On the bias of instantaneous FAPAR estimates in open-canopy forests.

Abstract Global products of the fraction of absorbed photosynthetically active radiation (FAPAR) are operationally available from a variety of space agencies. A proper validation of these products is essential and hinges on the acquisition of accurate ground-based FAPAR estimates of the vegetation contained within the field of view of the space sensor at the time of satellite overpass. Often remotely sensed FAPAR products are defined with respect to theoretical rather than ambient illumination conditions which complicates in situ validation efforts. Similarly, the spatial complexity and substantial heights of certain plant environments may prevent the reliable sampling of certain radiation fluxes. As a consequence, many field campaigns are carried out on agricultural crops or within young tree plantations where canopy height is not an issue. This contribution compares different approaches for estimating instantaneous FAPAR in tall, open-canopy forest stands under a variety of architectural, spectral and illumination related conditions. The bias associated with these estimations is separated into a sampling error and a transfer bias. The former relates to the impact of both the number and location of the measurements whereas the latter addresses the quality of the theory that relates these measurements to the actual canopy FAPAR. Among the various methods tested it was the 2-flux FAPAR estimator (1 − TPAR) that performs best in open forest canopies under typical summer conditions. The quality of the 1 − TPAR canopy FAPAR estimator changes, however, with illumination conditions, foliage colour and especially with the background brightness. Similarly, the smaller the size of the area for which the FAPAR is to be estimated the larger the variability of the bias is going to be (and this irrespective of the choice of in situ estimation techniques). Evidence is provided that working under overcast sky conditions will reduce the sampling error but may well increase the transfer bias when compared to clear sky conditions. A parametric relationship is developed that allows to predict the instantaneous canopy FAPAR for arbitrary diffuse-to-total-incident-radiation ratios (at any given solar zenith angle). This approach has a similar transfer bias as the 1 − TPAR method when the forest floor is dark but dramatically outperforms the 2-flux approach under snowy background conditions (RMSE = 0.9934 versus 0.5801, respectively). The number of samples acquired was found to be crucial in reducing the variability of the bias of a given FAPAR estimator. Both random and grid-based sampling schemes result in similar FAPAR biases but do not lend themselves easily to the acquisition of hundreds of data points needed for reliable estimations under direct-only illumination conditions. Transect sampling—which is shown to deliver best results if carried out at ninety degrees to the solar azimuth angle—appears ideally suited to acquire the necessary numbers of samples enabling the generation of accurate quasi-instantaneous FAPAR estimates in open-canopy forests.