Differentiating forest types using TerraSAR–X spotlight images based on inferential statistics and multivariate analysis

Abstract This study investigated the potential of applying statistical analysis tests, for example, two sample Z-testand the Factor Analysis (FA) tool, on the TerraSAR-X backscattering coefficient, for distinguishing between different types of forestsand detecting changes in distribution and extent of them. Two sample Z-testis an inferential statistical test that determines whether there is a statistically significant difference between the means in the data from two independent groups. FA is a multivariate analysis that can examine the structure or relationship between variables. Twelve pilot plots for forestsof 17 ha were surveyed in a water protection catchment near Hanover, Germany. The foresttypes were deciduous, coniferous, and mixed. In order to sustain groundwater quality, deciduoustrees were planted over a period of several years to gradually replace the coniferous trees in the catchment area. Regular forestobservations were required to ensure that the percentages of deciduousand mixed forestsin this catchment areawere increasing relative to coniferous forests. Fourteen dual-co-polarized TerraSAR-X (HH/VV) images were used to monitor the forestsin the period from March 2008 to January 2009. The values of the backscattering coefficient (σ 0 ) for the test plots were statistically analyzed using the two sample Z-testand the Factor Analysis tools. The study showed that Factor analysis tools succeeded in differentiating between the coniferous forestand both the deciduous forestand the mixed forest, but failed to discriminate between the deciduousand the mixed forest. Only one factor was extracted for each sample plot of the coniferous forestwith approximately equal loadings during the whole acquisition period from March 2008 to January 2009. However, two factors were extracted for each deciduousor mixed forestsample plot, where one factor had high loadings during the leaf-on period from May to October, and the other one had high loadings during the leaf-off period from November to April. Furthermore, the research revealed that the two sample Z-testdifferentiated the deciduousand mixed forestsfrom the coniferous forest, and discriminated between deciduous forestand mixed forest. Statistically significant differences were observed between the mean backscatter values of the HH-polarized acquisitions for the deciduous forestand the mixed forestduring the leaf-off period from November to April, but no statistically significant difference was found during the leaf-on period from May to October. Moreover, plot samples for the deciduous foresthad slightly higher mean backscattering coefficients than those for the mixed forestduring the leaf-off period. Applying the Factor Analysis and the two sample Z-teston the backscattering coefficient of multi-temporal TerraSAR-X data facilitates distinction of foresttypes, tracks changes in forestpatterns, and estimates the extent of environmental disasters in forestregions. This accomplishes sustainable forest management, which can play an important role not only in preserving groundwater quality but also in achieving climate change adaptationgoals.