River Floodplain Vegetation Scenario Development Using Imaging Spectroscopy Derived Products as Input Variables in a Dynamic Vegetation Model

AbstractRiver floodplains are becoming increasingly subject tomultifunctional land-use. In this contribution, we arelinking imaging spectrometer derived products with adynamic vegetation model to improve the simulationand evaluation of scenarios for a river floodplain in theNetherlands. In particular, we are using airborne HyMapimaging spectrometer data to derive Leaf Area Index (LAI),spatial distribution of Plant Functional Types (PFT), andmodel dominant species abundances as input for theecological model. We use the dynamic vegetation model(DVM) SMART2-SUMO to simulate vegetation successionunder scenarios of changing abiotic conditions and man-agement regimes. SMART2 is a soil chemical modelwhereas SUMO describes plant competition and resultingvegetation succession. We validate all remote sensingderived products and the DVM calibration independentlyusing extensive field sampling. We conclude that thedynamic vegetation models can be successfully initializedusing imaging spectrometer data at currently unprece-dented accuracy. However, all efforts undertaken forvalidation in this contribution may significantly exceedcapacities for national or continental scale application ofthe proposed method.

River Floodplain Vegetation ScenarioDevelopment Using Imaging SpectroscopyDerived Products as Input Variables in a

Dynamic Vegetation ModelM.E. Schaepman, G.W.W. Wamelink, H.F. van Dobben, M. Gloor, G. Schaepman-Strub,

L. Kooistra, J.G.P.W. Clevers, A. Schmidt, and F. Berendse

IntroductionRiver floodplains are becoming increasingly subject tomultifunctional land-use. In hydraulics, floodplains areassessed for the suitability of flood control strategies inrelation to their roughness for reducing and delaying floodpeaks (Ghavasieh et al., 2006). In climate change, flood-plains are under investigation due to changing dischargeregimes, potentially resulting in an increase in peak and lowflows (Middelkoop et al., 2001), as well as increased pressureon hard infrastructures protecting large fractions of a growingpopulation living in the affected areas (Kabat et al., 2005).However, a particular challenge remains that river floodplainsystems are among the most complex ecosystems on Earth.The lack of detailed information about functional relation-ships and processes at the landscape and catchments scalecurrently hamper assessment of their ecological status(Jungwirth et al., 2002). Regional-based investigations pointto complex interactions, affecting soils (Kooistra et al.,2004), vegetation (Clevers et al., 2004), as well as animal life(Kooistra et al., 2005) in floodplains, but may not be limitedto these.

An increasing number of studies point to the fact, thatmedium range projections of land-use in general (Feddemaet al., 2005), and river floodplain multifunctional use inparticular will be primarily driven by anthropogenic impact,affecting insect (Nickel and Hildebrandt, 2003) and fishpopulation (Peterson and Kwak, 1999). This human influ-ence may also occasionally override the expected mediumrange impact of climate induced changes (Schaepman et al., 2005) on floodplains (e.g., extreme events and risingtemperatures).

The goal of this contribution is to combine remotesensing-derived products with dynamic vegetation modeling(DVM) to improve the simulation and evaluation of futurescenarios for a river floodplain.

In order to do so, we link imaging spectrometer-derivedproducts as variables in a DVM to assess the development ofa river floodplain that has been taken out of agriculturalproduction and is allowed to undergo a natural succession.

PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Oc t obe r 2007 1179

M.E. Schaepman, L. Kooistra, and J.G.P.W. Clevers are withthe Wageningen University, Centre for Geo-Information,Droevendaalsesteeg 3, NL-6708 PB Wageningen, TheNetherlands, ([email protected]).

G.W.W. Wamelink and H.F. van Dobben are with Alterra,Landscape Centre, Wageningen, The Netherlands.

M. Gloor is with the Earth and Biosphere Institute, Schoolof Geography, Leeds Univ., UK.

G. Schaepman-Strub is with the Wageningen University,Nature Conservation and Plant Ecology, Wageningen, TheNetherlands, and KNMI, Atmospheric Research Division,DeBilt, The Netherlands.

F. Berendse is with the Wageningen University, NatureConservation and Plant Ecology, Wageningen, TheNetherlands.

A. Schmidt is with Alterra, Centre for Geo-Information,Wageningen, The Netherlands.

Photogrammetric Engineering & Remote Sensing Vol. 73, No. 10, October 2007, pp. 1179–1188.

0099-1112/07/7310–1179/$3.00/0© 2007 American Society for Photogrammetry

and Remote Sensing

PMSRS-07.qxd 9/14/07 11:22 PM Page 1179

We compare therefore the impact of two different manage-ment scenarios on biomass production in the floodplain forthe year 2050.

Materials and MethodsTest SiteWe focus on a large-scale nature development area betweenthe cities of Arnhem (NL), Nijmegen (NL) and Emmerich(GER), named the “Gelderse Poort” (Figure 1).

