Geolocation Assessment of MERIS GlobCover Orthorectified Products

2972 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011

Geolocation Assessment of MERIS GlobCoverOrthorectified Products

Patrice Bicheron, Virginie Amberg, Ludovic Bourg, David Petit, Mireille Huc, Bastien Miras, Carsten Brockmann,Olivier Hagolle, Steve Delwart, Franck Ranéra, Marc Leroy, and Olivier Arino

Abstract—The GlobCover project has developed a service ded-icated to the generation of multiyear global land cover maps at300-m spatial resolution using as its main source of data thefull-resolution full-swath (300 m) data (FRS) acquired by theMERIS sensor on-board the ENVISAT satellite. As multiple singledaily orbits have to be combined in one single data set, an accu-rate relative and absolute geolocation of GlobCover orthorectifiedproducts is required and needs to be assessed. We describe in thispaper the main steps of the orthorectification pre-processing chainas well as the validation methodology and geometric performanceassessments. Final results are very satisfactory with an absolutegeolocation error of 77-m rms and a relative geolocation error of51-m rms.

Index Terms—Image registration, land surface, spatialcoherence.

I. INTRODUCTION

W IDE field-of-view sensors such as the AdvancedVery High Resolution Radiometer (AVHRR),

VEGETATION, MODerate Imaging System (MODIS),Multiangle Imaging Spectro-Radiometer (MISR), and MediumResolution Imaging Spectrometer (MERIS) provide a neardaily global coverage of the earth with an appropriate resolutionto derive land cover or land cover change. These main sourcesof data have been used widely in the past to produce landcover maps with AVHRR at 8 km [1] and 1 km [2] spatialresolution (or ground sampling distance). Unfortunately, theproduct quality is often limited by the rather poor geometricaccuracy of the data. Significant progress in terms of geometric

Manuscript received April 5, 2009; revised December 24, 2009,August 28, 2010, and January 14, 2011; accepted January 22, 2011. Date ofpublication May 23, 2011; date of current version July 22, 2011. The MEDICISCNES tool has been developed with the support of the French company CS(Communication and System).

P. Bicheron, M. Huc, B. Miras, and M. Leroy are with thePOSTEL/MEDIAS-France, CNES-BPI 2102, 31401 Toulouse Cedex 9,France (e-mail: [email protected]; [email protected];[email protected]; [email protected]).

V. Amberg and D. Petit are with the MAGELLIUM, 31521Ramonville Saint Agne, France (e-mail: [email protected]; [email protected]).

L. Bourg is with the ACRI-ST, 06904 Sophia-Antipolis Cedex, France(e-mail: [email protected]).

C. Brockman is with the Brockmann-Consult, D-21502 Geesthacht,Germany (e-mail: [email protected]).

O. Hagolle is with the CNES/CESBIO, 31401 Toulouse Cedex 9, France(e-maiL: [email protected]).

S. Delwart is with the ESA/ESTEC, 2201 AZ Noordwijk, The Netherlands(e-mail: [email protected]).

F. Ranéra and O. Arino are with the ESA/ESRIN, 00044 Frascati, Italy(e-mail: [email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TGRS.2011.2122337

performance has been made more recently at 1 km withMODIS [3] or VEGETATION/GLC-2000 product [4], [5]. TheGlobCover product from the European Space Agency (ESA)goes beyond this with a land cover map using as its main sourceof data the full resolution (300 m) mode data (FRS) acquiredover the years 2005 and 2006 by the MERIS sensor on-boardthe ENVISAT satellite and a service capable of reproducingthis product on a multiyear scale.

Identification of different land cover types at global scalegenerally uses the seasonal characteristics of vegetation growththat can be captured by the temporal dynamics of spectral infor-mation acquired by the wide field-of-view sensors. However, anumber of physical effects such as cloud and atmospheric con-tamination and surface anisotropy require compositing multipledaily orbits into a single data set [6], [7]. Achieving a high levelaccuracy relative geolocation is therefore a critical step for eachorbit. In addition, even if absolute geolocation accuracy is notneeded in principle for such compositing, the use of the outputproducts with a geographical scope is strongly limited and sub-ject to additional errors, such as mislocation of control points[8], if the absolute geolocation accuracy is poor. Therefore,major efforts are made in geometric correction and the assess-ment of geolocation accuracy whatever the sensors—AVHRR[9], [10], ATSR [11], [12], VEGETATION [13], POLDER [14],MODIS [15], [16], MISR [17], [18], WindSat [19], SSM/I [20].The impact of mis-registration effects has also been studied oncomposited data [21], [22] as well as on land cover [23], [24]and land cover change [25], [26].

The MERIS mission has been designed with a primaryobjective to better understand the role of oceans and oceanproductivity in the climate system and as such the absolute geo-location accuracy specifications for MERIS was set to 2000 m.Thanks to its 15 spectral bands with a high radiometric res-olution in the optical domain and its 300-m spatial resolu-tion, MERIS also offers great opportunities for observationover terrestrial surfaces. However, the geolocation performanceneeds to be significantly improved in that case. This wasachieved by introducing a regular delivery of the ENVISATattitude parameters, as measured by the on-board software, andcorrecting for the remaining pointing biases. Simultaneously,short studies [27], [28] were performed to verify the geometricperformance over specific time periods and these showed thatabsolute geolocation root mean square errors stayed within theranges of 170 m up to 500 m. All the aforementioned geometricperformance assessments were based either on the originalnavigation and resampling (nearest neighbor) model of MERIS,or on limited scenes sampling, and not on a systematic

0196-2892/$26.00 © 2011 IEEE

BICHERON et al.: GEOLOCATION ASSESSMENT OF MERIS GLOBCOVER ORTHORECTIFIED PRODUCTS 2973

projection grid. To improve this, an Accurate MERIS Ortho-Rectified Geolocation Operational Software (AMORGOS) [29]was developed by ESA to provide precise geolocation informa-tion for every image pixel of the MERIS Full Resolution prod-uct (FR) using a Digital Elevation Model (DEM) named GlobalEarth Topography And Sea Surface Elevation at 30 arc secondresolution (GETASSE30) [30]. As the GlobCover processingintegrates the AMORGOS tool coupled with a cartographicprojection system taking into account the local elevation, anextensive study must be achieved to assess the performance ofthis approach.

