Remote Sensing of Environment 157 (2015) 48–57

Contents lists available at ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

Airborne hyperspectral data to assess suspended particulate matter andaquatic vegetation in a shallow and turbid lake

Claudia Giardino a,⁎, Mariano Bresciani a, Emiliana Valentini b, Luca Gasperini c,Rossano Bolpagni d, Vittorio E. Brando a,e

a Institute for Electromagnetic Sensing of the Environment, CNR-IREA, Milano, Italyb Institute for Environmental Protection and Research, ISPRA, Rome, Italyc Institute of Marine Sciences, CNR-ISMAR, Bologna, Italyd Department of Life Sciences, University of Parma, Parma, Italye Environmental Earth Observation Program, CSIRO Land and Water, Canberra, Australia

⁎ Corresponding author at: Institute for ElectromagneCNR-IREA, via Bassini 15, 20133 Milano, Italy. Tel.: +0223699300.

E-mail address: giardino.c@irea.cnr.it (C. Giardino).

http://dx.doi.org/10.1016/j.rse.2014.04.0340034-4257/© 2014 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 29 December 2013Received in revised form 26 March 2014Accepted 28 April 2014Available online 9 July 2014

Keywords:Imaging spectrometryBio-optical modellingShallow turbid lakeLake TrasimenoSuspended particulate matterSubmerged vegetationBottom depth

This paper presents an application of a physic-based method that relies on spectral inversion procedures to si-multaneously estimate concentrations of water constituents, water column heights (cH) and benthic substratetypes in Lake Trasimeno (Italy) from airborne imaging spectrometry. Complexwaters of this lake are challengingdue to the coexistence of optically-deep turbid waters and of optically-shallow waters, mostly characterised bydense submerged aquatic vegetation (SAV) beds. Airborne data acquired on 12May 2009 byMultispectral Infra-red and Visible Imaging Spectrometer (MIVIS) were converted into remote sensing reflectance Rrs(λ) with theatmospheric correction code ATCOR. A spectral inversion procedure implementing a bio-optical model (namelyBOMBER), parameterised with in situ data, was firstly run to retrieve concentrations of suspended particulatematter (SPM), chlorophyll-a (chl-a) and coloured dissolved organic matter (i.e. aCDOM(440)) in the optically-deep waters. The areas where the retrieved optimisation error was higher than 10% were instead assumed asoptically-shallow. In these areas two maps depicting the linear unmixing of three substrate types (i.e., silty-clay, Chara ssp. and other hydrophyte) and the water column heights were produced. The MIVIS-derived prod-ucts were validatedwith field data providing a reliable estimation of SPM, chl-a, aCDOM(440) and cH (determina-tion coefficients always R2 N 0.7). SPM concentrationswere also similar to a 5.4-km long transect offlow-throughturbidity data, and the SAV map was comparable to in situ observations. Generally, the colonisation patterns ofSAV were reflecting the spatial distribution of SPM concentrations. In particular, the positive role of Chara onkeeping SPM concentrations lowwas observed. Future research should extend this application to remote sensingdata acquired in other seasons to trace the dynamics of SAV and its effect on spatial water clarity.

© 2014 Elsevier Inc. All rights reserved.

1. Introduction

The optical properties of lakes are heavily influenced by catchmentphysiography and land use (Davies-Colley, Vant, & Smith, 2003).Given that rivers are the dominant inflows to lakes, lake waters mightbe expected to reflect inflowing materials. Indeed, the sediments origi-nating from land and channel erosion processes are transportedthrough the hydrographic network into lakes and oceans, as the finalsinks, modulating their physico-chemical proprieties substantially(Sheng & Lick, 1979; Xu, Xiangdong, Xuhui, & Qian, 2011). Apart fromthis, the type and abundance of primary producers can greatly influencethe characteristics of colonised lakes modulating the interactions

tic Sensing of the Environment,39 02 23699298; fax: +39

among biological, chemical, and physical processes, especially for shal-low lakes (Carpenter & Lodge, 1986). However, over the last century adramatic reduction in representativeness and extent of submergedaquatic vegetation (SAV) has been observed in a major portion oflakes worldwide (Hicks & Frost, 2011; Jeppesen et al., 2010). The maincauses are to be found in the shorelinemodification and reinforcement,water use and abuse for agricultural, industrial and human purposes,and eutrophication (Dudgeon et al., 2006; O'Hare et al., 2010). Hence,in eutrophic lakes with high concentrations of suspended matter, phy-toplankton and dissolved matter, the lake water transparency is verylowwith a rather poor lightfield. As a result, a rapid and progressive dis-appearance of macrophytes is expected.

In shallow lakes, SAV is the principal primary producer representinga key element of the entire aquatic ecosystem (Nõges et al., 2010). SAVactively participates in the cycling of nutrients, regulating the availabil-ity of macro-elements, in pollution control mechanisms, in the increas-ing habitat heterogeneity and in sustaining species diversity (Bolpagni

49C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

et al., 2007; Sollie, Coops, & Verhoeven, 2008; Wetzel, 1990). SAV alsocontributes to limit the resuspension of bottom materials that are ex-pected to be primarily driven by wind-induced currents and wave ac-tions; furthermore, their dominance results in higher concentrationsof organically-derived constituents (Barko & James, 1998; Madsen,Chambers, James, Koch, & Westlake, 2001). On the other hand, in bareor poorly-vegetated littorals of wind-exposed shallow lakes frequentresuspension of particulates is expected, resulting in the masking ofphytoplankton and increased water turbidity (Van Duin et al., 2001).In general, the sediment resuspension (1) enhances the release ofnutrients from sediment, promoting the replacement of aquatic phaner-ogams by algae and the increase of primary productivity; and (2) in-creases the concentrations of suspended particulate matter (SPM) thatpromotes light attenuation. As a general rule, fine layer clays, which at-tenuate light more intensely than similarly sized spherical particles,might remain suspended almost indefinitely (Kirk, 1994). This isespecially true in shallow lakes where an increased impact of water-sediment interface processes upon a lake ecosystem might occur(Scheffer, Hosper, Meijer, Moss, & Jeppesen, 1993).

It is therefore of paramount importance to support the SAV recoveryin order to control the regeneration of nutrients from sediments and tolimit sediment resuspension. Moreover SAV's allelopathy together withthe competition with phytoplankton for nutrients and light may inhibitphytoplankton growth (Blindow, Hargeby, Meyercordt, & Schubert,2006).

Due to its capability of synoptic spatial coverage, remote sensinghelps to understand the interactions between SPM (e.g., Lindell,Pierson, Premazzi, & Zilioli, 1999 and reference therein) and SAV(e.g., Dekker et al., 2011 and reference therein) in inland and coastalwaters ecosystems.

Recently, Odermatt, Gitelson, Brando, and Schaepman (2012) pro-vided a review of the algorithms that have been adopted to retrievewater quality parameters (including SPM) from satellite data inoptically-deep and optically-complex waters. They distinguished em-pirical and analytical methods (and in-betweens with the epithet“semi-”) from spectral inversion procedures; the latter build onmatching spectral measurements with bio-optical forward models bymeans of inversion techniques. The development of empirical or semi-empirical models generally needs coincident ground measurementsand the algorithms are generally scene- and/or sensor-dependent.Contrary, spectral inversion procedures are more generic and mightbe applicable independently of groundmeasurements and sensor char-acteristics. Nonetheless, they are less used due to the difficulties, or in-accuracies, of obtaining the parameters for model calibration (Ma,Duan, Tang, & Chen, 2010). As SPM is usually quantified outside the op-tical features of otherwater constituents (e.g., phytoplankton) (Binding,Jerome, Bukata, & Booty, 2010), the developments of empirical or semi-empirical models is feasible and many studies have applied thosemethods to satellite of inland waters (Wu, Cui, Duan, Fei, & Liu, 2013and reference therein). A successful example of spectral inversion pro-cedures for mapping total suspended matter from Landsat data inDutch lakes can be found in Dekker, Vos, and Peters (2001).