This border-crossing natural reserve is located along theriver Rhine where it splits into three branches, namely theWaal, the Nederrijn and the IJssel. The floodplain Millinger-waard (51.84° N; 05.99° E; WGS84; 700 ha) is part of theGelderse Poort. The site rises 12 m ASL with a minimum of8.8 m ASL and a maximum of 15.6 m ASL. Before the 1990s,the main function of the floodplain was agriculture, namelycultivated grassland and arable land (e.g., maize). In theperiod 1990 to 1993, the agricultural function was graduallychanged into a combined nature conservation and floodprotection function. Since then, the floodplain is allowed toundergo natural vegetation succession. Nature managementmeasures are limited to removing the fences between formeragricultural parcels and grazing by cattle and horses in lowdensities. This converted the Millingerwaard into a heteroge-neous landscape with river dunes stretching along the river,a large softwood forest in the eastern part along the winterdike, and in the intermediate area a pattern of differentvegetation succession stages (pioneer, grassland, and shrubs).In addition, several clay pits are present. Nature manage-ment (grazing) within the floodplain is aiming at increasingthe biodiversity (namely the diversity of species and ecosys-tems), given the condition that the discharge capacity of theriver should be above the critical safety levels duringflooding events. To stimulate the development of a heteroge-neous landscape (see Figure 2), a low stocking density ofone animal (e.g., Galloway, Koniks) per 2 to 4 ha has beenchosen. This stocking density allows grazing all year roundand allows the development of forested areas.

Data Acquisition on the GroundSimultaneous to the acquisition of airborne imaging spectrom-eter data, extensive ground measurements were carried out.These included various sampling schemes and measurement

techniques, to comply with the requirements to satisfyrelevant input variables for data preprocessing, radiativetransfer modeling (RTM), as well as DVM. Table 1 summarizesthe measurements performed for this study.

We performed sun-photometer measurements using aSolar Microtops II instrument to characterize the atmos-pheric conditions during the HyMap airborne data acquisi-tion. The setup of the instrument included the use of fivecollimators, each with a field of view (FOV) of 2.5 degrees(sun looking), covering the wavelengths 440 nm, 675 nm,870 nm, 936 nm, and 1,020 nm. Aerosol optical thickness(AOT) at 440, 675, 870, and 1,020 nm, respectively, isretrieved using the Bouguer-Lambert-Beer law:

(1)

where � � center wavelength, V(�) � measured detectorvoltage at �, Vo(�) � extraterrestrial voltage at �, D � Earth-sun distance, �(�) � total optical thickness, and M � airmass. AOT is obtained after subtraction of the optical depthdue to Rayleigh scattering from the total optical depth. AOTis subsequently used in the atmospheric correction of theHyMap data. Figure 3 depicts the locations of groundmeasurements performed. The sun-photometer was locatedin the center of the floodplain.

Nineteen reference sites consisting mainly of naturaland man-made materials (e.g., roads, artificial clay pit, sandybeach) have been selected and measured using an AnalyticalSpectral Devices FieldSpec® Pro FR spectroradiometer (ASD)for vicarious calibration purposes as well as validation ofthe atmospheric correction. ASD data were spectrally con-volved to HyMap spectral response functions for directcomparison.

Leaf and canopy reflectance spectra were measuredusing a second ASD incorporating a high intensity contactprobe with a leaf clip for non-destructive reflectance andtransmittance measurements of in vivo leaves. Leaf spectraad-axial and ab-axial of dominating species (e.g., Calama-grostis epigejos, Rubus caesius, and Urtica dioica) weremeasured using the contact probe. We have further chosento measure canopy reflectance at 21 vegetation relevés (2 m � 2 m each). The location of the 21 sites were derivedbased on an existing vegetation map created in 2002 for the

V(l) � Vo(l)D�2e�M�(l)

1180 Oc t obe r 2007 PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING

Figure 1. Location and current land use for the flood-plain Millingerwaard along the river Rhine in theNetherlands.

Figure 2. Oblique photograph of the Millingerwaardnatural reserve towards the East. Vegetation successionstages are ranging from grasslands (foreground) toshrubs and forests (background).

PMSRS-07.qxd 9/14/07 11:22 PM Page 1180

Millingerwaard (Van Geloof and de Ronde, 2002), comple-mented by a field survey in May 2004 to adjust for potentialchanges. In addition, the vegetation relevés served fordifferent purposes, such as performing a detailed vegetationdescription, visual estimates of the fractional coverage, aswell as destructive biomass sampling following all opticalmeasurements.

The vegetation description (Schmidt et al., 2005) wasperformed following the method of Braun-Blanquet (1951).Abundance per species was estimated optically as percentagesoil covered by living biomass in vertical projection, andscored in a nine-point scale. The vegetation relevés covered

the most important plant communities as described bySchaminée et al. (1998) present in the area. All bryophytesand lichens, and vascular species that were not readilyrecognizable in the field, were collected for later identifica-tion. Taraxacum species were taken together as T. vulgare,and Rubus species were taken together as R. fruticosus,except R. caesius. No subspecific taxa were used. Nomencla-ture follows van der Meijden et al. (1990), Touw and Rubers(1989), and Brand et al. (1988) for vascular species, mossesand lichens, respectively. Syntaxonomic nomenclaturefollows Schaminée et al. (1998). Following optical measure-ments at the vegetation relevés, destructive abovegroundbiomass sampling in three 50 cm � 50 cm subplots of each ofthe 21 sample sites was performed. Vegetation biomass wassampled in a relatively homogeneous (vegetation) cover,located at three of the corners of each main plot. Biomass wasclipped at 0.5 cm above the ground level and stored in plasticbags. The collected material was air-dried, first for five daysat room temperature in open bags, and subsequently dried for24 hours at 70°C, and weighed. Sampled vegetation materialfor the 21 vegetation plots was also chemically analyzed forN, P, K, Ca, and Mg content (mmol/kg).

