Identification of Acacia invasive species in protected areas of Spain using PlanetScope high-resolution satellite images and machine learning models in time series: an important action for protective management of forests

Natural protected areas are important ecosystems in the world which provide valuable habitats for many unique fauna and flora. However, they have been destroyed by many natural and anthropogenic factors in recent years. One of the main destructive factors of forests in these ecosystems is the establishment of Invasive Plant Species (IPS). This study aims to map the forest area and Acacia sp. distribution (as the main IPS) in Cíes Islands in northwestern Spain using high resolution satellite images, and machine learning methods in time series. For this purpose, PlanetScope satellite images of the study area were obtained for June of 2016, 2020, and 2024. In addition, some data of location of tree species in forest cover and distribution of Acacia sp. were collected during fieldworks to use them as ground truth data. Four machine learning algorithms (RF, SVM, DT, and ANN) were used to separate the forest cover in time series. Results showed that Random Forest (RF) algorithm was the most accurate method to classify forest cover in the study area (for 2016, OA: 89.16 %, $\hat{k}$: 0.86; for 2020, OA: 93.33 %, $\hat{k}$: 0.91; for 2024, OA: 95.83 %, $\hat{k}$: 0.93). In addition, results demonstrated that the forest area increased in Cíes Islands during 8 years based on RF algorithm (2016: 129.99 ha, 2020: 134.05 ha, 2024: 135.45 ha). Results showed that the area of Acacia sp. increased about 28.01 % in these islands over 4 years (2016–2020) and then decreased about 21.31 % in following 4 years (2020–2024). Validation of Acacia sp. distribution maps showed that RF has effectively classified Acacia sp. in Cíes Islands in different years (for 2016, OA: 86.66 %, $\hat{k}$: 0.82; for 2020, OA: 90, $\hat{k}$: 0.87; for 2024, OA: 95 %, $\hat{k}$: 0.91). This study presented an innovative methodology that combined satellite imagery and machine learning models to provide the practical tools for forest managers against invasive plant species. The maps produced in this study facilitate the conservation of biodiversity inside protected areas in context of challenges of climate change.