LAI and fAPAR CYCLOPES global products derived from VEGETATION. Part 2: validation and comparison with MODIS collection 4 products
Introduction
In the past ten years, various medium resolution sensors have been launched. Several land products derived from these sensors are, or will be soon available to the user community at both regional and global scales. As an example, Table 1 shows that, for year 2002, at least 6 different leaf area index (LAI) and 9 fraction of Absorbed Photosynthetically Active Radiation (fAPAR) products were developed. The user must therefore face the dilemma of choosing the most appropriate product to suit his application. All these products require therefore qualitative and quantitative accuracy assessment, which is, in addition, mandatory for many applications including those based on data assimilation into process models.
Evaluation of product accuracy is a very difficult task, particularly regarding the global extent of most of the products, their kilometric spatial resolution, as well as the dynamics of the vegetation. The Land Product Validation Subgroup (LPV) of the Committee Earth Observing Satellite (CEOS) was mandated to coordinate and standardize international validation activities (Justice et al., 2000, Morisette et al., 2006). Validation is generally achieved through direct validation, i.e. comparing satellite products to ground measurements of the corresponding biophysical variables. In the case of medium resolution sensors, the main difficulty relies in scaling local ground measurements to the extent corresponding to medium spatial resolution pixel size. A realistic sampling strategy has to be designed, considering the available manpower and measurement methods.
Multiple initiatives have been conducted among the community to provide ground leaf area index maps at medium resolution (BigFoot (Cohen & Justice, 1999), CCRS (Fernandes et al., 2003), MODLAND (Morisette et al., 2002), VALERI http://www.avignon.inra.fr/valeri). They are all based on the use of high spatial resolution imagery to extend ground sampling measurements. CEOS LPV subgroup proposed a framework to share the ground LAI maps among the entire community to support this international LAI validation activity (Morisette et al., 2006). These maps have been used within many studies for product validation. However, the validation process is generally restricted to few sites (Cohen et al., 2003, Gobron et al., 2006, Huemmrich et al., 2005, Morisette et al., 2002, Privette et al., 2000, Tan et al., 2005, Tian et al., 2002a, Tian et al., 2002b, Wang et al., 2004, Yang et al., 2006b) or limited to an area (Chen et al., 2002, Cohen et al., 2006b, Hill et al., 2006). Moreover, most of these LAI maps are a one shot effort: very few studies show repeated measurements along the year or between years. Up to now about 40 LAI maps are available. Fewer fAPAR and fCover validation maps are available mainly coming from the VALERI project.
To complement direct validation results which are limited by the small number of sites, indirect validation based on a larger ensemble of sites for which no ground measurements is required, provides a far better sampling both in space and time. Indirect validation includes inter-comparison between products as well as evaluation of their temporal and spatial consistency. Baret et al. (2006) propose the BELMANIP network of sites, designed to represent the variability of surface conditions over the Earth. This ensemble of sites is considered by CEOS/LPV as a benchmark for indirect validation.
Regardless direct or indirect validation and the considered product, particular attention must be paid to:
- •
Product definition: Although users generally agree on the definition of the product, departure from this definition could be observed both for ground measurement collection depending on the devices used, and for satellite products depending on assumptions embedded in the biophysical algorithm and the measurement configuration.
- •
Product geometry: to perform direct or indirect validation, particularly at the site level, the target must obviously match the same area, i.e corresponds to the same geographic location and size. Geolocation uncertainties, differences in projection systems and point spread functions have to be accurately accounted for.
- •
Temporal sampling: temporal compositing of the product differs from one algorithm to another. Again, when considering LAI products, this may vary from one single day (ground measurements), to daily (MERIS MGVI), via monthly (GLOBCARBON), weekly (MODIS) or every decade (CYCLOPES, CCRS). However, nominal temporal sampling interval could be hampered by missing data mainly due to cloud occurrence.
This paper focuses on the validation of CYCLOPES (version 3.1) LAI and fAPAR products derived from VEGETATION sensors (Baret et al., 2007-this issue) at 10 days temporal sampling (composited over 30 days windows) over a 1/112° plate–carrée spatial grid. Performances are compared to MODIS collection 4 LAI and fAPAR (Myneni et al., 2002) 8 days temporal sampling interval products (composited over 16 days windows) over a 1 km sinusoidal grid. However, for fAPAR, MODIS collection 4.1 fAPAR, delivered by the Boston University at monthly temporal sampling interval is used in place of fAPAR collection 4.0, since a bug was detected in the code generating the collection 4.0 product (FPAR under diffuse radiation produced instead of FPAR under direct solar radiation). In addition, ECOCLIMAP climatology values for LAI (Masson et al., 2003) are used as an independent reference. The validation is achieved from years 2000 to 2003. CYCLOPES products, developed in the setting up of the Land Surface Thematic Centre POSTEL, can be downloaded at http://postel.mediasfrance.org/en/DOWNLOAD/Biogeophysical-Products/, MODIS collection 4.0 products are available at ftp://e0dps01u.ecs.nasa.gov/MOLT/MOD15A2.004/ and ECOCLIMAP LAI is provided at http://www.cnrm.meteo.fr/gmme/PROJETS/ECOCLIMAP/page_ecoclimap.htm.