The forested part of the Millingwaard is dominated bywillow trees, having dominant species of Salix fragilis L.(crack willow), Salix alba L. (white willow), and Populusnigra L. (Lombardy poplar). A dense undergrowth is presentwith Urtica dioica L. (common nettle), Calamagrostisepigejos (L.) Roth (wood small-reed), and Rubus caesius L.(European dewberry) being the dominant plant species.Since canopy reflectance measurements in the forest werenot possible with the ASD due to presence of dense under-growth and water bodies, a hemispherical camera was usedto estimate the gap fraction, leaf inclination angle, and leafarea index (LAI). In the forest, thirteen sample plots wereselected following the VALERI sampling scheme (Baretet al., in review). Stratified random sampling was usedto position the sample plots in various softwood canopydensities. The coordinates of the sample plots were regis-tered using GPS at each center point. In total, 156 points(i.e., 13 elementary sampling units (ESU’s) according theVALERI sampling scheme with each ESU having 12 sub-sampling points) were measured (Mengesha et al., 2005b).


TABLE 1. OVERVIEW OF GROUND AND AIRBORNE DATA ACQUISITIONS IN THE MILLINGERWAARD

No. Sample Instrument Locations Date (2004) Measurements

Atmospheric Solar Microtops 1 02 August AOT (Aerosol condition sunphotometer optical thickness)

Vicarious ASD Fieldspec 19 (5 � 5 m) 28 July and Reflectance spectracalibration Pro FR 02 August (sand, clay, asphalt,

water)Leaf and canopy ASD Fieldspec 21 (5 � 5 m) 28 July TOC (Top-of-canopy)

reflectance spectra Pro FR and leaf spectraCanopy structure Hemispherical 13 (20 � 20 m), 28 July to LAI, gap fraction,

camera VALERI scheme 06 August leaf angle distributionVegetation relevés Braun-Blanquet 21 (2 � 2 m) 13–16 August Vegetation structure,

method species compositionDestructive biomass Laboratory 21 (0.5 � 0.5 m) 13–16 August Wet/dry biomass,

sampling + chemical analysis N and P concentrationanalysis

Soil characteristics Theta probe, 86 28 July Soil moisture and temperature gun temperature

Imaging HyMap 2 flight-lines 28 July and 126 spectral bands, 5 m spectrometer data 02 August spatial resolution,

512 pixels across track, acquisition time 11 38 UTC (13 38 MEST)

Figure 3. Geocoded HyMap image indicating samplelocations of ground based measurements for radiometriccorrection and calibration, as well as determining leafand canopy spectra, canopy structure, soil, and waterwithin the Millingerwaard floodplain.

PMSRS-07.qxd 9/14/07 11:22 PM Page 1181

Airborne Data AcquisitionAirborne imaging spectrometer data (HyMap (Cocks et al.,1998)) were acquired on 28 July and 02 August 2004 in 126spectral bands ranging from 400 to 2,500 nm (spectral band-width of 15 to 20 nm) over the Millingerwaard. The data isprocessed to surface reflectance or Hemispherical-DirectionalReflectance Factor (HDRF) following the terminology ofSchaepman-Strub et al. (2006). Preprocessing is partiallycompensating for adjacency effects as well as directionaleffects induced by the atmosphere using the model combina-tion PARGE/ATCOR-4 (Richter and Schlapfer, 2002; Schlapferand Richter, 2002). However, there is no particular treatmentof the surface induced anisotropy in this approach, resultingin the surface reflectance data to approximate HDRF. Using thesame correction scheme, HyMap data was also geocoded andremapped to UTM (Zone 31 N, geodetic datum WGS84) at anequally spaced ground sampling distance of 5 m in both axes.The flight was performed close to the local solar noon (11 38 hrs coordinated universal time (UTC), 13 38 hrs MiddleEuropean Summer Time (MEST)) at a solar zenith angle of 33°and solar azimuth angle of 178°.

SMART2-SUMO ModelThe model SMART2-SUMO (Kros 2002, Wamelink et al.,2005; Wamelink et al., 2003) is used to simulate vegetationsuccession under scenarios of changing abiotic conditionsand management regimes. SMART2 is a soil chemical modelthat describes chemical processes like weathering, adsorp-tion, desorption, mineralization, and immobilization. Fertil-ization and atmospheric deposition of nitrogen compoundsare also accounted for. SUMO describes plant competitionand resulting vegetation succession. In SUMO for a vegeta-tion structure type (e.g. grassland or forest), the biomass forfive functional types is simulated: grassland and herbs, dwarfshrubs, shrubs, pioneer tree, and climax tree. All fivefunctional types are always present, though the amount ofbiomass simulated for each functional type may vary enor-mously. SMART2-SUMO contains a full description of thenutrient cycle through root uptake, investment in biomass(divided over root, shoot and leaf), litter fall and nitrogenmineralization. Vegetation management (e.g., mowing andgrazing) is described as the removal of part of the biomass atthe end of the growing season. SMART2-SUMO is a pointmodel, which does not describe spatial (horizontal) interac-tion. Therefore, the model can be applied on various spatialscales, provided the necessary input data are available. TheSUMO-SMART2 model is calibrated using field data (onbiomass, soil pH, etc.). In order to run the SUMO-SMART2model at a spatially explicit scale, remote sensing-derivedproducts serving as input variables for the DVM are needed.We use imaging spectroscopy derived biomass for theinitialization of the model. We first use this data to validateand then to initialize the SMART2-SUMO model for twodifferent model runs. We predict the vegetation succession ofthe river floodplain by assuming different managementscenarios, resulting in estimating biomass productiondivergence in the year 2050. The two management scenariosthat were used are: (a) agricultural management, where thegrassland is mown once a year and 100 kg/ha nitrogen issupplied every year, and (b) extensive nature management,where the grassland is grazed by cows and horses at adensity of approximately one grazing unit per hectare.

