Exploring the utility of temporal information in improving invasive alien tree mapping using Sentinel-2 imagery in the Global South

This page contains the code used in the study "Exploring the utility of temporal information in improving invasive alien tree mapping with Sentinel-2 imagery in southern Africa ". The aim of this work was to compare different methods to improve the accuracy of mapping invasive alien trees. This study assessed the effectiveness of different temporal approaches using Sentinel-2 imagery in the Grassland Biome of southern Africa. Using a Random Forest classifier trained and validated with field-collected data, four temporal approaches were compared: a single date to match the fieldwork campaign (in this case in the dry season), a single date to maximize phenological differences (the flowering of invasive wattle), all the scenes available in the calendar year relating to the fieldwork, and monthly composites for the same year. Treatments 1. Fieldwork date: Single (Scenes before fieldwork mosaicked into one image) 2. Phenological date: Single (Scenes from August mosaicked to highlight phenological contrasts). 3. 2023: All scenes Multiple (All scenes used to generate a median annual composite). 4. 2023: Monthly composites (Monthly composites for 2023) Contents Python and Javascript: Codes used for each classification for each treatment . Training data: Field-collected training data points used to train the classifier. Classification results: GeoTIFF files showing the classification outputs for each dataset. Metadata document: A detailed explanation of each dataset. Data accessibily To ensure replicability, we provide the training datasets and code used in this study. Access to imagery is as follows: Sentinel-2: Freely available (for example on Google Earth Engine)