DSpace at VNU: Correlating mass physical properties with ALOS reflectance spectra for intertidal sediments from the Ba Lat Estuary (northern Vietnam): An exploratory laboratory study

Correlating mass physical properties with ALOS reflectance spectra for intertidal sediments from the Ba Lat Estuary northern Vietnam: an exploratory laboratory study Nguyen Thi Ngoc&Kats

Correlating mass physical properties with ALOS reflectance

spectra for intertidal sediments from the Ba Lat Estuary

(northern Vietnam): an exploratory laboratory study

Nguyen Thi Ngoc&Katsuaki Koike&Nguyen Tai Tue

Received: 31 October 2012 / Accepted: 27 March 2013

# Springer-Verlag Berlin Heidelberg 2013

Abstract Characterization of the sediment composition of

tidal flats and monitoring of their spatiotemporal changes has

become an important part of the sustainable management

of coastal environments To accurately classify sediments

through remote sensing, a comprehensive understanding of

sediment reflectance spectra is indispensable The present

laboratory-based study explores the performance of the high

spatial resolution (10×10 m) Advanced Land Observing

Sat-ellite (ALOS) launched in 2006 Relationships between

re-flectance spectra (bands 1 to 4) and four typical mass physical

properties were investigated under wet and dry experimental

conditions for intertidal sediments sampled near the Ba Lat

Estuary in northern Vietnam Reflectance in the near-infrared

region corresponding to ALOS band 4 (0.76–0.89 μm) was

found (1) to have a strong negative correlation with sand

content (dry wt%) under both wet and dry conditions (linear

correlation coefficient r=–0.7859 and –0.8094, respectively),

(2) to increase with decreasing relative water content (%) in a

given sediment type (r=–0.7748 to –0.9367 for mud, sandy mud, muddy sand, and sand), (3) to have a positive correlation with organic matter content (r=0.7610 and 0.6460 under wet and dry conditions for contents >0.20 dry wt%), and (4) to be insignificantly correlated with mineral composition assessed

in terms of contents (wt%) of quartz, clay minerals, and mica group minerals Positive relationships between reflectance and water content for the pooled data of all sediment types (r=0.6395) or organic matter content contrast with previous findings, and can be attributed to close interrelationships between these properties and the predominance of sand con-tent as controlling factor of reflectance This study clarifies that ALOS band 4 provides the most useful imagery for intertidal monitoring because its reflectance, as simulated using the laboratory data, shows the strongest correlation with sand content In a next step, these experimental findings should be verified by identifying the reflectance relationships

at satellite image scales, and also considering the effects of other tidal flat features on reflectance, such as microtopography and biological surface characteristics

Introduction Estuaries commonly act as traps for sediments, nutrients, and chemical contaminants (e.g., Dalrymple et al 1992; Eisma

1997; Roy et al 2001; Swaney et al.2008; Taljaard et al

2009) From an ecological point of view, tidal flats lining estuaries play a crucial role for estuarine ecosystems because they provide nesting, feeding, and spawning grounds for invertebrate animals, fish, and birds (Dyer et al.2000) The sediment composition of tidal flats is particularly important because of its strong correlation with species distribution patterns and the structure of animal communities—e.g., bi-valves and shellfish—which have been shown to vary with the

N T Ngoc

Graduate School of Science and Technology,

Kumamoto University, 2-39-1 Kurokami,

Kumamoto 860-8555, Japan

N T Ngoc

Faculty of Geology, Hanoi University of Science,

334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam

K Koike ( *)

Department of Urban Management, Graduate School

of Engineering, Kyoto University, Katsura C1-2-215,

Kyoto 615-8540, Japan

e-mail: koike.katsuaki.5x@kyoto-u.ac.jp

N T Tue

Graduate School of Science and Engineering,

Ehime University, 2-5 Bunkyo-cho,

Matsuyama, Ehime 790-8577, Japan

Geo-Mar Lett

DOI 10.1007/s00367-013-0327-1

type of bottom sediment (e.g., Ysebaert and Herman2002).

Furthermore, sediment distribution can also be a key factor for

understanding hydrodynamic conditions, biogeochemical

cy-cles, and morphological change (e.g., Dalrymple et al.1992;