In this paper, we estimate the absolute and relative geo-metric accuracy of GlobCover products. Section II providesan overview of the baseline processing of MERIS level 1B.The orthorectification modules are then described including theAMORGOS geolocation and cartographic projection modules.The validation process is presented in Section III. Absoluteand relative geometric accuracies of GlobCover products arerequested to be better than 150 m (i.e., half a pixel) so as todeliver a final land cover map of high quality. Our objective istherefore to evaluate whether these requirements are fulfilled ona global scale. These assessments are performed using disparitymeasurements in column and line shifts between the orthorec-tified images and reference independent images acquired oversites located at different latitudes, at different times and withdifferent topographies and cloud covers. The validation processis performed by an independent team of the GlobCover projectdifferent from the group responsible for the production of theorthorectified images. Section IV describes the results of therelative and absolute geometric accuracy using different simplestatistic criteria.

II. MERIS GLOBCOVER ORTHORECTIFICATION

A. Description of the MERIS Level 1B Processing

On-board ENVISAT launched in 2002, MERIS is a widefield-of-view push-broom imaging spectrometer measuring thesolar radiation reflected by the earth in 15 spectral bands from412.5 nm to 900 nm. Each of these 15 bands is programmablein position and in width. The instrument has a field of viewof 68.5◦ and covers a swath width of 1150 km at a nominalelevation of 800 km enabling a global coverage of the earth in3 days. The wide field-of-view is shared between five identicaloptical cameras arranged in a fan shape configuration, eachcamera covering a 14◦ field of view with a slight overlap(see Fig. 1). An image is constructed using the push-broomprinciple: a narrow strip of the earth is imaged onto the entranceslit of the spectrometer, defining the across track dimension andthe motion of the satellite provides the along track dimension.The spectral dimension is achieved by imaging the entrance slitof each spectrometer via a dispersing grating onto a 2-D CCD.

The MERIS instrument resolution in full spatial resolution(FR) is 290 m (along track) × 260 m (across track) at nadir.Data at a coarser resolution are systematically generated on-board by spatially (across-track) and temporally (along-track)averaging a group of 4 × 4 pixels producing a Reduced spatialResolution (RR) data set with a 1160 m by 1040 m resolution.

Fig. 1. MERIS sensor: FOV, camera tracks, pixel enumeration, and swathdimension.

The RR data are transmitted to ground on a global basis whereasthe FR data are limited to regional coverage, focusing on landsurfaces and coastal areas.

The level 1B MERIS full-resolution full-swath (FRS) prod-uct contains calibrated top of atmosphere gridded radiancesover the full sensor’s swath. The radiometric processing [31]includes several steps, namely detection of saturated pixels,stray light correction and estimation of spectral radiances. Thegeolocation processing is split in five steps (product limits, tiepoints on earth location, elevation retrieval, re-sampling, andsun glint) which are summarized below.

Due to the sharing of the field-of-view by five identicalcameras, there is no spatial continuity in the data acquired bythe instrument across track: the slight overlap between adjacentcameras, as well as the slight inter-camera misalignment, re-quires spatial re-construction to be provided for the users withspatially continuous and regularly sampled MERIS products, atlevel 1 or higher. This re-construction product grid is based onan ideal instrument acquisition grid: the along-track samplingis the actual instrument one while the across-track sampling isdefined as perpendicular to the satellite track and evenly spacedon-ground.

The MERIS product grid is computed from satellite naviga-tion and attitude data allowing computation of the intersectionof the instrument field-of-view with the earth surface repre-sented by the WGS 84 reference ellipsoid at zero elevation. Theretrieved instantaneous field-of-view swath is sampled with aconstant distance on-ground to build the product pixels (Fig. 2).Correspondence with instrument pixels can then be done on thebasis of the across-track pointing angle. In the nominal MERISprocessing, this is performed only on a sub-set of the productpixels, called the tie points so as to reduce the product size. Thetie points grid has 71 tie points across track. It corresponds to a16 × 16 sub-grid of the RR product grid and to a 64 × 64 sub-grid of the FR product grid. At tie points, the geolocation data(longitude λ, latitude φ) is complemented with illumination andobservation angles (θs, ϕs, θv, ϕv) and meteorological infor-mation. Elevation above the reference ellipsoid is also providedwith a first order parallax correction terms due to this elevation.The regular time sampling of the measurements in the along


Fig. 2. MERIS level 1B tie points grid along the ENVISAT orbit.

track direction provides only a quasi-even distance on earth.Variations can be up to 3% and are due to the orbital motion ofthe satellite and the ellipsoidal shape of the earth, the across-track inter-pixels distance being regular by definition.

It must be noted here that since the level 1B product grid isfilled by a nearest neighbor method from the instrument gridwith a slight spatial over-sampling, the same instrument samplecan be found several times in the same level 1B product (it isthen identified as a ‘duplicate’ pixel within the level 1B productflags).

B. Geolocation and Cartographic Projection Modules

The geolocation processing described in previous sectionwas developed to fulfill mission requirements aimed at theocean community i.e., 2000 m. Such requirements are clearlynot sufficient for land applications and therefore ESA hasput some efforts to improve the on-board attitude processingsoftware as well as geolocation monitoring. A slow degrada-tion in the MERIS absolute geolocation was observed beforeDecember 2003—the mean error was about 500 m—mainly inthe across-track direction [28]. On December 2003, the changein on-board attitude processing software resulted in an immedi-ate improvement of the geolocation to 270 m. A modification ofthe MERIS pointing auxiliary data took place on January 2005which further improved geolocation performance of standardproducts to about 170 m [27]. In parallel, ESA initiated thedevelopment of the AMORGOS tool whose purpose was toprovide accurate per-pixel geolocation information—longitudeλ, latitude φ, elevation h—accounting for the earth surfaceelevation and actual satellite navigation and attitude control,generally not available at the time of near-real time MERIS dataprocessing. Using MERIS FRS products as input, the MERISfull-swath geo-located (FSG) products are generated with theAMORGOS tool as follows:

1) identify the original instrument pixel, on the basis of FRSproduct information and of the auxiliary data used duringthe re-sampling step of the FRS product generation,

2) compute satellite location and actual attitude at acquisi-tion time,

3) using the instrument pixel’s characterized pointing direc-tion, follow its line-of-sight until it intersects the earth’ssurface, represented by the DEM GETASSE30 on top ofthe reference ellipsoid (Fig. 3). This elevation model is acomposite data set using the SRTM30 data set [32], ACEdata set [33], Mean Sea Surface (MSS) data [34] and theEGM96 ellipsoid [35].

Fig. 3. AMORGOS geolocation principle and parallax illustration: intersec-tion between the line of sight and the GETASSE 30 DEM. αmax, P0, and P1correspond respectively to the maximum field of view (34.25◦), the MERISstandard products geolocation, and the AMORGOS geolocation.

