January 13, 2009

LAI Validation

The validation process has first to verify the consistency of its spatial and temporal variations. Thus, it appears that the cloud contamination of surface reflectances creates a non-realistic high spatial variability. This will be corrected in the ADEOS-2/POLDER-2 advanced algorithm with a multi-temporal filtering module that eliminates the residual clouds and/or aerosols. The merging algorithm creates some spatial inconsistencies in the case of bi-modal distribution of LAI "track" favoring the values close to the mean global LAI equal to 1.5. Some developments are in progress to improve this aspect. However, the preliminary POLDER-1 LAI products well reproduce the vegetation gradients at the continental scale as shown on Figure 1. The inversion of seasons between the two hemispheres is clear for the transitional ecosystems of which vegetative cycles are well marked, while the equatorial African forest is stationary. LAI values are relevant, very low or even equal to 0 on the desert, around 5 on the wet broadleaf evergreen forest, and intermediary on various shrubland and woodland. This realism emphasizes the local characteristics such as the arid area of Kalahari crossed by the transect around 21° South.

The second step in the validation procedure consists in comparing POLDER LAI with other LAI products and in-situ measurements, considered as references. The ground measurements have been collected during international campaigns or regional studies. Data are available in the literature or in various databases and have been compiled by CNRM (Centre National de Recherches Météorologiques). The comparison is limited by the spatial representativity of in-situ measurements, the low resolution of POLDER (6km) and the time shift between the acquisitions. However, a large number of points allows estimating the relevance of POLDER LAI. Only data collected during the last decade have been used for the comparisons in order to limit the risk of major changes on the surface. Each ground data is associated with the POLDER LAI of the synthesis period closer to the measurement date. When a data is collected during the summer of the North hemisphere, it is associated with the POLDER LAI of June. The comparisons on Figure 2 show a general good agreement. The main discrepancies appear on pure conifer forests for which the Kuusk' model (1995) has some difficulties to retrieve relevant LAI. The difference is interpreted as a consequence of the algorithm hypothesis considering the canopy as a turbid medium, and not taking account for the foliage aggregation, which is important for conifers. That leads to a systematic underestimation of LAI, more than a factor of 2.

On other ecosystems, the POLDER LAI globally give correct values and time profiles as shown on Figure 3. The foliage growth during spring is well reproduces for crops, grassland, and deciduous forests in spite of the time shift between the respective acquisitions. The lowest values on evergreen forests are due to cloud contamination. On the pine forest of Landes (Figure 3m), POLDER LAI displays a vegetative cycle characteristic of the undergrowth. These time profiles sample all the terrestrial ecosystems and climatic environments. Their consistency demonstrates the good quality of these preliminary POLDER-1 LAI products.

LAI reference maps have generated in the frame of the VALERI (Validation of Land European Remote sensing Instruments) project from high-resolution images and ground measurements collected during the 2000 and 2001 campaigns. Instrumented sites are cropland, temperate pine forest and boreal forest, sahelian grassland, and palm tree plantation. Comparison between POLDER LAI and VALERI LAI are generally satisfactory despite the time shift of 3 or 4 years between the respective acquisitions, and the low spatial resolution of POLDER. Figure 4 shows an example of comparison for POLDER-1 pixels located on the large area of boreal forest and mixed agricultural fields instrumented in Estonia during the 2000 and 2001 campaigns. The POLDER-1 LAI displays consistent values even if they are lightly lower than VALERI LAI is.


Kuusk, A., A fast invertible canopy reflectance model, Remote Sensing of Environment, 51:342-350, 1995.