Dominant Species-Physiognomy-Ecological (DSPE) System for the Classification of Plant Ecological Communities from Remote Sensing Images

Abstract

This paper presents the Dominant Species-Physiognomy-Ecological (DSPE) classification system developed for large-scale differentiation of plant ecological communities from high-spatial resolution remote sensing images. In this system, the plant ecological communities are defined with the inference of dominant species, physiognomy, and shared ecological settings by incorporating multiple strata. The DSPE system was implemented in a cool-temperate climate zone at a regional scale. The deep recurrent neural networks with bootstrap resampling method were employed for evaluating performance of the DSPE classification using Sentinel-2 images at 10 m spatial resolution. The performance of differentiating DSPE communities was compared with the differentiation of higher, Dominant Genus-Physiognomy-Ecological (DGPE) communities. Overall, there was a small difference in the classification between 58 DSPE communities (F1-score = 85.5%, Kappa coefficient = 84.7%) and 45 DGPE communities (F1-score = 86.5%, Kappa coefficient = 85.7%). However, the class wise accuracy analysis showed that all 58 DSPE communities were differentiated with more than 60% accuracy, whereas more than 70% accuracy was obtained for the classification of all 45 DGPE communities. Since all 58 DSPE communities were classified with more than 60% accuracy, the DSPE classification system was still effective for the differentiation of plant ecological communities from satellite images at a regional scale, indicating its applications in other regions in the world.

Keywords

• dominant plant species
• plant communities
• ecological communities
• remote sensing images
• classification
• recurrent learning