Real-time ET Estimation based on Soil Moisture Sensor Array

Abstract:

Evapotranspiration (ET) – the combination of evaporation and transpiration – is a critical component in the water cycle, accounting for a significant portion of the water budget. ET can be estimated with various approaches, including weighing lysimeters based on the soil water balances, remote sensing techniques based on surface energy balance, and eddy-covariance flux towers based on atmospheric dynamics models. Each approach involves uncertainty since they rely on indirect measurements and certain assumptions. Although remote sensing techniques can provide spatially extensive data, the uncertainty is known to be quite high.

Recent advances in sensor and telecommunication technologies enable spatiotemporally resolved observations of key properties such as soil moisture at depth-discrete locations in real time, which provides great potential for estimating actual ET. A time series of the vertical profiles of soil moisture, when properly calibrated, can help to capture the water balance within the soil with a given rainfall, and therefore, provide the estimation of ET. The ET is equivalent to the rate of water loss when the downward leaking flow outside of a soil column can be neglected or quantified.

In this study, we aim to develop a hybrid machine learning-based inverse modeling framework for estimating ET in real time based on in situ soil moisture sensors and vadose-zone flow modeling. The inverse modeling framework couples an ecohydrological model – the Richards equation and plant root uptake model – with the ensemble Kalman filter. We demonstrate our approach using the soil moisture data at the East River watershed in Colorado. We also evaluate the effects of various model parameters and assumptions, such as root density distribution and soil hydraulic parameters.