Skip to main content
U.S. flag

An official website of the United States government

16) Ensemble Machine Learning Accurately Predicts Coastal Shoreline Change Rate

Presentation Date
Tuesday, August 15, 2023 at 12:00pm - Tuesday, August 15, 2023 at 3:00pm



Ensemble Machine Learning ThumbnailArctic coastal environments are experiencing rapid changes due to the changing climate. However, our knowledge of the existing environmental factors that control and predict coastal soil erosion rates is limited due to the lack of comprehensive field measurements and understanding of the processes involved. Previous efforts to predict coastal erosion rates have utilized various methods, resulting in a wide range of predictions. In this study, used 15 observed environmental factors and a large dataset of field observations (n=11,546) from the North Slope of Alaska. We used an ensemble of machine learning approaches to achieve three main objectives: (a) identify dominant environmental predictors influencing coastal erosion rates, (b) derive empirical relationships between dominant environmental factors and erosion rates, and (c) use the derived relationships to predict coastal erosion rates and compare the prediction accuracy of simpler model with the machine learning predictions. Our results show that the ensemble of three machine learning approaches accurately predicted coastal erosion rates (R2 = 0.88-0.98). Among the 15 environmental factors considered, latitude, aspect, elevation, and temperature emerged as dominant predictors of coastal erosion rates. Notably, the machine learning models performed better in predicting erosion rates at exposed shoreline points compared to sheltered areas. The non-linear equations we derived demonstrated a similar prediction accuracy to that of the machine learning approach. Overall, this study provides valuable insights into the environmental controls of coastal erosion rates and helps narrow the range in predicting coastal erosion rates. The empirical relationships we developed can serve as potential benchmarks for evaluating representations of environmental controls in process-based models.

Funding Program Area(s)