Remote sensing has also been used to make large-scale inventoriesof benthic substrates (e.g., Dekker, Brando, & Anstee, 2005; Fearns,Klonowski, Babcock, England, & Phillips, 2011; Phinn, Roelfsema,Dekker, Brando, & Anstee, 2008), including lagoons (Alberotanza,Brando, Ravagnan, & Zandonella, 1999; Alberotanza, Cavalli, Pignatti, &Zandonella, 2006; Marani et al., 2006) and lakes (Bresciani, Bolpagni,Braga, Oggioni, & Giardino, 2012; Giardino, Bartoli, Candiani, Bresciani,& Pellegrini, 2007). In particular, airborne imaging spectrometry hasbeen used to make large-scale inventories of macrophytes, sea grassesand corals (Heege, Bogner, & Pinnel, 2004; Hunter, Gilvear, Tyler,Willby, & Kelly, 2010; Kutser, Miller, & Jupp, 2006). Retrievalmethodsmay involve classification techniques that, although precise(e.g., Reguzzoni, Sansò, Venuti, & Brivio, 2003), are effectively limited toapplication on single scenes. Hence, as for SPM retrieval, spectral

inversion procedures have been developed (e.g., Lee, Carder, Chen, &Peacock, 2001; Lee, Carder,Mobley, Steward, & Patch, 1999) and adopted(e.g., Brando et al., 2009) to promote automation of scene-independentapproaches from the classification process. Overall, spectral inversionprocedures have the advantage of simultaneously measuring informa-tion about the concentrations of water quality parameters (includingSPM) in the water column and, in the case of optically-shallow waters,of enabling bathymetry and measuring substrate type (including SAV).

In this work, a spectral inversion procedure was adopted forassessing SPM concentrations and SAV colonisation patterns from at-mospherically corrected hyperspectral airborne data of a shallow lake(average depth 4.5 m). The lake is characterised by recurring sedimentresuspension phenomena (average Secchi disc depth 1.5 m) whichmakes its water optically-deep, except the areas colonised by thick ex-tensions of SAV in which the bottom is visible (and there is a measur-able water-leaving radiance signal from the substrate). As part of thisstudy, SPM concentrations and colonisation patterns of SAV mapswere analysed with the aim of describing the role of rooted macro-phytes on water clarity.

2. Materials and methods

2.1. Study area

Located in central Italy, Lake Trasimeno is the fourth largest lake forextension in the country (124 km2). The lake's catchment area coversapproximately 30,900 ha and it has not natural emissary (Orsomando& Catorci, 1991). Tourism, agriculture and livestock breeding representthe major pressures in Trasimeno catchment. Cultivated lands coverabout 70% of the catchment's area of the lake, with irrigated intensiveagriculture occurring in 28% of the area (Mearelli, Lorenzoni, &Mantilacci, 1990). The annual charge of organic carbon (500 t), nitrogen(550 t) and phosphorus (30 t), although not consistent, negatively af-fects water quality because of the specific seasonal temperature cyclecoupled with irregular lake bed properties (Bresciani, Giardino, &Boschetti, 2011;Mearelli et al., 1990). The lake is roughly circular,fluvialin origin and also tectonic, it is a closed lake, with un-stratified and veryshallowwaters. The actual annual rainfall is about 754mm;with amainmaximum in autumn and a secondary in the spring. The average annualair temperature is 14 °C and the maximum value is recorded in July.

Monthly to biweekly water monitoring programs carried out by thelocal water authority shows that Lake Trasimeno is a carbonate-richlake (Charavgis et al., 2011). In situ and satellite based observations pro-vide average values of chlorophyll-a (chl-a) and Secchi disc depth of8.5 mgm−3 and 1.1 m, respectively (Giardino, Bresciani, Villa, &Martinelli, 2010). Those conditionsmake the lake waters turbid, threat-ened by drought and anoxia. In particular, during the end of almostevery summer, the lake is affected by recurrent algal blooms, particular-ly of filamentous cyanobacteria (Cylindrospermopsis raciborskii andPlanktothrix agardhii), a phenomena that are depending on both watertemperature increase andwater level decreasing (Bresciani et al., 2011).

The lake sediments generally show a high organic content and grainsize classified as clay and silt-clay along the southern coast, and as thesands with varying percentages of clay and silt along the northerncoast (Charavgis et al., 2011). Those sediments are often re-suspendedinto the water column by wind actions; resuspension phenomena thatare also locally induced by navigation towards the islands. Overall, forthe entire lake, the average value of SPM in the lake is 11.7 gm−3

(Giardino et al., 2010).The southeast part of the lake is densely colonised by emergent spe-

cies (e.g., Phragmites australis, Typha angustifolia, Scirpus ssp., Carex ssp.)and submerged macrophytes (Potamogeton pectinatus, Chara globularis,Myriophyllum spicatum, Ceratophyllum demersum, Potamogeton crispus).The common reed (P. australis) is the widespread species capable offorming dense belts that are among the largest of peninsular Italy.Since 2000, the ecosystem is a protected area (Natura 2000 sites) for

50 C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

its exceptional value for flora and fauna richness and species diversity.The Lake Trasimeno represents the most important inland wetlandarea of the Italian Peninsula.

2.2. Field data

Fieldwork activities were undertaken in the study area to gatherdata for the purpose of calibrating both the bio-optical model and theatmospheric correction code used for removing atmospheric effects,and for validating the image-derived products. In particular, a field cam-paign conducted on 12May2009 (Fig. 1), synchronously to the airborneoverpass, was performed as follows.

• A total of 9 stations, distributed within the scene imaged by thehyperspectral airborne sensor, were investigated with the aim ofgathering data for validating the image-derived concentrations ofSPM, chl-a and the absorption coefficient of CDOM at 440 nm(aCDOM(440)) (Table 1). At each station,water sampleswere collectedof water samples of the top 1 m water layer and filtered in situ forsubsequent laboratory analysis. SPM and chl-a concentrations weredetermined gravimetrically (Strömbeck & Pierson, 2001) and accord-ing to Lorenzen (1967), respectively. aCDOM(440) was determinedspectrophotometrically according to Babin et al. (2003).

• A 5.4-km long transect of turbidity data was collected using a flow-through system. The system is composed of a hydraulic device, (es-sentially an intake pipe) continuously pumping water from 0.5 mdepth into a Turner Design Scufa-II turbidimeter, and a global posi-tioning system (GPS), both logged by a Campbell data-logger. Logged

Fig. 1. Location of fieldwork activities performed in the study area. Circles shows the 9 stationswshows the locations for registering the benthic substrate types. The flow-through system (blackposition of the sun-photometer. All data were acquired on 12 May 2009, synchronously with tsensor is outlined). Grey transects show the locations of the acoustic depth sounding performe

values of turbidity, in nephelometric turbidity units (NTU), werecorrected for delays caused by the flow-through system.

• 11 GPS points for locating the presence of SAV and for gathering theinformation on the submersed habitat (including uncolonised areas)were registered. At these stations the nadir-viewing reflectance ofSAV and of silty-clays substrates were derived ex-situ, above water,using an Analytical Spectral Device (ASD) FieldSpec radiometer anda Spectralon reflectance panel. The reflected radiation field overcollected sampleswas assumed to be Lambertian and the benthic sub-strates albedo was computed by averaging the nadir viewing reflec-tance values of the different measured species.

• Nearby the lake (asterisk in Fig. 1), the aerosol optical thickness (AOT)measurements were collected from the EKO MS-120 sunphotometerreadings for gathering data for correcting the image for the atmo-spheric effects.

Then, a comprehensive dataset of specific inherent optical properties(SIOPs), which has been collected in six days (between May and Sep-tember of 2008 and 2009) for a total of 31 investigated locations distrib-uted in the lake were used. At those stations, water samples near thesurface were collected to measure the absorption spectra of particles.The absorption spectra of particles retained on the filters ap(λ), wereobtained using the filter pad technique (Strömbeck & Pierson, 2001)and were calculated according to Babin et al. (2003). The filters werethen treated with acetone to extract pigments and the absorption spec-tra of non-algal-particle (aNAP(λ)) of these bleached filters were mea-sured. The absorption spectrum of phytoplankton aph(λ) was derivedby subtracting aNAP(λ) from ap(λ) spectra. The spectrophotometric

erewater sampleswere taken for deriving in situ data of chl-a, SPM and aCDOM(440); starsline) was cruised from north to south for a length of about 5.4 km. The asterisk shows the

he hyperspectral airborne survey (the 3.8 kmwide portion of lake imaged by the airborned on 11–12 June 2009.