Imaging Spectroscopy Derived Products as Input Variables forDynamic Vegetation ModelsThe spatially explicit biomass input variable for theSMART2-SUMO model is based on imaging spectrometerdata. Biomass derived from HyMap data is based first on asensitivity analysis comparing several LAI estimation methods

using the HyMap surface reflectance data in combinationwith four retrieval methods (Mengesha et al., 2005a) (Plate 2).These methods have been refined to account for the signifi-cant shadow fraction present in high-resolution airborne(imaging spectrometer) data as well as for spatial heterogeneityof the species composition.

Estimates of LAI were retrieved most successful in theforest using the method proposed by Chen et al. (2002),including a comparison to the methods proposed by Roujeanand Lacaze (2002) and Weiss et al. (2002). Validation wasperformed by using a simple kriging approach to interpolatewithin the ESUs (Baret and Rossello, 2006) in the softwoodforest stand and comparing the aggregated HyMap pixels(5 m to 20 m) with this interpolation.

For the estimates of biomass of non-forested areas(grasses and herbs, dwarf shrubs, and shrubs), a forwardspectral linear mixture modeling of modeled and measuredleaf spectra has been applied to approximate dominant plantspecies composition in the Millingerwaard. Leaf spectra wereeither measured in the field using the ASD contact probe, ormodeled using the leaf model PROSPECT (Jacquemoud andBaret, 1990). Missing input parameters for PROSPECT ofdominant plant species not measured in vivo were extractedfrom literature (Jacquemoud, 2006). The combined radiativetransfer model PROSPECT/SAIL (Verhoef and Bach, 2003)was used to create homogenous, single plant species TOCreflectance. Forward linear mixture modeling based ondominant plant species per plant community (Schaminéeet al., 1998) was used to create a spectral library at plantcommunity level. Finally, spectral unmixing (Hu et al., 1999;Keshava and Mustard, 2002) was used on the HyMap data tounmix for abundances of shadow, plant functional types, andother relevant endmembers (e.g., water, gravel, clay, sand,artificial/man-made materials).

The LAI product generated using the above approachwas directly compared to the destructive biomass sampledin the 21 plots.

The initialization of SMART2-SUMO assumed standardbiomass values for countrywide applications for bothagricultural and natural grasslands in 1975. We imple-mented two management scenarios to evaluate the effect ofmanagement measures on vegetation succession (biomassproduction): an agricultural management scenario, includingyearly mowing, and a natural succession scenario underextensive grazing. We also assumed the hydrology to beconstant over time. The soil type (fraction of sand or clay)was derived from spectral unmixing of ASD measurementsand using field observations per plot. The stocking densityof cattle and horses in the natural scenario was also esti-mated from field observations in the plots.

Model validation was performed by comparing thesimulated biomass in 2004 in the nature managementscenario with the actual measured biomass in the field.Next, the simulated biomass was improved by replacing thesimulated biomass of the grasses and herbs functional typein 2004 by the biomass estimated from the HyMap-derivedLAI, producing a forecast until 2050.

ResultsHyMap ProcessingHyMap data were evaluated for geometric and radiometricquality. Since most of the ground data acquired in the fieldwas registered in the Dutch reference system (Rijksdriehoek-stelsel (RD)), a co-location procedure was established tore-project the HyMap data (WGS84, UTM zone 31N coordinatesystem) into RD, and vice versa. The initial co-registrationuncertainty was found to be 1 to 2 pixels (5 to 10 m). Since


PMSRS-07.qxd 9/14/07 11:22 PM Page 1182

the vegetation maps have been established, using DGPS-basedmethods, HyMap data was re-projected to the ground datausing a polynomial transformation, eventually reducing theuncertainty to �1 m.

HyMap data did not suffer from saturation or excessnoise (Figure 4), and a principal component based approachwas performed including the use of Eigenvalues distributionto determine the dimensionality of the data. The dimension-ality was considered to be very high (10 at half the sum ofall Eigenvalues, 60 at a quarter of the maximum) for HyMap,with a few (4 to 6) noisy bands to be expected.

Surface reflectance data were retrieved by using the sun-photometer derived values for aerosol transmittance and

horizontal visibility (v � 15.5 km) as well as geometricalillumination and observation conditions. The ATCOR inher-ent iterative validation scheme (IFCALI) was used to determinethe quality of the atmospheric correction on the vicariouscalibration measurements (see Schaepman et al. (2004)).

Vegetation ClassificationIn total, 79 plant species were registered in the 21 vegetationrelevés made in the Millingerwaard. The relevés were syntax-onomically identified into plant communities (Schaminéeet al., 1998) by using the program ASSOCIA (Table 2; Figure 5:see Wamelink et al., 2003). The chemical analysis ofbiomass samples showed that the N-content was significantly


Figure 4. HyMap minimum, maximum, mean and standard deviation signal for 28 July 2004.

