Remote Sensing of Coastal Vegetation: Dealing with High Species Turnover by Mapping Multiple Floristic Gradients

Abstract

Abstract Aims Mapping gradual transitions in plant species composition via a combination of ordination and regression from remote sensing data is becoming an established approach. However, straightforward analysis of areas with high species turnover rates may result in a loss of information since a high level of generalization is required. In this study, we investigate whether analysis of more homogeneous subsets, in contrast to processing of the complete dataset, is a viable approach to mapping multiple floristic gradients. Location The coastal nature reserve ``Zwin’’ (Belgium). Methods The measured dataset is partitioned into more homogeneous subsets based upon species composition using hierarchical classification. The dataset and subsets are then processed separately. First, ordination is performed to extract floristic gradients in plant species composition; second, these gradients are related to airborne hyperspectral remote sensing data through regression models and mapped by projecting these models on image data. Regression validation and Mantel tests are used to compare the results within the study and to other studies. Results Hierarchical classification resulted in two homogeneous vegetation subsets. Ordination yielded four gradients in the area and all regression models compared favorably to similar studies in other areas with R textasciicircum2 values ranging from 0.47 to 0.74. The Mantel test showed that by dividing the dataset into subsets, higher resemblance to the original vegetation data can be achieved. Conclusion We showed that mapping gradual transitions in plant species composition across multiple subsets sampled from one measured vegetation dataset is a promising approach for retrospective analysis of areas with high species turnover rates. In addition to potential improvements in performance, this complementary analysis enables mapping of additional gradients, suggesting that all conventionally predicted maps remain available, valuable, and necessary for thorough understanding of plant species composition.