Lettuce evapotranspiration and crop coefficients using eddy covariance and remote sensing observations

Abstract

Lettuce (Lactuca sativa L.) is a high-value crop for irrigation districts in the low deserts of the USA Southwest. To ensure maximal crop quality, negligible soil salinity stress, minimal nutrient loss and reduced pathogen susceptibility, lettuce irrigation must meet, but not exceed, crop water use requirements. However, lettuce crop water use information is outdated in this region: prior studies were conducted at least four decades ago (1960–1980) and do not represent current varieties, management practices, and climate. To address this shortcoming, 12 commercial sites in Yuma, Arizona, USA were evaluated between 2016 and 2020 to update lettuce water use requirements and crop coefficients. The study measured crop evapotranspiration (ET_c) using eddy covariance observations at eight iceberg and four romaine sites, where planting dates varied throughout the fall. Observed ET_c and remote sensing data were used to model the daily soil water balance and derive crop coefficients: single (K_c), basal (K_cb), and soil evaporation (K_e). The analysis was supported by lettuce crop height estimates and fractional vegetative cover (f_c) via remote sensing. Days to maturity averaged 75 ± 15 and 89 ± 12 days for romaine and iceberg, respectively, where season lengths increased as planting dates progressed from early fall to late winter. Average planting date for romaine sites was about 20 days earlier than average iceberg sites. When growing intervals are cast in heat units, dependence on crop type and time of planting was reduced. Average cumulative growing-degree-day and enhanced-degree-day metrics were 1133 ± 87 and 754 ± 48 °C-days, respectively. Seasonal lettuce ET_c averaged 278 ± 24 mm. Cumulative irrigation applied, plus precipitation, averaged 355 ± 88 mm. Lettuce K_c for sites varied from 0.90 ± 0.13 to 1.19 ± 0.11 and K_cb from 0.20 ± 0.05 to 1.01 ± 0.11 for the initial and mid-season growth stages, respectively. These updates will help growers improve their irrigation efficiency for lettuce and provide important documentation needed by water managers.

Introduction

Effectively and efficiently managing irrigation water for vegetable crops is crucial for economic and sustainable farming. Vegetables are short-season, high value crops that cannot be deficit or over-irrigated and still maintain marketability (Luna et al. 2012, 2013). Effective water management is especially important for farms in the Southwestern USA where all production relies upon the severely stressed Colorado River. While many of the farms in this region have high priority water entitlements, continuous and future access is not assured. Given the high fraction of freshwater resources used by agriculture, 65–72% (ADWR 2014; Hutson et al. 2005; Mpanga and Idowu 2021) it is important to establish how much irrigation supplies are used by arid land crops and to promote effective water management tools.

Yuma, Arizona, and nearby low desert areas in the Southwest produce the majority of winter lettuce and other cool-season vegetables in the USA (Sanchez 2000). In Yuma, vegetable crops are a $730 million industry (USDA NASS 2021), of which the lettuce industry dominates. Lettuce varieties are grown on diverse soils ranging from light-textured soils to heavier clay soils and are selected for their optimal climate (Kerns et al. 1999; Sanchez 2000). Planting dates vary from early September to late December to cover the market time window of November to March (Bealmear and Nolte 2014). At Yuma, there is only one lettuce cycle each year. The length of the growing season to harvest varies from about 60 to 130 days for early and late plantings, respectively (Simko et al. 2014). Typically, iceberg lettuce is seeded in twin rows on raised beds, at 1.02–1.07-m bed spacings, whereas romaine will sometimes be seeded in wider beds with 5 or 6 seed lines per bed (Simko et al. 2014). Because lettuce quality and yield can be significantly reduced by soil salinity, lettuce fields normally require pre-plant irrigation to reduce salt concentrations in the lettuce rooting depth (Kerns et al. 1999). After planting, solid set sprinklers are usually employed to irrigate near-daily for about a week to cool beds and allow germination. It is customary in Yuma to switch to closed-end level furrow irrigation once seedlings have emerged (Sanchez et al. 2009), which usually includes furrow irrigation applied 14–21 days after planting to sufficiently soften the soil prior to thinning and cultivation operations. Thereafter, furrow irrigation is managed to avoid saturating the middle of the lettuce bed, which can increase the occurrence of plant disease (Turini et al. 2011; Kerns et al. 1999). Typically, three to four more furrow irrigations are applied after thinning (Sanchez et al. 2009).

