Vegetation mapping and its spatial accuracy based on drone multispectral images of secondary vegetation in southwestern Japan

Abstract

Monitoring the type and coverage of vegetation using remote sensing data is important for ecological management and research. In this study, we used drone multispectral imagery to produce a fine-scale vegetation map of secondary vegetation within part of the campus of Hiroshima University, Japan. We examined the spatial accuracy of the vegetation map through comparison with ground-based survey data in which the vegetation was classified into six physiognomic units: secondary grassland, bamboo, and woodland dominated by broadleaved deciduous, broadleaved evergreen, pine, and coniferous trees. The drone images were acquired in June, September, October, and November 2020. We analyzed the improvement in vegetation classification accuracy achieved using a composite seasonal image instead of the single season images. We also analyzed the improvement regarding the spatial accuracy of vegetation mapping obtained using various vegetation indices instead of the raw multispectral images. The overall accuracy of the seasonal composite data was 90 %, i.e., significantly higher than that of the single season images (70 %; two-way ANOVA, p < 0.01). In contrast, none of the vegetation indices produced significant improvement in terms of spatial accuracy for any of the vegetation units (one-way ANOVA, p > 0.01). The findings of the present study revealed that the advantage of capturing phenological change in multiseasonal drone data is essential for accurate vegetation mapping.