Compositing to 10-daily composites
This section describes the compositing performed in the pre-processing of the images of all sensors except MSG-SEVIRI.
For a given period (day or dekad), all corrected segments are searched with at least partial coverage over the specified output ROI. This generally implies that for each output pixel, different observations are available, from different segments or registration dates. By means of the selected compositing rule, these are combined into a single "best" observation, which is then written to the out-IMG. As the "observation" comprises different "layers" (reflectances, angles, SM, derived VI's or fAPAR,...), the same will be true for the composite IMG.
The compositing rule is a "constrained" Max-NDVI for AVHRR (NOAA and METOP), which means that the observation with the highest NDVI is selected and transferred (with all its spectral values) to the out-IMG. In this way clouds, snow and water (which all have low NDVI) are suppressed. Instead of NDVI another vegetation indicator fAPAR is selected for the images of SPOT-VEGETATION and TERRA-MODIS. This mostly gives identical outcomes. Over sea the Max-NDVI | fAPAR has adverse effects, as it mostly selects the clouds instead of the cloudfree water (lowest NDVI | fAPAR). This is the case for the standard SPOT-VGT syntheses of CTIV. Therefore, when the pre-processing is tailored for MCYFS, the water surfaces are composited with a Min-NIR rule, which always promotes the water (near-zero NIR-reflectance).
The adjective "constrained" means that other factors are taken into account, such as the view and solar zenith angles and the settings of the status map. In practice, the available observations for each pixel are first classified according to the scheme of the table below, which holds for the AVHRR and MODIS-chains. First all observations are completely discarded if s>75° or v>T2 (45° for AVHRR, 35° for MODIS). The remaining data is grouped in six classes, following the table (for MODIS T1=T2, so only three classes remain: B, C, D). Then the highest non-empty class is searched, with the hierarchy D2>D1>C2>C1>B2>B1. If this best class only contains a single observation, that one is selected. If more than one, the best is selected with the above rule (Max-NDVI or fAPAR for land, Min-NIR for water). If the pixel belongs to class A (no good observations) its position in the composite is flagged (in all spectral layers). So even if all observations are cloudy, the composite still will contain values, but the SM will indicate the cloudy nature.
Table Classification scheme for the compositing of AVHRR (T1/2=40/45°) and MODIS (T1=T2=35°)
Images with meteorological data (temperature, rainfall,...) or combined info such as DMP (RS+meteo) can be composited as well. However, here the most logical "rule" is to compute the simple mean or sum over all valid (cloud/snow-free) observations. Compositing is a crucial step: whereas the individual segments contain a lot of clouds and occupy different and scattered areas, the composite or synthesis images are better "filled", less contaminated by clouds and they always cover the full area of the ROI. Of course, a longer compositing step improves the quality of the resulting syntheses, but it also decrease the temporal detail of the final time series of composite images. For the MCYFS, the 10-daily step is the most appropriate (though the post-processing also creates monthly composites).
Quality of the compositing
The different compositing steps (Segments→S1→S10 →S30) can be evaluated by means of the following two ancillary images, which are also created by the involved programs:
- IMG with Ngood: This gives per pixel the number of unflagged measurements available for the compositing. The higher Ngood for a certain pixel, the better the choice.
- IMG with Segment_ID: This is the ID of the raw segment whose values were selected by the compositer for the concerned pixel. Via this ID, also the other characteristics of the segment can be traced back (registration data/time, cloud cover, RMSE,...). The quicklooks with the registration date are derived in this way.