Table 1Fitting statistics betweenMIVIS and in situdata:mean absolute error (MAE,which is computed as:1

�n∑

n

i−1MIVISi−in situij j; i.e. themodule of thedifference betweenMIVIS-retrieved and in

situmeasured value in each station,with thenumber of samplen equal to 7), relative rootmeans square error (rRMSE,which is computed as

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1=n∑

n

i−1MIVISi−in situið Þ

s 2

; i.e. the square root

of the average of the sumof the squares of the differences betweenMIVIS-retrieved and in situmeasured value in each station,with n = 7), slope and intercept of the linear regression line,coefficient of determination (R2), statistical significant (p*** b 0.001, **p b 0.01, p* b 0.05). The average values (av.) from both in situ and MIVIS data in the seven stations are given to-gether with their concentrations ranges (min–max) in the last four columns.

Parameter MAE rRMSE Slope Intercept R2 p In situ av. In situ ranges (min–max) MIVIS av. MIVIS ranges (min–max)

SPM (gm−3) 0.36 11.18 1.16 −0.68 0.91 *** 4.45 2.29–6.59 4.43 2.30–5.75chl-a (mgm−3) 0.58 38.94 0.71 0.73 0.71 ** 2.17 0.75–4.30 2.04 0.25–4.45aCDOM(440) (m−1) 0.16 68.42 0.25 0.17 0.78 ** 0.28 0.22–0.34 0.44 0.30–0.72

51C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

determination and processing of the absorption spectra of CDOM, a-CDOM(λ), were derived according to Babin et al. (2003). In a subset ofthese multi-temporal 31 stations, a HydroScat-6 backscattering sensorwas used to estimate the backscattering coefficient of the particles(bbp(λ)) (Maffione & Dana, 1997). For each water sample the concen-trations of SPMand chl-a,were determinedwith themethods describedabove. Furthermore, the suspended particulate inorganic matter (SPIM)was determined by combusting, at 550 °C, the filters previously used toassess SPM (Strömbeck & Pierson, 2001).

In order to gather conventional echographic profiles of the studyarea, a high-resolution bathymetric survey of the lake in the sametime-period of image acquisition (11-12 June 2009) was carried out. Avertical incidence, 200 kHz, narrow (8°, conical) beam echosounder,characterised by a very short pulse length (350 μs), was used to obtainbathymetric profiles up to a minimum depth of 0.25 m. In order tolimit errors in the depth estimate due to the possible penetration ofthe signal into the soupy water–sediment interface, we digitally sam-pled the echograms at 2 Mhz sampling frequency, and stored the datainto SEGY files that where subsequently processed using SeisPrho(Gasperini & Stanghellini, 2009). The depth estimates were carried outaccording to the procedure described in Gasperini (2005), performinga tracemixing between adjacent pings and computing an amplitude en-velope trace by convolving the squared values of the original data tracewith a rectangularwindow of the samewidth of the source pulse. Final-ly, using a simple threshold-time delay method the bottom detectionfrom the amplitude envelope trace was achieved. Conversion oftravel-times into water-depth was performed assuming a constantsound velocity of 1490 ms−1, the average of several calibration pointscollected in the lake.

2.3. MIVIS data and preprocessing

The airborne data used in this study were acquired by the Multispec-tral Infrared and Visible Imaging Spectrometer (MIVIS), which is ahyperspectralwhiskbroomscanner fromDaedalus.MIVIS is amodular in-strument which consists of four spectrometers that simultaneously mea-sure the radiation coming from the Earth's surface in the visible and in thenear, middle and thermal infrareds. Radiance values are automaticallyand simultaneously recorded in 102 channels, on different tracks of themass memory connected to the sensor. MIVIS has a total field of view(FOV) of 71° and operates with a global positioning and inertial naviga-tion system collecting data used to create geospatially corrected imagery.

On 12May 2009, aMIVIS image of eastern portion of Lake Trasimenowas collected from an altitude of 2000m, producing a spatial resolutionof 4 m, within a swath of 3.8 km (cf. Fig. 1). The airborne image was ac-quired at 10.12 UTC with the flight path oriented along the principalsolar plane to minimise directional reflectance effects in the across-track direction. The atmospheric correction of MIVIS data collected bythe first spectrometer (i.e., that is the one relevant for this study: 20channels from 440 to 820 nm, with a full width half maximum of10 nm), was carried out using the ATCOR-4 code (Richter, 2009).ATCOR-4 is an atmospheric correction code used for the atmosphericcorrection of small and wide FOV airborne sensors. ATCOR-4 uses

look-up tables generated by MODTRAN (Berk, Bernsten, & Robertson,1989), relating sensor radiances and albedo for various atmosphericand geometric conditions. ATCOR-4 was run with a maritime aerosolmodel and a horizontal visibility of 16 km, the latter derived from insitu measurements of the AOT value (0.3 at 550 nm). The ATCOR-4-derived atmospherically corrected reflectance was then converted intoremote sensing reflectance Rrs(λ) dividing by π.

According to Brando and Dekker (2003), the environmental noiseequivalent reflectance difference (NEΔRrsE) was computed aiming theassessment of an integrated measure of sensor signal-to-noise ratioand scene-specific characteristics. By applying the method developedby Wettle, Brando, and Dekker (2004) an average value of NEΔRrsE of0.0001 sr−1 from 440 to 820 nm was achieved, consistently with thecharacteristics of MIVIS scenes acquired in other lakes (Giardino et al.,2007). Just in the first channel at 440 nm and in the last two at 798and 820 nm, NEΔRrsE increased to 0.00025 sr−1. As distinguishablelevels of at least 0.00025 sr−1 are desirable for assessing water proper-ties in optically complex waters (Brando & Dekker, 2003; Dekker et al.,2001), the quality of the MIVIS image was considered appropriate forfurther analysis.

Once image pre-processing was complete, a linear spectral mixtureanalysis (LSMA) (Combe et al., 2005; Small, 2001; Ustin, Roberts,Gamon, Asner, & Green, 2004)was applied in order to detect themacro-phytes emerging from thewater surface. Those areas, which correspondto the 6% of the lake area imaged by MIVIS, are mostly colonised byP. australis (88 ha)while less pixels (30 ha)were colonised by emergingSAV (mainly P. pectinatus). Those areas were left out from the spectralinversion procedure used for producing the MIVIS-derived maps.

2.3.1. Bio-optical modelling and spectral inversion procedureThe spectral inversion procedure used in this study was performed

with the tool BOMBER (Giardino et al., 2012), which implements the al-gorithms of Lee, Carder, Mobley, Steward, and Patch (1998); Lee et al.(1999) for deriving bottom depths and water properties by optimiza-tion techniques. The software package makes use of bio-optical modelsfor both optically-deep and -shallow waters. The optimization tech-nique (Lasdon & Waren, 1978) allow the maps of SPM, chl-a, andaCDOM(440) to be retrieved. Then, in case of optically-shallow waters,bottom depth and distributions of up to three different types of sub-strate are achieved as well.

