597.

Brakenridge, G.R., Nghiem, S.V., Anderson, E., and Chien, S.2005. Space-based measurement of river runoff. EOS: Trans-actions of the American Geophysical Union 86(19): 185–192.

Brasington, J., Langham, J., and Rumsby, B. 2003. Methodolog-ical sensitivity of morphometric estimates of coarse fluvialsediment transport. Geomorphology 53(3/4): 299–316.

Bustamante, J., Pacios, F., Diaz-Delgado, R., and Aragones,D. 2009. Predictive models of turbidity and water depth inthe Donana marshes using Landsat TM and ETM+ images.Journal of Environmental Management 90(7): 2219–2225.

Carbonneau, P.E. 2005. The threshold effect of image resolutionon image-based automated grain size mapping in fluvialenvironments. Earth Surface Processes and Landforms 30(13):1687–1693.

Carbonneau, P.E., Lane, S.N., and Bergeron, N.E. 2004.Catchment-scale mapping of surface grain size in gravelbed rivers using airborne digital imagery. Water ResourcesResearch 40(7): Article No. W07202.

Carbonneau, P.E., Bergeron, N.E., and Lane, S.N. 2005. Auto-mated grain size measurements from airborne remote sensing

for long profile measurements of fluvial grain sizes. WaterResources Research 41(11): Article No. W11426.

Carbonneau, P.E., Fonstad, M.A., Marcus, W.A., and Dugdale, S.2011. Making riverscapes real. Geomorphology. 137(1): 74–86.

Davies-Colley, R.J. and Smith, D.G. 2001. Turbidity, suspendedsediment, and water clarity: A review. Journal of the AmericanWater Resources Association 37(5): 1085–1101.

Dekker, A.G., Hoogenboom, H.J., Goddijn, L.M., andMalthus, T.J.M. 1997. The relationship betweeninherent optical properties and reflectance spectrain turbid inland waters. Remote Sensing Reviews 15:59–74.

Dieck, J.J and Robinson, L.R 2004. Techniques and MethodsBook 2, Collection of Environmental Data, Section A, Biolog-ical Science, Chapter 1, General classification handbook forfloodplain vegetation in large river systems: U.S. GeologicalSurvey, Techniques and Methods 2 A–1, 52 p.

Federal Geographic Data Committee. 2005. Final Guidelines forProviding Appropriate Access to Geospatial Data in Responseto Security Concerns. Federal Geographic Data Committee:Reston, VA. http://www.fas.org/sgp/othergov/fgdc0605.pdf.Accessed 30 November 2007.

Fonstad, M.A. and Marcus, W.A. 2005. Remote sensing of streamdepths with hydraulically assisted bathymetry (HAB) models.Geomorphology 72(1–4): 107–120.

Feurer, D., Bailly, J.S., Puech, C., LeCoarer, Y., and Viau, A. 2008.Very high resolution mapping of river immersed topographyby remote sensing. Progress In Physical Geography 32(4):1–17.

Fox, M. and Bolton, S. 2007. A regional and geomorphicreference for quantities and volumes of instream wood inunmanaged forested basins of Washington State. North Amer-ican Journal of Fisheries Management 27(1): 342–359.

Gilvear, D.J. and Bryant, R. 2003. Analysis of aerial photographyand other remotely sensed data. In Tools in Fluvial Geomor-phology, Kondolf, G.M. and Piegay, H. (eds). Wiley: London,133–168.

Gilvear, D.J., Hunter, P., and Higgins, T. 2007. An experi-mental approach to the measurement of the effects of waterdepth and substrate on optical and near infra-red reflectance:a field-based assessment of the feasibility of mapping sub-merged instream habitat. International Journal of RemoteSensing 28(10): 2241–2256.

Goovaerts, P. 2002. Geostatistical incorporation of spatial coor-dinates into supervised classification of hyperspectral data.Journal of Geographical Systems 4(1): 99–111.

Graham, D.J., Reid, I., and Rice, S.P. 2005. Automated sizingof coarse-grained sediments: image-processing procedures.Mathematical Geology 37(1): 1–28.

Graham, D.J., Rice, S.P., and Reid, I. 2005. A transferable methodfor the automated grain sizing of river gravels. Water ResourcesResearch 41(W07020): doi:10.1029/2004WR003868.

2 Management Applications of Optical Remote Sensing in the Active River Channel 39

Han, Z., Jin, Y-Q., and Yun, C-X. 2006. Suspended sedimentconcentrations in the Yangtze River estuary retrieved fromthe CMODIS data. International Journal of Remote Sensing27(19): 4329–4336.

