*2.2. ET Monitoring and Data Processing*

The actual evapotranspiration (ETa) was measured using the residual of energy balance (REB) method with a combination of surface renewal (SR) and eddy covariance (ECov) techniques. The SR and ECov are well-recognized methods to estimate the sensible heat flux density (H) and to calculate the latent heat flux density (LE), using the REB approach [23–28].

A full flux density tower was set up in the plots under normal farmers' irrigation practice at each of the experimental site, totaling four towers (Figure 1). In each tower, several sensors were set up. An NR LITE 2 net radiometer (Kipp & Zonen, Ltd., Delft, The Netherlands) was used to measure net radiation (Rn). Two 76.2 μm diameter, type-E, chromel-constantan thermocouples model FW3 (Campbell Scientific, Inc., Logan, UT, USA) were used to measure high frequency temperature data for computing uncalibrated sensible heat flux (H0), using the SR technique. An RM Young Model 81000RE sonic anemometer (RM Young Inc., Traverse City, MI, USA) was used to collect high frequency wind velocities in three orthogonal directions at 10 Hz, to estimate H for the latent heat flux density calculations using the ECov technique.

Each tower also consisted of three HFT3 heat flux plates (REBS Inc., Bellevue, WA, USA) inserted at a 0.05 m depth below the soil surface, to measure soil heat storage at three different locations; three 107 thermistor probes (Campbell Scientific, Inc., Logan, UT, USA) to measure soil temperature at three depths in the soil layer above the heat flux plates; three EC5 soil moisture sensors (METER Groups Inc., Pullman, WA, USA) to measure soil volumetric water content at soil depths and locations near the heat flux plate and the thermistor probes; EE181 temperature and RH sensor (Campbell Scientific, Inc., Logan, UT, USA) to measure air temperature and relative humidity; an SP LITE 2 Pyranometer (Kipp & Zonen, Ltd., Delft, The Netherlands) to measure solar radiation; and a TE525MM tipping-bucket rain gauge with magnetic reed switch to measure precipitation.

**Figure 1.** A fully automated surface renewal and eddy covariance evapotranspiration (ET) tower in the plot under normal farmer irrigation practice (NI) at site A3.

Except for the soil sensors, all other sensors were set up at 1.8 m above the ground surface. The data were recorded using a combination of a Campbell Scientific CR1000X data logger and a CDM-A116 analog input module. Direct two-way communication with each monitoring flux tower was possible using a cellular phone modem model CELL210 (Campbell Scientific, Inc., Logan, UT, USA). The data of the sonic anemometer and fine wire thermocouples were collected at a 10 Hz sampling rate and the data of the other sensors were sampled once per minute. Half-hourly data were archived for later analysis.

Both the EC and SR techniques were individually employed to determine sensible heat flux density. The available energy components, Rn and G (ground heat flux density) were also measured throughout the study period. After acquiring the half-hourly H0 data, a calibration factor (α) was established by determining the slope through the origin H values from the ECov technique versus H0 from the SR technique, separately for the positive and negative values of H0. The calibrated SR H value was finally estimated as H = α·H0. The SR H and the ECov H were used to determine the LE values. The advantage from using both the ECov and SR methods was that they are independent and similar results provide a high level of confidence in the data used [26,28,29]. Latent heat flux density was calculated using the Residual of Energy Balance equation, as follows:

$$\text{LE} = \text{Rn} - \text{G} - \text{H} \tag{1}$$

where LE, G, H are positive away from the surface, and Rn is positive towards the surface. G is the ground heat flux density at the soil surface. It is assumed that Rn, G, and H are measured accurately. While use of the full eddy covariance method often does not demonstrate closure, Twine et al. [30] recommended that ECov results could be forced to have closure by holding the measured Bowen ratio (H/LE) constant, and increasing the H and LE values until Rn − G = H + LE. Twine et al. [30] also reported that using the REB method provides nearly the same accuracy as using the Bowen ratio correction. After determining LE, ETa in mm d−<sup>1</sup> was calculated by dividing the LE in MJ m−<sup>2</sup> d−<sup>1</sup> by 2.45 MJ kg<sup>−</sup>1, to obtain the ET values in kg m−<sup>2</sup> d<sup>−</sup>1, which was equivalent to mm d−1.

A Tule sensor (Tule technologies, Inc., Oakland, CA, USA) was used in each of the deficit irrigation plots at the experimental sites, to estimate ETa using a surface renewal technique. The estimated daily ETa from the Tule sensors at each site was verified by comparing with the ETa measured from the full flux density towers.

