Riparian zones are areas of land around rivers, and have unique ecosystems with distinct vegetation and soil characteristics. As global river ecosystem loss continues to steadily increase, restoring vegetation in riparian zones has been a common way to counter the degradation. However, restoration efforts of these regions face challenges such as low vegetation survival rates, small-scale topographical complexities and region variability.

In this article, SEI scientist Romina Díaz-Gómez, in collaboration with University of California, Davis, studied the use of a random forest machine learning model for identifying locations to best plant vegetation in riparian zones. The model was trained on Light Detection and Ranging (LiDAR) data, using light-based remote sensing to measure topographic distances to the Earth. Using the Lower Yuba River in California, US, as the study area, the research team found that the random forest-powered machine learning model accurately predicted the presence of riparian vegetation around the river. The model also identified small-scale variation in surface elevation correlated most with riparian vegetation growth.

The team suggests four applications in which predictive models in vegetation can be used for improving riparian planting projects:

• adapting the framework and customizing it based on local LiDAR data for other rivers;
• providing a cost-effective opportunity for quickly identifying a wider set of sites for planting;
• identifying low-quality terrain conditions that can be improved upon; and
• improved understanding in interpreting probability values for managers and personnel.