Hestir, E.L., Khanna, S., Andrew, M.E., Santos, M.J., Viers, J.H.,Greenberg, J.A., Rajapakse, S.S., and Ustin, S.L. 2008. Iden-tification of invasive vegetation using hyperspectral remotesensing in the California Delta ecosystem. Remote Sensing ofEnvironment 112: 4034–4047.

Hick, P., Jernakoff, P., and Hosja, W. 1998. Algal bloom researchusing airborne remotely sensed data: Comparison of highspectral resolution and broad bandwidth CASI data withfield measurements in the Swan River in Western Australia.Geocarto International 13(3): 19–28.

Hodge, R.A., Brasington, J., and Richards, K.S. 2009. Charac-terisation of grain-scale fluvial morphology using TLS. EarthSurface Processes and Landforms 34: 954–968.

Hoogenboom, H.J., Volten, H., Schreurs, R., and de Haan,J.F. 1997. Angular scattering functions of algae and silt: ananalysis of backscattering to scattering fraction. Ackleson,S.G. and Frouin, R.J. (eds). Ocean Optics XIII, SPIE vol. 2963:392–400.

Hughes, M.L., McDowell, P.F., and Marcus, W.A. 2006. Accuracyassessment of georectified aerial photographs: Implications formeasuring lateral channel movement in a GIS. Geomorphology74: 1–16.

Hunter, P.D., Tyler, A.N., Gilvear, D.J., and Willby, N.J. 2008.The spatial dynamics of vertical migration by Microcystisaeruginosa in a eutrophic shallow lake: A case study usinghigh spatial resolution time-series airborne remote sensing.Limnology and Oceanography 53: 2391–2406.

Ip, F., Dohm, J.M., Baker, V.R., Doggett, T., Davies, A.G.,Castano, B., Chien, S., Cichy, B., Greeley, R., and Sherwood, R.2006. Flood detection and monitoring with the AutonomousSciencecraft Experiment onboard EO-1. Remote Sensing ofEnvironment 101(4): 463–481.

Jensen, J.R. 2007. Remote Sensing of the Environment: An EarthResource Perspective, 2nd ed., Upper Saddle River, NJ: PrenticeHall, 592 pages.

Jochem, K., Daniel, H., Muhar, S., Gerhard, M., and Preis, S.2007. The use of large wood in stream restoration: experiencesfrom 50 projects in Germany and Austria. Journal of AppliedEcology 44(6): 1145–1155.

Jordan, D.C. and Fonstad, M.A. 2005. Two-dimensional map-ping of river bathymetry and power using aerial photographyand GIS on the Brazos River, Texas. Geocarto 20(3): 1–8.

Kaab, A. and Prowse, T. 2011. Cold-regions river flow observedfrom space. Geophysical Research Letters 38(L08403): doi:10.1029/2011gl047022.

Kishi, S., Song, X., and Li, J. 2001. Flood detection in Changjiang1998 from Landsat-TM data. Space Technology 20(3):99–105.

LaCapra, V.C., Melack, J.M., Gastil, M., and Valeriano, D. 1996.Remote sensing of foliar chemistry of inundated rice withimaging spectrometry. Remote Sensing of Environment 55(1):50–58.

Lane, S.N., Westaway, R.M., and Hicks, D.M. 2003. Estimationof erosion and deposition volumes in a large, gravel bed,braided river using synoptic remote sensing. Earth SurfaceProcesses and Landforms 28(3): 249–271.

Lane, S.N., Widdison, P.E., Thomas, R.E., Ashworth, P.J., Best,J.L., Lunt, I.A., Sambrook Smith, G.H., and Simpson, C.J.2010. Quantification of braided river channel change usingarchival digital image analysis. Earth Surface Processes andLandforms 35: 971–985

Legleiter, C.J. 2003. Spectrally driven classification of high spatialresolution, hyperspectral imagery: A tool for mapping in-stream habitat. Environmental Management 32(3): 399–411.

Legleiter, C.J. and Goodchild, M.F. 2005. Alternative repre-sentations of in-stream habitat: classification using remotelysensed data, hydraulic modeling, and fuzzy logic. InternationalJournal of Geographical Information Science 19(1): 29–50.

Legleiter, C.J., Marcus, W.A., and Lawrence, R. 2002. Effects ofsensor resolution on mapping in-stream habitats. Photogram-metric Engineering and Remote Sensing 68(8): 801–807.

