Measuring and Modelling Evapotranspiration over Complex Terrain

This contribution deals with the spatial and temporal scales involved in the processes that control evapotranspiration (ET) and confront these with the merits and limitations of various observation and modelling techniques. We make a strong case for integrated approaches to further develop our understanding of evapotranspiration.

The most challenging, but at the same time most relevant conditions to accurately represent ET are found in semi-arid regions, specifically complex terrains with strong thermal contrasts between dry and wet (irrigated) areas. We will present three cases with different objectives in terms of processes that control ET and the methods used to study them. First is the LIAISE campaign, where we will focus on how to describe ET depending on the spatial scale considered ranging from regional to landscape to local scale. Second is the E-DATA campaign, where ET is controlled by a thermally driven and topographically enhanced regional flow that alters the turbulent mixing and the structure of the atmospheric boundary layer. Third deals with a machine learning approach to determine ET based on standard weather station data.

LIAISE took place during the summer of 2021 in the Pla d’Urgell region of the Ebro River Valley in Catalonia, Spain. The surface was homogeneous at the field scale (e.g. fields of alfalfa). However, the surface was heterogeneous at the regional scale (~10-100km) because of the spatial distribution of irrigated crops and dry natural vegetation. We examined the impact of the boundary layer on surface fluxes at two of the LIAISE sites: one in the irrigated, crop-covered area and one in the dry, naturally-vegetated area.  We use an atmospheric mixed-layer column model that is heavily constrained by the surface and boundary layer observations from the LIAISE experiment.

The E-DATA experiment took place during November 2018 and focussed on quantifying the processes that drive ET in a shallow lake surrounded by extremely dry conditions in a salt flat (Salar del Huasco) of the Chilean Atacama desert. We use the WRF model at 100-m resolution to represent the local processes as well as the heterogeneity and regional transport to understand the evaporation and ABL dynamics over the water.

The machine learning study explores whether a physics-informed machine learning  approach can be used to improve the estimated evapotranspiration for irrigated fields located in a desert environment, without arbitrary tuning after training  and only using readily available data (standard meteorological data and satellite derived vegetation indices). We focus on an irrigated pecan orchard in Northwest Mexico. Multi-year eddy-covariance ET estimates are used to train and validate the model.