Mapping urban tree species by integrating multi-seasonal high resolution pleiades satellite imagery with airborne LiDAR data

Mapping individual tree species is critical to understand the ecosystem services value of the urban forest and improve management. The main aim of this study is to assess the effectiveness of integrating multi-seasonal high resolution Pleiades satellite images with airborne LiDAR data for improving urban tree species mapping in the City of Tampa, FL, USA. In this study, we evaluated and compared the abilities of multi-seasonal individual and combined Pleiades satellite images with airborne LiDAR data for classifying seven urban tree species to understand effects of season and tree canopy height information on tree species mapping accuracy. A multi-level classification system was adopted to classify image objects of the tree species and groups. To efficiently utilize seasonal change information of tree species, a seasonal trajectory difference index (STDI) was developed with multi-seasonal images. Species mapping accuracy was compared and evaluated among individual seasonal Pleiades images and combined dry-wet season images, and between feature types: (i) 6 canonical variables only, transformed by canonical discriminant analysis algorithm; ii) 6 canonical variables plus 2 normalized Digital Surface Model (nDSM) derived variables; and iii) 6 canonical variables plus 2 nDSM variables and 4 STDI band indices. The 6 canonical variables were extracted from all spectral and textural features (SFs) including tree species fraction features. Random Forest and Support Vector Machine classifiers were used to classify the seven tree species with tree crown image objects. The research results demonstrate that (1) with additional 2 nDSM variables and extracted 4 STDI band indices, the accuracy of mapping urban tree species could be further improved (p < 0.09); and (2) with the 6 canonical variables, tree species mapping accuracies could be significantly improved using late spring season (April) image compared to all other seasonal images (p < 0.01), combined dry-wet season images performed even better. Our results further confirm that there exists a significant seasonal effect on tree species classification and that the 2 nDSM variables and the newly developed STDI in this study are useful to improve mapping urban tree species.