The main differences between the MERIS standard productsgeolocation and the AMORGOS computed geolocation are:

1) satellite ephemeris and attitude are re-computed frombest possible quality sources to ensure best achievableaccuracy,

2) information (longitude λ, geodetic latitude φ and eleva-tion h) are provided for each image pixel,

3) pixel location is derived taking into account the actualearth surface elevation along the viewing direction,

4) information is retrieved according to the original instru-ment pixel, regardless of the image re-gridding.

The MERIS FSG images are then projected on a cartographicsystem. The plate–carrée coordinate reference system (CRS)has been chosen as it is the most commonly used projectionfor land cover product. The reference ellipsoid is WGS 84assuming, respectively an equatorial (Re) and polar radius (Rp)equal to 6378137 m and 6356752.3 m. The grid cells have anangular pixel resolution Resdeg and a spatial size defined by itsheight and width

height = r × 2π/360×Resdeg

width =Re × cos(φ)× 2π/360×Resdeg

with r =Re ×Rp√

R2e × sin2(θc) +R2

p × cos2(θc)

and θc = arctan(1− f)2 × tan(φ)

with f =(Re −Rp)

Reand Resdeg = 1/360. (1)

The general concept used for the image georefencing hasalready been described in the literature [9], and [36]. And here,one must consider:

— a direct geolocation function f(l, p, h) producing thegeographical coordinates (λ, φ) in the reference systemassociated to a cell of the georeferenced image for anypoint located in the raw image by its line l, column p andelevation h. In the MERIS case, it is derived through theAMORGOS tool.


Fig. 4. Determination of line and column shifts with the MEDICIS tool.

— a reverse geolocation function f−1(λ, φ) applied to everycell of geographic coordinates determining at which rawimage line l and column p the cell is imaged.

As the grid of MERIS FSG product is not evenly spacedin angle, the reverse function is not strictly defined and mayonly be predicted. We use two 4th degree polynomial functionsL(λ, φ) and P (λ, φ) linking line l and column p of any pixelfrom the MERIS FSG image as a function of latitude φ andlongitude λ. The coefficients of these predictive polynomialfunctions are determined over a sub-sampling of the MERISFSG grid through an equation system built considering onepoint out of ten and solved using a least-squares minimisationprocedure. As recommended in [37], [38], these functionsare computed on two grids of constant elevation Hmin andHmax representing, respectively the minimum and maximumelevation value on the DEM.

The retrieval of the pixel line l and column p in the MERISFSG image of the current cell (λ, φ) is done through an iterativeapproach. First, an estimation of line and column is processedusing the reverse location function defined by the predictorpolynomial functions L(λ, φ) and P (λ, φ). For the estimatedline and column, the corresponding latitude and longitude areinterpolated through a bilinear sampling over a 2 × 2 pixelsneighborhood in the MERIS FSG image. These geographicquantities are then used for a new estimate of the line andcolumn of the current cell. Successive iterations are repeateduntil a predefined tolerance is reached. The final estimate ofline l and column p at the current elevation h is derived froma linear interpolation between the results obtained at Hmin

and Hmax. In case the polygon crosses the meridian ±180◦,the continuity in longitude is ensured adding a modulo 360◦.The final radiometry DN(l, p) is computed with a bi-cubicinterpolation algorithm over a 4 × 4 pixels neighborhood.

III. VALIDATION PROCESS

A. Methodology

Two kinds of validation are performed: 1) verificationof co-registration accuracy of orthorectified MERIS images

(or relative geolocation accuracy); 2) verification of absolutegeolocation accuracy. In both cases, the validation is performedby comparing the ortho-rectified MERIS images with somegeo-located reference data. For each MERIS-reference datapair, the methodology is based on an automatic selection ofsampling points (thousands of points by pair) on a regular gridwith tuneable sampling rate. This detection is performed bycomputing similarity measurements between the two images.Quantitative assessments of relative and absolute geolocationsare then performed in terms of disparity measurements (col-umn and line shifts) between sampling points of the twoimages.

The correlation measurement is performed using theMEDICIS CNES correlation tool. For each sampling point,the principle is to compute a similarity measurement betweena master image (or reference image) and a slave image, thatis translated by incremental steps with respect to the masterimage (Fig. 4). Deformations applied by MEDICIS are onlytranslations here. Shifts in line and column are estimated in alocal window (50 × 50 pixels) centered on each correlationpoint by evaluating in an iterative process the translation thatmaximizes the similarity criterion between the master imageand the slave image [39], [40]. For our study, we have chosenthe standard correlation as the similarity criterion.

The disparity measurement between two images is generallyperformed on a great number of sampling points. To assure theaccuracy of the disparity estimate, several types of pixels are nottaken into account in the final statistics. First, ocean flagged pix-els are removed considering that the similarity measurementsover ocean are often not accurate due to the target temporalvariability between two acquisitions and its lack of texture. In-valid flagged pixels, including especially clouds present in bothimages, are not taken into account. Sampling points with a lowsimilarity (i.e., low confidence in disparity measurements) arealso excluded from the final statistics: the accuracy of similaritymeasurements between two images decreases when changes ofthe target have occurred between the two acquisitions. Globalshifts between the images are finally computed by averaginglocally measured shifts obtained on all sampling points afterthe removal of outliers described above. This averaging reduces


the correlation bias and produces accurate shift measurementestimates.

For relative geolocation accuracy assessment, the referenceproducts are naturally MERIS images acquired over the samearea at a different time. For absolute geolocation accuracy as-sessment, the reference images are data produced by the ETM+multi-spectral sensor on-board LANDSAT 7. The resolution is30 m and only band B4 (760–900 nm) is used. The choice ofB4 is made to ensure a maximum spectral overlap with the865 nm spectral channel B13 of MERIS and to maximize thesimilarity between the images [41]. LANDSAT images havebeen geolocated and orthorectified with RMSE accuracy in-ferior to 50 m [42], [43]. As LANDSAT products have a higherspatial resolution than the MERIS products, LANDSAT imageshave been rescaled to 300 m (MERIS products resolution) byspatial averaging. Thus absolute validation is performed at theMERIS 300-m resolution.