In the bio-opticalmodel, the abovewater remote sensing reflectanceRrs(λ), is expressed as a function of the subsurface radiance reflectancer0−(λ), which is approximated as the sum of contributions from thewater column r0

−(λ)C and the bottom r0−(λ)B:

Rrs λð Þ ¼ Π � r−0 λð Þ1−r−0 λð Þ� � ; ð1Þ

r−0 λð Þ ¼ r−0 λð ÞC þ r−0 λð ÞB ¼ r−0 λð Þdp 1−A0e−D0κ λð ÞcHð Þ� �

þ A1ρ λð Þe −D00κ λð ÞcHð Þ; ð2Þ

with k λð Þ ¼ a λð Þ þ bb λð Þ; ð3Þ

52 C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

r−0 λð Þdp ¼ g � u λð Þ; ð4Þ

with u λð Þ ¼ bb λð Þa λð Þ þ bb λð Þ ; ð5Þ

whereΠ in Eq. (1) take into account the effects of air–water interface onthe transmission of radiance; r0−(λ)dp, in Eqs. 2 and 4 is the contributionof optically-deep waters in sr−1; cH is the effective free water columnheight measured from the top of the canopy/substrate to the watersurface; g is a scalar used to relate the total absorption and scatteringcoefficients to the subsurface radiance reflectance by means of u(λ)(Eq. 5). u(λ) and k(λ) in Eqs. 3 and 5, respectively are functions of thetotal absorption coefficient a(λ) and the total backscattering coefficientbb(λ), which are modelled depending on concentrations of water con-stituents (cf. Eqs. 9–15). It should be noticed that, for increasing valueof cH, the waters becomes optically-deep and Eq. (2) is simplified tor0−(λ) = r0

−(λ)dp. Then, for the definition of the model parameters A0,A1, D′ and D″ and for a more detailed description of the bio-opticalmodel refer to Lee et al. (1998; 1999) and Giardino et al. (2012). Thebottom albedo ρ(λ) in Eq.(2) is expressed as linearmixing of three ben-thic covers:

ρ λð Þ ¼ b0ρ0 λð Þ þ b1ρ1 λð Þ þ b2ρ2 λð Þ; ð6Þ

with b0 þ b1 þ b2 ¼ 1; ð7Þ

where bi = 0,1,2, represent the relative distribution (Eq. 7) of threedifferent substrate albedo ρ(λ)i = 0,1,2.

The optimization process is then performed through the minimiza-tion of a target function δ, which measures the spectral distance be-tween modelled RMrs(λ) and image derived RIrs(λ):

δ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXλmax

λmin

RMrs λð Þ−RI

rs λð Þ� �2

;

vuut ð8Þ

There are six variables for Eq. (8): concentrations of three waterquality parameters (i.e. SPM, chl-a and aCDOM(440) as shown inEqs. 9–15), cH and the bottom distribution b0 and b1 (as b2 is a functionof b0 and b1, cf. Eq. 7) of pre-defined albedo classes. In case of optically-deep waters the variables decrease to the three concentrations of SPM,chl-a and aCDOM(440). Starting from initial values of the unknowns,the predictor-corrector procedure changes these values computing thedistance function δ, at each step. The algorithm repeats this processuntil δ reaches a minimum. Values corresponding to the δ minimumrepresent the best (optimum) set for the unknown variables.

Fig. 2. The chl-a specific absorption

The parameterization of the bio-optical model implemented inBOMBER was carried out based both on in situmeasurements collectedin 2008 and 2009 and on literature data. The total absorption coefficienta(λ) was modelled as:

a λð Þ ¼ aw λð Þ þ aph λð Þ þ aNAP λð Þ þ aCDOM λð Þ; ð9Þ

where aw(λ), aph( λ) aNAP(λ) and aCDOM(λ) are the absorption coeffi-cients of pure water, phytoplankton, non-algal particles and CDOM, re-spectively. aw(λ) was taken from Pope and Fry (1997) and Smith andBaker (1981). aph(λ) was modelled according to similar studies(Brando & Dekker, 2003; Keller, 2001) by means of the specific phyto-plankton absorption coefficient a⁎ph(λ) (Fig. 2).

aph λð Þ ¼ a�ph λð Þchl‐a; ð10Þ

The non-algal-particle absorption coefficient aNAP(λ) wasparameterised as follows.

aNAP λð Þ ¼ aNAP 440ð Þe‐SNAP λ−440ð Þ ¼ GaSPMþ Oað Þe‐SNAP λ−440ð Þ; ð11Þ

where Ga and Oa are the numerical values used to set the slope andthe intercept of the regression line between SPM concentrationand aNAP(440). In situ data analysis provided Ga and Oa equals to0.0248 and 0, respectively (Fig. 3). SNAP, which is a scalar definingthe slope of the exponential curve of aNAP(λ) (Babin et al., 2003)was equal to 0.013.

The absorption coefficient of CDOM aCDOM(λ) wasmodelled accord-ing to similar studies (Brando & Dekker, 2003; Keller, 2001) as a func-tion of aCDOM(440) and SCDOM, the slope factor of the absorptionspectra of CDOM. In situ data analyses provided a value of SCDOM of0.016.

aCDOM λð Þ ¼ aCDOM 440ð Þe−SCDOM λ−440ð Þ; ð12Þ

The spectral total backscattering coefficient bb(λ) was computed as:

bb λð Þ ¼ bbw λð Þ þ bbph λð Þ þ bbNAP λð Þ; ð13Þ

where bbw(λ), bbpy(λ) and bbNAP(λ) are the backscattering coeffi-cients of pure water, phytoplankton and non-algal particles, respec-tively. bbw(λ), was taken from Dall'Olmo and Gitelson (2006).bbpy(λ) was modelled as a function of the specific backscattering co-efficients due to phytoplankton bb⁎ph(λ), with bb⁎ph(440) = 0.00138and γph = 1.359

bbph λð Þ ¼ bb�ph λð Þ chl‐a ¼ bb

�ph 440ð Þ 440= λð Þ‐γphchl‐a; ð14Þ

spectrum of Lake Trasimeno.

Fig. 3. Scatterplot of aNAP(λ) as a function of SPM. The linear regression gives a slope of0.0248, used to set Ga, with a null intercept, used to set Oa.

Fig. 5. Albedo spectra of the three substrates sampled in the study area: silty-clay, Charassp. (mainly Chara globularis) andother submerged hydrophyte beds (mainly Potamogetonpectinatus andMyriophyllum spicatum).

53C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

Similarly, the backscattering coefficient of non-algal particle bbNAP(λ)was modelled as follows:

bbNAP λð Þ ¼ bb�NAP λð Þ f SPM ¼ bb

�NAP 440ð Þ 440= λð Þ−γNAP f SPM; ð15Þ

where bb⁎NAP(440) is equals to 0.01271, γNAP = 0.602 and f is a scalarused to describe the fraction of SPM (in our study f = 0.67 in Fig. 4)that contributes to the backscattering of NAP according to Strömbeckand Pierson (2001).

The bottom albedo spectra used to parameterise BOMBER (cf. Eq. 6)are shown in Fig. 5: ρ(λ)0 is the spectra of silty-clay substrates, ρ(λ)1 in-dicates the spectra of Chara ssp. beds (mainly C. globularis) and ρ(λ)2 isthe spectra of others submerged hydrophyte beds growing in the studyarea (dominated by P. pectinatus andM. spicatum).

Since the detection in underwater environment can be complicatedby the effect of waters absorption and scattering (Silva et al., 2008) dueto water constituents (Lee & Carder, 2004) and, in particular, becausethe ability to accurately recognize SAV is strongly dependent on thedepth and turbidity of the overlying water column (Hunter et al.,2010), a distinguish between optically-deep and -shallow waters hasto be adopted for running the spectral inversion procedure. To theaim, BOMBER was run with the optically-deep model for the wholelake area, allowing the classification of pixelswith an acceptable optimi-zation error (i.e., the target function δ in Eq. 8) from those with higheroptimization errors, where bottom signals were expected to be detect-able. In particular, an optimization error threshold δ b 10% was used toclassify optically-deep water pixels; for those pixels SPM, chl-a and

Fig. 4. Scatterplot of SPIM as a function of SPM. The linear regression gives a slope of 0.67used to set f in Eq. (15).

aCDOM(440) concentrations were provided by BOMBER. On contrary,pixels with δ ≥ 10% were assumed as optically-shallow waters. Inthose areas (that correspond to the 40% of the lake area imaged byMIVIS) a further run of BOMBER was accomplished. In this case, thesimplifying assumption of constant water optical properties wasassumed before running the code for estimating cH and the linearunmixing of the three benthic covers. According to previous studies(e.g., Adler-Golden, Acharya, Berk, Matthew, & Gorodetzky, 2005;Dierssen, Zimmerman, Leathers, Downes, & Davis, 2003) this assump-tion avoids some ambiguities in separating the effects of differentdepths, bottom materials, and water type (e.g., the presence of chl-aboth in phytoplankton organisms and in the macrophytes). To runBOMBER, the concentrations of water constituents were assumed in-variable within all the shallow water area. SPM, chl-a, and aCDOM(440)werefixed to 1.0 gm−3, 2.5mgm−3 and0.4m−1, respectively accordingto in situ values measured in two stations (cf. Fig. 1, stations 8 and 9)where the signal from the bottom was achievable.