Ryu et al.2004; Taljaard et al.2009) Despite these important

functions, tidal flats in many parts of the world have been

degraded or destroyed by overexploitation, reclamation,

in-dustrialization, pollution, but also by detrimental effects of

climate change (Dyer et al.2000) The mass physical

sedi-ment characteristics of tidal flats and the monitoring of

their spatiotemporal change has therefore become an

indis-pensable part of the sustainable management of such coastal

environments

Remote sensing is one of the most effective monitoring

tools because of its relatively low cost, rapid information

updating, synoptic view, and repetitive observation capability

In this context, a number of remote sensing studies have been

devoted to clarifying the sediment composition of intertidal

environments (e.g., Yates et al.1993; Rainey et al.2003; Ryu

et al.2004; Sørensen et al.2006; van der Wal and Herman

2007; Choi et al.2011) In addition, geomorphological

fea-tures such as the surface roughness, and tidal channel density

of tidal flats (e.g., Bartholdy and Folving1986; Doerffer and

Murphy1989; Tyler et al.1996; van der Wal et al.2005; Ryu

et al.2008; Choi et al 2011) have been characterized with

reasonable accuracy These studies essentially examined the

usefulness of image classification techniques for monitoring

tidal flats by integrating mainly field spectra and satellite

imagery By factor analysis of Landsat TM (Thematic

Map-per) data, Doerffer and Murphy (1989) revealed that three

factors—namely, topography, water content, and surface

tem-perature (in this order)—have the greatest effect on the

reflec-tance spectra of main sediment types in the Wadden Sea Yates

et al (1993) found that three image classification techniques

(maximum likelihood classification, regression modeling, and

spectral mixture modeling) can distinguish mudflats more

precisely than sandy flats using Landsat 5 TM bands 1–3 data

The spectral unmixing technique can also be appropriate for

classifying sediment types into wet mud, dry mud, wet sand,

and dry sand from Airborne Thematic Mapper (ATM) images

(Rainey et al.2003), and for classification of heterogeneous

tidal flats (van der Wal and Herman2007)

In other approaches, the abundance of microphytobenthos

on tidal flats has been assessed by means of the absorption

spectra of chlorophylla at about 0.675 μm using laboratory

and/or field data in combination with remotely sensed data

(e.g., Doerffer and Murphy 1989; Paterson et al 1998;

Deronde et al.2006; Kromkamp et al 2006; Murphy et al

2008; Adam et al 2011) These studies made use of

hyperspectral and hyperspatial data such as aerial

photogra-phy with a ground resolution of 0.4 m (Doerffer and Murphotogra-phy

1989) and a HyMap TM Scanner at 4×4 m pixel resolution

(Deronde et al.2006)

In addition to satellite image analyses, fundamental stud-ies on the reflectance spectra of intertidal sediments under laboratory conditions are indispensable as ground-truthing for the accurate classification of sediment composition Inter-tidal sediment consists of three main components: mineral grains, organic matter, and interstitial water Together with the geometric size of the mineral grains, these components strongly affect the reflectance characteristics of the sediment (Asrar1989; Rainey et al.2000) As shown by Hunt (1977) and Clark (1995,1999), the wavelength and depth of absorp-tion of reflectance also varies over the visible to infrared part

of the spectrum as a function of mineral class (e.g., silicates, oxides and hydroxides, carbonates, and borates) Furthermore, organic carbon content influences the reflectance spectra in the visible to near-infrared region (Stoner and Baumgardner

1981; Sinha1987; Korsman et al.1999)

Several recent studies have considered the effect of grain size and water content on the reflectance of intertidal sedi-ments: the reflectance was measured in the field and/or laboratory, or calculated from satellite images (e.g., Bryant

et al.1996; Rainey et al.2000; Ryu et al.2004; van der Wal and Herman2007; Small et al 2009) Through in situ and laboratory reflectance experiments, Rainey et al (2000) demonstrated that reflectance in the simulated ATM band

9 is positively correlated with sand content, but negatively with interstitial water content Ryu et al (2004) revealed that the Landsat ETM bands 4 and 5 are effective for detecting grain-size composition and surface water content, respec-tively From relationships between grain size and water content of sediments and laboratory reflectance, Small et

al (2009) suggested that grain size and water content may

be mapped by duplicating the laboratory reflectance and hyperspectral imagery Most of these earlier studies, how-ever, focused on individual sediment parameters, the effects

of interrelationships among them being rarely considered, especially in the case of remotely sensed satellite data

In the past, satellite-derived reflectance spectra were mostly based on Landsat data (Landsat TM and ETM+), which had (and still have) a moderate spatial resolution (30×30 m; e.g., Bartholdy and Folving1986; Doerffer and Murphy1989; Yates et al.1993) Due to the relatively low resolution, the mapping accuracy was also low, possibly because of the effect of variable water contents (Ryu et al

2004), and mixtures of reflectance spectra caused by foot-print overlaps between adjacent sedimentary facies (Sørensen et al 2006) This shortcoming can today be reduced by using satellite imagery from the ALOS (Ad-vanced Land Observing Satellite) AVNIR-2 (Ad(Ad-vanced Vis-ible and Near Infrared Radiometer type 2), launched in

2006 The higher spatial resolution (10×10 m) of this satel-lite would greatly improve the precision of spatial mapping However, applications for the monitoring of tidal flats are currently still very limited