According to the USDA National Agricultural Statistics Service (NASS 2019), the average seasonal irrigation for romaine and iceberg in Arizona farms was about 740 mm in 2018. Earlier studies indicated that lettuce irrigation in USA desert areas can exceed 1000 mm per season (Sanchez et al. 2009; Turini et al. 2011). However, applicable crop water requirement information to guide water planning and efficient irrigation scheduling for marketable lettuce in Yuma is limited. Erie et al. (1982) provided daily and total consumptive use (i.e., crop evapotranspiration, ET_c) for lettuce in 1960–1962 studies in Mesa, AZ, using periodic gravimetric measurements of soil water depletion. They reported a seasonal ET_c of 216 mm. However, details on irrigation methods, application rates, and lettuce spacings for that study are unavailable and results may not be comparable to present-day Yuma lettuce production. Other evapotranspiration studies pertinent to Yuma conditions are scarce. Gallardo et al. (1996) reported lettuce ET_c in a Salinas, CA study using only sprinkler irrigation. Oliveira et al. (2005) used subsurface drip irrigation in a Maricopa, AZ study. Neither study included the ET_c that occurred in approximately the first 1/3 of the season and consequently reported low total ET_c of approximately 150 mm (Gallardo et al. 1996) and 100–134 mm (Oliveira et al. 2005). In a San Joaquin Valley (SJV), CA study, Johnson and Trout (2012) used remote sensing to quantify lettuce crop transpiration (T_c), which totaled 125–138 mm for ≈ 60-day seasons, planted in late summer.

In addition to total lettuce crop water requirements, recent guidance on lettuce ET_c demands over the course of the season is lacking for Yuma. Irrigation scheduling could be done using daily curves from Erie et al. (1982) and recommendations provided by Sanchez (2000). Alternatively, seasonal ET_c dynamics could be determined in fields using detailed soil moisture monitoring. Options for ground-based ET_c observations include soil–water balance measurement from gravimetric sampling (Tanner 1967), soil moisture profiling with neutron probes (Holmes et al., 1967), plot specific monitoring with weighing lysimeters (e.g., Slatyer and McIlroy 1961) and surface energy flux observations via Eddy Covariance (EC) stations. Optimal lettuce production is achieved when the upper surface soil moisture profile is maintained slightly less than field capacity (FC), an approach that maximizes yield and minimizes nutrient loss (Sutton and Merit 1993; Luna et al. 2013; Michelon et al. 2020). The upper surface layer extends to 0.5 m with most feeder roots within the top 0.15 m (Jackson and Stivers 1993).

Another approach estimates daily lettuce ET_c by applying weather-based crop coefficient (K_c) methods, as done in FAO56 (Allen et al. 1998). These have been widely adopted over the past two decades (Pereira et al. 2021). In the single crop coefficient approach for FAO56, daily ET_c is determined by multiplying the daily K_c and ET_o. K_c values are derived from crop growth stages and from condition end-member values representing bare soil and full cover. ET_o values are calculated with weather data using a standardized equation for a short reference crop (Allen et al. 1998). Alternatively, the FAO56 dual crop coefficients could be more effective for lettuce because they quantify the soil evaporation (E_s) contribution of ET_c more accurately than the single K_c procedure. In the dual procedures, K_c is separated into coefficients of basal (K_cb), representing T_c, and (K_e), representing E_s. Simplified ET_c methods exist to monitor lettuce, e.g., by Gallardo et al. 1996, which uses ET_o observations to estimate crop growth. These, or similar, approaches might be effective management tools. More recently, however, the free and wide availability of satellite data, notably Sentinel 2, makes possible the monitoring of crops, and their water requirements almost every day.