Two main advantages of such a study can be highlighted:first, a high precision of disparity measurements is provided byMEDICIS. In fact, the intrinsic correlation error of MEDICIS(internal CNES studies) is evaluated to 0.025 pixels (versus0.3 pixel at best for a manual selection of sampling point). Sec-ond, the estimate of the disparity between images is made veryaccurate by averaging on a high number of sampling points.In our study, the sampling rate is 10 pixels corresponding toapproximately 150 000 sampling points per pair automaticallyselected and evaluated. Since ocean pixels, invalid pixels orsampling points with a low similarity rate are excluded, onlyabout 20% are kept in the final statistics.

The disparity measurements provided by the correlation toolare produced by several error contributions distinguishableaccording to the assessment case. For relative and absolutelocation error assessments, the correlation error of MEDICISmust be considered. In the case of relative error assessment,we must add the co-registration error of MERIS data. In thecase of absolute error assessment, we must take into account theregistration error of MERIS data with respect to the referencedata as well as their own geolocation error (< 50 m). Thegeolocation precision is directly derived from the mean shiftmeasurements in line and column.

B. Reference Images Selection

Geolocation errors can be induced theoretically by severalphenomena such as measurement errors of satellite ephemerisand attitude, lack of DEM precision when processing geolo-cation and projection, and instrumental drifts. As MERIS iscomposed of five independent CCD cameras observing fiveadjacent areas, some independent geolocation errors can alsooccur between the different parts of the scene imaged by eachsensor. This last source has not been investigated in our study.Finally, the test sites have been carefully selected according to:

• Different latitudes and observation dates for the monitor-ing of the temporal instrumental drifts in latitude.

The measurements quality of MERIS instrument have beenrigorously studied and regularly monitored in flight [29]. How-ever, some instrumental drifts could occur in latitude and intime, principally because of thermo-elastic effects of the solar

illumination on the satellite. Such instrumental drift effectson the geolocation accuracy can be highlighted by comparingrelative and absolute validation results obtained on sites locatedat different latitudes and products acquired at different dates.

• Different topographies for the study of the potential reliefinfluence.

Geolocation errors induced by topography can be highlightedby looking for some relations between these errors and theelevation. This study can be performed by computing the jointprobability between geolocation error and elevation. From thisprobability, it is then possible to analyze the geolocation error asa function of elevation and then to highlight a possible relationbetween topography and geolocation errors. A faster approachconsists of comparing geolocation accuracy results obtained onseveral sites presenting different relief (desert areas, flat lands,mountainous areas, etc.).

• Cloud detection quality and data availability.Cloud pixels are not taken into account in the similarity

process (see Section III-A). Thus the selected images mustpresent as few clouds as possible to keep enough samplingpoints for the average of accurate disparity estimates. Sinceseveral images of different sites and for different time periodsmust be selected, sites with a high revisit rate must be favored.

• Sites taken at different longitudes.This aims at verifying that sites located at different longi-

tudes present the same geolocation error.For what concerns the determination of the relative geo-

location accuracy, about five images per site and per timeperiod are selected providing a total amount of about100 orthorectified MERIS images—5 images × 5 sites ×4 time periods—representing theoretically 200 MERIS pairs.For each period, the images must be acquired on differentorbits. The maximum common area between two images is10◦ × 10◦ due to GlobCover tiling [44]. For each site, thefour following time periods have been selected over the timewindow 2004 and 2005:

• from December 1st to 15th and from January 1st to 7thfor the winter season taking the northern hemisphere asreference,

• from April 8th to 30th for spring,• from June 1st to 30th for summer,• from October 1st to 21th for autumn.According to the criteria outlined above, five sites

have been chosen: Madagascar, Tunisia, Spain–Morocco,Romania–Ukraine, Poland-Sweden. Table I represents the num-ber of selected MERIS images and derived pairs per site thathave been successfully geo-located and orthorectified. Finally,146 pairs out of a maximum of 200 pairs have been computedwhich is deemed sufficient for the relative geolocation accuracyestimate. Note that we were not able to select the full initial setof 5 MERIS images for each site and season for the followingreasons:

• For some sites and seasons (e.g., northern Europe in winteror autumn), it is very difficult to select five images withlimited cloud cover.

• At the time of this study, the data collection for the year2005 was not completely delivered. Particularly, the winter


TABLE ISELECTED DATASET

TABLE IIGLOBAL RELATIVE ACCURACY RESULTS (UNIT: METER)

Fig. 5. Global relative accuracy results (unit: meter).

season (January, February, and March) was very poorlycovered.

IV. RESULTS

A. Relative Geolocation Accuracy Results

1) Global Results: Each pair is built by associating oneorthorectified MERIS image for one site and one season withanother image present for the same site at the same season butnot on the same day. For each pair, quality criteria (arithmeticmeans, standard deviations, and RMSE errors in latitude andlongitude as well as the global RMSE) have been estimatedby averaging results computed with the MEDICIS tool on allsampling points: it results in 146 disparity measurements. On

this experimental material, global statistics have been computedas described below and summarized in Table II:

• Mean features represent the mean value of the 146 esti-mated quality criteria.

• Standard deviation features represent the standard devia-tion of the 146 estimated quality criteria.

• Minimum features represent the minimum value of the146 estimated quality criteria.

• Maximum features represent the maximum value of the146 estimated quality criteria.

The mean RMSE total has a value of 51.6 m with valuesin latitude and longitude quite similar showing that no specificproblem seems to exist in one of these directions. This meansthat the GlobCover requirement of 150 m is globally satisfiedconcerning the co-registration quality of FR MERIS products.Standard deviations of RMSE are satisfactory with valuesabout 20 m (i.e., small) in latitude and longitude. Moreover,RMSE obtained on the whole data set are very similar. Wemay therefore be confident in the assessment of the relativegeolocation accuracy.

Fig. 5 represents the RMSE in longitude as a functionof RMSE in latitude for the 146 pairs: each symbol repre-sents a pair. The bold semi-circle represents the geolocationrequirement set at the beginning of the GlobCover project:pairs located inside the bold semi-circle satisfy the geometricproperty requirements (i.e., global RMSE < 150 m). Thisrepresentation shows that only 3 pairs out of the 146 selecteddo not satisfy the GlobCover specifications for the MERISimage co-registration: this is a very satisfactory result. Thedashed semi-circles correspond to a RMSE value of 100 m.Only 12% of the pairs are located between 100 and 150 m.Furthermore, these pairs correspond to desert areas on which itis difficult to perform accurate similarity measurements because


Fig. 6. Influence of time on the relative geolocation results. A symbolrepresents a result for one pair of MERIS images.

of the lack of features. If these pairs are removed from the dataset, the mean RMSE decreases significantly to a mean valueof 35 m.