Within this general context, the Ba Lat Estuary (BLE) in

northern Vietnam was selected as a case study site, mostly

because the BLE was proclaimed a Ramsar site in 1989

(Ramsar Convention Bureau1997) and subsequently given

the status of a nature reserve in 1995, owing to its rich

biodiversity as a wetland ecosystem Nevertheless, the tidal

flats of the BLE have in recent decades been negatively

affected by both natural processes associated with coastal

erosion and deposition (van Maren 2007), and human

in-terventions in the form of coastal development, especially

land reclamation, shrimp farming, and the exploitation of

intertidal benthic organisms Monitoring the tidal flats of the

BLE is therefore indispensable for conserving its natural

resources and the environment in general

Based on the above background, this article aims to

clarify comprehensive characteristics of reflectance spectra

of intertidal sediments under laboratory conditions with

respect to sediment grain-size and mineral composition, as

well as water and organic matter contents, as fundamental

parameters in remote sensing studies The results were then used to specify which of the ALOS band(s) were the most suitable for obtaining good correlations between the reflec-tance data and particular sediment types Such specifications can contribute substantially to the mapping of sediment types by means of satellite imagery

Study area The Red River originates in the Yunnan highlands of China where the widespread occurrence of red laterite soils give local river waters its reddish-brown color It has a length of 1,200 km and a catchment area of 160,000 km2(Milliman et

al.1995) The Red River flows southeastward through large cities and industrial centers (e.g., Phu Tho, Ha Noi, Thai Binh, and Nam Dinh) before draining into the Gulf of Tokin The main discharge period coincides with the summer monsoon season The sediment load of the Red River (approx 160×106

a

b

c

d

Fig 1 Location of the study area showing the Ba Lat Estuary in northern Vietnam, and the positions of the 101 sampling stations on the tidal flat Photos are examples of a a shoal, b a shrimp pond, c a mangrove forest, and d a clam farm

Geo-Mar Lett

metric tons per year) is ranked 9th worldwide Siltation in the

distributaries of the lower delta region resulting from the high

sediment load in combination with tidal pumping, for

exam-ple, is a constant threat to the waterway leading to the harbor

of Haiphong (Lefebvre et al.2012)

The BLE, which is the largest estuary in the Red River

system of northern Vietnam (Fig.1), is shaped roughly like a

broad arrow jutting out into the sea It acts as a gateway for

sediment currently transported from the land to the sea at a rate

of 31×106m3annually This volume is estimated to be

equiv-alent to 38 % of the total sediment transported by the Red River

system per year (Duc et al.2007) The water discharge of the

Red River Delta is characterized by seasonal variations Daily

discharges are low in the dry season, ranging from 1,500 to less

than 1,000 m3/s at the hydrological station of Son Tay, which is

located on the Red River at the entrance of the delta High

discharges occur in the rainy season with an average of

14,000 m3/s, a maximum value of 33,600 m3/s having been

recorded in August 1971 The mean concentration of sediment

varies from 0.2 kg/m3in the dry season to 1.4 kg/m3in the

rainy season, reaching 7 kg/m3 during floods (van Maren

2005) The tides in the BLE are diurnal and classified as

mesotidal, averaging at about 2.3 m and ranging from 0.1 to

3.7 m (Marine Hydro-meteorological Center2009) The tidal

influence, in particular salinity intrusion, is conspicuous in the

Red River Delta in the dry season Seawater intrudes up to

20 km landward from the Ba Lat mouth (Luu et al.2010)

For the purpose of this study, a tidal zone within 2 m

water depth at high spring tides was selected (Fig.1) There

are numerous islets and shoals, the latter composed mainly

of well-sorted and relatively homogenous fine sand (Duc et

al.2007) Mangroves grow naturally in the northern part of

Con Lu and along tidal creeks, new mangroves having been

planted in the muddy flats of the southern part of Con Lu

since 1997 The mangroves on the Con Vanh and Con Ngan

islets were converted to shrimp ponds starting in the 1980s

(Fig.1) The tidal flats in the northeastern and southwestern

parts of the estuary are occupied by clam farms (Meretrix

lyrata) Rapid deposition has occurred along the BLE

coast-line in the last 50 years, extending the land several

kilome-ters downstream and offshore (van Maren2007)

Materials and methods

Field sampling and laboratory treatment

Duplicates of about 200 g of intertidal sediment were

col-lected at 101 stations (Fig 1) from the uppermost layer

(<1 cm) during a neap tide in June 2010, when the tidal

flats were completely exposed The sample locations were

selected to cover a wide grain-size range Each sample was

collected using a stainless steel spade To avoid loss of

water, the samples were sealed tightly in double polyethyl-ene bags and then stored in iceboxes After transport to the laboratory, they were kept frozen at–20 °C until analysis Before defrosting, one set of samples was split into sub-samples for the analyses of grain size, mineral composition, water content, and reflectance measurements After defrosting

at room temperature, each subsample was prepared for the determination of its designated parameter as described in detail below

