An insight into machine learning models to predict the distribution of Leucanthemum vulgare Lam. in northwestern rangelands of Iran

Invasions by non-native species are an increasing problem, especially at valuable rangelands. In the present research, the suitable habitat of the invasive species of Leucanthemum vulgare Lam., in the rangelands of the Namin County in northwest Iran, was assessed using the topographic, climatic, and soil variables. Four machine learning models, including random forest (RF), boosted regression trees (BRTs), generalized linear model (GLM), and generalized additive models (GAM), were used in the R environment. This research was conducted in May and June 2019–2020. The presence and absence of the L. vulgare were recorded using a stratified random sampling method using a global positioning system. The soil samples were taken at a depth of 0 to 30 cm from the presence and absence sites of L. vulgare . The results showed that GAM performed with 95% Kappa, 91% AUC, 88% TSS better than others and, followed by GLM, BRTs, and RF in decreasing order among the implemented models. The predictive performance of the GAM model using tenfold cross-validation in the Caret package indicate that clay, organic matter, elevation, and phosphorus were the critical factors influencing the spread of L. vulgare . According to the GAM prediction, 12.95% (13,799.79 ha) of the Namin County is potentially suitable for the L. vulgare . The spatial pattern of invasive species in rangelands can help make sound decisions in controlling the expansion of invasive species and related ecological implications.