Artificial intelligence applied to big data reveals that lake invasions are predicted by human traffic and co-occurring invasions

Abstract

Preventing the spread of aquatic invasive species is an important management action. Identifying the characteristics of lakes that are susceptible to invasion creates an opportunity for management groups to prioritize limited resources for high-risk areas. In this study, we leveraged big data from a popular fishing app and other publicly available sources of environmental and human-use exposure measurements to develop machine learning models to predict aquatic invasive species presence in 30,375 lakes in the upper Mississippi river basin of the United States. Our results predicted that an additional 665, 771, 544, 703, and 638 lakes in the basin are invaded or at high risk of invasion by Eurasian watermilfoil, curly-leaf pondweed, rusty crayfish, Chinese mystery snail, and dreissenid mussels, respectively. Lake invasions were predicted by a combination of environmental, human-use exposure, and community dynamics variables. Features that made a lake more attractive to recreationists were consistently important across our models including the presence of a boat ramp, larger lake size, and surrounding natural landscape. The importance of co-occurring invasive species in some models could reflect several scenarios including invasional meltdown, facilitation among species, similar pathways for introduction, or similar response to the environment. Our models predicted a higher proportion of invasions in less popular lakes compared to known invasions. The finding underscores the potential importance of less popular lakes in the invasion process and suggests that the detection of invasions may be lower in these lakes. These results serve as a valuable tool for data-driven management decisions and can provide actionable insights for effective aquatic invasive species management.