Grain size, mineral composition, and organic matter content Grain-size distributions of the sediment were measured by sieving in the case of the sand fraction (0.063 to 2 mm) and pipette analysis in the case of the mud fraction (<0.063 mm), following the procedure of Tue et al (2012) For this purpose, sand-rich subsamples (20 g each) were dried, weighed, and then wet-sieved through a 0.063 mm sieve to separate the sand and mud fractions The sand fractions were then dry-sieved at one-phi sieve intervals (0.50, 0.25, 0.125, and 0.063 mm), each size fraction being subsequently weighed The

previous-ly collected mud fractions were poured into a glass cylinder along with a constant volume of distilled water, the suspension being thoroughly stirred before analysis Using a pipette,

25 ml of suspension were then extracted at a constant depth after 40 s, 16 minutes, 59 minutes, and 15 h, the time

i n t e r v a l s— a f t e r c o r r e c t i o n f o r t e m p e r a t u r e fluctuations—corresponding to the four size ranges of 63–

10, 10–5, 5–1, and <1 μm The extracted suspensions were evaporated at 105 °C, weighed, and expressed as percentages

of the total sample Sediment type classification followed the ternary scheme of Folk (1968)

To determine mineral contents and abundances, X-ray diffraction (XRD) analyses were carried out on 57 represen-tative samples selected on the basis of the grain size and color of each sediment type, but also considering the dis-tance between sampling points The dried samples were pulverized and prepared for XRD analysis following stan-dard procedures Individual minerals and their relative abun-dances were determined from the diffraction patterns using the “quantitative analysis” software program developed by ICDD (International Centre for Diffraction Data)

Organic matter contents were estimated on the basis of organic carbon measurements carried out by Tue et al (2012) on samples recovered from the same locations as those of the present study, using a multiplication factor of 1.724 (Nelson and Sommers1996)

Reflectance spectra and water content Reflectance spectra of water-saturated subsamples were in each case measured in a dark laboratory to prevent contami-nation from ambient scattered light For this purpose, each wet

sample was placed in a 2-cm-deep glass dish of 7 cm diameter.

Measurements were made at ten randomly chosen points by

rotating the dish, the average value being used as the

repre-sentative reflectance spectrum value for the sample The

irra-diated surface area (footprint) amounted to 7.07 cm2in each

case, which corresponds to the area of a circle 3 cm in

diameter A FieldSpec®3 spectrometer (Analytical Spectral

Devices Inc.) was used for the measurements over a

wave-length range from visible to short-wave infrared (0.35 to

2.50 μm) Original reflectance was calibrated to absolute

reflectance using the data from a white reference panel

After the measurements were completed, each vessel was

dried in an oven at 105 °C for 24 h and re-weighed This drying

procedure served the purpose of gradually evaporating surface

and pore waters from the samples in order to accurately

eval-uate the effect of varying water contents on the reflectance

spectra Relative water contents were calculated from the

dif-ference in weights between the wet and dry samples, being

expressed as percentages in terms of the mass of water relative

to the mass of dry solids (cf Flemming and Delafontaine

2000) After drying, the reflectance spectra of the dried samples

were also determined The dry condition is analogous to that of

eolian sand dunes and upper tidal flats exposed for extended

periods in the course of spring–neap cycles Taken together, the

reflectance measurements can be used to distinguish between

wet sand, dry sand, wet mud, and dry mud in the study area

Using the reflectance spectra determined in the laboratory,

the reflectance corresponding to the ALOS AVNIR-2 bands 1,

2, 3, and 4 was calculated by averaging these values within

each of the wavelength ranges, i.e., 0.42–0.50, 0.52–0.60,

0.61–0.69, and 0.76–0.89 μm, respectively

Statistical analyses

Strength of relationships between each mass physical property

and its reflectance was expressed by correlation coefficients

(r) The statistical significance of r at the level <0.05 was

confirmed by the one-tailed test Multivariate regression

anal-ysis was then carried out to consider the combined effects of

all three mass physical properties on the reflectance The

SPSS software package was used for all the statistical

analyses

Results Characterization of intertidal sediments The samples were classified into four sediment types: mud (M), sandy mud (sM), muddy sand (mS), and sand (S), depending on sand contents that ranged from 1 to 99 % Sand contents (Cs) averaged 6 % for M, 26 % for sM, 75 % for mS, and 97 % for S (Table1and Fig.2a) The mean and standard deviation of the sand contents for all samples was 64±37 % The mean grain sizes of the four sediment types were 0.015 mm (M), 0.037 mm (sM), 0.101 mm (mS), and 0.154 mm (S)