2) Influence of Time, Latitude, and Topography: In this sec-tion, we analyze the behavior of MERIS images co-registrationerrors with time, latitude and topography. To highlight a pos-sible degradation of co-registration accuracy with time, thetemporal evolution of the RMSE in latitude and longitude isrepresented for the 5 sites (Fig. 6). As expected, the relativegeolocation accuracy over the desert site of Tunisia is not asgood as those obtained on the other sites. When considering allsites, the co-registration errors of MERIS images do not showany increase with time: no relation between seasons and errorsis identified. However, RMSE differences do exist within eachseason. This may arise from directional effects generating ra-diometric differences that are not corrected in daily images. The

Fig. 7. Influence of latitude on the relative geolocation results. A symbolrepresents a result for one pair of MERIS images.

similarity measurement may be exaggerated by shifting artifi-cially successive images so as to counterbalance these effects,all the same, the RMSE magnitudes stay globally within theGlobCover specifications. Potential radiometric instrumentaldrifts in time have no effect on co-registration accuracy ofMERIS images.

Potential latitude dependant co-registration accuracy hasbeen analyzed by representing the evolution of the same cri-teria with latitude (Fig. 7). The main conclusion is that theco-registration errors of MERIS images for the 4 seasons


Fig. 8. Influence of elevation on the RMSE (unit: meter) for the relativegeolocation results obtained over Madagascar site.

TABLE IIIGLOBAL ABSOLUTE ACCURACY RESULTS WITH

LANDSAT PRODUCTS (UNIT: METER)

Fig. 9. RMSE in latitude as a function of RMSE in longitude for the absolutegeolocation assessment (unit: meter).

do not vary with this parameter: no relation between lati-tude and errors seems to exist. Thus potential instrumentaldrifts in latitude show no effect on co-registration accuracy ofMERIS products. Over all seasons, the difference of relativevalidation results between the site of Tunisia and the other

Fig. 10. Arithmetic mean (shifts) in latitude and in longitude for absolutegeolocation accuracy with LANDSAT products (unit: meter). The squaresymbol corresponds to the mean value.

sites, as already observed in the previous sections, is clearlyvisible.

The possible influence of the topography on co-registrationaccuracy of MERIS images is represented on Fig. 8. Thesite of Madagascar has been selected because of its complextopography. For a single pair of images, local disparity mea-surements of all sampling points located within an interval of10 m of elevation are averaged so as to relate the RMSE toelevation. Elevation is related negatively to geolocation error(higher elevation has smaller error). At the same time, the noiseincreases with elevation since similar RMSE levels are some-times observed at high elevation (between 1250 and 2000 m)and low elevation (< 500 m). These results are not surprising.First, steeper slopes are generally present at higher elevationgenerating stronger geolocation errors [23]. Secondly, RMSEis probably estimated less accurately at high elevation than atlow elevation. In fact, for the tested site, less sampling points arelocated at high elevation than at low elevation resulting in fewerresults on average for high altitude. Nevertheless, the RMSElevel remains within the 150 m requirement and influence ofelevation on geolocation error is therefore acceptable. Finally,residual thin clouds may be present in both images around thecorrelation point but at different locations. The RMSE peakobserved at 1300 m represents an example of this.

B. Absolute Geolocation Accuracy Results

In this section, a cross-validation of FR MERIS imagesgeolocation accuracy is done using LANDSAT images. Thispart of the study is not done on the whole data set. Only imagesacquired in summer over Spain and Madagascar are used lead-ing to 10 disparity measurements (one for each pair). For allthe pairs of LANDSAT and MERIS images, the same qualitycriteria as Section III-A have been estimated by averaging localresults computed on all sampling points.

Table III summarizes the results obtained on the 10 pairs.The global RMSE has a value of 77.1 m well below the 150 m


requirement. Fig. 9 represents the RMSE in longitude asa function of latitude for the 10 pairs. All pairs are locatedinside the semi-circle of geolocation validity required by theGlobCover project. A shift in longitude is observed with a meandifference higher in longitude than in latitude. This remark isconfirmed by Fig. 10 representing the mean difference in longi-tude as a function of latitude. On this diagram, the scatter plot isalmost centered on zero in latitude but not in longitude. It is alsoseen in Table III that extreme values of RMSE are uncommonlymore tightened than for the the relative accuracy assessment(104.2 m for the maximum value versus 214.4 m). First, thenumber of disparity measurements is here significantly lowerwith only two sites over a single season. In addition, LANDSATimages do not cover entirely MERIS images resulting in a lowernumber of sampling points per pair.

V. CONCLUSION

The co-registration and absolute accuracies of orthorectifiedMERIS GlobCover products are in agreement with specifica-tions i.e., well below the 150 m requirement. This level ofaccuracy could only be achieved with the recent improvementsof the MERIS pointing characterization and attitude processingsoftware as well as the development of additional processingmodules. First of them, the AMORGOS tool provided geolo-cation information for every image pixel whereas it was onlyavailable at tie points from the MERIS FRS products. Thecartographic projection tool allowed for the computation ofsurface reflectances in a common grid and for all the MERISFRS products used.

For what concerns the assessment method, we have demon-strated that a very accurate control points positioning has beenperformed with the MEDICIS tool. For each MERIS image,a large number of sampling points (typically many thousands)have been selected ensuring an accurate estimate of the absolutegeolocation and co-registration accuracy of MERIS images.The study has been done on numerous pairs (146 for the relativevalidation). No effect on geolocation generated by instrumentaldrifts in time and latitude has been identified. This study alsoshowed that the influence of elevation on the geolocation accu-racy remains acceptable implying that the use of GETASSE 30DEM is therefore sufficient. This result may be compared withsimilar results previously obtained [36] where orthorectificationmade with SRTM or another DEM at a better spatial resolutiongave similar results. Finally, the respective relative and absolutegeolocation accuracies of 51.6 m and 77.1 m are well withinthe GlobCover requirements. The AMORGOS tool coupledwith a cartographic projection system integrating the localelevation may be recommended as a standard for the MERISimage processing so as to enhance the development of usefulapplications over terrestrial ecosystems.

ACKNOWLEDGMENT

We are very grateful to the European Union Joint ResearchCentre for having provided the LANDSAT products used in thisstudy. P. Bicheron also thanks Judy Wallace for having carefullyread this paper.

REFERENCES

[1] R. Defries, M. Hansen, and J. R. G. Townshend, “Global discriminationof land cover types from metrics derived from AVHRR pathfinder data,”Remote Sens. Environ., vol. 54, no. 3, pp. 209–222, Dec. 1995.

