Spectral and textural information for the detction of the invasive African tulip across a Samoan landscape

In this study, we examine whether combining high-resolution Planet multispectral satellite imagery and LiDAR-derived elevation data into two machine learning models can effectively detect African tulip (Spathodea campanulata P. Beauv.). A total of 43 feature datasets, and a spectral-based subset within, were examined using a random forest (RF) feature selection to determine the most suitable features and reduce dimensionality. Two feature dataset scenarios (combined and spectral) were then masked using a tree/non-tree dataset to demarcate tree pixels that could further be classified as African tulip and non-African tulip. This repo holds five jupyter notebooks that were used to generate the lidar-derived tree/non-tree mask, spectral and textural features (from lidar and multispectral satellite imagery), as well as conduct pre-processing (training data) to construct the classification models.