3. Results and discussion

Fig. 6 presents the pseudo true colour MIVIS image, the SPMmap inoptically-deep waters and, in optically-shallow waters, bottom depthand benthic substrate, those last three all retrieved by applyingBOMBER to the atmospherically corrected MIVIS image.

The pseudo true colour MIVIS image (Fig. 6) qualitatively describesthe diversity of waters within the lake. The turquoise-cyan colours aredue to SPM scattering; the dark-blue colour on the south-western sidedepicts the lake's area where macrophytes beds absorb most of thelight. In the pseudo true colour MIVIS image, 10 transects at the borderbetween optically-deep and -shallow waters, each of them of about400 m of length, are outlined for the subsequent discussion on inter-action between SAV and SPM.

The BOMBER retrieved products in the optically-deep waters de-scribe the concentrations of SPM, chl-a and aCDOM(440), respectively.The average values of SPM, chl-a and aCDOM(440) were respectively5.9 gm−3 (standard deviation (sd) = 1.0 gm−3), 1.1 mgm−3 (sd =0.5 mgm−3), and 0.35 m−1 (sd = 0.06 m−1). These values are reliablefor the season, as in late spring the lake is in an acceptable level of qual-ity that instead get worst from mid summer until mid autumn, whenthe lake is affected by algal blooms (Giardino et al., 2010).

BOMBER-derived concentrations were then compared to in situ dataacquired synchronously with MIVIS. In particular, in the optically-deepwaters, seven of the nine stations (cf. Fig. 1 stations 1 to 7) were avail-able for the validation. MIVIS data were averaged on a 3 by 3 pixel re-gion of interest centred on the GPS location of in situ stations. Overall,the MIVIS-derived estimations of SPM, chl-a and aCDOM(440) agreed in

Fig. 6. The pseudo true colour MIVIS image (with, in grey, locations of transects used for subsequent discussions on interaction between SAV and SPM) and the three BOMBER-retrievedproducts obtained fromMIVIS data acquired on 12 May 2009: a) SPM concentration in the optically-deep waters, b) cH in optically-shallow waters and c) benthic substrate in optically-shallow waters. The grey areas in b) and c) indicate the emerging macrophytes estimated by LSMA.

54 C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

situ data (Table 1), showing a coefficient of determination (R2) alwayshigher than 0.7. In particular, the highest statistical significant wasachieved for SPM that showed the lowest relative root means squareerror (rRMSE). Results for aCDOM(440) were the less satisfactory proba-bly due to the lack ofMIVIS bands at shorterwavelengths (thefirst bandis at 440 nm), which are the most sensitive to variation of CDOM(Kutser, Pierson, Kallio, Reinart, & Sobek, 2005). However, the resultsfor chl-a and aCDOM(440), will be not further discussed in this paper asdo not provide additional information to examine the interaction be-tween SPM and SAV.

Fig. 7.Comparison ofMIVIS-derived SPM concentrations (left y axis) andflow-throughNTU roudata (pixel ID 304), with respect to the time of MIVIS overpass, is indicated with the vertical b

TheMIVIS-derived SPM image (Fig. 6, a) quantitatively describes thediversity of SPM patterns within the lake. SPM concentrations higherthan 7 gm−3 were observed in the northern part, while lowest valuesoccurred in the southern area, nearby the area colonised by SAV (cf.Fig. 6, c). To qualitatively evaluate the spatial variation of MIVIS-derived SPM concentrations, image data were plotted together withflow-through NTUmeasurements scaled by a factor of 50 (Fig. 7). Over-all, bothMIVIS-based estimations and transectmeasurements show lowvariability (SPM ranged 4–6 gm−3, 200–300 NTU) as the transect (cf.Fig. 1) didn't cover the areas close to station 1 where the higher SPM

gh data along the horizontal transect (cf. Fig. 1). The approximate location of transect in situar.

Fig. 8. Scatter plot ofMIVIS-derivedwater columnheight cH and in situ acoustic bathymet-ric datameasured in two zones (cf. Fig. 1 grey transects). Thefitting statistics are describedin Table 1 caption; the 1:1 line is plotted as a dotted line.

55C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

occurred nor the lower SPM values in the southern portion of thelake.

The effective water column height cH assessed from BOMBER in theoptically-shallow waters is shown in Fig. 6, b. MIVIS-derived cH valuesvaried between 0.1 and 3.8 m; lowest cH are located in the southernarea, where an average depth of 50 cm was observed, while highestwater column heights were observed in the northern area. Validationof bottom depth by means of beam acoustic data is shown in Fig. 8: ahigh match was achieved within the range from 1.5 to 3 m.

Fig. 6, c describes theMIVIS-derived bottom types, defined accordingto the three types of benthic substrates (cf. Fig. 5). The patterns ofbenthic substrates are shown as R,G,B:b1,b2,b0 so that the result containslinear unmixing of the three benthic covers: Chara ssp. (mainlyC. globularis) in red, submerged hydrophytes (mainly P. pectinatus andM. spicatum) in green and silty-clay substrates in blue.Mixed picture el-ements contain more than only one class and the sum of the bottomcoverage in each pixel is always 1 (cf. Eq. 7). Those patterns were in ac-cord with field observations performed during the airborne campaign(cf. Fig. 1, GPS locations marked with stars). SAV covered an area of ca.634 ha, accounting for 76% of the shallow waters zone, while the restwas characterised by silty-clay substrate. Most of the submerged

Fig. 9. Gradient of average value (with sd) of MIVIS-derived SPM depending on the distance fr

meadows occurred in the southern part and were represented bydense formation with a macrophyte cover higher than 60% (424 ha,equal to the 84.7% of the total area colonised by SAV). On the otherhand, only a minor part of colonised substrate exhibited macrophytecovers less than 20% (30 ha, equal to the 4.7% of the total area colonisedby SAV). On the whole, the percentage of pixels showing a dense colo-nisation (macrophyte covers N 60%), characterised alternatively bypure populations of Charophytes or submerged hydrophytes, were al-most equally represented (169.9 ha and 149.8 ha, respectively). In con-trast, the sectors of shallow waters covered by SAV with abundancesranging between 20 and 60% were characterised by the simultaneouspresence of Charophyte and submerged hydrophytes beds.With respectto the growth depth of macrophytes, considering the height of the freewater column above them, Charophyte beds exhibited depths steadilylower than those of submerged hydrophytes. The pixels dominated byCharophytes exhibited average water column height cH of 0.34 m(±0.07 m), whereas the pixels dominated by submerged hydrophytesexhibited values of 0.83 m (±0.41 m). In general, with the increasingof water column heights, a clear reduction in macrophyte cover per-centages was observed; dense SAV areas shown for water columnheights of about 0.6 m, while the sparse ones were characterised bycH of about 2.0 m.

Considering the spatial arrangement of SPM in the lake sectors adja-cent to the submerged macrophyte meadows along all the investigatedtransects (cf. Fig. 6, grey lines in the pseudo true colour MIVIS image),the SPM concentrations constantly increased the further we moveaway from the macrophyte stands edge. On average, in the range 0–50 m the SPM concentrations were about 1.5 ± 0.5 gm−3, whereas inthe outer belt (350–400 m) reached values of 5.1 ± 1.2 gm−3 (Fig. 9).

