EcoImaging: Advanced sensing and machine learning for investigating aboveground and belowground interactions in natural and managed ecosystems

Abstract:

Capturing the interactions between vegetation processes, land surface, and subsurface can provide valuable information about the functioning of terrestrial ecosystems. By using remote sensing techniques, plants can be efficiently characterized at different spatiotemporal scales. Additionally, geophysical methods can image subsurface physical and hydrological properties at high resolutions. By combining aboveground and belowground sensing techniques (referred to here as EcoImaging), we are able to link environmental factors that drive ecological outcomes at an unprecedented scale.

We present results from case studies in both natural and managed ecosystems. The first case study focuses on natural ecosystems within a mountainous watershed, where plant distribution and characteristics play a key role in water, carbon and nutrient cycles. While ecological studies are often performed at the plot scale, the spatial characterization of plant community distributions and identification of key environmental factors driving such heterogeneity provides essential insights and is enabled by advanced imaging. Here, we developed a machine learning approach to map plant communities at high-resolution using optical, hyperspectral, and LiDAR remote sensing data. By performing a multivariate analysis to identify co-variabilities among plant communities, topographical metrics, and soil properties derived from geophysical measurements, we identify soil moisture and microtopographical features as primary controls on the distribution of specific plant communities. The second case study focuses on a managed ecosystem, at farm scale, located in Arkansas. The EcoImaging approach is used to investigate soil-plant spatiotemporal co-variability and assess the impact of soil characteristics and water management on plant development and crop yield. UAV-based imaging at a cm scale is performed since the early vegetative stage to map plant density and plant vigor which is needed to identify low productivity areas and guide near-real-time farming decisions. In addition, geophysical imaging of soil properties allow us to associate areas of low productivity with increased soil clay content, while the temporal analysis of geophysical data show the impact of soil moisture and irrigation practice on plant dynamics and yield.