In either FAO56 procedure, daily K_c and K_cb curves for a crop are derived as segmented lines over four growth stages: initial, development, mid-season, and late season (Allen et al. 1998). Difficulties, however, emerge in applying the time-based K_c procedures for lettuce and other crops because growth stage lengths can be highly variable due to climate and cultural factors, such as plant density (Grattan et al. 1998). Consequently, some have promoted using ground cover observations to adjust the crop coefficient in time (Allen and Pereira 2009; Johnson and Trout 2012; Fernandez-Pacheco et al. 2014), an approach that removes the growth stage-based segmentation. The results for lettuce are inconsistent. Johnson and Trout (2012) present a tight relationship between K_cb and ground cover fraction for data collected in an experimental field in SJV, CA. Conversely, Grattan et al. (1998) found large differences in measured K_c as a function of percent cover in commercial fields, one in the low desert near Yuma, AZ, and the other on the central coast of California. The K_c variability was high at all levels of cover and varied between the locations.

Sustaining the economic and societal benefits of Yuma lettuce production is crucial locally and nationally for farms, growers, distributors and employees. Achieving sustainability is especially timely considering the current drought and reduced irrigation water supplies. Answers lie in part with achieving efficient irrigation management and alignment with actual crop water needs at each growth stage. In this regard, a project was designed with a main objective of establishing present-day crop water use, crop coefficients, and irrigation practices for two major Yuma commodity crops, iceberg and romaine lettuce.

This lettuce study proceeds as follows. In "Methods", observations and data sets needed to measure and estimate ET_c are described. This section consists of five parts: "Study sites" a description of the Yuma study sites, "Evapotranspiration measurements" a presentation of deployment and data collection procedures using eddy covariance stations, "Fractional cover estimation" a description of remote sensing for fractional vegetative cover (f_c) estimation methods, and related field experiments, "Estimation of lettuce soil evaporation and basal crop coefficients with EC Data" a presentation of soil water balance methods to evaluate single and dual crop coefficients, and "Estimation of total ET_c with heat units" a description of the use of two heat unit metrics to estimate crop growth stages and total ET_c. In "Results", results are summarized: "Evapotranspiration from iceberg and romaine lettuce" reports irrigation water use and ET_c results from eddy covariance collections, "Plant heights" presents techniques used to estimate plant height, "Fractional cover" describes f_c estimation via remote sensing, in "Measurement and modeling crop coefficients" FAO56 crop coefficient analyses are described, and "Application of heat unit modeling"  shows the use of heat units to estimate growth stages, time to maturity, and total ET_c. In “Conclusion”, the significance of the ET_c and crop coefficient findings from EC and the remote sensing results are discussed.

Methods

Study sites

Eddy covariance studies were conducted for lettuce (Lactuca sativa L.) in Yuma, Arizona, a hot, arid, and low-lying (~ 35 m above sea level) region receiving about 80 mm of precipitation annually (https://www.usclimatedata.com/climate/yuma/arizona). The study consisted of 12 lettuce field sites, located on private commercial farms, distributed within three Yuma County water regions: Yuma County Water Users Association (YCWUA), Yuma Irrigation District (YID), and North Gila Irrigation and Drainage District (NGIDD) (Fig. 1, image background is Normalized Difference Vegetation Index, NDVI). Site areas ranged from 5.7 to 14.5 ha, where romaine was grown at four sites and iceberg at eight sites (Table 1). To preserve growers’ anonymity, each site identifier is coded by its corresponding irrigation district, deployment years, and a deployment sequence letter. For example, YID19-20a means: YID—Yuma Irrigation District; 19–20—deployment period from fall 2019 to spring 2020; a—first deployment. The study years extended from the fall of 2016 through winter 2020. The EC deployment dates (Table 1) are indicative of lettuce seeding dates, which varied for sites from mid-September to late November. Lettuce harvest dates varied from mid-November to early March. Lengths of lettuce seasons were shorter for September and October plantings (60–80 days) than November plantings (90–106 days).