The separation between the areas dominated by macrophytes andthe water zones characterised by high concentrations of SPM is evenmore pronounced. This is not a surprising result; many authors havehighlighted the central role played by the SPM in attenuating the avail-ability of underwater light field, inmany cases in amore significantwaythan chl-a (Havens, 2003; Zhang, Qin, Chen, Chen, &Wu, 2005). In otherwords, the macrophyte distribution seems to reflect the influence ofSPMand vice versa: the outer sectors of the colonised area are quite exclu-sively colonised by angiosperms, whereas Charophyte vegetation ismain-ly confined in inner part of the submerged belt close to the shoreline. Thisis probably due to the lower sensitivity of P. pectinatus andM. spicatumto water turbidity and eutrophication compared to Charophyte species(Azzella, Rosati, Iberite, Bolpagni, & Blasi, 2014; Blindow, 1992). Further-more, the external macrophyte stands are more exposed to the action ofwaves, favouring the affirmation of anchored macrophytes.

om MIVIS-derived stand edges of SAV. SPM concentrations increase with such a distance.

56 C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

4. Conclusions

A physic-based retrieval of water quality parameters and bottomproperties in the turbid Lake Trasimeno (Italy) was achieved bymeans of airborne hyperspectral MIVIS data. A MIVIS image acquiredon 12 May 2009 was converted into above water Rrs(λ) with ATCOR-4, a MODTRAN-based atmospheric correction code for wide FOV air-borne scanners. A bio-optical model, parameterised with in situ datawas used to convert the image-derived Rrs(λ) values both into concen-trations of optically active constituents in the water column (SPM, chl-aand aCDOM(440)), and into substrate types and bottomdepths. The spec-tral inversion procedure implementing the bio-optical model (namelyBOMBER) was run with two different configurations (i.e., optical deepand optical shallow modes) depending on the optimization error pro-vided by the inversion technique. Despite an average bathymetry of4.5 m, recurrent wind-induced re-suspension sediments phenomenamake the Lake Trasimeno waters optically-deep. Only in certain areas,mainly located in the southern parts, important beds of SAV keep thewater clear (with SPM and chl-a concentrations much lower than inthe pelagic areas) and the bottom visible. The double run of BOMBERallowed the uncertainties in separating the effects of different depths,benthic/substrate features andwater types (e.g., composition and vege-tation cover detection in underwater environment can be complicatedby the physical effect ofwaters absorption and scattering) to be success-fully avoided (Lee & Carder, 2004; Silva, Costa, & Melack, 2010). In theoptical deep waters, the spectral inversion procedure provided rangesof SPM, chl-a and aCDOM(440) comparable to in situ data collected theday of the airborne campaign. A further evaluation of image-derivedSPM was based on high spatial resolution transect in situ data: about5.4 km of flow-through-derived measurements of NTU qualitativelymatch the SPM spatial patterns derived fromMIVIS. In the shallow wa-ters areas, the MIVIS-derived water column heights above the sub-strates were matching the acoustic soundings. The benthic coverpatterns of Chara ssp. (mainly C. globularis), submerged hydrophytes(mainly P. pectinatus and M. spicatum) and silty-clay substrates(i.e., the mainly benthic substrates characterising the study area) de-rived from MIVIS, were also comparable to in situ observations.

MIVIS-derived maps of SPM concentrations and colonisation pat-terns of SAV were then analysed with the aim of describing the role ofrooted macrophytes on water clarity. It is widely recognized that theloss of vegetation cover or the rarefactions of SAV promote importanteffects on water body metabolism, on sediment chemical proprieties,and nutrient cycling (Pinardi et al., 2009; Soana& Bartoli, 2013). Overall,MIVIS-derived products showed how SPM concentrationsmodulate themacrophyte spatial patterns. Simultaneously, the presence of densesubmerged stands limits the onset of sediment resuspension. These ob-servations are in total agreement with the results obtained by Van denBerg, Scheffer, Van, and Coops (1999) for Veluwemeer. Furthermore,this study indirectly confirms (1) the positive role of Chara vegetationon the maintenance of water clarity, much more than P. pectinatusthat is restricted to the marginal sectors of SAV in Lake Trasimeno;and (2) that Charophytes emerge as the stronger competitors in clearwater compared to submerged hydrophytes. The repetition of the pro-posed physics based approach to remote sensing data acquired inother seasons would permit to confirm such behaviour.

The approach presented in this study is transferable to other lakesfor which the optical properties of water column and bottom character-isation are known. Nevertheless, the retrieval performance will dependon spectral and spatial resolutions of earth observation data. Lee andCarder (2002) indicated that the total number of channels providedby both MERIS and MODIS are adequate for most coastal and oceanicremote-sensing applications but for optically shallow waters, sensorswith 20-nm contiguous bands (as MIVIS) still provide better resultsthan MERIS.

More recently Botha, Brando, Anstee, Dekker, and Sagar (2013) rec-ommended using the highest spectral resolution sensors available as

the classification accuracies are higher for differentiating substrata aswell as providing better discrimination at increased depth. For shallowinlandwater applications high spatial resolution is needed, hence usual-ly as a compromise multispectral imagery is used (e.g. Dekker et al.,2005; Lindell et al., 1999). Therefore, it is expected that Sentinel-2 willhave improved ability for discrimination of benthic composition overLandsat (Hedley, Roelfsema, Koetz, & Phinn, 2012).

Acknowledgements

We are very grateful to A. Taramelli, F. Zucca, F. Filipponi andL. Pizzimenti for useful discussions on image data processing andin situ data handling. We are also very grateful to M. Bartoli andA. Martinelli for providing feedbacks and knowledge on the studyarea. This work would not be possible without assistance provided inthe field by M. Musanti. MIVIS data were acquired by CGR Blom Parma(Italy). This study was co-funded by ARPA Umbria (prot. n. 8624),Italian Space Agency (CLAM-PHYM project, contract n. I/015/11/0),European Union (FP7-People Co-funding of Regional, National and In-ternational Programmes, GA n. 600407) and the CNR RITMARE FlagshipProject.We thank three anonymous reviewerswhose comments helpedto improve the manuscript.

References

Adler-Golden, S. M., Acharya, P. K., Berk, A., Matthew, M.W., & Gorodetzky, D. (2005). Re-mote bathymetry of the littoral zone from AVIRIS, LASH, and QuickBird Imagery. IEEETransactions on Geoscience and Remote Sensing, 43, 337–347.

Alberotanza, L., Brando, V. E., Ravagnan, G., & Zandonella, A. (1999). Hyperspectral aerialimages. A valuable tool for submerged vegetation recognition in the Orbetello La-goons, Italy. International Journal of Remote Sensing, 20(3), 523–533.

Alberotanza, L., Cavalli, R. M., Pignatti, S., & Zandonella, A. (2006). Classification of sub-mersed aquatic vegetation of the Venice lagoon using MIVIS airborne data. Annalsof Geophysics, 49(1), 271–276.

Azzella, M. M., Rosati, L., Iberite, M., Bolpagni, R., & Blasi, C. (2014). Changes in aquaticplants in the Italian volcanic-lake system detected using current data and historicalrecords. Aquatic Botany, 112, 41–47.

Babin, M., Stramski, D., Ferrari, G. M., et al. (2003). Variations in the light absorption coef-ficients of phytoplankton, nonalgal particles, and dissolved organic matter in coastalwaters around Europe. Journal of Geophysical Research, 108, 3211.

Barko, J. W., & James, W. F. (1998). Effects of submerged aquatic macrophytes on nutrientdynamics, sedimentation, and resuspension. In E. Jeppesen, M. Søndergaard, & K.Christoffersen (Eds.), The structuring role of submerged macrophytes in Lakes(pp. 197–217). New York: Springer.

Berk, A., Bernsten, L. S., & Robertson, D. C. (1989).MODTRAN: A moderate resolution modelfor LOWTRAN7. Air Force Geophysics Laboratory, Hanscom AFB, MA, GL-TR-89-0122(38 p.).

Binding, C. E., Jerome, J. H., Bukata, R. P., & Booty, W. G. (2010). Suspended particulatematter in Lake Erie derived from MODIS aquatic colour imagery. InternationalJournal of Remote Sensing, 31(19), 5239–5255.

Blindow, I. (1992). Decline of charophytes during eutrophication: Comparison with an-giosperms. Freshwater Biology, 28, 9–14.

