![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Error structure in simulated and measured snow cover information |
VerfasserIn |
S. Kolberg, K. Engeland |
Konferenz |
EGU General Assembly 2009
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 11 (2009) |
Datensatznummer |
250027992
|
|
|
|
Zusammenfassung |
The use of satellite data in calibration and updating of snow cover models require an
assessment of the error structure in order to assimilate the remotely sensed information with
other types of data. In particular for grid distributed models, the spatial covariance needs to
be modelled in order to avoid over-conditioning on the potentially very high nominal number
of measurements.
It is shown that use of the Normalised Difference Snow Index (NDSI) directly as the
interface variable, rather than re-scaling and truncating to fractional snow covered area
(SCA), facilitates the use of a Normal error model, and removes the most dramatic
heteroscedasticity. The dependency of simulated and measurement errors on forest cover,
elevation and terrain exposure is analysed, as is the spatio-temporal correlation structure of
these errors.
Table 1 summarises the most important reasons why an assumption of independent
errors (like when multiplying single-observation likelihood terms) is likely to cause
over-conditioning. An alternative error model attempting to provide a more realistic
assessment of the information content in the data is proposed.
Table 1: Imperfections degrading the performance of a simple multiplicative error
model. Spatial connectivity of terrain attributes also yields spatially correlated
errors.
Redundancy source Measured values Simulated values
Spatial
autocorrelation Imperfect atmospheric
correction (image specific)
Errors in the snow storage or melt depth
GMRF surfaces
Bias in elevation gadients
Temporal
autocorrelation Temporally stable (but
spatially heterogeneous) bare ground
reflectance Biased melt rate
Albedo-melt feedback
Non-Gamma distribution
Sub-grid heterogeneous melt
Correlation with
terrain attributes Heterogeneous snow reflectance (fresh
/ ripe snow)
Heterogenous illumination
Varying forest cover Biased elevation gradient in storage or
melt depth.
Errors in albedo or radiation
simulations
Heteroscedasticity Varying NDSI sensitivity to snow and
bare ground reflectance Varying SCA sensitivity to melt depth
Deviation from NDSI-SCA
transformation model |
|
|
|
|
|