| Zusammenfassung |
This study investigated the statistical relationship between defoliation in pine
forests infested by nun moths (Lymantria monacha) and the spectral bands of
the RapidEye sensor, including the derived normalized difference vegetation
index (NDVI) and the normalized difference red-edge index (NDRE). The
strength of the relationship between the spectral variables and the ground reference
samples of percent remaining foliage (PRF) was assessed over three
test years by the Spearman’s ρ correlation coefficient, revealing the following
ranking order (from high to low ρ): NDRE, NDVI, red, NIR, green, blue, and rededge.
A special focus was directed at the vegetation indices. In both discriminant
analyses and decision tree classification, the NDRE yielded higher classification
accuracy in the defoliation classes containing none to moderate levels
of defoliation, whereas the NDVI yielded higher classification accuracy in the
defoliation classes representing severe or complete defoliation. We concluded
that the NDRE and the NDVI respond very similarly to changes in the amount of
foliage, but exhibit particular strengths at different defoliation levels. Combining
the NDRE and the NDVI in one discriminant function, the average gain of
overall accuracy amounted to 7.8 percentage points compared to the NDRE
only, and 7.4 percentage points compared to the NDVI only. Using both vegetation
indices in a machine-learning-based decision tree classifier, the overall
accuracy further improved and reached 81% for the test year 2012, 71% for
2013, and 79% for the test year 2014. |
