Predictability of Cottonwood Recruitment Along a Dynamic, Regulated River

Abstract

Riparian vegetation planting and management are vital to river engineering projects. To inform of these activities, science needs to provide practitioners with a better understanding of influences on recruitment and where vegetation will most likely establish and survive. This study investigated whether the spatially explicit recruitment of Populus fremontii (Fremont cottonwood), a dominant riparian species in the western United States, could be predicted along a dynamic, alluvial regulated river. We used a ~34-km segment of the Yuba River in California, United States, which was mapped in 2017 after a large flood reset the terrain. Five years later from August through November 2022, a field campaign characterised precise locations of juvenile cottonwoods. We evaluated predictions from deterministic and statistical models. For the deterministic test, a spatially distributed riparian seedling recruitment model was used with expert-estimated parameters. The model was not accurate in this case but was informative. For the statistical approach, a supervised classification random forest (RF) algorithm, driven by 2017 hydrophysical and topographic variables, was trained and cross-validated using 2022 cottonwood presence and absence observations. The RF model had an overall accuracy of 87% and an AUC-ROC value of 94%, with the most important variables being the detrended DEM, channel proximity and inundation survival. Topographic variables were much more significant than hydrophysical ones. Further developments to understand underlying governing equations and recruitment model parameters will draw on lessons from the RF model. Both deterministic and statistical models are recommended to evaluate riparian vegetation restoration designs, as each yields unique insights.