Predictive Analysis of Invasive Plants

Overview

Invasion by non-native plants costs the US over $21 billion annually and can cause devastating effects on whole ecosystems. The best strategy of solving this problem efficiently and cost-effectively is through early detection and eradication, which relies on the accurate prediction of invading species. However, studies in this area are extremely limited. Hence, this study aims to utilize Machine Learning (ML) to develop a novel method of predicting the spread of invasive plants at an early stage. Using three Machine Learning algorithms: Random Forest (RF), Multivariate Multiple Linear Regression (MMLR), and Support Vector Machines (SVM), the desired models can be created. Trained on an invasive plant dataset from Africa and California, and correlated with environmental data from the Visual Crossing, the models produced proved highly accurate and reproducible in their ability to identify and predict the distribution patterns of the most invasive plants and the regions most at risk from invasion. When the methods applied on the national Tamarisk genus, the results were comparable to the Terra and Aqua satellite vegetation data from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors. Additionally, these models acted as a novel way of creating plant specific invasion curves, allowing greater knowledge of how to minimize and manage the spread of these invasive plants. These methods and models, applicable to any other taxa and location, represents an unparalleled opportunity to implement timely and proactive management strategies against biological invasions.