1 Vito 16/12/2004 VALERI Meeting 10 March 2005, INRA, Avignon, France A new VALERI validation site in North-Western China: The Shandan grassland Four years of bilateral cooperation between the Peoples Republic of China and the Flemish Community in the RESPOM project F. Veroustraete 2, M.G. Ma 1, 2, J. Bogaert 1,3, 4, L. Lu 1, 2, X. Li 1, T. Che 1, C.L. Huang 1, Q.H. Dong 2 and R. Ceulemans 3 1 Cold and Arid Regions Environmental and Engineering Research Institute (CAREERI), Chinese Academy of Sciences (CAS), China. Chinese Academy of Sciences (CAS), China. 2 Vito/TAP, Centre for Remote Sensing and Earth Observation Processes, Belgium. 3 University of Antwerp, Department of Biology, Belgium. 4 Universit Libre de Bruxelles, cole Interfacultaire de Bioingnieurs, Belgium. 2 Vito 16/12/2004 Contents The project. The project. The spatial sampling strategy. The spatial sampling strategy. Shandan site description. Shandan site description. The spatial sampling strategy at Shandan. The spatial sampling strategy at Shandan. Field instruments Field instruments Results Field measurements LAI Results Field measurements LAI Results Field measurements Gap fraction Results Field measurements Gap fraction Results Field measurements Albedo Results Field measurements Albedo Results Up-scaling with Landsat ETM+ Results Up-scaling with Landsat ETM+ Results Aggregation of ETM+ imagery to 1x1 km Results Aggregation of ETM+ imagery to 1x1 km Results Validation of the MODIS LAI product Results Validation of the MODIS LAI product Conclusions. Conclusions. 3 Vito 16/12/2004 The project The objectives of the project To evaluate the absolute accuracy of biophysical products (LAI, fAPAR, fCover) derived with a range of algorithmsfrom large swath sensors (e.g. AVHRR, POLDER, VEGETATION, SEAWIFS, MSG, MERIS, AATSR, MODIS, MISR,). To evaluate the absolute accuracy of biophysical products (LAI, fAPAR, fCover) derived with a range of algorithms from large swath sensors (e.g. AVHRR, POLDER, VEGETATION, SEAWIFS, MSG, MERIS, AATSR, MODIS, MISR,). To inter-compare products derived from different sensors and algorithms. For this purpose, the project develops: For this purpose, the project develops: A network of sites distributed globally. A standard methodology designed to directly measure the biophysical variables of interest at the proper spatial and temporal scales A standard methodology designed to directly measure the biophysical variables of interest at the proper spatial and temporal scales. 4 Vito 16/12/ Spatial definition of the sampling strategy. 2.Site related georeferenced field measurements of LAI, Albedo and Gap fraction. 3.Up-scaling of field measurements using high resolution Landsat ETM+ imagery. 4.Aggregation of high resolution validation fields to large swath sensors resolution o validate bio- geophysical products from MODIS (VEGETATION upcoming). 4.Aggregation of high resolution validation fields to large swath sensors resolution (1x1 km) to validate bio- geophysical products from MODIS (VEGETATION upcoming). 5.Evaluation of bio-geophysical product accuracy over an ensemble of Valeri sites and campaign dates available (the future). The project Specific objectives in RESPOM 5 Vito 16/12/2004 The project core sites Valeri core sitesShandan site MODLAND Core sites 6 Vito 16/12/2004 The spatial sampling strategy 7 Vito 16/12/2004 The Shandan site 8 Vito 16/12/2004 The Shandan site is located in the footprint of the Qilian Mountains where the Heihe river originates. Due to the fertile soil and appropriate climate, this grassland became a high quality horse feedlot for over more than 2000 years. The yearly mean precipitation level is approximately 150 mm and evaporation is 2531 mm. This area belongs to the semi-arid regions in China. The Shandan site is centered at N, E with an elevation of 2700 m. The field campaign was executed from July 11th to July 15th, The location of the 38 ESUs is randomly determined and the ESUs localized with GPS. The vegetation is characterized by a very homogenous semi-arid grassland with a high fractional cover. The following data were collected: LAI2000 data (LAI, gap fraction), albedo, TRAC data, soil temperature and vegetation diversity (species richness). Shandan site description 9 Vito 16/12/2004 The spatial sampling strategy at Shandan 37 crosses (ESUs) Spatial distribution and Representation of each cover class Few measurement transects 3 km Landsat ETM+ false-color RGB of channels 1, 2, and 3 for the Shandan site 10 Vito 16/12/2004 The spatial sampling strategy at Shandan TRAC LAI and Albedo Site ESU distribution 11 Vito 16/12/2004 Field instruments used in the campaign PAR Albedometer (Patent pending,Bogaert J.