Towards an ensemble of RS-based SEB models to constrain the uncertainty in daily ETc monitoring in nut orchards

Woody crops such as almond and pistachio orchards are proliferating very fast in arid and semi-arid agricultural regions. This is the case of the southeastern Spanish region of Castilla-La Mancha, where the shortage of water resources and the low rainfall during the crop growing season under these conditions, makes it necessary to conduct efficient use of irrigation water in order to improve the sustainability of these crops.

A variety of Remote Sensing based (RS-based) surface energy balance (SEB) models have been shown effective to estimate crop evapotranspiration (ETc), and capture water stress conditions, using satellite imagery. Although their performance sometimes depends on the crop type or the environmental conditions. In addition, some limitations remain for an operational and continuous monitoring of daily ETc at a fine spatial and temporal resolution for water management or irrigation scheduling purposes, particularly on nut orchards. A model ensemble might help in overtaking these shortcomings. 

Recent efforts in the framework of the WATERSNUTS project (“remote sensing and digital farming for sustainable water use in almond and pistachio orchards”) have combined computational design with well-stablished SEB approaches into a Python environment to generate daily maps of distributed ETc covering Castilla-La Mancha region, for a selected time period and a predefined spatial resolution, starting with 20 m x 20 m. Up to now, two models, the Mapping Evapotranspiration with Internalized Calibration (METRIC) and Two-Source Energy Balance (TSEB), were implemented for testing, and a time series of Landsat 8-9, and Sentinel 2-3 were used as inputs. Whereas METRIC stands on VNIR and TIR data from Landsat series at 30-m pixel size, the implemented version of TSEB adopts a disaggregated Land Surface Temperature (LST) at 20-m spatial resolution, that has already shown good results in previous research applied to the tandem Sentinel-2 (S2)/Sentinel-3 (S3).

The reference evapotranspiration, ETo, plays a key role in this computational framework to fill the daily gaps with no available satellite images. A layer of 5-km gridded observational daily ETo values was provided by the Spanish State Meteorological Agency (AEMET). A self-derived crop classification map was used to focus the analysis on the nut orchards and discern between irrigated and rainfed plots, and look into the differences between water regimens.

Before upscaling, a local assessment was conducted in an agricultural area located in Tarazona de La Mancha, Spain (39º 15’ 58’’ N, 1º 56’ 23” W), for the period 2021-2024, using data from two full-equipped eddy-covariance towers installed at the center of an almond and a close by pistachio orchards.

The ensemble results are promising for nut orchards such as almonds or pistachio plantations, since S3-S2 disaggregated LST can help in increasing the frequency of daily ETc estimates through TSEB modeling in reduced size plots, while METRIC can outperform for those days with Landsat overpass. Further integration of additional SEB approaches, or RS-based water balance estimates, would enrich the ensemble, and foster the constrain of the uncertainty in evapotranspiration monitoring in nut orchards.