TABLE 2. CLASSIFIED VEGETATION SYNTAX FOR THE INDIVIDUAL PLOTS IN THE MILLINGERWAARD

PlotRD-coordinates

Number X Y Syntax Code Syntax Name

02 196826,397 431032,034 16BC01 Lolio-Cynosuretum03 196814,358 431034,655 31CA01B Echio-Melilotetum typicum04 196795,228 431076,810 31CA01B Echio-Melilotetum typicum05 196817,681 431082,001 31A Chenopodio-Urticetalia06 196863,64 431170,789 33RG03 RG Petasites hybridus-[Galio-Urticetea]07 196905,185 431185,534 29AA03C Chenopodietum rubri rorippetosum08 196900,856 431209,847 31CA01B Echio-Melilotetum typicum09 196852,201 431313,554 31CA01B Echio-Melilotetum typicum10 196768,893 431466,838 38AA01B Artemisio-Salicetum agrostietosum

stoloniferae11 196791,704 431480,965 37AB01A Pruno-Crataegetum typicum12 196764,162 431620,923 37AC02A Hippophao-Ligustretum typicum13 196712,193 431648,921 31CA Dauco-Melilotion14 196640,748 431448,257 31CA02 Bromo inermis-Eryngietum campestris15 196643,353 431367,244 31CA02 Bromo inermis-Eryngietum campestris16 196504,145 431230,224 31AA01 Bromo-Corispermetum17 196512,661 431189,218 31CA02 Bromo inermis-Eryngietum campestris18 196485,196 431115,084 14CA Tortulo-Koelerion19 196559,239 431122,375 31CA02 Bromo inermis-Eryngietum campestris20 196587,126 431140,347 31CA02 Bromo inermis-Eryngietum campestris21 196528,714 430948,963 31RG08 RG Cichorium intybus-[Agropyretalia

repentis/Arrhenatheretalia]22 196360,958 430753,776 31A Chenopodio-Urticetalia

PMSRS-07.qxd 9/14/07 11:22 PM Page 1183

(P � 0.001) higher in the grazed plots (2.0 percent) thanin the un-grazed plots (1.3 percent). For other nutrients (P, K, Mg), these differences were not significant.

Corresponding TOC reflectance measurements of theplots are listed in Plate 1.


Plate 1. Canopy reflectance spectra measured for the 21 vegetation plots in the Millingerwaard.

Figure 5. Detrended Correspondance Analysis (DCA) of the 21 vegetation samples in the Millingerwaardusing the CANOCO software (Ter Braak and Smilauer, 2002). Number of species: 79, Eigenvalues: �1 � 0.64, �2 � 0.58, �� 4.78; the plot therefore represents 26 percent of the variance in the speciesdata. Detrending performed by using second order polynomials. The species plot (a) shows the position ofthe 58 species whose weight is 5 percent of the maximum weight; after overlaying the species plot withthe sample plot (b), the Euclidian distance of a species to a sample is an inverse measure for theprobability to find a species in a plot. An explanation of the abbreviated species names is in Appendix A.

Ground and Airborne-derived LAIGround based hemispherical photographs of forest commu-nities were processed using the CAN_EYE® software (Baretand Weiss, 2006), and resulted in an estimation of both,effective and true LAI. LAI values were ranging from 4.7 to

PMSRS-07.qxd 9/14/07 11:22 PM Page 1184


Plate 2. Spatially distributed LAI of the Millingerwaard based on four different retrievalschemes: (a) Weighted Difference Vegetation Index (WDVI), (b) Reduce Simple Regression(RSR), (c) Vegetation Continuous Fields (FVC), and (d) Normalized Difference Vegetation Index(NDVI). LAI units are in m2/m2.

6.5 m2/m2 and 2.9 to 4.0 m2/m2 for true and effective LAIrespectively (see Plate 2). The simple kriging approachcomparing interpolated in situ LAI with HyMap-derived RSRLAI proved to be of very good quality (r2 � 0.88).

Linking Remote Sensing and the Dynamic Vegetation ModelThere appears to be a satisfying agreement between themeasured by clipping and weighing and modeled above-ground biomass, although at very low biomass, the simu-lated values are sometimes significantly lower than theactually measured ones (Figure 6). This may be due to anover-estimation of the stocking intensity (note that theplots with low biomass values are the grazed ones). Therelationship is improved, when relating the HyMap-derived LAI and the biomass (r2 � 0.61) while excludingthe grazed plots.

Next, we attempted to improve the biomass simulated bySUMO for 2050 by re-initializing SUMO in 2004 using the

HyMap derived biomass. Plate 3 shows the result for twoscenarios: continuation of the agricultural management andnatural succession; the latter with and without re-initializationin 2004.

Figure 7 shows that grazing also influences the relationbetween reflectance and biomass; the grazed plots appear tohave a much higher LAI at a given biomass compared to theun-grazed ones. Subsequently, these plots were excludedfrom our analysis because their low number prevented aseparate calibration. An explanation for the high LAI of thegrazed plots might be their N-content; a chemical analysisof plant material showed that its N-content was signifi-cantly higher (P � 0.001) in the grazed plots (2.0 percent)than in the un-grazed plots (1.3 percent). For other nutri-ents (P, K, Mg) theses differences were not significant.Apparently management (e.g. grazing, mowing) will alsohave to be taken into account if biomass is to be estimatedfrom reflectance.