-UA) LAI-2000 (Licor) TRAC (CCRS) 12 Vito 16/12/2004 Results: LAI field measurements LAI2000 Clear-cut increase in standard deviation with increasing LAI 13 Vito 16/12/2004 Results: LAIield measurements - TRAC vs LAI2000 Results: LAI field measurements - TRAC vs LAI2000 Comparison between TRAC and LAI-2000 LAI data elicits a relatively good agreement. 14 Vito 16/12/2004 Results: Field measurements - Gap fraction vs LAI LAI2000 sampling Small zenith angles are suboptimal 15 Vito 16/12/2004 Results: Albedo field measurements Very weak relationship between standard deviation and albedo. Probably a linear increase of standard deviation with increasing albedo. 16 Vito 16/12/2004 Results: Upscaling using VIs from Landsat ETM+ VIField LAI measurementsEquationR2R2 Author NDVI/SR Allometric method NDVI = LAI LAI LAI Turner et al. (1999) SR = LAI LAI NDVI/SR Allometric method NDVI = LAI Fassnacht et al. (1997) SR = LAI SR Allometric method SR = 1.92LAI Peterson et al. (1987) SR = LAI0.83 NDVI/SR LAI-2000 and TRAC NDVI = 0.032LAI Chen and Cihlar (1996) SR = 1.014LAI SRLAI-2000 and TRAC SR = 1.153LAI Chen et al. (2002) SR Ceptometer SR = log(LAI)0.97Spanner et al. (1994) NDVI/SR LAI-2000NDVI = /(1/LAI ) Gong et al. (1995) SR = 0.96/(1/LAI-0.066) NDVI Allometric method LAI = 33.99NDVI Curran et al. (1992) SR Allometric method SR = 0.614LAI Running et al. (1986) NDVIAllometric method NDVI = 0.03LAI Nemani et al. (1993) 17 Vito 16/12/2004 Results: Upscaling using VIs from Landsat ETM+ NDVI: R = 0.53; RMSE = 0.24 SR: R = 0.54; RMSE = 0.20 SAVI: R = 0.67; RMSE = 0.18 18 Vito 16/12/2004 Results: Upscaling using Landsat ETM+ NDVI 19 Vito 16/12/2004 Results: Upscaling LAI using ETM+ SAVI Transfer function 20 Vito 16/12/2004 Results: Upscaling Gap fraction using ETM+ NDVI Transfer function 21 Vito 16/12/2004 Results: Upscaling Albedo using ETM+ NDVI Transfer function 22 Vito 16/12/2004 Results: Aggregation of ETM+ LAI HR to 1x1km Pixel frequency at the 30x30 m scale is different for each ESU. To aggregate, the mean value of the pixel distribution was selected as a first proxy for the pixel value at the 1x1 km scale. 23 Vito 16/12/2004 Results: Aggregation of ETM+ LAI HR to 1x1km Aggregated 1x1 km LAI data from ETM+ NDVI data using mean values Aggregated 1x1 km LAI data from ETM+ SR data using mean values Aggregated 1x1 km LAI data from ETM+ SAVI data using mean values 24 Vito 16/12/2004 Results: Validation of MODIS LAI product Validation of MODIS LAI data with aggregated LAI data from ETM+ NDVI Validation of MODIS LAI data with aggregated LAI data from ETM+ SR Validation of MODIS LAI data with aggregated LAI data from ETM+ SAVI Best fit (highest R), best slope (closest to 1) and smallest intercept (systematic bias) 25 Vito 16/12/2004 The complete procedure has been applied at the Shandan site in Northwestern China. The Shandan site is very suitable since its vegetation is a homegeneous grassland. Three bio- geophysical variable fields have been produced for validation purposes, e.i., LAI, gap fraction and albedo. SAVI seems to be the best of three VIs to perform up-scaling. Nevertheless, the aggregated 1x1 km, obtained with NDVI based up-scaling, elicits the best results in the MODIS LAI product validation. The comparison between LAI measurements with the LAI-2000 and TRAC is satisfactory. However, LAI-2000 measurements are of beter quality. Suboptimal sampling is observed for the between 10 and 30 cm high grassland Shandan site, when small view zenith angles are selected with the LAI The MODIS LAI product elicits a systematic bias with higher LAI values. A plausible reason can be s.o.p. How about,. conclusions 26 Vito 16/12/2004 Conclusions: Suboptimal LAI sampling LAI2000 sampling Small zenith angles are suboptimal when the vegetation has a low height. In that case the effective LAI will be underestimated. This can explain the higher MODIS LAI with respect to the up-scaled field measurements. It should be investigated whether the elimination of the smallest view zenith angles gives better validation results, or the application of DHP. 27 Vito 16/12/2004 Shandans flowers are grateful for your attention