Fig. 1

Yuma, Arizona irrigation districts for this study: Yuma County Water Users Association (YCWUA, orange), Yuma Irrigation District (YID, blue), and North Gila Irrigation and Drainage District (NGIDD, red). Mexico lies immediately to the South and Southwest of the districts, and California to the Northwest. Background image is NDVI from Sentinel 2 (colour figure online)

Table 1 Lettuce eddy covariance site deployment locations, crop types, field sizes, and schedules

Sites were chosen, 5 ha or larger, within three irrigation regions- Yuma Irrigation District, Yuma County Water Users Association, and North Gila Irrigation and Drainage District- that sampled different soil types and included strong land-owner support. Twelve lettuce sites were occupied over 3 growing seasons (2016–2020). Selection criteria for observations needed at Yuma were feasibility, spatial representativeness, and high temporal resolution. Observations based on soil water data can meet these requirements, but spatial representation is low. Eddy covariance observations, adopted for this study, produce much better temporal and spatial sampling of weather conditions at the field sites: samples are representative of several hectares and can be nearly time-continuous for entire growing seasons. By combining daily EC observations of ET_c with logs of irrigation/precipitation, a soil water balance can be calculated at the lettuce field sites. When used in conjunction with the FAO56 dual crop coefficient computations, K_cb and K_e can be modeled. In turn, the E_s and T_c contributions of the daily ET_c at sites can be tracked.

Lettuce irrigation and salinity management practices and field configurations for the sites were generally similar. All were furrow irrigated prior to planting to reduce salt concentrations in the lettuce rooting depth. The lettuce at iceberg sites and two romaine sites (YID17b and NGIDD19-20) were seeded in dual parallel lines on raised beds, at bed spacings of 1.02 m, center to center. Romaine at site YID16 was seeded in three lines at bed spacing of 1.07 m, while romaine at site YID17a was seeded in seven lines at bed spacing of 2.2 m. In all but one instance (YID17a), fields were initially sprinkler irrigated, then furrow irrigated after establishment. Because of wide beds at YID17a, irrigation at this site was sprinkler applied throughout the season. Otherwise, sprinklers were used for at least the first 5–8 days after planting. Distribution uniformity for sprinkler irrigation at Yuma is generally ~ 85% (Sanchez and French 2023). For the subsequent 9–16 days, fields were again irrigated, either by sprinkler or furrow irrigation, prior to lettuce thinning. All sites thinned the lettuce seedling lines at a standard spacing of ≈ 0.23 m between plants. Following thinning, the field sites were irrigated using closed end level furrows. Furrows were built and compacted to a trapezoidal profile to speed furrow irrigation advance times and to promote irrigation uniformity, which exceeds 90% at Yuma (Sanchez et al. 2009). Furrow irrigation lengths for sites varied from ≈ 200 to 300 m. Irrigation water amounts applied to all fields were measured. For sprinkler irrigation systems, in-line meters and both automatic and manual rain gauges were used. Except at site YID17a, all rain gauges were removed from the field at sites just before furrow irrigation was applied. Thus, following initial sprinkler irrigations, precipitation at the 11 other sites was not measured. For furrow irrigation, we used meters at the farm canal gates when they were available (i.e., at YID sites). Where these were not available, furrow irrigation deliveries were determined by measured water velocity, water depth, and irrigation time during each irrigation event based on canal ditch geometry.

Reference climate data used in this study were provided by the University of Arizona, Yuma Agricultural Center AZMET weather station and denoted as the Yuma Valley station (32° 43′ N; 114° 42′ W; 36 m above sea level) (https://ag.arizona.edu/azmet/02.htm). AZMET data are collected over short, clipped and well-watered grass, approximately 50 ft², and are considered representative of reference ET. The station was located near the northern boundary of the YCWUA district and the proximity of the station relative to the 12 field sites varied from 1 to 10 km. Two other AZMET stations in the Yuma area were also available. One, located about 16 km east of Yuma Valley was in the lower NGIDD district and had close proximity to the YID and NGIDD sites (6 km) but was far away from some YCWUA sties (26 km). The other station in the southern portion of YCWUA (Yuma South) was further away from the majority of sites. Table 2 lists the Yuma Valley station data of monthly means for the period between planting and harvest for the 12 sites. These include: monthly means of maximum, minimum air temperature (T_max, T_min), minimum relative humidity (RH_min), solar radiation (R_solar), wind speed at 2-m height (u₂; re-calculated from AZMET wind speed measurements at 3 m), growing degree days (GDD), and the Penman–Monteith reference evapotranspiration (ET_o) for a short crop (Allen et al. 1998). GDD values were computed using the single-sine method with horizontal cutoff (Baskerville and Emin 1969; https://ipm.ucanr.edu/WEATHER/ddconcepts.html) with basal and maximum temperature thresholds, respectively, of 4.5 °C (Sanders et al. 1980; Dufault et al. 2009) and 28 °C (Wurr et al. 1992; Wheeler et al. 1993). Cumulative GDD (CGDD) between seeding and harvest for the 12 sites varied from about 965 to 1270 °C-days.

