Estimation of citrus water requirements by means of water and energy balance models driven by in situ, reanalysis and remote sensing data

Evapotranspiration (ET) knowledge is crucial for evaluating crop field water budgets and agricultural water resources management. To monitor crop water requirements various data sources are used such as: in situ (meteorological and soil water content data) measurements, reanalysis database, remote sensing observations, and models. Two approaches can be implemented: the Soil Water Balance (SWB) and the Surface Energy Balance (SEB).

This research aimed to evaluate these two approaches, by combining in situ or reanalysis meteorological data with remotely sensed images to explore the possible synergies between the approaches to propose an operational ET estimation in the context of future Thermal InfraRed (TIR) missions (TRISHNA and LSTM). With a SWB model, both actual evapotranspiration (ET_a) and soil water content (SWC) were daily estimated; whereas, with a SEB model latent heat flux (LE) was instantaneously evaluated.

Among the available SWBs, the SAtellite Montoring for Irrigation (SAMIR) is a FAO-2Kc-based model integrating remotely sensed images of vegetation cover for evapotranspiration spatialization and water balance. SAMIR can be forced by irrigation either measured or simulated employing specific rules based on the simulated SWC. Alternatively, the Soil Plant Atmosphere and Remote Sensing Evapotranspiration (SPARSE) is a two-source SEB model driven by remotely sensed Land Surface Temperature (LST) and vegetation cover. Both SWB and SEB were investigated by using different input variable combinations. For SAMIR, two combinations were employed: a) using in situ and b) using ERA5-Land reanalysis meteorological variables to estimate crop reference evapotranspiration and precipitation depth. Both incorporated farmer irrigation scheduling and Sentinel-2 NDVI-derived vegetation cover. For SPARSE, three combinations were employed: a) using in situ meteorological data, LST, and albedo; b) replacing LST and albedo with Landsat-8/9 data; c) replacing in situ data with ERA5-Land reanalysis while maintaining Landsat-8/9 inputs.

The experiments occurred during seven irrigation seasons, from 2018 to 2024, in a Mediterranean citrus orchard (Citrus reticulata Blanco cv. Mandarino Tardivo di Ciaculli), located near Palermo, Italy (38° 4’ 53.4’’ N, 13° 25’ 8.2’’ E) in which different irrigation systems and management strategies were applied. The field was equipped with a standard weather station, an Eddy Covariance tower, and four “drill and drop” probes to acquire: meteorological variables, energy fluxes, and SWC, respectively.

SAMIR best performance was obtained using the a-combination with Root Mean Square Error (RMSE) always less than 0.54 mm d^-1 and 0.02 cm³ cm^-3 for ET_a and SWC, respectively. These metrics were achieved excluding data from 2021 during which worse metrics (ET_a RMSE equal to 0.87 mm d^-1) were probably caused by the presence of weeds due to the lack of maintenance provided by the farmer. SPARSE best performance was obtained using a-combination with LE RMSE equal to 53 W m^-2. Noticeably, b- and c- combinations were implemented using a limited number of data (contextually to satellites acquisitions) thus achieving worse metrics (RMSE equal to 66 W m^-2 and 93 W m^-2 for b- and c- combinations, respectively).

Satisfactory results gained permit this work to keep on being updated toward the synergies between the approaches for better ET estimation.