A key contributing factor in biodiversity loss is the lowering of intraspecific genetic variation within species. Lowering genetic diversity of a population causes increased susceptibility for diseases, reduced evolutional potential and lower fitness of a next generation. Species and individuals with different genotypes can be differentiated based on their morphological and physiological traits. In this study, we establish a direct connection between spectral data and intraspecific genetic variation of individual trees in a temperate forest system (Laegern, 47°28N, 8°21E), located in the Swiss midlands. The dataset includes genetic and spectral information. Genetic data contain microsatellite analyses of 77 F.sylvatica individuals. Remote sensing data encompass 7 multitemporal acquisitions of the APEX (Airborne Prism Experiment) imaging spectrometer with 2 m spatial resolution and 284 spectral bands. Imaging spectrometer data were obtained at various times during growing season between 2009 and 2016. We used Partial Least Squares Discriminan Analysis (PLS–DA) to classify genotypes based on spectral information. Different strategies to select the PLS-DA models input data were tested and compared.
Results/Conclusions
Based on Bayesian analyses of microsatellite data and K-means clustering three genotypes within sampled F.sylvatica trees were distinguished. Using all the spectral bands as input for the PLS-DA models result in a poor discrimination of the genotypes. Therefore, we selected spectral bands with the highest discrimination power based on Analysis of Variance (ANOVA), Variable Importance in Projection (VIP) score and Principal Component Analysis (PCA). The wavelength regions between 1.05 µm and 2 µm revealed the highest potential to discriminate the three genotypes. Depending on input data selection based on ANOVA, VIP or PCA analyses, different PLS-DA model performances were observed. Identifying the specific wavelength regions of the electromagnetic spectrum that are related to genotypic variation and using these data as input for PLS-DA models may enable intra-specific discrimination based on in-situ spectral information. Furthermore, the distinction of trees with different genotypes based on remote sensing time-series helps to trace back genetic diversity loss in time and could contribute to the conservation of the species and habitats.