Table 2 Monthly weather summaries for years as observed at the Yuma Valley AZMET site

Also calculated from AZMET data, though not shown in Table 2, were effective degree days (EDD), which adjusts GDD values when light levels are growth limiting, as might occur for cloudy intervals or winter months (Scaife et al 1987; Wurr et al. 1988):

$$1/EDD=1/GDD+a/R$$

(1)

where a is an empirical constant (0.5 [deg-day-m²/MJ] used for this study), and R is total daily solar radiation (MJ/m²).

Between the seeding and harvest dates for sites planted in 2016 and 2017 (Table 1), the cumulative precipitation recorded at the AZMET Yuma Valley station was 2.0 mm or less. Most of the 2016–2017 sites were in the YID district, closer to the NGIDD AZMET station, which recorded between 0 and 4.0 mm for those growing seasons. During lettuce seasons planted in 2018 and 2019, differences in cumulative precipitation recorded at the Yuma Valley and NGIDD stations were small, between ± 5 and 10 mm.

Measured soil texture fractions were used to estimate the soil type, soil water holding, and near-surface heat capacity parameters needed for determining surface energy balance corrections for shallow heat storage (Table 3). Soil texture values represent three samples from each site from the top 0.15 m depth, the interval corresponding to heat stored above soil heat flux plates. Samples were analyzed using a laser diffraction particle size analyzer (LS13320, Beckman Coulter Life Sciences, Indianapolis, IN, USA).¹ The sampled soil depths are also relevant to soil evaporation (E_s) parameters, namely, readily evaporable water (REW) and total evaporable water (TEW). The USDA soil texture triangle (USDA 2017) was used to classify the measured textures to soil type (Table 3). Values of field capacity (FC) and wilting point (WP) were estimated with typical FC and WP values for the given site soil type, as provided in FAO56 (Allen et al. 1998; Table 19). Values of FC varied from 0.26 to 0.40 m³ m⁻³ for loam and clay soils, respectively. Similarly, the REW for the 0.12-m evaporation layer (Z_e) was assigned to fields according to soil type following FAO56 (Allen et al. 1998) and Allen (2011) guidelines. The TEW for sites were calculated by the FAO56 equation for TEW, assuming Z_e = 0.12 m, and varied from 23 to 32 mm (Table 3).

Table 3 Measured sand, silt, and clay fractions^a, USDA soil type^b, and estimated values of field capacity (FC), wilting point (WP), readily evaporable water (REW) and total evaporable water (TEW) for the 12 Yuma lettuce study sites

Lettuce plant height (h) was observed almost every 12 days (3–7 times per site and season) using a meter stick at randomly selected points near the EC sites. The h measurements started about three weeks after planting and corresponded to approximately a 4-true-leaf stage.

