Constraints in range predictions of invasive plant species due to non-equilibrium distribution patterns: Purple loosestrife (Lythrum salicaria) in North America

Abstract Predicting distribution patterns of invasive species in regions outside of their native range is a fundamental component of early warning systems. The first aim of this study was to analyse some of the constraints and limitations concerning the applicability of results obtained from predictive, eco-geographical modelling methods. The next main objective was to evaluate the minimum monitoring-time requirements for reliable range predictions based on non-indigenous occurrences. This was achieved by comparing departures in ‘model quality improvement’ with ‘elapsed time’ after initial species establishment and subsequently increasing levels of data completeness. Incomplete sampling or small population numbers are common problems when dealing with recently established non-indigenous species. To account for this, this study compared results from two recently developed methods which are supposedly able to deal with the ‘few known occurrences’ factor when predicting potential geographical distributions. Time series re-sampling was used as an historical simulation approach in order to apply a more realistic scenario of data situations typical for early stages of invasion processes. The well-documented invasion history of Purple loosestrife (Lythrum salicaria L.) in North America has provided an appropriate case study which highlights the effects of varying spatio-temporal trends, in terms of predictability of the currently invaded range. The results of both methods congruently showed that it would be unreasonable to predict the potential distribution of this species, to any acceptable degree of accuracy, on the base of the first few recorded data points. Based on a realistic scenario of spatial invasive spread, a reliable prediction of the current non-native distribution in North America was only possible after an elapsed time span of approximately 150 years. Even a prediction precision of only 50% of the current occurrences would require at least 100 years after naturalization. Generally, the predictive capacities of correlative models are conspicuously decreased when underestimated niche dimensions are included—irrespective of the used method. As invasive spread is irreversible in most cases, monitoring time requirements of 100–150 years have to be regarded hazardous and unacceptable. Consequently, large scale spatial predictions cannot rely on the analysis of currently known non-indigenous occurrences alone. One suggestion, therefore, may be to incorporate species’ ‘native range distribution’ data into the models in order to achieve more reliable spatial predictions over a shorter timescale.