MAISRC researchers are working to develop a first-of-its-kind eco-epidemiological model that will forecast the potential risk of spread of zebra mussels and starry stonewort across Minnesota. The model will take into account introduction probability, establishment probability, and levels of management interventions. This model will be used as a decision-making tool to generate effective intervention strategies and design cost-effective surveillance programs to mitigate and prevent the spread of AIS.
To establish introduction probability, pathways among lakes will be evaluated based on water connectivity, boater movement, and geographic proximity. To understand the establishment probability, researchers will use next-generation ecological niche modeling techniques with remote sensing data. Cumulatively, this will identify lakes or areas of the state that are at higher risk for AIS, including lakes that are highly vulnerable and lakes that may be “super-spreaders,” both of which will help prioritize management efforts.
Data from this project is now available for download and use:
This project resulted in a predictive risk model for zebra mussels and starry stonewort that estimates the probability of a lake becoming infested by 2025.
The model evaluated all 25,000 bodies of water in Minnesota that are recognized by the Minnesota DNR and took into account the three abovementioned factors for each. The simulation was run 10,000 times to produce a percentage probability of whether a lake will become infested with either invasive species. For example, a score of 0.3245 means that when the model was run 10,000 times, the lake became infested 3,245 times by 2025 – a 32.45% chance.
Note that each county is on a separate tab of the Excel spreadsheet.
Final report summary:
Ecological Niche Models: We created ecological niche models for starry stonewort under current and future climate scenarios, zebra mussels, and Heterosporis sutherlandae. These models provide projections of theoretically suitable lakes in Minnesota, based on known environmental conditions of the species in the native and invaded ranges. While the potential range varies for each species, it is clear from this project that there are many suitable, but not yet invaded, lakes in Minnesota. Efforts to prevent spread should remain a high priority for managers. It is also important to note that not all lakes are considered suitable and efforts to more strategically target intervention is warranted.
Network Models: We created network matrixes for watercraft movement, water connectivity and geographic proximity. These matrixes provide a robust dataset from which we can estimate connectivity through known high-risk human-mediated (watercraft) and natural (water and proximity) pathways. The watercraft and water connection data both provide directionality and weighted edge (e.g. number of boats and river distance) and were further developed for use in other parts of the project. Significant effort was made to adjust the sampling bias inherent in the watercraft inspection data to account for unequal sampling effort and sampling locations. This is achieved through a series of statistical models, including random forest (biased effort), logistic regression (biased location), gamma regression (number of boats) and linear regression (number of boats staying on the same lake, e.g. self-loop). Understanding and evaluating connections between lakes will help managers to prioritize prevention and early detection efforts. These data are now being used to inform the MAISRC project Decision-making tool for optimal management of AIS.
We successfully built predictive risk models for zebra mussels and starry stonewort to estimate the probability of a lake getting infested by 2025. The risk models quantified the risk by incorporating both watercraft and water connectivity and lake suitability (see above for data). The simulation allowed an infested lake to spread the AIS to uninfested lakes through these pathways following a stochastic process. Even with introduction, an uninfested lake could only become infested if the lake was suitable for the specific AIS. Once a lake switched from uninfested to infested, the lake became a new source from which the AIS could continue to spread in the model. Using known location data (confirmed zebra mussel infestations as of 2011 or starry stonewort infestations as of 2015) and a Bayesian modeling approach, the model was calibrated for the time period 2012-2017 (zebra mussels) or 2016-2017 (starry stonewort) with 26-time steps per year. The outputs were the number of infested lakes for each species each year. The model calibration performed very well and was used to project future risk scores for uninfested lakes from 2018-2025 using 10,000 simulations.
In addition to overall trends, the results provided a lake-level breakdown of likelihood of introduction via boats or water, and combined for both. For example, a score of 0.3245 means that 3,245/10,000 times the lake became infested by 2025. While the model is not perfect (no models are), the results are robust and provide useful information from which to make decisions. In isolation, this may not be entirely useful, but when considered across a watershed, county or state, the ability to rank waterbodies based on actual, not perceived, risk is a game changer. These data will also be useful overtime to assess the trends of AIS invasion (i.e. what happens if we do nothing?). In both cases, the scenarios project a future with more infested lakes, albeit not equally distributed. For example, NE Minnesota remains at relatively low risk, while Central Minnesota increases significantly.