Evapotranspiration measurements

Eddy covariance deployments (Table 1) were conducted for sites in YID, YCWUA, and NGIDD. Water allocations, priorities, and delivery sources are noted in Noble (2015). Deployments of EC stations were similar to those as described for wheat experiments (French et al. 2020): following shaping and planting of beds, but just prior to the first irrigation, equipment tripods and sensors were positioned mid-field. Systems were mostly Campbell Scientific manufacture (Logan, UT, USA), but also included sensors from: LI-COR Biosciences (Lincoln, NE, USA; infrared gas analyzers), Kipp & Zonen (Delft, Netherlands; 4-way net radiometers), Hukseflux (Delft, Netherlands; soil heat flux plates), and Vaisala (Helsinki, Finland, temperature/humidity probes). Six unique stations were used for the study, 4 of which were new as of 2017. All loggers and covariance sensors were calibrated by the manufacturer in 2016 and 2017. Zero and span of infrared gas analyzers (IRGA) were done in July 2017 and again in July 2018. Each station consisted of two tripods, one for measurements and the other for power supply. The measurement tripod included a sonic anemometer, infrared gas analyzer (IRGA), net radiometer (all 2 m above the ground), two soil heat flux plates, data logger, and cell modem. A storage battery was placed at the tripod base. The power supply tripod was placed approximately 5 m northeast of the measurement tripod and contained two 60 W solar panels, a solar power regulator, and connecting power cables. Sonic units were set horizontally- all sites were precision leveled. Due to short maximum h (~ 0.33 m) and short season (< 100 days), EC assemblies did not need to be raised during the season. EC azimuths were set due west to reduce instances of self-obstructed airflow: predominant winds were from the north and secondarily from the south-southwest.

Each station collected multiple micrometeorological observations (~ 108 variables per time step) at 20 Hz sample rates, configured under Campbell Scientific’s EasyFlux DL ™ (Logan, UT) program to allow continuous data measurements during the cropping cycle. Simultaneously 30-min block-averaged fluxes, adjusted with Webb–Pearman–Leuning (Webb et al. 1980) corrections, were stored. Computation of 30-min evapotranspiration (ET) estimates used WPL fluxes. With few exceptions, EC stations were visited weekly to inspect alignment, assess instrument operations, check for rodent damage, farm equipment damage, vandalism, and cleared of bird debris. Station functioning was monitored daily via cell-phone modem links. Data were stored on compact flash cards and changed approximately every 2 weeks.

Subsequent processing used EddyPro V.7.0.8 software (LI-COR Biosciences, Lincoln, NE, USA) in Express Mode to compute 30-min fluxes sourced from the original 20 Hz time series files (Fratini and Mauder 2014). Results from this processing updates values previously reported in Dhungel et al. (2023). Changes in energy storage within the soil mass above the two heat flux plates were computed using logged soil temperature and volumetric water content. R scripts were applied to remove data spikes and fill short-duration data gaps. Spike removal followed the methodology described by Vickers and Mahrt (1997). The on-line gap filling tool (http://www.bgc-jena.mpg.de/~MDIwork/eddyproc/method.php; employing techniques as described in Falge et al. 2001) was used to fill time gaps not previously filled. The tool’s friction velocity (U*) filtering with a moving point test method (Papale et al. 2006) was chosen.

Assessment of energy balance closure was done at daily time steps by regressing advective energy against radiative energy and correcting for net energy storage (Anderson and Wang 2014). Considering that daily average Bowen ratio for all stations ranged between − 0.09 and + 0.14, and that preferred closure methodology is an open question (Wohlfahrt et al. 2010), closure errors were corrected at the daily time step by assigning residuals to LE fluxes. Two stations, YID17a and YID17b, had IRGA failure for most of the deployment periods and closure errors for them could not be assessed. In this case LE fluxes were computed via residual of the energy budget, i.e., LE = Rn − G − H − ΔS. 2-D Flux footprint analyses were done using a modification of an R script provided by Kljun (footprint.kljun.net; Kljun et al. 2015).