[2] M. Hansen, R. DeFries, J. R. G. Townshend, and R. Sohlberg,“Global land cover classification at 1 km resolution using a decisiontree classifier,” Int. J. Remote Sens., vol. 21, no. 6, pp. 1331–1365,Apr. 2000.

[3] M. A. Friedl, D. K. McIver, J. C. F. Hodges, X. Y. Zhang, D. Muchoney,A. H. Strahler, C. E. Woodcock, S. Gopal, A. Schneider, A. Cooper,A. Baccini, F. Gao, and C. Schaaf, “Global land cover mapping fromMODIS: Algorithms and early results,” Remote Sens. Environ., vol. 83,no. 1/2, pp. 287–302, Nov. 2002.

[4] E. Bartholomé and A. Belward, “GLC2000: A new approach to globalland cover mapping from Earth observation data,” Int. J. Remote Sens.,vol. 26, no. 9, pp. 1959–1977, May 2005.

[5] S. Fritz, E. Bartholomé, A. Belward, A. Hartley, H. J. Stibig, and H. Eva,“Harmonization, mosaicking, and production of the global land cover2000 database,” Office Official Publ. Eur. Communities, Luxembourg,Germany, EUR 20849, 2003.

[6] B. Holben, “Characteristics of maximum value composite images fromtemporal AVHRR data,” Int. J. Remote Sens., vol. 7, no. 11, pp. 1417–1434, Nov. 1986.

[7] M. Leroy and J. L. Roujean, “Sun and view angle corrections on re-flectances derived from NOAA/AVHRR data,” IEEE Trans. Geosci. Re-mote Sens., vol. 32, no. 3, pp. 684–697, May 1994.

[8] A. H. Strahler, L. Boschetti, G. M. Foody, M. A. Friedl, M. C. Hansen,M. Herold, P. Mayaux, J. T. Morisette, and S. V. Stehman, “Global landcover validation: Recommendations for evaluation and accuracy assess-ment of global land cover maps,” Office Official Publ. Eur. Communities,Luxembourg, Germany, GOFC-GOLD Rep. 25, Mar. 2006.

[9] G. W. Rosborough, D. Baldwin, and W. J. Emery, “Precise AVHRR imagenavigation,” IEEE Trans. Geosci. Remote Sens., vol. 32, no. 3, pp. 644–657, May 1994.

[10] J. F. Moreno and J. Melia, “A method for accurate geometric correction ofNOAA AVHRR HRPT data,” IEEE Trans. Geosci. Remote Sens., vol. 31,no. 1, pp. 204–226, Jan. 1993.

[11] D. Shin, J. K. Pollard, and J. P. Muller, “Accurate geometric correction ofATSR images,” IEEE Trans. Geosci. Remote Sens., vol. 35, no. 4, pp. 997–1006, Jul. 1997.

[12] C. T. Mutlow, A. M. Zavody, and D. T. Llewellyn-Jones, “The AlongTrack Scanning Radiometer (ATSR)—Global validation results,” in Proc.2nd ERS-1 Symp., Space Service Environ., Hamburg, Oct. 1993, pp. 1245–1249.

[13] S. Sylvander, P. Henry, C. Bastien-Thiry, F. Meunier, and D. Fuster,“VEGETATION geometrical image quality,” in Proc. VEGETATIONConf., 2 Years Oper. Prepare Future, G. Saint, Ed., Lake Maggiore, Italy,Apr. 2000, pp. 1–12.

[14] P. Y. Deschamps, F. M. Bréon, M. Leroy, A. Podaire, A. Bricaud,J.-C. Buriez, and G. Sèze, “The POLDER mission: Instrument charac-teristics and scientific objectives,” IEEE Trans. Geosci. Remote Sens.,vol. 32, no. 3, pp. 598–615, May 1994.

[15] R. Wolfe, M. Nishihama, A. J. Fleig, J. A. Kuyper, D. P. Roy, J. C. Storey,and F. S. Pratt, “Achieving sub pixel geolocation accuracy in support ofMODIS land science,” Remote Sens. Environ., vol. 83, no. 1/2, pp. 31–49,Nov. 2002.

[16] X. Xiong, N. Che, and W. Barnes, “Terra MODIS on-orbit spatial charac-terization and performance,” IEEE Trans. Geosci. Remote Sens., vol. 43,no. 2, pp. 355–365, Feb. 2005.

[17] V. Jovanovic, M. A. Bull, M. M. Smyth, and J. Zong, “MISR in flightcamera geometric model calibration and geo-rectification performance,”IEEE Trans. Geosci. Remote Sens., vol. 40, no. 7, pp. 1512–1519,Jul. 2002.

[18] V. Jovanovic, C. Moroney, and D. Nelson, “Multi-angle geometric pro-cessing for globally geo-located and co-registered MISR image data,”Remote Sens. Environ., vol. 107, no. 1/2, pp. 22–32, Mar. 2007.

[19] W. E. Purdy, P. W. Gaiser, G. A. Poe, E. A. Uliana, T. Meissner, andF. J. Wentz, “Geolocation and pointing accuracy analysis for the WindSatsensor,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 3, pp. 496–505,Mar. 2006.

[20] G. A. Poe and R. W. Conway, “A study of the geolocation errors of thespecial sensor microwave/imager (SSM/I),” IEEE Trans. Geosci. RemoteSens., vol. 28, no. 5, pp. 791–799, Sep. 1990.

[21] D. Roy, “The impact of mis-registration upon composited wide fieldof view satellites and implications for change detection,” IEEE Trans.Geosci. Remote Sens., vol. 38, no. 4, pp. 2017–2032, Jul. 2000.


[22] B. Tan, C. E. Woodstock, J. Hu, P. Zhang, M. Ozdogan, D. Huang,W. Yang, Y. Knyazikhin, and R. B. Myneni, “The impact of griddingartefacts on the local spatial properties of MODIS data: Implications forvalidation, compositing, and band to band registration across resolutions,”Remote Sens. Environ., vol. 105, no. 2, pp. 98–114, Nov. 2006.

[23] Y. Carmel, “Characterizing location and classification error patterns intime-series thematic maps,” IEEE Trans. Geosci. Remote Sens. Lett.,vol. 1, no. 1, pp. 11–14, Jan. 2004.

[24] K. T. Weber, J. Théau, and K. Serr, “Effect of co-registration error onpatchy target detection using high resolution imagery,” Remote Sens.Environ., vol. 112, no. 3, pp. 845–850, Mar. 2008.

