Examples of pkkalman
Assimilate fine spatial resolution NDVI time series acquired at (100 m) with median spatial resolution NDVI time series (300 m). Both time series consist of ten acquisitions in time (provided as a multi-band raster dataset), but fine spatial resolution data are sparse due to cloud coverage. Missing data are characterized by the option -obsnodata. Coarse spatial resolution data have occasional cloud cover too (-modnodata 255). The spatial coverage of the model must be at least the spatial coverage of the observation. Here, we run the kalman filter in forward mode only.
pkkalman -obs fine_10band.tif -mod coarse_10band.tif -obsnodata 255 -modnodata 255 -dir forward -ofw fine_filled.tif
Same as above, but run in forward, backward and smooth mode. Here we provide output filenames for the uncertainties of the predicted time series.
pkkalman -obs fine_10band.tif -mod coarse_10band.tif -obsnodata 255 -modnodata 255 -dir forward -ofw fine_forward_10band.tif -u_ofw fine_forward_uncert_10band.tif -dir backward -ofb fine_backward_10band.tif -u_obw fine_backward_uncert_10band.tif -dir smooth -ofw fine_smooth_10band.tif -u_ofb fine_smooth_uncert_10band.tif
If the observation time series is not complete, i.e., has less acquisitions than the model, we must explicitly provide the time sequence. Here, the observation at fine spatial resolution is based on two acquisitions only (at times 3 and 8). The model at coarse spatial resolution is based on ten acquisitions (at times 1-10).
pkkalman -obs fine_2band.tif -mod coarse_10band.tif -obsnodata 255 -modnodata 255 -dir forward -ofw fine_forward_10band.tif -tmod 1 -tmod 2 -tmod 3 -tmod 4 -tmod 5 -tmod 6 -tmod 7 -tmod 8 -tmod 9 -tmod 10 -tobs 3 -tobs 8
In the following example the missing data are not coded within the observations, but read from an external mask:
pkkalman -obs fine_10band.tif -mod coarse_10band.tif -obsmask finemask_10band.tif -modmask coarsemask_10band.tif -msknodata 1 -mskband 0 -dir forward -ofw fine_forward_10band.tif
The observations and model can be provided as multiple single-band raster datasets. The time sequence for the two observations and ten model inputs are explicitly provided via the options -tobs and tmod.
pkkalman -obs fine_1.tif -obs fine_2.tif -obsmask finemask_1.tif -obs finemask_2.tif -mod coarse_1.tif -mod coarse_2.tif -mod coarse_3.tif -mod coarse_4.tif -mod coarse_5.tif -mod coarse_6.tif -mod coarse_7.tif -mod coarse_8.tif -mod coarse_9.tif -mod coarse_10.tif -modmask coarsemask_1.tif -modmask coarsemask_2.tif -modmask coarsemask_3.tif -modmask coarsemask_4.tif -modmask coarsemask_5.tif -modmask coarsemask_6.tif -modmask coarsemask_7.tif -modmask coarsemask_8.tif -modmask coarsemask_9.tif -modmask coarsemask_10.tif -mskband 0 -msknodata 1 -dir forward -ofw fine_forward.tif -tmod 1 -tmod 2 -tmod 3 -tmod 4 -tmod 5 -tmod 6 -tmod 7 -tmod 8 -tmod 9 -tmod 10 -tobs 3 -tobs 8