Spatial predictions of soil moisture across a longitudinal gradient in semiarid ecosystems using UAV and RGB sensors

Unmanned aerial vehicles (UAVs) can be used to monitor soil moisture (SM) in semiarid regions. We collected high resolution RGB imagery concurrently with hundreds of SM samples across nine sites over a latitudinal gradient in the western USA. We used RGB bands, texture metrics, and vegetation indices to predict SM using machine learning. The model showed a moderately acceptable accuracy (R2 = 0.63 with cross validation, R2 = 0.53 using an independent validation). Texture metrics (‘mean’ and ‘entropy’), and the Excess Green (ExG) index had high predictive power while RGB bands performed poorly. Unlike Idaho and Montana, the spatial predictions for Utah and California showed high reliability (α < 0.01). Novel linear equations for the conversion of raw digital number (DN) values to reflectance are provided and facilitate remote sensing applications that build upon UAV-RGB imagery as a robust and cost-effective pathway to model SM for monitoring semiarid ecosystems.