M. WEISS/F. BARET 15 th NOVEMBER, R. Bosseno CESBIO E. Mougin, F. Gascon, L. Jarlan Y. Tarcol, E. Martin

Transcription

1 REPORT ON THE MEETING M. WEISS/F. BARET 15 th NOVEMBER, 2001 Participants: INRA (CSE Avignon) F. Baret INRA (Bioclimatologie Bordeaux) D. Guyon Noveltis M. Weiss IRD R. Bosseno CESBIO E. Mougin, F. Gascon, L. Jarlan Y. Tarcol, E. Martin MEDIAS M. Leroy 1. F. Baret: VALERI Context 1.1 TAOB Committee F. Baret has given the conclusion driven by the CNES TAOB (Earth, Atmosphere, Ocean, Biopsphere) Group concerning the VALERI activities from June 2000 to June 2001: «Le comité TAOB constate que le projet subit une évolution notable. Outre le sous-dimensionnement en moyens humains, des questions se posent sur l'utilisation des paramètres mesurés (nombre de sites, investissement des Le comité TAOB réaffirme l intérêt de validation de paramètres géophysiques et recommande une évolution prudente du nombre de sites ; il souhaite qu une la méthodologie end to end soit établie avant de multiplier le nombre de sites, et demande de traiter en priorité les capteurs existants (Vegetation, AVHRR ). Il apprécie par ailleurs l utilisation de données SPOT pour piloter l intégration spatiale des mesures sol.» The TAOB committee notes that the project is really well developed. Additionally to the underestimation of human means, some question are raised concerning the use of the measured parameters (number of sites, participation of the different international teams). The TAOB committee insists on the interest of biophysical parameter validation and recommend to take care of the number of sites ; the committee wishes an end-to-end methodology to be developed before increasing the number of VALERI sites and asks to treat as prior existing sensors (VEGETATION, AVHRR). They appreciate the use of SPOT data for the spatial interpolation of the ground data. F.Baret noted that a variety of canopy types is however necessary for a good validation activity.. Marc Leroy insisted on the fact that the TAOB committee whished an end-to-end methodology. Marc advices that VALERI should provide high resolution LAI map for some sites for the end of march (see 3). F. Baret proposes to concentrate on the sites sampled with LAI2000 with a preliminary up-scaling process based on co-kriging and simple transfer functions. This will tehn be used to validate VEGETATION and AVHRR products such as fapar derived through NDVI (actual products) for the nenxt TAOB meeting. 1.2 CEOS F. Baret and M.Weiss have participated to the CEOS meeting in June at Rome. You can find some presentations, as well as the report published in The Earth Observer hold at this meetings at The CEOS supports the VALERI activities. In that frame, Jeff Morrissette who is responsible of the LPV (Land Validation Products) proposes to order ETM+ data for free: LPV can work with site contact to either: 1) order additional ETM+ data for the site or 2) post ETM+ (or SPOT, or other) data that has been ordered but not through the MODIS land team, if investigators can supply a copy of that data to LPV. If it has not already been done, each site manager should therefore select available images (see VALERI web Site, Contact page, click LANDSAT) and send an to M. Weiss with all the image ID required.

2 1.3 Other initiatives: validation product development In a parallel direction, two projects have been submitted to provide the real end-to-end methodology, i.e to validate large swath sensor products: RTE (Réseau Terre-Espace, French research ministery financial support) and Cyclopes submitted to the FP5 program of the European Community. We hope that these initiatives will improve the (wo)man power for the processing of the biophysical maps of products. For the moment nothing has come from ESA!!! 1.4 Thesis in the frame of VALERI Sébastien Garrigues begins a Phd in January supported by the CNES and Alcatel Space Industry on the mixed pixel effect. His thesis will be monitored by F. Baret and Denis Allard (INRA, Biométrie). Li : we hope that INRA (not yet decided) will support this thesis which aims at developing a method of aerosol optical thickness from the BF2 sensor and methods for satellite data atmospheric correction based on AERONET data mainly for VEGETATION and MERIS Measurement campaigns Globally, all the measurement campaigns were successful with a lot of measurements acquired. We thank all the people who have invested a lot of time and energy to collect the data, as well as to have followed the sampling scheme. A copy of the transparencies presented for the campaign measurement will be soon available on the valeri web site (page documents): 1. Puechabon, France: this campaign has been conducted in collaboration with the MODLAND people. Participants were: Marie Weiss, Nadine Bruguier, Bruno Combal, Jean-François Hanocq, Roland Bosseno, Hervé Bohbot, Serge Rambal, Alain Rochetaud, Sébastien Garrigues, Camille Lelong, Fabien Dauriac, Yuri Knyazikhin, Grace Smith, Jennifer Flax, Jiarui Dong, Jianan Hu, Oleg Panferov 2. Aek Loba, Sumatra, Indonesia: Camille Lelong, Vincent Abt, Jean-Charles Jacquemard, Edyana Suryana 3. Nézer, France: Dominique Guyon, Sébastien Garrigues, Laurent Franchisteguy 4. Gourma, Mali: Eric Mougin 5. Alpilles, France: Nadine Bruguier, Bruno Combal, Iñaki Garcia, Marie Weiss, Ignacio Negri, Roland Bosseno 6. Fundulea, Romania: Frédéric Baret, Nadine Bruguier, Laurent Prévot, Nadia Rochdi, Catalin Lazar, Goreges Petcu, Elena Petcu 7. Turco, Bolivia : Roland Bosseno, Jean-François Hanocq, Freddy Loza de la Cruz, Rufian Villa Vilca 8. Laprida, Argentina : Marie Weiss, Carlos Di Bella, Frederico Bock, Victoria Feler 9. Järvselja, Estonia : Tiit Nilson, 10. Counami, Guyana : Frédéric Baret, Valery Gond, Onoefé Ngwete, Boris Ruelle, Richard Sante, Sacha Weber 11. Sierra Chincua, Mexico : Frédéric Baret, Marisa España, Jean-François Hanocq, Raúl Cardenas and Luis López-Pérez. The hemispherical photographs have not been yet processed. The software developed by Noveltis ( 3.2) is now evaluated. However, people should notice that one single hemispherical picture is not sufficient to sample the plot. It is thus necessary to sample the same way as for LAI2000 measurements. The digital camera must always be put in Automatic mode and in RGB mode, FINE resolution and the flash of!!. The Noveltis software is indeed based on a color classification. For canopies with understorey or adventice, it is also necessary to place the LAI2000 (or digital camera) at the ground level. An other is possible: take 2 photographs, one looking upward, the other downward. Some problems are also encountered in Counami and Mexico due to the soil topography. Concerning SPOT images, two problems have been encountered in Estonia and Bolivia, for which SPOT image has forgotten to launch the programming. Before the campaign week, please check that SPOT image has not forgotten your existence! We ask Hervé JeanJean if it is possible to order a SPOT image red programming for the same price when using ISIS. 3. Methodology