Legleiter, C.J., Kinzel, P.J., and Overstreet, B.T. 2011. Evaluatingthe potential for remote bathymetric mapping of a turbid,sand-bed river: 2. Application to hyperspectral image datafrom the Platte River. Water Resources Research 47(W09532):doi: 10.1029/2011wr010592.

Legleiter, C.J., Roberts, D.A., and Lawrence, R.L. 2009. Spec-trally based remote sensing of river bathymetry. Earth SurfaceProcesses and Landforms 34: 1039–1059.

Legleiter, C.J. and Roberts, D.A. 2005. Effects of channel mor-phology and sensor spatial resolution on image-derived depthestimates. Remote Sensing of Environment 95: 231–247.

Legleiter, C.J., Roberts, D.A., Marcus, W.A., and Fonstad, M.A.2004. Passive remote sensing of river channel morphologyand in-stream habitat: physical basis and feasibility. RemoteSensing of Environment 93(4): 493–510.

Lejot, J., Delacourt, C., Piegay, H., Fournier, T., Tremelo, M.L.,and Alleman, P. 2007. Very high spatial resolution imageryfor channel bathymetry and topography from an unmannedmapping controlled platform. Earth Surface Processes andLandforms 32(11): 1705–1725.

Marcus, W.A. 2002. Mapping of stream microhabitats with highspatial resolution hyperspectral imagery. Journal of Geograph-ical Systems 4(1): 113–126.

Marcus, W.A. 2012. Remote sensing of the hydraulic envi-ronment in gravel-bed rivers. In M. Church, P. Biron, A.Roy (eds.), Gravel-bed rivers: Processes, tools, environments.Chichester, John Wiley and Sons. p. 261–285.

Marcus, W.A., Aspinall, R., Boardman, J., Crabtree, R., Despain,D., Halligan, K., and Minshall, W. 2001. Validation of

40 Fluvial Remote Sensing for Science and Management

High-Resolution Hyperspectral Data for Stream and RiparianHabitat Analysis. Annual Report (Phase 2) to NASA EOCAPProgram, Stennis Space Flight Center, Mississippi, 127 p. plusfigures.

Marcus, W.A and Fonstad, M.A. 2008. Optical remote mappingof rivers at sub-meter resolutions and watershed extents. EarthSurface Processes and Landforms 33: 4–24.

Marcus, W.A., Legleiter, C.J., Aspinall, R.J., Boardman, J.W.,and Crabtree, R.L. 2003. High spatial resolution, hyperspec-tral (HSRH) mapping of in-stream habitats, depths, andwoody debris in mountain streams. Geomorphology 55(1–4):363–380.

Maruca, S. and Jacquez, G.M. 2002. Area-based tests for associa-tion between spatial patterns. Journal of Geographical Systems4(1): 69–83.

Mertes, L.A.K. 2002. Remote sensing of riverine landscapes.Freshwater Biology 47: 799–816.

Mertes, L.A.K., Dekker, A., Brakenridge, G.R., Birkett, C., andLe’tourneau, G. 2002. Rivers and lakes. In S.L. Ustin (ed.), Nat-ural Resources and Environment, Manual of Remote Sensing,John Wiley and Sons, New York.

Mertes, L.A.K., Smith, M.O., and Adams, J.B. 1993. Estimatingsuspended sediment concentrations in surface waters of theAmazon River wetlands from Landsat Images. Remote Sensingof Environment 43: 281–301.

Milan, D., Heritage, G., Large, A., and Entwistle, N. 2010.Mapping hydraulic biotopes using terrestrial laser scan data ofwater surface properties Earth Surface Processes and Landforms35(8): 918–931.

Morse, B. and Hicks, F. 2005. Advances in river ice hydrology1999–2003. Hydrological Processes 19(1): 247–263.

Pavelsky, T.M. and Smith, L.C. 2004. Spatial and temporalpatterns in Arctic river ice breakup observed with MODIS andAVHRR time series. Remote Sensing of Environment 93(3):328–338.

Pavelsky, T.M. and Smith, L.C. 2009. Remote sensing ofsuspended sediment concentration, flow velocity, and lakerecharge in the Peace-Athabasca Delta, Canada. Water Re-sources Research 45(W11417): doi:10.1029/2008WR007424.

Penuelas, J., Gamon, J.A., Griffin, K.L., and Field, C.B. 1993.Assessing community type, plant biomass, pigment com-position and photosynthetic efficiency of aquatic vegetationfrom spectral reflectance. Remote Sensing of Environment 46:110–118.