Using the daily ETa determined in each experimental site and the daily reference ET (ETo) retrieved from the spatial CIMIS (California Irrigation Management Information System) data [31] for the coordinates of the monitoring station, the daily actual crop coefficient Ka (=Ks × Kc) was calculated using Equation (2):

$$\mathbf{K\_a = ET\_a/ET\_o} \tag{2}$$

The daily stress coefficient (Ks) represents water and salt stresses, management, and environmental multipliers. To obtain the actual ET, Ks is needed to adjust crop coefficient (Kc). Spatial CIMIS combines remotely sensed satellite data with traditional CIMIS stations data, to produce site-specific ETo on a 2-km grid, which provides a better estimate of ETo for the individual sites.

#### *2.3. Canopy Temperature and Soil Moisture Monitoring*

Two SI-411fixed view-angle infrared thermometers (IRTs, Apogee Instruments, Logan, UT, USA) were used to measure canopy temperature in each experimental plot. The IRTs were installed on a pole with a 47.5◦ angle below horizon, in opposite direction, viewing north and south to match for consistency. The IRTs were installed 1.8 m from the ground surface. The average temperatures of IRTs viewing north and south were considered to be the canopy temperatures. Canopy temperature was scanned with the IRTs units every minute and readings were averaged over a 30-min interval, using ZL6 cellular data logger (METER Groups Inc., Pullman, WA, USA).

Crop Water Stress Index (*CWSI*) was estimated using the difference between measured canopy and air temperatures (*dTm*), using Equation (3):

$$\text{CWSI} = \frac{(dT\_m - dT\_{LL})}{(dT\_{IL} - dT\_{LL})} \tag{3}$$

The *dTm* was compared against lower (*dTLL*) and upper (*dTUL*) limits of the canopy–air temperature differential, which could be reached under non-water-stressed and non-transpiring crop conditions. The Idso et al. approach [32] was used for estimating *dTLL* and *dTUL*.

Watermark Granular Matrix Sensor (Irrometer company, Inc., Riverside, CA, USA) was used to measure soil water tension at multiple depths of 15, 30, 45, 60, 90, and 120 cm, on a continuous basis. The data of Watermark sensors were recorded by a 900M Monitor data logger (Irrometer company, Inc., Riverside, CA, USA), on a 30-min basis.

#### *2.4. Soil Salinity Assessment*

Soil properties were surveyed and characterized within an approximate footprint area of 200 m × 200 m, around the ET monitoring stations in each plot, to assess soil salinity following deficit irrigation regimes. Surveys of apparent soil electrical conductivity (ECa) were conducted in October 2019 (right after the alfalfa harvest), using mobile electromagnetic induction (EMI) equipment, following the guidelines developed by the U.S. Salinity Laboratory of the United States Department of Agriculture, for field-scale salinity assessment [33–36]. ECa measurements were taken with a dual-dipole EM38 sensor (Geonics Ltd., Mississauga, ON, Canada), in horizontal (EMh) and vertical (EMv) dipole modes, to provide shallow (0 to 0.75 m) and deep (0 to 1.5 m) measurements of ECa, respectively. At each of the plots, soil cores at four distinct depth ranges (0–0.3, 0.3–0.6, 0.6–0.9, and 0.9–1.2 m) were taken from 6 sampling locations, which were selected using the ESAP software to reflect the spatial variability of root zone soil salinity. A comprehensive laboratory analysis was conducted on all soil samples.

#### *2.5. Yield Measurements and Plant Stand Evaluation*

Yield sampling from the sub-plots was conducted on the same day or the day before, when the participating growers scheduled to harvest the entire experimental fields. In each irrigation plot, yield samples were taken from 12 sub-plots with a dimension of 1.5 m wide and 2.0 m long (Figure 1). The sub-plots were harvested using a hand cutter. A portable PVC quadrate was used to accurately sample uniform sub-plot sizes. Plant cutting height was 6–8 cm. Fresh weights of plants harvested

within the quadrate was recorded, after which samples were dried for three days in conventional oven at 60 ◦C and recorded for alfalfa dry matter (DM). The significance of deficit irrigation strategies on mean dry matter yields was evaluated using a *t*-test.

Forage quality test for each collected sample was conducted using the Near Infrared Reflectance Spectroscopy (NIRS) method [37], to determine Crude Protein (CP), Acid Detergent Fiber (ADF), and Lignin percentage.

All irrigation plots were evaluated for plant stand density in February 2020, four months after switching the deficit irrigated plots back to normal irrigation practices. A portable PVC quadrate of 0.6 m wide and 0.6 m long was used to count the number of plants from the center of the 12 sub-plots that were used for yield measurements. Mean plant numbers per hectare were compared to the plots under normal farmers' irrigation practices.
