![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Validation of soil organic carbon predictions based on airborne imaging spectroscopy and multivariate regressions |
VerfasserIn |
Antoine Stevens, Bas van Wesemael, Isabel Miralles |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250051237
|
|
|
|
Zusammenfassung |
Multivariate regressions are frequently used to predict soil properties from airborne imaging
spectroscopy data. Samples are taken in the field and their properties analysed using
conventional methods. The predictive ability of the models is often estimated by randomly
splitting samples into a calibration and validation set. This procedure is likely to overestimate
the true prediction accuracy for pixels that are spectrally different to the calibration set. Based
on a hyperspectral image acquired over the Grand-Duchy of Luxembourg (flight line of 420
km2), we propose to validate multivariate calibration models (Partial Least Square and
Penalized-spline Signal regressions) of Soil Organic Carbon (SOC) content using an
independent set of samples. We observed an increase in prediction error between the
calibration and validation set in the order of 8 to 138 %, the best model reaching
Root Mean Square Error (RMSE) of 4.5 gCkg-1 and Ratio of Performance to
Deviation (RPD) of 1.75. An analysis of the pattern of the errors revealed that this
phenomenon can be related to the location of the validation samples in the score space. In
order to produce a reliable SOC map, multivariate models were refitted through a
cross-validation of the entire sampling database. The distributions of predicted and
observed SOC were compared at both regional (difference between mean values
 1 gCkg-1) and field scales (differences < 5gCkg-1). Finally, the potential of
imaging spectroscopy for SOC monitoring were explored with a Power Analysis. |
|
|
|
|
|