Disentangling water flux dynamics on an eroded cropland using an automated chamber system, water stable isotopes, and novel data-driven machine learning approaches

Understanding the water cycle is increasingly crucial to assess ecosystem resilience and ensure sustainable management and food security. Within the terrestrial water cycle, Evapotranspiration (ET) plays a pivotal role returning 60% of terrestrial precipitation back to the atmosphere. In agricultural systems, especially in water-scarce regions, understanding the water use of crops relative to their productivity (water use efficiency, WUE) is of paramount importance.

The AgroFlux sensor platform, including an automatic, robotic FluxCrane, is part of a long-term experiment in an agricultural system. We combine three years of ET measurements and two years of fully automated water stable isotope measurements coupled with campaign-based soil and plant measurements. The system is measuring along an erosion gradient with three different soil types to examine small scale heterogeneity of soils and their effect during various environmental conditions on different crops. The automated system generates data with high temporal and spatial resolution resulting in a new class of data that both enables and demands modern, efficient data analysis approaches. We use data-driven machine learning modeling approaches as an interface between the high-resolution monitoring networks and campaign-based measurements to provide better predictive results.

With our research we try to improve the knowledge of evapotranspiration by using novel modeling approaches coupled with measurements of common environmental parameters, plant specific parameters and water stable isotopes. We are investigating the potential of evapotranspiration estimation and modeling, and the possibility of automatically measuring and modeling the isotopic signature of evapotranspiration to decompose the water cycle into its components.