3 Here, we have discussed about how to derive high resolution LAI maps from a limited number of ground measurements ( 3.1 and 3.2). The methodology is based on geostatistics, i.e the use of kriging methods ( 3.4). Two of the three kriging methods are based on the use of ancillary information provided by a SPOT image. This ancillary information is based on the use of transfer functions ( 3.3). Considering that the kriging techniques have to be more investigated, and due to the TAOB comments who ask for an end-to-end methodology, Marc Leroy proposed to derive high resolution LAI maps for a maximum number of 2000 and 2001 sites (all the data must be processed), using only the SPOT images and the results from Spatial Sampling Concerning the sampling scheme which was proposed for the 2001 campaigns (Table 1), it has been noticed that transects are not representative of a SPOT pixel (20mx20m). This is the reason why a new sampling strategy can be proposed. This issue will be also discussed in the next meeting (March). We therefore suggest to sample at least three locations in a 1km 2 square. Each sample MUST be representative of the SPOT pixel. The locations must be distributed so that: 1. all surface types are sampled 2. the samples are spatially distributed: in order to properly compute the LAI variogram, all the distances between points must be represented (from 0 to 3000m). We therefore recommend to sample 2 more locations in the central square so that intermediate distances are represented 3. People can use the natural paths existing in the landscape (dense vegetation). It is however recommended not to sample too close to the path so that the path do not disturb the signal of the sampled SPOT pixel. 4. In continuous landscapes, transect can be performed but measurements must be representative of the SPOT pixel. Discrete landscape 3 20 m Km Hemispherical or LAI 2000 (1 above/4 below) points GPS central point 27 x 12 local cross-shaped measurements composed of one above canopy and four below acquisitions for LAI2000 and/or four above canopy picture (hemispherical photograph). Measurements are distant from 4m and thus, each part of the cross is 20m long. In case or row crops, the 20m cross is always placed transversely to the rows. The GPS positioning is achieved at the center of the cross. The measurements are performed at more than 40m from the field edges 2 1km-transects in the central pixel (East/West and North/South) with LAI2000 measurement (1 above, 10 below) each 50m. Each measurement must be georeferenced. In case of large fields (i.e. larger than 300m long), transects along single fields (around 500m) should be measured. 2 schemes are proposed with 12 local measurements points each and one central GPS measurement: - cross-shaped measurement: 2 perpandicular branches 28m long with 6 equally spaced local measurements - square shaped measurements: 1 central point, 8 peripheric measurements and 3 randomly distributed points

4 Continuous landscape 9 1km-transects in each pixel (1km x 1km), oriented perpendicularly to the vegetation gradient. Visual notes every meter to distinguish 4 classes (bare soil, low, medium, high vegetative area). Sub-sampling in the 4 classes to perform ground measurements (LAI2000, hemispherical photographs). Each ground measurements is geo-referenced with GPS positioning. Table 1. Sampling scheme proposed for 2001 campaigns (see the report of March 2001 meeting available on the document page of the VALERI web site) 3.2 LAI measurements LAI2000 Calibration The 3 VALERI LAI2000 have been calibrated on the th October They were horizontally placed side by side on a bare soil, in clear sky conditions (Figure 1). Above measurements were acquired every 15s, from 3.30 to 5.30 PM. Figure 2 shows good agreement between two sensors for the first ring. Same features are observed for the other rings except for the largest zenith angle, and for the three instruments. We therefore propose to re-process all the LAI2000 data without using the largest view angle. A document with the calibration coefficient will be soon available on the web site (methodology page). Figure 1. LAI2000 Calibration experiment design 2001/10/10 - View Angle y = x R 2 = PCH /10/10 - View Angle y = x R 2 = PCH-0979 Figure 2. Example of LAI2000 inter-calibration results.