[25] J. R. G. Townshend, C. O. Justice, C. Gurney, and J. McManus, “Theimpact of mis-registration on change detection,” IEEE Trans. Geosci.Remote Sens., vol. 30, no. 5, pp. 1054–1060, Sep. 1992.

[26] X. Dai and S. Khorram, “The effects of image mis-registration on theaccuracy of remotely sensed change detection,” IEEE Trans. Geosci.Remote Sens., vol. 36, no. 5, pp. 1566–1577, Sep. 1998.

[27] P. Goryl and S. Saunier, “MERIS Absolute Geolocation Status,” ESA-GAEL, Paris, France, Contract no. 15993/02/I-LG, Tech. Rep. Issue 3,Apr. 2006.

[28] P. Goryl et al., “MERIS products quality status report: MEGS7.4and IPF5,” Eur. Space Agency, Paris, France, Tech Rep. Issue 1,Mar. 2006.

[29] L. Bourg and F. Etanchaud, “The AMORGOS MERIS CFI (AccurateMERIS Ortho Rectified Geolocation Operational Software),” Tech. Rep.PO-ID-ACR-GS-003, Feb. 2007, Issue 3.

[30] M. Bouvet, Intercomparison of ACE and SRTM 30 Global DigitalElevation Model at Global and Regional Scale, no. 2.0, Oct. 2004,ESA/ESTEC.

[31] L. Bourg, “MERIS level 1 detailed processing model, parameters datalist,” Eur. Space Agency, Paris, France, Tech. Rep. PO-TN-MEL-GS-002,Oct. 2006, ACRI MERIS Team.

[32] T. G. Farr, P. A. Rosen, E. Caro, R. Crippen, R. Duren, S. Hensley,M. Kobrick, M. Paller, E. Rodriguez, L. Roth, D. Seal, S. Shaffer,J. Shimada, J. Umland, M. Werner, M. Oskin, D. Burbank, andD. Alsdorf, “The Shuttle Radar Topography Mission,” Rev. Geophys.,vol. 45, p. RG2 004, May 2007. DOI:10.1029/2005RG000183.

[33] P. A. M. Berry, R. D. Hilton, C. P. D. Johnson, and R. A. Pinnock,“ACE: A new global digital elevation model incorporating satellite altime-ter derived heights,” in Proc. Eur. Space Agency, ERS-ENVISAT Symp.,Gothenburg, Sweden, Oct. 2000, pp. 1–9.

[34] G. Sotis, “EnviSat-1 Product Specifications,” RA-2 Product Specifica-tions, vol. 14, no. 3, PO-RS-MDA-GS-2009, Revision P, Dec. 2006.

[35] F. G. Lemoine, S. C. Kenyon, J. K. Factor, R. G. Trimmer, N. K. Pavlis,D. S. Chinn, C. M. Cox, S. M. Klosko, S. B. Lutchke, M. H. Torrence,Y. M. Wang, R. G. Williamson, E. C. Pavlis, R. H. Rapp, and T. R. Olson,“The development of the joint NASA GSFC and NIMA geopotentialmodel EGM96,” NASA Goddard Space Flight Center, Greenbelt, MD,20771, Jul. 1998.

[36] S. Riazanoff, S. Mbaye, C. Demange, and S. Delwart, “ENVISAT MERISgeometry handbook,” VisioTerra, Paris, France, Tech. Rep. VT-P194-DOC-001-E, Jul. 2005, Issue 1.5.

[37] S. Baillarin, A. Bouillon, M. Bernard, and M. Chikhi, “Using a threedimensional spatial database to orthorectify automatically remote sensingimages,” in Proc. ISPRS Workshop Service Appl. Spatial Data Infrastruc-ture, Hangzhou, China, Oct. 2006, pp. 89–93.

[38] S. Baillarin, J. P. Gleyzes, J. M. Delvit, A. Bouillon, E. Breton, andL. Cunin, “Validation of an automatic image orthorectification process-ing,” in Proc. IGARSS, Anchorage, AK, 2004, pp. 1398–1401.

[39] J. Inglada and A. Giros, “On the possibility of automatic multisensorimage registration,” IEEE Trans. Geosci. Remote Sens., vol. 42, no. 10,pp. 2104–2120, Oct. 2004.

[40] J. Inglada, V. Muron, D. Pichard, and T. Feuvrier, “Analysis of artifacts insub pixel remote sensing image registration,” IEEE Trans. Geosci. RemoteSens., vol. 45, no. 1, pp. 254–264, Jan. 2007.

[41] K. V. Khlopenkov and A. P. Trishchenko, “Implementation and evaluationof concurrent gradient search method for reprojection of MODIS level 1Bimagery,” IEEE Trans. Geosci. Remote Sens., vol. 46, no. 7, pp. 2016–2027, Jul. 2008.

[42] M. Vanda Nunes de Lima, IMAGE2000 and CLC2000: Products andMethods, EUR21757, 92-894-9862-5, EC-JRC, 2005.

[43] C. J. Tucker, D. Grant, and J. D. Dykstra, “NASA’s Global ortho-rectifiedlandsat data set,” Photogramm. Eng. Remote Sens., vol. 70, no. 3, pp. 313–322, Mar. 2004.

[44] P. Bicheron, M. Huc, C. Henry, S. Bontemps, and J.-P. Lacaux, Glob-Cover: Products Description Manual. Paris, France: Eur. Space Agency,Dec. 2008, Issue 2, Rev. 2.

Patrice Bicheron received the Ph.D. degree oncontinental biosphere dynamic from UniversityPaul Sabatier, Toulouse, France, in 1997.

He is currently working with Spot Image, a sub-sidiary of EADS/Astrium where his work is directedtoward the development of land cover and agro-environmental applications. From 1998 to 1999,he was a Postdoctoral Fellow with the FrenchSpace Agency (CNES), Toulouse. His main researchincluded the use of the POLDER instrument forvegetation monitoring. In 2000, he joined SCOT,

a subsidiary of CNES, to develop applications using low resolution earthobservations (VEGETATION, AVHRR). From 2002 until 2008, he participatedin the French Land Surface Thematic Centre POSTEL where he was deputymanager of the EU/FP7-CYCLOPES project for the development of globalbiophysical products from multi-sensor observations. He was also manager ofthe ESA-GlobCover project providing the first global land cover map at 300-mspatial resolution using ENVISAT/MERIS observations.

Virginie Amberg received the Ph.D. degree in signalprocessing field, with a special interest in the use ofSAR images, in 2005.