Blindow, I., Hargeby, A., Meyercordt, J., & Schubert, H. (2006). Primary production in twoshallow lakes with contrasting plant form dominance: A paradox of enrichment?Limnology and Oceanography, 51(6), 2711–2721.

Bolpagni, R., Pierobon, E., Longhi, D., et al. (2007). Methane and carbon dioxide water-atmosphere daily exchanges in an oxbow lake with a Trapa natans. Aquatic Botany,87, 43–48.

Botha, E. J., Brando, V. E., Anstee, J. M., Dekker, A. G., & Sagar, S. (2013). Increased spectralresolution enhances coral detection under varying water conditions. Remote Sensingof Environment, 131, 247–261.

Brando, V. E., Anstee, M., Wettle, M., Dekker, A. G., Phinn, S. R., & Roelfsema, C. (2009). Aphysics based retrieval and quality assessment of bathymetry from suboptimalyperspectral data. Remote Sensing of Environment, 113, 755–770.

Brando, V. E., & Dekker, A. G. (2003). Satellite hyperspectral remote sensing for estimatingestuarine and coastal water quality. IEEE Transaction on Geoscience and RemoteSensing, 41, 1378–1387.

Bresciani, M., Bolpagni, R., Braga, F., Oggioni, A., & Giardino, C. (2012). Retrospective as-sessment of macrophytic communities in southern Lake Garda (Italy) from in situand MIVIS (Multispectral Infrared And Visible Imaging Spectrometer) data. Journalof Limnology, 71(1), 180–190.

Bresciani, M., Giardino, C., & Boschetti, L. (2011). Multi-temporal assessment of bio-physical parameters in Lakes Garda and Trasimeno from MODIS and MERIS. ItalianJournal of Remote Sensing, 43(3), 49–62.

Carpenter, S. R., & Lodge, D.M. (1986). Effects of submersed macrophytes on ecosystemprocesses. Aquatic Botany, 26, 341–370.

Charavgis, F., Ciccarelli, E., Cingolani, L., et al. (2011). Quadro delle conoscenze ambientalipregresse. In A. Martinelli (Ed.), Tutela Ambientale del lago Trasimeno (pp. 43–68).ARPA Umbria978-88-905920-03 (In Italian).

57C. Giardino et al. / Remote Sensing of Environment 157 (2015) 48–57

Combe, J. P., Launeau, P., Carrère, V., et al. (2005). Mapping microphytobenthos biomassby non-linear inversion of visible-infrared hyperspectral images. Remote Sensing ofEnvironment, 98(4), 371–387.

Dall'Olmo, G., & Gitelson, A. A. (2006). Effect of bio-optical parameter variability and un-certainties in reflectance measurements on the remote estimation of chlorophyll-aconcentration in turbid productive waters: Modeling results. Applied Optics, 45,3577–3592.

Davies-Colley, R. J., Vant, W. N., & Smith, D.G. (2003). Colour and clarity of natural waters.New York: Ellis Horwood.

Dekker, A. G., Brando, V. E., & Anstee, J. M. (2005). Retrospective seagrass change detec-tion in a shallow coastal tidal Australian lake. Remote Sensing of Environment, 97,415–433.

Dekker, A. G., Phinn, S. R., Anstee, J., et al. (2011). Intercomparison of shallow water ba-thymetry, hydro-optics, and benthos mapping techniques in Australian and Caribbe-an coastal environments. Limnology and Oceanography: Methods, 9, 396–425.

Dekker, A. G., Vos, R. J., & Peters, S. W. M. (2001). Comparison of remote sensing data,model results and in situ data for total suspended matter (TSM) in the southernFrisian lakes. The Science of the Total Environment, 268(1–3), 197–214.

Dierssen, H. M., Zimmerman, R. C., Leathers, R. A., Downes, T. V., & Davis, C. O. (2003).Ocean color remote sensing of seagrass and bathymetry in the Bahamas Banks byhigh-resolution airborne imagery. Limnology and Oceanography, 48(1, part 2),444–455.

Dudgeon, D., Arthington, A. H., Gessner, M.O., et al. (2006). Freshwater biodiversity: Im-portance, threats, status and conservation challenges. Biological Reviews, 81, 163–182.

Fearns, P. R. C., Klonowski, W., Babcock, R. C., England, P., & Phillips, J. (2011). Shallowwater substrate mapping using hyperspectral remote sensing. Continental ShelfResearch, 31, 1249–1259.

Gasperini, L. (2005). Extremely shallow-water morphobathymetric surveys: The ValleFattibello (Comacchio, Italy) test case. Marine Geophysical Researches, 26, 97–107.

Gasperini, L., & Stanghellini, G. (2009). SEISPRHO: An interactive computer program forprocessing and interpretation of high-resolution seismic reflection profiles.Computers & Geosciences, 35, 1497–1507.

Giardino, C., Bartoli, M., Candiani, G., Bresciani, M., & Pellegrini, L. (2007). Recent changesin macrophyte colonisation patterns: An imaging spectrometry-based evaluation ofsouthern Lake Garda (northern Italy). Journal of Applied Remote Sensing, 1, 011509.

Giardino, C., Bresciani, M., Villa, P., & Martinelli, A. (2010). Application of Remote Sensingin water resource management: The case study of Lake Trasimeno, Italy. WaterResources Management, 4, 3885–3899.

Giardino, C., Candiani, G., Bresciani, M., Lee, Z., Gagliano, S., & Pepe, M. (2012). BOMBER: Atool for estimating water quality and bottom properties from remote sensing images.Computers & Geosciences, 45, 313–318.

Havens, K. E. (2003). Submerged aquatic vegetation correlations with depth and light at-tenuating materials in a shallow subtropical lake. Hydrobiologia, 493, 173–186.

Hedley, J., Roelfsema, C., Koetz, B., & Phinn, S. (2012). Capability of the Sentinel 2 missionfor tropical coral reef mapping and coral bleaching detection. Remote Sensing ofEnvironment, 120, 145–155.

Heege, T., Bogner, A., & Pinnel, N. (2004). Mapping of submerged aquatic vegetation witha physically based process chain. Proc. SPIE 5233, Remote Sensing of the Ocean and SeaIce 2003. 43, (February 26, 2004).

Hicks, A. L., & Frost, P. C. (2011). Shifts in aquatic macrophyte abundance and communitycomposition in cottage developed lakes of the Canadian Shield. Aquatic Botany, 94,9–16.

Hunter, P. D., Gilvear, D. J., Tyler, A. N., Willby, N. J., & Kelly, A. (2010). Mapping macro-phytic vegetation in shallow lakes using the Compact Airborne Spectrographic Imag-er (CASI). Aquatic Conservation: Marine and Freshwater Ecosystems, 20, 717–727.

Jeppesen, E., Kronvang, B., Olesen, J. E., et al. (2010). Climate change effects on nitrogenloading from cultivated catchments in Europe: Implications for nitrogen retention,ecological state of lakes and adaptation. Hydrobiologia, 663, 1–21.

Keller, P. A. (2001). Comparison of two inversion techniques of a semianalytical model forthe determination of lake water constituents using imaging spectrometry data.Science of the Total Environment, 268, 189–196.

Kirk, J. T. O. (1994). Light and photosynthesis in aquatic ecosystems. New York, NY, USA:Cambridge University Press.

Kutser, T., Miller, I., & Jupp, D. L. B. (2006). Mapping coral reef benthic substrates usinghyperspectral space-borne images and spectral libraries. Estuarine, Coastal and ShelfScience, 70, 449–460.

Kutser, T., Pierson, D. C., Kallio, K. Y., Reinart, A., & Sobek, S. (2005). Mapping lake CDOMby satellite remote sensing. Remote Sensing of Environment, 94, 535–540.

Lasdon, L. S., & Waren, A.D. (1978). Generalized reduced gradient software for linearlyand non linearly constrained problems. In H. J. Greenberg (Ed.),Design and implemen-tation of optimization software (pp. 335–362). Alphen aan den Rijn: Sijthoff &Noordhoff International Publishers.

Lee, Z., & Carder, K. L. (2002). Effect of spectral band numbers on the retrieval of watercolumn and bottom properties from ocean color data. Applied Optics, 41,2191–2201.