The water content (Cw), which showed a strong neg-ative correlation with the sand content (Fig 3a), varied from 2 to 94 % with a mean of 37 ± 19 % for all samples The mean water contents of the four sediment types were

64 % for M, 53 % for sM, 29 % for mS, and 24 % for S (Table 1 and Fig 2b) The organic matter content (Com) varied with the sediment type in the range of 0.01 to 2.78 %, with a mean for all samples (total mean) of 0.78

± 0.88 % Similar to the water content, the organic matter content increases as the sand content decreases (Fig.3b), the maximum being 1.54 % in the M type sediment, the minimum 0.20 % in the S type sediment (Table 1 and Fig 2c)

Quartz was common in all samples and was identified as the dominant mineral, its content varying from 21–92 % with a total mean of 71±18 % Other major minerals iden-tified in more than a quarter of the samples were clay minerals, including illite (1–56 %, with a total mean of 11

±13 % inn=46 samples), kaolinite (1–13 %, 2±3 %, n=20), and montmorillonite (1–16 %, 5±4 %, n=16), mica group minerals, including muscovite (1–47 %, 8±8 %, n=37) and phlogopite (1–42 %, 10±12 %, n=18), and feldspars such as albite (2–16 %, 5±4 %, n=16) Minor minerals (<2 %) found in a few samples were orthoclase, anorthite, andesine, chamosite, gibbsite, tremolite, pyrophyllite, and saponite Feldspars were absent in the M type samples No correlation was found between the sand content and quartz, clay min-erals, and mica group minmin-erals, the respective correlation coefficients being 0.0136,–0.1363, and 0.1522

Table 1 Descriptive statistics

for sand content (dry wt%),

relative water content (%), and

organic matter content (dry wt%)

of each sediment type ( n number

of samples, SD standard

deviation)

Sediment type Sand content (%) Water content (%) Organic matter content (%)

Geo-Mar Lett

Characterization of reflectance spectra Reflectance spectra from a given sediment type (M, sM, mS, and S) were averaged for wet or dry laboratory conditions The data show that the spectral patterns of the four sediment types are similar under the same conditions (wet or dry), whereas they show different trends between the wet and dry conditions: the reflectance under dry conditions is larger than that under wet conditions in each case (Fig 4) The reflectance decreases strongly at wavelengths >1.3μm, par-ticularly for two strong absorption bands at about 1.4 and 1.9μm under wet conditions Despite a large decrease in the absorption depth, the absorption at 1.9μm still appears in all the sediment types examined under dry conditions These absorptions can be attributed to the presence of hydroxyl groups and water molecules in the crystal structure of clay minerals (Clark 1995), or to surface and interstitial water

By contrast, the reflectance spectra become almost flat at wavelengths >0.8μm under dry conditions As a result, the spectral difference between the wet and dry conditions is most conspicuous at wavelengths >1.3μm

Common to both the wet and dry conditions, the spectra increase from 0.35 to 0.8 μm but decrease at wavelengths

>2.2 μm for all sediment types, the reflectance being in-versely proportional to sediment composition The M type sediment has the largest reflectance under both wet and dry conditions, whereas the S and mS types maintain relatively small reflectance A marked difference in reflectance between the M and S types is observed within the 0.75– 1.3μm wavelength region

In a next step, the original spectra illustrated in Fig 4 were converted to simulated reflectances corresponding to ALOS bands 1 to 4 by the method described above, the reflectance being denoted asRB1,RB2,RB3, andRB4, respec-tively As estimated from the original spectra, the simulated reflectance increases monotonously from bands 1 to 4 for all sediment types under both wet and dry conditions (Fig.5) Interestingly, band 4 shows a distinct difference in reflec-tance between the M and S type sediments Correlations between the reflectance spectra and the mass physical prop-erties are expressed as the goodness of fit in terms of the correlation coefficient between the simulated ALOS band

0

20

40

60

80

100

b

0.00

0.50

1.00

1.50

2.00

2.50

3.00

Sediment types c

0

20

40

60

80

100

a

Fig 2 Box plots representing variations of a sand content (dry wt%),

b relative water content (%), and c organic matter content (dry wt%) in

each of four sediment types classified as mud (M), sandy mud (sM),

muddy sand (mS), and sand (S) The bottom and top whiskers denote

the minimum and maximum values The bottom, middle, and top

horizontal lines in a box and the black circle denote the lower quartile,

median, upper quartile, and mean, respectively

0.00 0.50 1.00 1.50 2.00 2.50 3.00

0 10 20 30 40 50 60 70 80 90 100

r= 0.8221

0 10 20 30 40 50 60 70 80 90 100

0 10 20 30 40 50 60 70 80 90 100

Fig 3 Scatter diagrams

showing the relationships

between sand content (dry

wt%) and a relative water

content (%), and b organic

matter content (dry wt%), the

correlation being negative in

both cases

reflectance and the sand content, water content, organic

matter content, or mineral composition (cf Table2)

