Presentation Type
Oral Presentation
Category
STEM (science, technology, engineering, mathematics)
Abstract/Artist Statement
A winter snowpack is not runoff capable until it is warmed to 0 degree Celsius and saturated in liquid water, the achievement of this state is called ripening Whether a snowpack is ripened or not is a critical information of the snowpack state and behavior regarding water availability and floods. Before a snowpack is ripened, it is a buffer to rain and intense melt. Once a snowpack ripens, rain and high temperature will cause water releases that may be available for soil and ecosystems, but can also cause floods. In addition, hydroelectric dams rely on gradual release of water caused by gradual ripening of the snowpack in the spring in order to use and store water resources efficiently. Yet, the variability of the snowpack ripening is poorly constrained, especially within a watershed with complex topography and considering inter-annual variability. UsingK-Means clustering on the output of SNODAS, a widely used operational model, we identify a tipping point in the behavior of the snowpack. Warm and low-snow years experience chaotic behavior and mid-winter releases of water instead of ordered progressive spring stream and soil recharge. Using a Long Short-Term Memory Neural Network and K-Means clustering, we develop a methodology to predict the spring ripening of a snowpack using information on the snowpack during the first half of winter, with an error of 14 days, corresponding to the natural variability of weather.
Mentor Name
Joel T. Harper
Personal Statement
Numerous companies and scientists aim to predict floods that can have death tolls and billions-worth of damages, especially in a warmer world. Yet most groups neglect the state of the snowpack that is rained on. Poor snowpack years also have numerous implications on droughts and wildfires, issues that Montana already suffers of, for which earlier preparation saves lives, wildlife and infrastructures. This work uses a scientific approach and Machine Learning methods, with never seen before performances, to create, test and deploy operational forecasting procedures. In addition, this work extracted and confirmed the previously found malfunction of ecosystems in a warmer and drier work. This has immediate consequences on our life in Montana, where aquifers, ecosystems and recreation rely on the stabilization induced by a strong snowpack. The significance of this work and the results exposed above make me consider my application worth consideration for the "Best of GradCon" award.
Video presentation
Patterns and prediction of the snowpack ripening using Machine Learning methods
UC 333
A winter snowpack is not runoff capable until it is warmed to 0 degree Celsius and saturated in liquid water, the achievement of this state is called ripening Whether a snowpack is ripened or not is a critical information of the snowpack state and behavior regarding water availability and floods. Before a snowpack is ripened, it is a buffer to rain and intense melt. Once a snowpack ripens, rain and high temperature will cause water releases that may be available for soil and ecosystems, but can also cause floods. In addition, hydroelectric dams rely on gradual release of water caused by gradual ripening of the snowpack in the spring in order to use and store water resources efficiently. Yet, the variability of the snowpack ripening is poorly constrained, especially within a watershed with complex topography and considering inter-annual variability. UsingK-Means clustering on the output of SNODAS, a widely used operational model, we identify a tipping point in the behavior of the snowpack. Warm and low-snow years experience chaotic behavior and mid-winter releases of water instead of ordered progressive spring stream and soil recharge. Using a Long Short-Term Memory Neural Network and K-Means clustering, we develop a methodology to predict the spring ripening of a snowpack using information on the snowpack during the first half of winter, with an error of 14 days, corresponding to the natural variability of weather.