非表示:
キーワード:
-
要旨:
Spatiotemporal observations in Earth System sciences
are often affected by numerous and/or systematically
distributed gaps. This data fragmentation is inherited from
instrument failures, sparse measurement protocols, or unfavourable
conditions (e.g. clouds or vegetation thickness in
case of remote-sensing data). Missing values are problematic
as they may cause analytic biases and often inhibit advanced
statistical analyses. Hence, gapfilling is an undesired
but necessary task in Earth System sciences. State-of-the-art
gapfilling algorithms based on Singular Spectrum Analysis
(SSA) exploit the information contained in periodic temporal
patterns to fill gaps in the observations. Here we propose
an extension of this method in order to additionally consider
the spatial processes and patterns underlying most geoscientific
datasets. The latter has been made possible by including
a recently developed 2-D-SSA approach. Using both artificial
and real-world test data, we show that simultaneously
exploiting spatial and temporal patterns improves the gapfilling
substantially.We outperform conventional approaches
particularly for large and systematically recurring gaps. The
new method is reasonably fast and can be applied with a minimum
of a priori assumptions regarding the structure of the
data and the distribution of gaps. The algorithm is available as a ready-to-use open source software package.