Relationship between reflectance spectra and sand content

Correlation trends between reflectance and grain-size

com-position were also assessed in terms of the relationship

between the reflectance spectra in the different ALOS bands

and the sand contents of all sediment types The negative

correlation between reflectance and sand content (i.e.,

re-flectance increases as sand content decreases) is particularly

notable as wavelengths increase from bands 1 to 4 under

both wet and dry conditions, the correlation under the dry

condition being slightly higher The relationships between

RB4 and sand content (Cs) under wet and dry conditions

show the best correlations (Fig.6), and can be expressed by

the following equations:

RB4wet¼ 0:88  103 Csþ 0:25 ð1Þ

RB4dry¼ 1:37  103 Csþ 0:43 ð2Þ

These equations adequately approximate the relationships

for both wet and dry conditions, the respective correlation

coefficients of−0.7859 and −0.8094 being statistically

signif-icant at the p<0.01 level For comparison, the relationships

between reflectance and sand concentration (mass per unit

volume) for both wet and dry conditions were also

investigat-ed, but the correlations were found to be inferior to those

based on sand content (mass per unit mass; r=−0.5865 and

−0.7162, respectively)

Relationship between reflectance spectra and water content Because the reflectance values for dried samples were higher than the corresponding values for wet samples (Figs.4and5), samples representative of the four sediment types were selected and the change in their reflectance spectra measured as the water contents (Cw) decreased The sand contents of these samples were 8 % (M), 41 % (sM), 72 % (mS), and 96 % (S), respectively Using ALOS band 4 for the simulated reflectance, this procedure reveals strong negative correlations betweenRB4andCwfor all four sediment types (Fig.7) The correlation coefficients ofRB1

toRB4together with theCwof the four datasets are listed in Table3 Although all bands show high correlations and the differences in the correlation coefficients are small, band 4 shows the highest correlations in all cases except for the mud dataset: at high water contents, the latter tends into the opposite direction, i.e., the correlation becomes weaker as the wavelength increases

Next, the complete dataset was used to determine a general relationship between the simulated reflectance (RB1

toRB4) and Cw The data reveal thatRB4 has the strongest correlation (r =0.6395, significant at p < 0.01), as evident from Table2 The reflectance ofRB4can be correlated with

Cwby the following empirical equation (cf also Fig.8):

In this case, however, the positive correlation contrasts with the negative correlations betweenRB4andCwin Fig.7 This apparent conflict may be caused by interactions between the effects ofCsandCw Thus, theCwdata can be subdivided

0.00 0.05 0.10 0.15 0.20 0.25 0.30

Wavelength (µm)

0.00 0.10 0.20 0.30 0.40 0.50

Wavelength (µm)

Fig 4 Average reflectance

spectra of the four sediment

types under a wet conditions

typical of in situ tidal flats at

low tide, and b dry conditions

typical of high-lying tidal flats

exposed for several days in the

course of neap –spring cycles

0.00 0.10 0.20 0.30 0.40 0.50

Simulated ALOS bands

a

RB1

RB2

0.00 0.10 0.20 0.30 0.40 0.50

Simulated ALOS bands

b

RB1

RB2

RB3

RB4

Fig 5 Simulated reflectance

corresponding to ALOS bands

1 to 4, obtained by averaging

the reflectance data of samples

belonging to a given sediment

type under a wet and b dry

conditions Note the marked

difference in reflectance

between mud and sand in

ALOS band 4

Geo-Mar Lett

into two classes based on sediment type: a highCw class,

ranging from 30–94 %, and a low Cwclass ranging from 2–

30 % (Fig.8) These correspond to mud-dominated (M and

sM) and sand-dominated (mS and S) sediments, respectively

Near-infrared electromagnetic waves are absorbed by water

Although the water content of the mud-dominated sediments

is higher than that of the sand-dominated sediments, their

reflectance is higher than that of the latter, as demonstrated

in Figs.4 and5 This signifies that the effect ofCwon the

reflectance is negligible in comparison with that ofCs

Relationships with organic matter content and mineral

composition

The data on organic matter content (Com) can be subdivided

into two groups using 0.20 % as a cutoff TheRB4 values

scatter widely for the total dataset comprising all samples

(Fig.9) There is no correlation withCombelow 0.20 %, this

being the case under both wet and dry conditions By contrast,

positive linear correlations can be identified forCom

exceed-ing 0.20 % TheRB4values increase with increasingCom, the

correlation being higher (r=0.7610, p<0.01) under wet than

under dry conditions (r = 0.6460, p < 0.01) Moreover, the

correlations between theRB1,RB2, andRB3reflectances and

Comare all weaker than forRB4

No correlation was found between the weight-percentages

of individual mineral groups andRB4, probably because the

r-values linkingRB4with quartz, clay minerals, and mica group

minerals were exceedingly low (0.0317, 0.0637, and−0.2995,

respectively) for the wet samples This lack of a correlation

was confirmed for the other three ALOS bands, and also for the dry samples Furthermore, there can be no direct correlation between quartz content and reflectance, because quartz is insensitive to reflectance spectra—there are no specific absorp-tion features in the visible to near-infrared region (Clark1995) The non-observance of any effects of other minerals on the reflectance spectra may be attributed to the small amount of material available for XRD analysis Consequently, reflectance

is not particularly useful in the identification of the mineral composition of intertidal sediments

