Making choices that matter - use of statistical regularization in species distribution modelling for identification of climatic indicators - a case study with Mikania micrantha Kunth in India.
Biological invasions by alien non-indigenous species are one of the major problems of the present era which impose massive environmental and socio-economic costs. In India about 40% of the floral species have long been recognized to be aliens, but the need for priority conservation efforts has only been felt since the turn of the century. Thus, it is now of utmost importance to predict the potential distribution of invasive alien species and identify suitable environmental conditions that allow the species to spread rapidly. The invasive plant Mikania micrantha was chosen as the test species. Native occurrence records (longitude and latitude) were obtained from the Global Biodiversity Information Facility (GBIF). For Indian occurrences, GBIF records were supplemented with occurrences from herbaria label data and information gathered from published literature. Nineteen climatic variables were obtained from World-Clim database. To predict the potential distribution, species distribution models (SDMs) were built by using logistic regression and the climatic variables were chosen by using two cross-validated regularization methods induced by least absolute shrinkage and selection operator (lasso) and the ridge penalty function. This approach has twofold benefits; it deals with the multicollinearity problem efficiently and selects the raw environmental covariates. Fβ-score was utilized to measure the models' performance. Combining the data from both native and alien ranges, seven environmental predictors were selected using four different background choices. Using lasso penalty, mean diurnal range (mean of monthly (max temp-min temp)) (BIO2), Isothermality (BIO2/BIO7) (×100) (BIO3), Temperature Annual Range (BIO5-BIO6) (BIO7), Precipitation of Wettest Month (BIO13), Precipitation Seasonality (Coefficient of Variation) (BIO15) and Precipitation of Warmest Quarter (BIO18) were found to be strong correlates for all four backgrounds. The predicted probabilities from the model containing these seven selected variables, demonstrated higher invasion risk in the central part of India than the model containing all the predictors. Accurate analysis of present distributions and effective predictive modelling of future distributions of invasive alien species is of vital importance for the early detection of the invasion and rapid remedial actions downstream. This study may aid in the adoption of management initiatives like early detection and rapid response. This could result in identifying both new populations and established populations to be prioritized for management.