Methodology to validate MSG vegetation products. Part Ⅱ: Geostatistical approaches for upscaling field data to high-resolution satellite scales

摘要

This paper is the second part of two-part set which proposes a methodology in order to validate the LSA SAF vegetation products (LAI/FVC/fAPAR) derived from SEVIRI/MSG. The main objective of this methodology refers to assessing the uncertainty of SEVIRI/MSG products by analytical comparison to in situ measurements. The scaling problem is solved in this work by considering high-resolution maps in order to make the direct comparison between ground truth and coarse-resolution products. A detail description of the measurement acquisition and estimates was presented in a first document whereas the estimation of the high-resolution biophysical maps from this in situ data set is undertaken in this paper. This work attempts to evaluate the capabilities of a geostatistical approach in the estimation of high-resolution LAI/FVC/fAPAR maps. The geostatistical approach is based on collocated cokriging, which allows to derive high-resolution maps from in situ measurements over a small area (5x5 km~2), centred at Barrax test site. This technique takes into account the spatial dependence of the data, the neighbouring information, densely sampled auxiliary information and the variance estimation as opposed to empirical functions. The method has shown to be appropriate for the spatial extension of in situ measurements. An important contribution of this work is the analysis of the uncertainties associated to the method which provides an appreciation of the varying precision of the cokriged estimates due to the irregular disposition of informative points by means of the estimated variance. On the other hand, a flag image is also provided by using the convex hull tool in order to account for possible uncertainties in previous steps to the final cokriging output.