Multivariate analysis

In a final step, the data on simulated reflectance versusCs,

Cw, and Comwere combined into a multivariate regression analysis Based on the results presented above, band 4 was chosen for this analysis The best-fit regression model shows that RB4 can be expressed by a linear function for

Cs(r=–0.7906, p<0.001) as:

Consequently, sand content is the predominant factor con-trolling the reflectance of these intertidal sediments The in-terrelationships between the three properties show that Cs

has strong negative correlations with Cw (r=–0.8464) and

Com(r=–0.8221)

Discussion For the Ba Lat estuarine tidal flats, mud samples have been shown to have the highest average reflectance over all four ALOS bands examined in the laboratory (Figs.4and5), the reflectance increasing with decreasing particle size (Table2) These findings contrast with those of several studies based

on the spectroscopy of intertidal sediments (e.g., Bryant et

al 1996; Ryu et al.2004; van der Wal and Herman2007; Small et al 2009) However, observations similar to the present ones can be found in other primary literature Thus,

an in situ study of intertidal sediments by Rainey et al (2000) reported muddy sediments to have a higher

Table 2 Correlation coefficients for simulated reflectance in ALOS

bands 1 to 4, based on averaging all data for sand content (dry wt%)

under wet and dry conditions, and for relative water content (%)

Wet sediment Dry sediment

0.00 0.05 0.10 0.15 0.20 0.25 0.30

0 10 20 30 40 50 60 70 80 90 100

RB4

0.00 0.10 0.20 0.30 0.40 0.50 0.60

0 10 20 30 40 50 60 70 80 90 100

RB4

Sand content (dry wt%)

r= 0.8094

r= 0.7859 Fig 6 Relationship between

sand content (dry wt%) and

simulated reflectance of ALOS

band 4 under a wet and b dry

conditions The regression lines

illustrate the strong negative

correlations for both

conditions

reflectance than sandy sediments under completely

water-saturated conditions Supporting evidence has also been

presented by Baumgardner et al (1985) and Clark (1999)

who demonstrated that the reflectance of pyroxene, silicate

rock, and soil samples generally decreases as grain size

increases over the visible to near-infrared band In fact, the

results of the present study can be explained by the

reflec-tance theory of Clark (1999), which states that grain size

controls the amount of scattering and absorption of

electro-magnetic waves in a set of particles As the size of grains

increases, absorption becomes stronger and scattering

weak-er, which results in a decrease in reflectance

The accuracy of estimates of the grain-size composition of

intertidal sediments from reflectance spectra is strongly

de-pendent on the water content, which varies with grain size,

exposure duration, and topography (Rainey et al.2000; Ryu et

al.2004) The increase in reflectance accompanying the loss

of water (Fig.7) is consistent with the laboratory observations

of Rainey et al (2000) and Small et al (2009) Rainey et al

(2000) observed that sediment samples collected on dried-out

tidal flats showed a strong correlation between sand content and reflectance A similar relationship was observed in the present study under dry laboratory conditions Accordingly, discriminating the type of intertidal sediment becomes possi-ble through the reflectance spectra when a tidal flat has suffi-ciently dried out in the course of extended exposure This study has shown that sand content is a predominant factor in the reflectance of intertidal sediments Variations in reflectance between wet and dry laboratory conditions are maximized at wavelengths >1.3μm (Fig 4) This confirms that reflectance is most sensitive to water content in the short-wave infrared region Band 5 (1.55–1.75 μm) Landsat TM and ETM+ sensors cover this region of the spectrum, this band

0.00 0.10 0.20 0.30 0.40 0.50

RB4

Water content (%)

Mud

0.00 0.10 0.20 0.30 0.40 0.50

RB4

Water content (%) Sandy mud

0.00 0.10 0.20 0.30 0.40 0.50

RB4

Water content (%)

Muddy sand

0.00 0.10 0.20 0.30 0.40 0.50

RB4

Water content (%) Sand

r= 0.9088 r= 0.9090

Fig 7 Diagram showing the

increase in simulated

reflectance (in ALOS band 4)

with decreasing relative water

content (%) in the course of

drying The four datasets ( n=7

in each case) represent the four

sediment types examined in this

study

0.00 0.05 0.10 0.15 0.20 0.25 0.30

RB4

Water content (%)

r= 0.6395

Fig 8 Relationship between relative water content (%) and simulated reflectance (in ALOS band 4) using all samples of the four sediment types Note that finer sediments (M and sM) have higher reflectance than coarser sediments (mS and S)