Lee, Z., & Carder, K. L. (2004). Absorption spectrum of phytoplankton pigments derived fromhyperspectral remote-sensing reflectance. Remote Sensing of Environment, 89, 361–368.

Lee, Z., Carder, K. L., Chen, R. F., & Peacock, T. G. (2001). Properties of the water columnand bottom derived from Airborne Visible Infrared Imaging Specrometer (AVIRIS)data. Journal of Geophysical Research, 106(11), 639–651.

Lee, Z., Carder, K. L., Mobley, C. D., Steward, R. G., & Patch, J. S. (1998). Hypespectral re-mote sensing for shallow waters: 1. A semianalytical model. Applied Optics, 37,6329–6338.

Lee, Z., Carder, K. L., Mobley, C. D., Steward, R. G., & Patch, J. S. (1999). Hyperspectral re-mote sensing for shallow waters: 2. Deriving bottom depths and water propertiesby optimization. Applied Optics, 38, 3831–3843.

Lindell, T., Pierson, D., Premazzi, G., & Zilioli, E. (1999). Manual for monitoring Europeanlakes using remote sensing techniques. EUR Report n.18665 EN. Luxembourg: Officefor Official Publications of the European Communities.

Lorenzen, C. J. (1967). Determination of chlorophyll and pheo-pigments: Spectrophoto-metric equations. Limnology and Oceanography, 12, 343–346.

Ma, R., Duan, H., Tang, J., & Chen, Z. (2010). Remote sensing of lake water environment.Beijing: Science Press (in Chinese).

Madsen, J.D., Chambers, P. A., James, W. F., Koch, E. W., & Westlake, D. F. (2001). The in-teraction between water movement, sediment dynamics and submersed macro-phytes. Hydrobiologia, 1–3, 71–84.

Maffione, R. A., & Dana, D. R. (1997). Instruments and methods for measuring thebackward-scattering coefficient of ocean waters. Applied Optics, 36, 6057–6067.

Marani, M., Belluco, E., Ferrari, S., et al. (2006). Analysis, synthesis and modelling of high-resolution observations of salt-marsh eco-geomorphological patterns in the Venicelagoon. Estuarine, Coastal and Shelf Science, 69(3–4), 412–426.

Mearelli, M., Lorenzoni, M., & Mantilacci, L. (1990). Il Lago Trasimeno. Rivista diIdrobiologia, 29, 353–389 (in Italian).

Nõges, T., Luup, H., & Feldmann, T. (2010). Primary production of aquatic macrophytesand their epiphytes in two shallow lakes (Peipsi and Võrtsjärv) in Estonia. AquaticEcology, 44, 83–92.

O'Hare, M. T., Clarke, R. T., Bowes, M. J., et al. (2010). Eutrophication impacts on a rivermacrophyte. Aquatic Botany, 92(3), 173–178.

Odermatt, D., Gitelson, A., Brando, V. E., & Schaepman, M. (2012). Review of constituentretrieval in optically-deep and complex waters from satellite imagery. RemoteSensing of Environment, 118, 116–126.

Orsomando, E., & Catorci, A. (1991). Carta della vegetazione del comprensorio Trasimeno(scala 1:50.000). Associazione Comuni del Trasimeno. Dipartimento di Botanica edEcologia, Università degli studi di Camerino. SELCA, Firenze. (in Italian).

Phinn, S., Roelfsema, C., Dekker, A. G., Brando, V. E., & Anstee, J. (2008). Mapping seagrassspecies, cover and biomass in shallow waters: An assessment of satellite multi-spectral and airborne hyper-spectral imaging systems in Moreton Bay (Australia).Remote Sensing of Environment, 112, 3413–3425.

Pinardi, M., Bartoli, M., Longhi, D., et al. (2009). Benthic metabolism and denitrification ina river reach: A comparison between vegetated and bare sediments. Journal ofLimnology, 68(1), 133–145.

Pope, R. M., & Fry, E. S. (1997). Absorption spectrum (380–700 nm) of pure water. II. In-tegrating cavity measurements. Applied Optics, 36, 8710–8723.

Reguzzoni, M., Sansò, F., Venuti, G., & Brivio, P. A. (2003). Bayesian classification by dataaugmentation. International Journal of Remote Sensing, 24(20), 3961–3981.

Richter, R. (2009). Atmospheric/topographic correction for satellite imagery. DLR reportDLR-IB 565-01/09, Wessling, Germany.

Scheffer, M., Hosper, S. H., Meijer, M. L., Moss, B., & Jeppesen, E. (1993). Alternative equi-libria in shallow lakes. Trends in Ecology & Evolution, 8, 275–279.

Sheng, Y. P., & Lick, W. (1979). The transport and resuspension of sediments in a shallowlake. Journal of Geophysical Research, 84(C4), 1809–1826.

Silva, T. F., Costa, M. F., Melack, J., & Novo, E. L. M. (2008). Remote sensing of aquatic veg-etation: theory and applications. Environmental Monitoring and Assessment, 140,131–145.

Silva, T. S. F., Costa, M. P. F., &Melack, J. M. (2010). Spatial and temporal variability of mac-rophyte cover and productivity in the eastern Amazon floodplain: A remote sensingapproach. Remote Sensing of Environment, 114, 1998–2010.

Small, C. (2001). Estimation of urban vegetation abundance by spectral mixture analysis.International Journal of Remote Sensing, 22(7), 1305–1334.

Smith, R. C., & Baker, K. S. (1981). Optical properties of the clearest natural waters(200–800 nm). Applied Optics, 20, 177–184.

Soana, E., & Bartoli, M. (2013). Seasonal variation of radial oxygen loss in Vallisneriaspiralis L.: An adaptation to sediment redox? Aquatic Botany, 104, 228–232.

Sollie, S., Coops, H., & Verhoeven, J. T. A. (2008). Natural and constructed littoral zones asnutrient traps in eutrophicated shallow lakes. Hydrobiologia, 605, 219–233.

Strömbeck, N., & Pierson, E. (2001). The effects of variability in the inherent optical prop-erties on estimations of chlorophyll a by remote sensing in Swedish freshwater. TheScience of the Total Environment, 268, 123–137.

Ustin, S. L., Roberts, D. A., Gamon, J. A., Asner, G. P., & Green, R. O. (2004). Using imagingspectroscopy to study ecosystem processes and properties. BioScience, 54(6),523–534.

Van den Berg, M. S., Scheffer, M., Van, Nes E., & Coops, H. (1999). Dynamics and stability ofChara sp. and Potamogeton pectinatus in a shallow lake changing in eutrophicationlevel. Hydrobiologia, 408(409), 335–342.

Van Duin, E. H. S., Blom, G., Los, F. J., et al. (2001). Modeling underwater light climate inrelation to sedimentation, resuspension, water quality and autotrophic growth.Hydrobiologia, 1–3, 25–42.

Wettle, M., Brando, V. E., & Dekker, A. G. (2004). A methodology for retrieval of environ-mental noise equivalent spectra applied to four Hyperion scenes of the same tropicalcoral reef. Remote Sensing of Environment, 93, 188–197.

Wetzel, R. G. (1990). Land–water interfaces: Metabolic and limnological regulators.Verhandlungen des Internationalen Verein Limnologie, 24, 6–24.

Wu, G., Cui, L., Duan, H., Fei, T., & Liu, Y. (2013). An approach for developing Landsat-5TM-based retrieval models of suspended particulate matter concentration with theassistance of MODIS. ISPRS Journal of Photogrammetry and Remote Sensing, 85, 84–92.

Xu, C., Xiangdong, Y., Xuhui, D., & Qian, L. (2011). Nutrient dynamics linked to hydrolog-ical condition and anthropogenic nutrient loading in Chaohu Lake (southeast China).Hydrobiologia, 661, 223–234.

Zhang, Y. L., Qin, B. Q., Chen, W. M., Chen, Y. W., & Wu, S.C. (2005). Biological and opticalproperties of Meiliang Bay in Lake Taihu. Acta Ecologica Sinica, 25, 454–460.