Lake, Thomas2023-11-302023-11-302023-09https://hdl.handle.net/11299/258876University of Minnesota Ph.D. dissertation. September 2023. Major: Plant and Microbial Biology. Advisor: David Moeller. 1 computer file (PDF); x, 231 pages.Biological invasions, a prevailing consequence of global change, have produced remarkable insights into the mechanisms of species spread. Despite improvements in managing the economic and ecological impacts of invasive species, accurate predictions of their geographic distribution and potential for range expansion into new habitats remains an ongoing challenge. Inaccurate predictions can distort our perception of economic and ecological risks and mislead management strategies. Recent advances in statistical models and forecasting algorithms show promise in estimating invasion risk; however, these models often deviate from key ecological and evolutionary assumptions, limiting their predictive accuracy and biological relevance. As such, exploring multidisciplinary approaches that integrate ecological and evolutionary principles with predictive modeling can significantly enhance our understanding of spread mechanisms, evolutionary dynamics, and ecological impacts.In the following four chapters, I examine leafy spurge (Euphorbia virgata), among the most pernicious invasive plant species in North America, to address persisting challenges linked to the prediction of biological invasions. In Chapter I, I analyze the use of species distribution models (SDMs) to forecast potential range expansion of invasive plant species. I focus on several invasive plant species in North America and examine alternative methods of spatial bias correction and multiple methods for model evaluation. Models developed without bias correction are often overly complex and do not transfer well to expanding range fronts. On the other hand, models that employ bias correction measures tend to be less complex and project into incipient areas more effectively. Invasion history was associated with the effectiveness of bias correction techniques. These findings highlight the importance of considering model fit and complexity in building biologically realistic SDMs for invasive species. Using multiple metrics for model evaluation can improve a users’ confidence in predictions of potential invasion risk. In Chapter II, I explore the use of remote sensing techniques, specifically satellite imagery, to detect leafy spurge across a heterogeneous landscape in Minnesota, USA. I compare two types of satellite imagery: Worldview-2 with high spatial and spectral resolution but limited availability, and Planetscope with lower resolution but daily imaging coverage. Using convolutional neural networks (CNNs), the Worldview-2 imagery model achieved an accuracy of 96.1% in detecting leafy spurge, while the Planetscope model achieved 89.9% accuracy. To enhance the Planetscope model, I incorporate a time series of images using long short-term memory networks (LSTMs) to capture phenological information. This modified model achieved a detection accuracy of 96.3%, on par with the high-resolution Worldview-2 model. I find that early and mid-season phenological periods in the Planetscope time series, as well as specific spectral bands (green, red-edge, and near-infrared), are crucial for accurate detection of leafy spurge. These results demonstrate the potential of modest resolution satellite imagery to accurately identify individual invasive species over complex landscapes if a temporal series of images is incorporated. This research highlights the use of remote sensing and deep learning techniques for invasive plant detection, particularly in large-scale, remote, and data-sparse areas. In Chapter III, I address key challenges related to forecasting biological invasions: the prevalence of spatial bias in species occurrence data and the inclusion of population dynamics. Precise predictions rely on unbiased data and may benefit from including how populations respond to their environment. To examine bias, I expand the remote sensing analyses to nine states in the U.S., providing a comprehensive record of leafy spurge occurrences. I predict the probability of leafy spurge occurrence across nearly 2.2 million square kilometers using Landsat satellite scenes from 2000 to 2020 and temporal convolutional neural networks (TempCNNs). Compared to the remotely-sensed occurrences, community science records often contained substantial spatial biases. To examine how bias interacts with data on population dynamics, I develop demographic models to estimate changes in the probability of occurrence over time, which provides insights into regions experiencing population growth, stability, or decline. I then construct SDMs in a factorial manner, considering community science versus remotely-sensed occurrences and including/excluding information on population growth/decline. The results demonstrate that reducing spatial bias and incorporating population dynamics leads to expanded predictions of suitable habitat, which likely reflects a more accurate understanding of the potential for range expansion of invasive species over time. By accounting for population dynamics, the models effectively capture environmental variation underlying invasion rates and provide more reliable predictions of suitable habitat. These findings also provide valuable insights to land managers to identify regions with a high potential for rapid expansion. In Chapter IV, I investigate the role of rapid evolution in the range expansion of invasive species. I examine the consequences of range expansion on population genomic diversity, niche breadth, and the evolution of germination behavior in leafy spurge. I find limited evidence for population structure in one historically well-documented introduction into southern Minnesota, with other populations sampled throughout northern Minnesota, the Dakotas, and Iowa being largely panmictic. Counter to expectations, range expansion from southern to northern Minnesota resulted in only low losses in sequence diversity, with isolated populations at the northern range edge retaining similar levels of diversity. However, the climatic niche expanded during range expansion, with populations at the leading edge occupying cooler climates. Guided by these findings, I test for differences in germination behavior over the time course of range expansion using a common garden experiment and temperature manipulations. Germination behavior diverged during range expansion, with populations from later phases having higher dormancy at lower temperatures. While the biological mechanism explaining this trait divergence may be a topic for future research, these findings highlight the importance of considering rapid evolution in invasion dynamics and suggest that distribution models may underestimate invasion potential if populations are assumed to be homogeneous and evolution is not considered.enAdaptationInvasion biologyMachine learningThesis or Dissertation