PMSRS-07.qxd 9/14/07 11:22 PM Page 1185


Plate 3. Total Biomass simulated by SUMO for two plant functional types, under ascenario of fertilization and mowing (agricultural management) or grazing (naturemanagement), the latter both with (dotted) and without (drawn) re-initialization in2004 (“re-init”). The left y axis denotes the biomass of the herbs and grasses in tonsper hectare, the right vertical axis plots that of all other plant functional types (woodyspecies, also in tons/ha). The starting point of the dotted line represents the re-initialized value. The oscillations in the first years after (re-) initialization are dueto model instability.

Figure 6. Measured versus modeled abovegroundbiomass. Modeled biomass values taken for 2004 fromthe SMART2-SUMO run for natural succession; meas-ured values were determined by clipping and weighing.

Figure 7. Measured biomass for the plots versus HyMap-derived LAI. The regression line pertains to the un-grazedplots only.

ConclusionsOur results demonstrate that imaging spectrometer-derivedproducts can be used for the initialization of the dynamicvegetation model SMART2-SUMO at a regional scale. Inparticular the preprocessing using all relevant informationavailable increased the reliability of the input data, andtherefore the final result. The separation of general PFT intothe functional types herbs and grasses, dwarf shrubs, shrubs,and forest (trees) proved to have a major effect on the

simulations for the development of the vegetation. However,since imaging spectrometer data are not yet available on anextended, multi-temporal basis, the performance of the modelin predicting the temporal development can only be judgedby expert knowledge. After re-initialization, the increase ingrass and herb biomass over time decreased, while the woodybiomass increases more rapidly. This seems to agree betterwith the actual vegetation succession in the area, where scrubcover rapidly increases at the expense of grass cover.

PMSRS-07.qxd 9/14/07 11:22 PM Page 1186

An extensive regional-based study was carried out tolink imaging spectrometer products to a dynamic vegetationmodel. Due to the regional flavor of the work performed, itsscalability to national or even continental scales needs to beinvestigated. A refinement of the regional-based approachwill include in the future more accurate vegetation mappingusing advanced spectral libraries and directional leaf opticalproperties measurements. A PFT-based unmixing approach isproposed, weighting the abundances per pixel, allowing formultiple PFTs present in one pixel (in particular the mixtureof shrub and grassland was identified to be a challenge). Thedata driven estimate of the LAI-biomass relation should bereplaced by a quantitative, physical based approach (e.g.,inversion of a radiative transfer based approach), as well asestimating NPP-based on a model including the use of localclimatology. The detailed assessment of PFTs may also serveto estimate their spatial frequency distribution helping toestimate the ecosystem’s roughness for reducing and delay-ing flood peaks.

AcknowledgmentsThe airborne data acquisition was performed in the frame ofthe STEREO-I HyEco project. The support of the BelgianSpace Office (Belspo) and VITO (B) is acknowledged. Thecontribution of G. Schaepman-Strub is supported by anEuropean Space Agency (ESA) external fellowship. We thankthe reviewers for their substantial comments.

ReferencesBaret, F., and P. Rossello, 2006. VALERI – VAlidation of Land

European Remote sensing Instruments, INRA, URL:http://www.avignon.inra.fr/valeri/ (last date accessed: 27 June2007).

Baret, F., and M. Weiss, 2006. CAN_EYE – Processing digitalphotographs for canopy structure characterization, INRA, URL:http://www.avignon.inra.fr/can_eye/ (last date accessed: 27 June2007).

Baret, F., M. Weiss, S. Garrigue, D. Allard, J.P. Guinot, M. Leroy,H. Jeanjean, H. Bohbot, R. Bosseno, G. Dedieu, C. Di Bella,M. Espana, V. Gond, X.F. Gu, D. Guyon, C. Lelong, E. Mougin,T. Nilson, F. Veroustraete, and R. Vintilla, In review, VALERI:A network of sites and a methodology for the validation ofmedium spatial resolution land satellite products, RemoteSensing of Environment.

Brand, A.M., A. Aptroot, A.J. De Bakker, and H.F. Van Dobben,1988. Standaardlijst van de Nederlandse korstmossen, Wet MedKNNV, Utrecht.

Braun-Blanquet, J., 1951. Pflanzensoziologie, Grundzüge derVegetationskunde, Springer, Wien.

Chen, J.M., G. Pavlic, L. Brown, J. Cihlar, S.G. Leblanc, H.P. White,R.J. Hall, D.R. Peddle, D.J. King, J.A. Trofymow, E. Swift, J. Vander Sanden, and P.K.E. Pellikka, 2002. Derivation and valida-tion of Canada-wide coarse-resolution leaf area index mapsusing high-resolution satellite imagery and ground measure-ments, Remote Sensing of Environment, 80:165–184.

Clevers, J.G.P.W., L. Kooistra, and E.A.L. Salas, 2004. Study ofheavy metal contamination in river floodplains using the red-edge position in spectroscopic data, International Journal ofRemote Sensing, 25:3883–3895.

Cocks, T., R. Jenssen, A. Steward, I. Wilson, and T. Shields, 1998. TheHyMap airborne hyperspectral sensor: The system, calibration andperformance, Proceedings of the 1st EARSeL Workshop on ImagingSpectroscopy (M.E. Schaepman, D. Schlapfer and K.I. Itten,editors), EARSeL, Zurich (CH), pp. 37–42.