She is currently with the French Space Agency(CNES), Toulouse, France, in the field of optical im-age quality. From 2005 to 2007, she was working forMagellium, participating in several projects dealingwith geometric images accuracy.

Ludovic Bourg received the Ph.D. degree in remote sensing from the Univer-sity of Paris VII, Paris, France, in 1995.

He joined ACRI-ST, a French R&D SME, in 1995 and has been workingon optical remote sensing projects since then, with particular focus on theMERIS mission for the European Space Agency (ESA). He was in particularin charge of the definition of the level 1 and Calibration algorithms, and thedevelopment of the associated software. In 2000, he took the overall lead of theMERIS processing chain definition up to Level 2, including the coordination ofseveral contributing laboratories. He was deeply involved in the commissioningof MERIS and is still presently in charge of managing the CAL/VAL activitiesand the evolutions of the MERIS algorithms. He has defined in particular thealgorithms of the AMORGOS post-processing tool described in the presentpaper and supervised the tool development for ESA. These past three years,he has increased his involvement in the definition of next generation Europeanoptical sensors, like OLCI on-board Sentinel-3.

David Petit received the Ph.D. degree in high res-olution radar interferometry from University PaulSabatier, Toulouse, France, in 2004.

His current position as Head of Research andDevelopment at Magellium leads him to manageseveral projects in various fields of remote sensingapplication, 1-D, 2-D and 3-D image processing, androbotics methods.

Mireille Huc, photograph and biography not available at the time ofpublication.

Bastien Miras, photograph and biography not available at the time ofpublication.


Carsten Brockmann has a M.Sc. diploma inoceanography in 1988 and the Ph.D. degree in earthscience. He specialised in remote sensing and infor-matics early in his professional career.

He was working at the University Hamburg,Hamburg, Germany, the GKSS Research Centre andthe European Space Agency as remote sensing ex-pert before he founded SCICON—Brockmann undKleeberg GbR in 1994 and later Brockmann Con-sult, in 1999. In 2001, he founded the BrockmannConsult GmbH which provides commercial software

solutions for environmental processing systems. As managing director of thecompany, he is responsible for all strategic decisions. He is actively involvedin many projects for the company, notably all projects related with software formanaging the data of the spaceborne sensor MERIS onboard the ESA satelliteENVISAT.

Olivier Hagolle received the engineer degree fromthe Ecole Supérieure d’Electricité, Gif-Sur-Yvette,France, in 1990.

Since 1990, he has been with the French SpaceAgency (CNES), Toulouse. From 1990 to 2002,he focused on satellite image quality and opticalsensors calibration methods. In 2002–2004, he de-veloped algorithms for the correction of directionaleffects for low resolution optical instruments such asVEGETATION, AVHRR, MERIS. Since 2004, he iswith the Centre d’Etudes Spatiales de la Biosphère

(CESBIO), Toulouse, France, where he coordinates the level 2 specificationsfor the Venμs sensor, which combines high resolution, frequent revisit, andconstant viewing angles. He focuses on the development of methods forestimating the aerosol optical depth over terrestrial surfaces.

Steve Delwart, photograph and biography not available at the time ofpublication.

Franck Ranéra received the Specialized Mastersdegree in remote sensing from the University PaulSabatier, Toulouse, France, in 1999.

He is currently with Spot Image, a subsidiaryof EADS/Astrium where his work is directed to-ward the management of projects related to riskfields. From 1999 to 2008, he was with Serco inItaly where he supported many ESA projects re-lated to risk and Application fields (Globcarbon,GLOBCOVER, Medspiration, Urbex, CoastChart).His main task was to ensure the correct implemen-

tation of the project with respect to the technical requirements. He wasresponsible of the project technical side, reporting directly to the ESA projectofficer. He conducted several acceptance test of the projects production’s chainand liaised frequently with scientific groups to ensure product reliability andconsistency.

Marc Leroy received the Ph.D. degree in radiativelydriven stellar winds from Paris VII University, Paris,France, in 1980.

He became Research Associate of Observatoire deParis, in 1981 and worked in the field of theoreticalplasma physics of the earth’s bow shock, both at Ob-servatoire de Paris and at the University of Maryland,College Park. He joined the French Space Agency(Centre National d’Etudes Spatiales), Toulouse, in1985 and has worked since then as a CNES Engineer.He worked first on calibration activities associated

to the SPOT program, and became in 1990 the Head of the Department ofImage Quality in the CNES Image Processing Division. He joined CESBIO,Toulouse, in 1993 with a specific interest in the physics of remote sensingmeasurements in the optical domain, and in charge there of the developmentof algorithms of land surface products of the POLDER/ADEOS instrument. Hejoined MEDIAS-France, in 2001 as Head of the Land Surface Thematic CentrePOSTEL. From 2001 to 2008, he developed POSTEL and its Service Centreand took a significant role in several important European projects, in particularGEOLAND and GLOBCOVER.

Olivier Arino received the Ph.D. degree in physicswith speciality in remote sensing from the Insti-tut National Polytechnique de Toulouse, Toulouse,France, in 1990.

After two years as a Postdoctoral Fellow withthe International Geosphere Biosphere Program andthe European Commission, he joined the EuropeanSpace Agency, in 1991. His current position as Headof the Projects Section in the Science Applicationsand Future Technologies Department of the EarthObservation Program Directorate leads him to man-

age the Data User Program in close collaboration with user communities (e.g.,300+ user organisations involved) and the relevant institutional communities,such as the United Nations Food and Agricultural Organisation, the UnitedNation Environment Program and the European Environment Agency. Heinitiated the GlobSeries that are considered as precursors to the climate changeinitiative called “Global Monitoring of the Essential Climate Variable” recentlyfunded by the Agency’s ministers. He also worked closely with the Interna-tional Environmental Conventions secretaries and users on climate change,desertification, biodiversity, wetlands and world heritage. He has authored orcoauthored 110 scientific publications in the field of albedo, fire, vegetation,agriculture, land cover, sea surface temperature, and was nominated for thebest European paper in open literature in 1992. He has acted as a reviewerfor the International Journal of Remote Sensing, IEEE TRANSACTIONS ON

GEOSCIENCE AND REMOTE SENSING, Remote Sensing of the Environmentand others. He is a member of international science groups in sea surfacetemperature, forest and land cover. He was appointed Senior Advisor in ESA,in 2008.

Geolocation Assessment of MERIS GlobCover Orthorectified Products

Documents

Transcript of Geolocation Assessment of MERIS GlobCover Orthorectified Products