Table 3 Correlation coefficients between the simulated reflectance in

ALOS bands 1 to 4 and the water content (%) of the four datasets ( n=7

in each case) used in Fig 7 , these being representative of the four

sediment types

Geo-Mar Lett

having been considered useful for estimating the water content

of sediments (Ryu et al.2004; van der Wal and Herman2007),

and which is supported by the above interpretation

Regarding control by organic matter, the maximum

con-tent in sediments has been suggested to be 15 % by Ben Dor

et al (1999), those authors also reporting that organic matter

contents <2 % have no effect on reflectance, whereas

con-tents exceeding 9 % have a strong effect Although the

organic matter contents in the present study were small

(0.01–2.78 %), their linear correlation with reflectance in

ALOS band 4 suggests a pivotal >0.20 % threshold (Fig.9)

The observation that such small amounts of organic matter

can have a strong effect on reflectance is new However, the

signature disappeared in the multivariate regression model

(Eq.4) due to the overriding effect of the sand content The

present study therefore emphasizes the necessity of

compre-hensively analyzing several mass physical sediment

proper-ties when trying to identify intertidal sediment types on the

basis of remotely sensed reflectance spectra

The most important finding of this study is that intertidal

sediment types with different sand contents can be best

discrim-inated by ALOS band 4 data because of the high correlation

between sand content and reflectance The wavelength range of

this band (0.76–0.89 μm) is roughly equivalent to that of band 4

(0.77–0.90 μm) of the Landsat TM and ETM+ sensors, which

Ryu et al (2004) found to be the most useful for classifying

intertidal sediment types on the Gomso tidal flats in Korea

However, the ALOS band 4 has a much higher spatial resolution

than the Landsat band 4, and should therefore produce superior

results in the present context Monitoring of intertidal sediments

should be implemented by using a combination of ALOS band

4 imagery and in situ samples for ground-truthing This

combi-nation should prove to be particularly useful in future studies

concentrating on the effect of surface water on reflectance, a

topic that was not investigated in the present study

Furthermore, the large difference in the size of the

foot-prints between the laboratory (approx 7.07 cm2) and ALOS

images (100 m2) needs attention in future studies In the

ALOS footprint, several elements of tidal flats such as tidal

channel networks, microtopography (e.g., ripples, holes, and mounds created by organisms), and biological surface covers (e.g., dense patches of diatoms and microalgae) may change from place to place, and thereby affect local reflectance spec-tra In addition to the mass physical properties investigated in this study, the relationships between these elements and the reflectance spectra should be clarified with the aim of improv-ing the precision in monitorimprov-ing tidal flats by remote sensimprov-ing Conclusions

Using material from the Ba Lat estuarine tidal flat, this laboratory study has comprehensively investigated the char-acteristics of the reflectance spectra of intertidal sediments

in relation to grain-size composition, water content, organic matter content, and mineralogical composition

1 General patterns of the reflectance spectra in the wave-length range of 0.35–2.50 μm were similar in mud, sandy mud, muddy sand, and sand for both wet and dry labora-tory conditions Differences in the reflectance spectra between wet and dry conditions were most conspicuous for wavelengths >1.3 μm In general, reflectance was found to be inversely related to grain size (i.e., reflectance increases as grain size decreases), mud having the largest reflectance under both wet and dry conditions

2 Reflectance was simulated using ALOS AVNIR-2 spec-tral bands 1 to 4 Band 4 (near-infrared) had the stron-gest negative correlation with sand content under both wet and dry conditions This suggests that scattering and absorption of electromagnetic waves in a set of particles

is controlled by grain size As grain size increases, absorption becomes stronger and scattering becomes weaker, which results in decreasing reflectance

3 Reflectance was found to increase as water content decreased in the various sediment types, but this trend

is reversed when pooling the data on all four sediment types Compared to sand content, water content has a subordinate role in governing reflectance

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.00 0.50 1.00 1.50 2.00 2.50 3.00

RB4

Organic matter content (dry wt%)

OM content > 0.20% OM content <0.20%

0.00 0.10 0.20 0.30 0.40 0.50

0.00 0.50 1.00 1.50 2.00 2.50 3.00

RB4

Organic matter content (dry wt%)

OM content > 0.20% OM content < 0.20%

Fig 9 Relationship between organic matter content (dry wt%) and simulated reflectance (in ALOS band 4) under a wet and b dry conditions The regressions lines illustrate the positive correlations for data with >0.2 % organic content under both conditions