Irrigation management with a time-continuous two-source modelling of tree crops calibrated with satellite LST data

Agriculture will progressively require more and more attention as changing climatic conditions and reduced water availability threaten food security worldwide. The optimization of the agricultural production is obtained with constant monitoring of the plant health (in terms of e.g., soil moisture, leaf temperature or evapotranspiration), which can be challenging if crop fields are too extensive.

Thermal observations from remote sensing are extensively used in agricultural monitoring to power (mostly-residual) energy balance model that provide evapotranspiration estimates. Two main issues hinder the quality of the results from these models: (a) sub-pixel heterogeneity, in particular related to mixed crops (e.g. row and tree crops), which can be captured only partially by the available LST information and (b) temporal frequency of the information, which for most freely-available products is usually at odds with spatial resolution (e.g., 1 km data from MODIS is available daily, whereas 90 m data from Landsat only once every 7-8 days). Furthermore, tree crops draw water from deep layers of soil, further disconnecting the satellite information from the biophysical processes involved in plant growth.

In this work, the use of a continuous, two-source, double-soil-layer coupled energy-water balance model is displayed as a solution of these issues. The link between the two balances allows to compute surface temperature internally, meaning that satellite LST observations are used, only when available, for the calibration process. Furthermore, the use of a double source in the energy exchanges allows to properly address the intra-pixel heterogeneity. Finally, the double soil layer allows to address the soil water and energy vertical gradient in complex systems, properly framing the surface observation from remote sensing within the overall environment.

Two pear tree fields in the Po Valley have been chosen as focus to study the effectiveness of this model, via a monitoring of the 2022 irrigation season, employing Sentinel 2 observations for the vegetation data and Landsat 8 LST for the calibration process. ET estimates are evaluated against flux tower observations. The increased accuracy of these estimates is key to enforce a more precise and effective irrigation and optimize the use of the water resource.