Feddema, J.J., K.W. Oleson, G.B. Bonan, L.O. Mearns, L.E. Buja,G.A. Meehl, and W.M. Washington, 2005. Atmospheric science:The importance of land-cover change in simulating futureclimates, Science, 310:1674–1678.

Ghavasieh, A.R., C. Poulard, and A. Paquier, 2006. Effect of roughenedstrips on flood propagation: Assessment on representative virtualcases and validation, Journal of Hydrology, 318:121–137.

Hu, Y.H., H.B. Lee, and F.L. Scarpace, 1999. Optimal linear spectralunmixing, IEEE Transactions on Geoscience and RemoteSensing, 37:639–645.

Jacquemoud, S., 2006. OPTICLEAF – The Database on Leaf OpticalProperties, University of Paris 7, URL: http://teledetection.ipgp.jussieu.fr/opticleaf/ (last date accessed: 27 June 2007).

Jacquemoud, S., and F. Baret, 1990. Prospect – A model of leaf optical-properties spectra, Remote Sensing of Environment, 34:75–91.

Jungwirth, M., S. Muhar, and S. Schmutz, 2002. Re-establishing andassessing ecological integrity in riverine landscapes, FreshwaterBiology, 47:867–887.

Kabat, P., W. Van Vierssen, J. Veraart, P. Vellinga, and J. Aerts,2005. Climate proofing the Netherlands, Nature, 438:283–284.

Keshava, N., and J.F. Mustard, 2002. Spectral unmixing, IEEE SignalProcessing Magazine, 19:44–57.

Kooistra, L., M.A.J. Huijbregts, A.M.J. Ragas, R.S.E.W. Leuven, andR. Wehrens, 2005. Spatial variability and uncertainty ecologicalrisk assessment: A case study on the potential risk of cadmiumfor the little owl in a Dutch river flood plain, EnvironmentalScience and Technology, 39:2177–2187.

Kooistra, L., R.S.E.W. Leuven, P.H. Nienhuis, R. Wehrens,L.M.C. Buydens, E.A.L. Salas, and J.G.P.W. Clevers, 2004.Exploring field vegetation reflectance as an indicator of soilcontamination in river floodplains, Environmental Pollution,127:281–290.

Kros, J., 2002. Evaluation of Biogeochemical Models at Local andRegional Scale, Ph.D. Thesis, Alterra Scientific Contributions 7,Alterra, Wageningen.

Mengesha, T., L. Kooistra, R. Zurita-Milla, S. De Bruin, andM. Schaepman, 2005a. Methodology comparison of quantitativeLAI retrieval using imaging spectroscopy and geo-spatial inter-polation in a softwood forest, Proceedings of the 4th Workshopon Imaging Spectroscopy (B. Zagajewski, M. Sobczak, andW. Prochnicki, editors), EARSeL, Warsaw, Vol. 1, pp. 141.

Mengesha, T., M.E. Schaepman, S. De Bruin, R. Zurita-Milla, andL. Kooistra, 2005b. Ground validation of biophysical productsusing imaging spectroscopy in softwood forests, Proceedings ofthe VALERI Workshop (F. Baret and M. Weiss, editors), INRA,Avignon, France, unpaginated CD-ROM.

Middelkoop, H., K. Daamen, D. Gellens, W. Grabs, J.C.J. Kwadijk,H. Lang, B.W.A.H. Parmet, B. Schädler, J. Schulla, and K. Wilke,2001. Impact of climate change on hydrological regimes andwater resources management in the Rhine basin, ClimaticChange, 49:105–128.

Nickel, H., and J. Hildebrandt, 2003. Auchenorrhyncha communitiesas indicators of disturbance in grasslands (Insecta Hemiptera) –A case study from the Elbe flood plains (northern Germany),Agriculture, Ecosystems and Environment, 98:183–199.

Peterson, J.T., and T.J. Kwak, 1999. Modeling the effects of land useand climate change on riverine smallmouth bass, EcologicalApplications, 9:1391–1404.

Richter, R., and D. Schlapfer, 2002. Geo-atmospheric processing ofairborne imaging spectrometry data, Part 2: Atmospheric/topo-graphic correction, International Journal of Remote Sensing,23:2631–2649.

Roujean, J.L., and R. Lacaze, 2002. Global mapping of vegetationparameters from POLDER multiangular measurements forstudies of surface-atmosphere interactions: A pragmatic methodand its validation, Journal of Geophysical Research D: Atmos-pheres, 107(D12):1261–1282.

Schaepman-Strub, G., M. Schaepman, T. Painter, S. Dangel, andJ. Martonchik, 2006. Reflectance quantities in optical remotesensing – Definitions and case studies, Remote Sensing ofEnvironment, 103:27–42.

Schaepman, M., R. Zurita-Milla, M. Kneubühler, J.G.P.W. Clevers, andS. Delwart, 2004. Assessment of long-term vicarious calibrationefforts of MERIS on land product quality, Sensors, Systems, andNext-Generation Satellites VIII (R. Meynart, S.P. Neeck, andH. Shimoda, editors), SPIE, Maspalomas, Vol. 5570, pp. 363–371.


PMSRS-07.qxd 9/14/07 11:22 PM Page 1187

Schaepman, M.E., G.W.W. Wamelink, H. van Dobben, M. Gloor,G. Schaepman-Strub, L. Kooistra, A. Schmidt, and F. Berendse,2005. Regional scale ecosystem modeling for vegetationscenario development – Demonstrated in a river floodplain(The Netherlands) using imaging spectroscopy, Proceedings ofthe 9th International Symposium on Physical Measurementsand Signatures in Remote Sensing (ISPMSRS) (S. Liang,J. Liu, X. Li, R. Liu, and M.E. Schaepman, editors), ISPRS,Beijing, Vol. XXXVI, pp. 667–670.

