Context 

Shifts in precipitation regimes due to climate change are significantly impacting dryland ecosystems, including vegetation composition and structure. Unoccupied aerial vehicles (UAVs) are widely used to monitor vegetation, but whether models built to predict changes in these characteristics are robust under extreme precipitation regimes is unclear.

Objectives

We aimed to predict key vegetation characteristics under three precipitation regimes (ambient, drought, and water addition) and assess model performance across these moisture conditions. We also evaluated how models built under ambient conditions predicted vegetation characteristics under extreme precipitation regimes.

Methods

UAV surveys were conducted at five sites subject to long-term precipitation manipulation along an elevation gradient in northern Arizona, United States (U.S.). Twenty-one vegetation indices and point cloud data from the UAV imagery were used to develop models to predict vegetation structure and composition characteristics. Model performance and transferability were assessed via error and directional bias within each treatment (i.e., in situ) and from ambient to precipitation treatments (i.e., model transfer).

Results

UAV-based models accurately measured vegetation characteristics across all regimes, but maximum height showed significantly higher error under drought conditions. Models developed under ambient precipitation and applied to extreme precipitation treatments exhibited significant differences in the error and directional bias, indicating they may not be suitable under climate change.

Conclusions

UAV-based models are effective for monitoring vegetation characteristics but may lose accuracy under extreme precipitation regimes expected under climate change. This study emphasizes the need to improve model transferability and suggests refining landscape monitoring approaches to consider extreme changes in precipitation and associated vegetation responses.