5 Hemispherical Photographs At this time, no hemispherical image is processed. A software was developed this summer by Noveltis (Olivier Hautecœur, Arnaud Tournier, Marie Weiss) and INRA (F.Baret). At this time, it is still in the testing phase. The software is based on an image classification instead of using a binarization threshold. We assume that for each location sampling, N images are acquired. The software first generates a learning image composed of a collection of representative parts extracted from each of the N images. Two different neural networks (called principal and secondary Nnets) are then applied to this learning image to learn the color topography and obtain a reduced color palette (Figure 3). Once this is achieved, the software needs the user intervention to assign colors in the learning image for which the probability of belonging to a vegetative element is 1. Then, the software applies these probabilities for each of the N images without requiring the user intervention. When the distance between a pixel color and the neural net palettes is two high, the pixel is not taken into account to compute the gap fraction. Figure 3. Example of a learning image on a maize canopy. On the left, in the upper left corner, the principal Nnet reduced color palette is presented, the secondary one is shown at the bottom of the left figure. The colors selected as being vegetative elements are selected in red. On the right, the classified image is presented: black = probability of being a vegetative element=0, green = probability of being a vegetative element = 1 (principal Nnet), blue = probability of being a vegetative element = 1 (secondary Nnet). The gap fraction is then computed by computing the ratio between the number of probability 1 pixels to the total number of pixels for 5 zenith angle ranges (the gap fraction is azimuthally integrated). The leaf area index and average leaf inclination angle are then computing using a look-up table. 3.3 Transfer Functions of the local measurements In the kriging method, the ancillary information correspond to the LAI derived from SPOT images. Grace Smith from the geography department of Boston University spent 3 months at INRA, working on the transfer functions of local measurements (Alpilles 2001 Site), i.e, developing reflectance-lai relationships using the ground measurements and the SPOT data. We assume that those relationships are site specific. 3 techniques have been considered: 1. Empirical relationships: multi-linear regression (SPOT bands) or NDVI relationships 2. the Miracle Fred Method, based on the assumption that, for an image and a class within a site, the reflectance variability is only due to LAI variability. The image is first classified and the average measured LAI (LAI ) and reflectance (ρ ) computed on the local measurements for each class. The derivative of the SAIL model is then used within each class to compute LAI = LAI LAI, knowing ρ = ρ ρ by simply applying a Taylor decomposition to the first order. 3. Hybrid methods: look-up-table and neural nets At this time, only partial results are available. The work should be finished at the end of January and a paper will be submitted to remote sensing of the environment.

6 3.4 Spatial Interpolation of ground measurements A study report (in French) as well as a paper describing the kriging methods presented in Aussois (Weiss, M. et al., Mapping leaf area index measurements at different scales for the validation of large swath satellite sensors: first results of the VALERI project. In: CNES (Editor), 8th International Symposium in Physical Measurements and Remote Sensing, Aussois, France, (2001) pp ) are available on the web site (see documents and methodology pages). 4. Web Site We remind you that you can visit the VALERI website at any time. Please, if you find some errors inside the site, send an to the webmaster (Jean-Pierre Guinot: jpg@avignon.inra.fr or Marie Weiss weiss@avignon.inra.fr). We try to update the web site at the time you send the data, which is not always possible due to overbooked planning. We apologize for the delay concerning some sites. Before sending the data, please, check that all the files follow the format of the data already available on the web (see Järvselja 2000 or Alpilles 2001). Don t forget to write a campaign report. It is highly recommended to send a CD containing AT LEAST the whole SPOT data to INRA (Marie Weiss or Frédéric Baret) so that all the available data can be archived in the same place Budget The 2001 budget is presented hereafter in Francs. Country Site Date SPOT Image Mission Camera + fish-eye Campbell 9667 Argentina Laprida BF Bolivia Turco BF Estonia Järvselja GPS 2235 France Alpilles Isatis 8160 Nezer Miscellaneous Campbell Puechabon VEGA French Guyana Counami Mexico Mali Gourma Romania Fundulea Sumatra Aek Loba Rome (CEOS) 9073 Romilly TOTAL Frais de Gestion (3% Hors salaire) 7799 M.Weiss salary Total This budget is preliminary and additional expenses will arrive. Further, early 2002 campaingns could be partly funded by this budget (Mali and Bolivia in 2002). A meeting is scheduled in March 2002 for the detailed plans of 2002 experiments, current status of the production of the biophysical maps and methodological progress (transfer functions, hemispherical photographs processing ).