Schaminée, J.H.J., E.J. Weeda, and V. Westhoff, 1998. Plantenge-meenschappen van de kust en van binnenlandse pioniermilieus,Opulus Press, Uppsala – Leiden, 346 p.

Schlapfer, D., and R. Richter, 2002. Geo-atmospheric processing ofairborne imaging spectrometry data, Part 1: Parametric orthorecti-fication, International Journal of Remote Sensing, 23:2609–2630.

Schmidt, A., H. van Dobben, G.W.W. Wamelink, L. Kooistra, andM.E. Schaepman, 2005. HYECO’04: Using hyperspectralreflectance data to initialize ecological models, Imaging Spec-troscopy – New Quality in Environmental Studies (B. Zagajewski,and M. Sobczak, editors), EARSeL, Warsaw, Vol. 1, pp. 247–254.

Ter Braak, C.J.F., and P. Smilauer, 2002. CANOCO ReferenceManual and Canodraw for Windows User’s Guide: Software forCanonical Community Ordination, (Version 4.5), MicrocomputerPower, Ithaca, New York.

Touw, A., and W.V. Rubers, 1989. De Nederlandse bladmossen –Flora en verspreidingsatlas van de Nederlandse Musci (Sphag-num uitgezonderd), KNNV Uitgeverij, Utrecht.

van der Meijden, R., E.J. Weeda, W.J. Holverda, and P.H. Hovenkamp,1990. Heukel’s flora van Nederland, Wolters-Noordhoff, Groningen.

Van Geloof, I., and I. de Ronde, 2002. De vegetatie van de Millinger-waard na 10 jaar “natuurontwikkeling,” Wageningen University,Wageningen.

Verhoef, W., and H. Bach, 2003. Simulation of hyperspectral anddirectional radiance images using coupled biophysical andatmospheric radiative transfer models, Remote Sensing ofEnvironment, 87:23–41.

Wamelink, G.W.W., J.J. de Jong, H.F. Van Dobben, and M.N. VanWijk, 2005. Additional costs of nature management caused bydeposition, Ecological Economics, 52:437–451.

Wamelink, G.W.W., C.J.F. Ter Braak, and H.W. Van Dobben, 2003.Changes in large-scale patterns of plant biodiversity predictedfrom environmental economic scenarios, Landscape Ecology,18(5):513–527.

Weiss, M., F. Baret, M. Leroy, O. Hauteceur, C. Bacour, L. Prevot,and N. Bruguier, 2002. Validation of neural net techniques toestimate canopy biophysical variables from remote sensing data,Agronomie, 22:547–554.

Appendix A: Explanation of the abbreviated species namesin Figure 5.Achilmil � Achillea millefolium; Agrossto � Agrostisstolonifera; Arctilap � Arctium lappa; Brassnig � Brassicanigra; Bromuine � Bromus inermis; Calamepi � Calama-grostis epigejos; Calyssep � Calystegia sepium; Cardunut� Carduus nutans; Carexhir � Carex hirta; Cerasfon� Cerastium fontanum; Cirsiarv � Cirsium arvense; Cirsivul � Cirsium vulgare; Cynoddac � Cynodon dactylon;Dactyglo � Dactylis glomerata; Elymurep � Elymus repens;Epilohir � Epilobium hirsutum; Epilotet � Epilobiumtetragonum; Erigecan � Erigeron canadensis; Eryngcam� Eryngium campestre; Euphoesu � Euphorbia esula;Festurub � Festuca rubra; Galeotet � Galeopsis tetrahit;Galiuapa � Galium aparine; Galiumol � Galium mollugo;Glechhed � Glechoma hederacea; Hernigla � Herniariaglabra; Loliuper � Lolium perenne; Lycopeur � Lycopuseuropaeus; Lythrsal � Lythrum salicaria; Matrimar � Matricaria maritima; Mediclup � Medicago lupulina;Melilalt � Melilotus altissima; Menthaqu � Mentha aquatica;Odontver � Odontites vernus; Oenotbie � Oenotherabiennis; Phleupra � Phleum pratense; Plantlan � Plantagolanceolata; Poa ann � Poa annua; Poa pra � Poa pratensis;Poa tri � Poa trivialis; Polynper � Polygonum persicaria;Potenans � Potentilla anserina; Potenrep � Potentillareptans; Ranunrep � Ranunculus repens; Rubuscae� Rubus caesius; Rumexace � Rumex acetosa; Rumexcri� Rumex crispus; Rumexobt � Rumex obtusifolius;Saponoff � Saponaria officinalis; Senecina � Senecioinaequidens; Senecjac � Senecio jacobaea; Solidcan� Solidago canadensis; Stellaqu � Stellaria aquatica;Tanacvul � Tanacetum vulgare; Taraxoff � Taraxacumofficinale s.s.; Triforep � Trifolium repens; Urticdio� Urtica dioica; Verbanig � Verbascum nigrum.)


PMSRS-07.qxd 9/14/07 11:22 PM Page 1188

River Floodplain Vegetation Scenario Development Using Imaging Spectroscopy Derived Products as Input Variables in a Dynamic Vegetation Model

Documents