Penuelas, J., Filella, I., and Gamon, J.A. 1997. Assessing photo-synthetic radiation-use efficiency of emergent aquatic vegeta-tion from spectral reflectance. Aquatic Botany 58: 307–315.

Rainey, M.P., Tyler, A.N., Gilvear, D.J., Bryant, R.G., andMcDonald, P. 2003. Mapping intertidal estuarine sedimentgrain size distributions through airborne remote sensing.Remote Sensing of Environment 86(4): 480–490.

Quibell, G. 1991. The effect of suspended sediment on reflectancefrom freshwater algae. International Journal of Remote Sensing12: 177–182.

San Miguel-Ayanz, J., Vogt, J., De Roo, A., and Schmuck, G.2000. Natural hazards monitoring: Forest fires, droughts,and floods-The example of European pilot projects. SurveyGeophysics 21: 291–305.

Schumann, G., Di Baldassarre, G., and Bates, P.D. 2009. The util-ity of spaceborne radar to render flood inundation maps basedon multialgorithm ensembles (Part 2). IEEE Transactions onGeoscience and Remote Sensing 47(8): 2801–2807.

Sheng, Y., Gong, P., and Xiao, Q. 2001. Quantitative dynamicflood monitoring with NOAA AVRR. International Journal ofRemote Sensing 22: 1709–1724.

Shibayama, M., Kanda, K., and Sugahara, K. 2007. Water tur-bidity estimation using a hand-held spectropolarimeter todetermine surface reflection polarization in visible, near andshort-wave infrared bands. International Journal of RemoteSensing 28(16): 3747–3755.

Silva, T.S.F., Costa, M.P.F., Melack, J.M., and Novo, E.M.L.M.2008. Remote sensing of aquatic vegetation: theory andapplications. Environmental Monitoring and Assessment 140:131–45.

Smith, L.C. 1997. Satellite remote sensing of river inundationarea, stage, and discharge: A review. Hydrologic Processes 11:1427–1439.

Teodoro, A.C., Goncalves, H., and Veloso-Gomes, F. 2008.Statistical techniques for correlating total suspended matterconcentration with seawater reflectance using multispectralsatellite data. Journal of Coastal Research 24(4) SUPPL. (200807 01): 40–49.

Underwood, E., Ustin, S., and DiPietro, D. 2003. Mappingnonnative plants using hyperspectral imagery. Remote Sensingof Environment 86: 150–161.

United Nations General Assembly. 1986. Principles Relatingto Remote Sensing of the Earth from Space, A/RES/41/65.http://www.un.org/documents/ga/res/41/a41r065.htm.Accessed 30 November 2007.

Wang, J.J., Lu, X.X., Liew, S.C., and Zhou, Y. 2009. Retrieval ofsuspended sediment concentrations in large turbid rivers usingLandsat ETM+: an example from the Yangtze River, China.Earth Surface Processes and Landforms 34(8): 1082–1092.

Westaway, R., Lane, S.N., and Hicks, D.M. 2003. Remote surveyof large-scale braided rivers using digital photogrammetryand image analysis. International Journal of Remote Sensing24: 795–816.

Westaway, R., Lane, S.N., and Hicks, D.M. 2001. Airborneremote sensing of clear water, shallow, gravel-bed rivers usingdigital photogrammetry and image analysis. PhotogrammetricEngineering and Remote Sensing 67: 1271–81.

Westaway, R., Lane, S.N., and Hicks, D.M. 2000. Developmentof an automated correction procedure for digital photogram-metry for the study of wide, shallow gravel-bed rivers. EarthSurface Processes and Landforms 25: 200–26.

Williams, D.J., Rybicki, N.B., Lombana, A.V., O’Brien, T.M., andGomez, R.B. 2003. Preliminary investigation of submergedaquatic vegetation mapping using hyperspectral remote sens-ing. Environmental Monitoring and Assessment 81: 383–392.

2 Management Applications of Optical Remote Sensing in the Active River Channel 41

Yamagata, Y., Wiegand, C., Akiyama, T., and Shibayama, M.1988. Water turbidity and perpendicular vegetation indicesfor paddy rice flood damage analyses. Remote Sensing ofEnvironment 26(3): 241–251.

Yuan, L. and Zhang, L-Q. 2007. The spectral responses of asubmerged plant Vallisneria spiralis with varying biomassusing spectroradiometer. Hydrobiologia 579: 291–299.

Zhang, X. 1998. On the estimation of biomass of submergedvegetation using Landsat thematic mapper (TM) imagery:A case study of the Honghu Lake, PR China. InternationalJournal of Remote Sensing 19(1): 11–20.