A brief description of the considered products is first given along with geometrical considerations (effect of Point Spread Function (PSF) and projection). Then, the temporal continuity of products is investigated to evaluate the fraction of valid pixels. Indirect validation exercise is described, including temporal consistency and statistical distributions. Direct validation is then considered. Conclusions are finally drawn on both the methodology applicable to any land product, and the performances of LAI/fAPAR products.
Section snippets
Product definition
The commonly accepted definition of LAI corresponds to half the developed green foliage area per unit horizontal ground area (Chen and Black, 1992, Stenberg, 2006). This definition corresponds to LAI measured using a planimeter and all possible allometric relationships (Frazer et al., 1997). It agrees generally with user community requirements (NPP estimation, carbon modeling, and global change). Production of ground LAI validation maps is most of the time achieved using indirect methods (Weiss
Indirect validation
Indirect validation consists in evaluating the performances of the different products, without comparing them to actual ground measurements. The temporal continuity and consistency of CYCLOPES and MODIS are first investigated. Then, statistical distributions for the main surface classes are compared.
Direct validation
Thanks to the international effort lead by the CEOS LPV subgroup, Garrigues et al. (in preparation) gathered about 40 ground LAI measurement maps. Although the number of available sites is relatively low, Fig. 9 shows that their distribution both in location and surface type is quite large with however an over-representation of the Northern hemisphere. Note that about 10 sites were sampled several times from 2000 to 2003. LAI derived from ground measurements may correspond to different
Conclusion
The main objective of this paper was the validation of CYCLOPES version 3.1 LAI and fAPAR products, by comparison with MODIS and ECOCLIMAP products, as well as ground measurements.
Results of this study showed that CYCLOPES products have reached a reasonable level of maturity: as compared to MODIS similar products, CYCLOPES achieved the best performances both for LAI (RMSE = 0.73 for effective, 0.84 for true) and fAPAR (RMSE = 0.10). CYCLOPES products show also very good temporal consistency,
Acknowledgments
This research was made funded by the European Union throughout the FP5/CYCLOPES and FP6/GEOLAND projects and the French Research Ministry throughout the Réseau Terre-Espace. The authors would also like to thank the many individuals who contributed to the different projects (BigFoot, VALERI, CCRS activities, MODLAND activities, …) that allow to gather all the ground measurements which represents a huge effort and manpower investment. We also thank the teams who contributed to make the BELMANIP
References (61)
- et al.
Neural network estimation of LAI, fAPAR, fCover and LAIxCab, from top of canopy MERIS reflectance data: Principles and validation
Remote Sensing of Environment
(2006) - et al.
LAI, fAPAR and fCover CYCLOPES global products derived from VEGETATION. Part 1: Principles of the algorithm
Agronomie
(2007) - et al.
Shortwave radiation budget of sahelian vegetation. 2. Radiative transfer model
Agricultural and Forest Meteorology
(1996) - et al.
Derivation and validation of Canada-wide coarse-resolution leaf area index maps using high-resolution satellite imagery and ground measurements
Remote Sensing of Environment
(2002) - et al.
Validating MODIS terrestrial ecology products: linking in situ and satellite measurements
Remote Sensing of Environment
(1999) - et al.
Comparisons of land cover and LAI estimates derived from ETM+ and MODIS for four sites in North America: A quality assessment of 2000/2001 provisional MODIS products
Remote Sensing of Environment
(2003) - et al.
Evaluation of MODIS LAI, fAPAR and the relation between fAPAR and NDVI in a semi-arid environment using in situ measurements
Remote Sensing of Environment
(2004) - et al.
Assessment of the MODIS LAI product for Australian ecosystems
Remote Sensing of Environment
(2006) - et al.
Impact of sensor's point spread function on land cover characterization: assessment and deconvolution
Remote Sensing of Environment
(2002) - et al.
Estimation of leaf area index from transmission of direct sunlight in discontinuous canopies
Agricultural and Forest Meteorology
(1986)