Spatiotemporal reconstruction of gaps in multivariate fields using the direct sampling approach
Article first published online: 5 OCT 2012
©2012. American Geophysical Union. All Rights Reserved.
Water Resources Research
Volume 48, Issue 10, October 2012
How to Cite
2012), Spatiotemporal reconstruction of gaps in multivariate fields using the direct sampling approach, Water Resour. Res., 48, W10507, doi:10.1029/2012WR012115., , and (
- Issue published online: 5 OCT 2012
- Article first published online: 5 OCT 2012
- Manuscript Accepted: 30 AUG 2012
- Manuscript Revised: 7 JUL 2012
- Manuscript Received: 12 MAR 2012
- gap filling;
- remote sensing;
- soil moisture
 The development of spatially continuous fields from sparse observing networks is an outstanding problem in the environmental and Earth sciences. Here we explore an approach to produce spatially continuous fields from discontinuous data that focuses on reconstructing gaps routinely present in satellite-based Earth observations. To assess the utility of the approach, we use synthetic imagery derived from a regional climate model of southeastern Australia. Orbital tracks, scan geometry influences, and atmospheric artifacts are artificially imposed upon these model simulations to examine the techniques' capacity to reproduce realistic and representative retrievals. The imposed discontinuities are reconstructed using a direct sampling technique and are compared against the original continuous model data: a synthetic simulation experiment. Results indicate that the multipoint geostatistical gap-filling approach produces texturally realistic spatially continuous fields from otherwise discontinuous data sets. Reconstruction results are assessed through comparison of spatial distributions, as well as through visual assessment of fine-scale features. Complex spatial patterns and fine-scale structure can be resolved within the reconstructions, illustrating that the often nonlinear dependencies between variables can be maintained. The stochastic nature of the methodology makes it possible to expand the approach within a Monte Carlo framework in order to estimate the uncertainty related to subsequent reconstructions. From a practical perspective, the reconstruction method is straightforward and requires minimum user intervention for parameter adjustment. As such, it can be automated to systematically process real time remote sensing measurements.