Fractional cover estimation

Modeling of ET_c requires knowledge of fractional vegetative cover (f_c) at all stages of growth. Vegetation cover controls partitioning between evaporation and transpiration (Skaggs et al. 2018). Use of spectral vegetation indices (VI) from remote sensing is a practical way to obtain f_c because correlations between some of them are strong and the corresponding data are readily and freely available (e.g., Landsat and Sentinel 2). Studies have quantified correlations between fractional vegetation cover and NDVI (e.g., Gao et al. 2020; Tenreiro et al. 2021). Considering lettuce, Johnson and Trout (2012) collected fractional ground-based observations with a pole-mounted camera to calibrate cover vs. NDVI. For their San Joaquin Valley, CA study they found a strong linear relation: f_c = 1.26 × NDVI − 0.18, with coefficient of determination (r²) = 0.96. Crucial to know for this study, however, is if this relationship is accurate for Yuma, AZ sites—climate, soils, and cropping practices in southwestern Arizona are not the same as for the Central Valley, CA. Considering that the spectral sampling provided by satellites is richer than before—while Landsat 9 has 4 VNIR bands, Sentinel 2 has 8 corresponding bands—this is an opportune time to consider other chlorophyll-related indices besides NDVI. To answer these questions, we collected fractional cover data with high resolution drone imagery and close-in-time multispectral, 5–20 m satellite remote sensing data over 7 Yuma sites (Table 4). To obtain f_c vs. time, we conducted 26 flights over two romaine sites from 29 September to 6 November 2020, and composited 18 flights over five iceberg sites 4 December 2019 to 28 January 2021. Plot-averaged cover for each sortie was computed, linearly interpolated to daily time steps, smoothed using a 7-day convolution time filter, and matched with results from satellite-based VI time series.

Table 4 Drone acquisition dates for fractional cover measurements at Yuma, AZ

UAS Processing

Ground-truth f_c were obtained from unmanned aircraft systems (UAS) data. Remotely sensed images were acquired with a DJI Inspire 1 drone, flown 15 m above the ground with 70% front overlap and 50% side overlap. An Inspire X3 camera (4000 × 3000 pixels) was used to acquire red–green–blue (RGB) images in JPG format. The average image resolution was ~ 0.7 cm/pixel. During field acquisition, three to four geo-referenced targets were placed at field corners prior to each flight. Pix4Dcapture (Pix4D, Switzerland) software was used to design and control flight sorties. Pix4Dmapper (Pix4D, Switzerland) was used to create ortho-mosaics.

The f_c classification was obtained with a simple-to-implement unsupervised spectral classification approach based on the Green/Red (G/R) spectral ratios. For lettuce grown in Yuma, AZ, the statistical distributions of G/R data are strongly bimodal, as illustrated in Fig. 2. This meant that discrimination between bare soil and vegetation could be done by finding the inter-modal minimum and segregating pixels less than or greater than this minimum. Addition of Blue band data, as might be done with the Canopeo approach (Patrignani and Ochsner 2015), did not improve classification results and were not used.

Fig. 2

Drone imagery of 2-row iceberg (left) and romaine (center). Bed spacing was 1.06 m (42 in), plants within each bed were separated by 0.31 m (12 in), planting density after thinning was 4.3 plants/m (~ 14 plants/ft). Crop classification was based on the ratio of green to red channels (right). Shown is a 2-D histogram for mature romaine where the 1:1 line (white) discriminates between plants and bare soil. Vegetation instances lie above the line, while soil occurred in two clusters: one just below the line representing shaded and wet soil elements, and a second cluster much further below and representing surface dry soil (colour figure online)

Satellite-based remote sensing processing

Besides NDVI (Rouse et al. 1974), numerous options exist for candidate indices to map vegetation cover. This study considered ten vegetation indices (Table 5): CIRE (Gitelson et al. 2005), CIGN (Clevers and Kooistra 2012), MTCI (Dash and Curran 1994), REIP (Guyot and Baret 1988), IRECI (Frampton et al. 2013), NDI45 (Delegido et al. 2011), MCARI (Daughtry et al. 2000), TCARI (Haboudane et al. 2002), OSAVI (Rondeaux et al. 1996), and TCARI/OSAVI (TO, Haboudane et al. 2002). Reasons for choosing one index over another depend on objectives and consider correlation with chlorophyll content, soil background, other non-photosynthetic matter, and index saturation. Criteria for index selection in this study were the following: lettuce cover prediction accuracy with a linear model, range of applicability, and availability.

Table 5 Vegetation indices evaluated for fractional vegetation cover

Vegetation indices were calculated from two satellite sensors: Sentinel 2 a/b, (http://www.esa.int/Applications/Observing_the_Earth/Copernicus/Sentinel-2) for 2017–2020, and Venus (https://www.theia-land.fr/en/product/venus/) for parts of 2018–2020. Sentinel 2 (a/b) are a pair of identical satellites collectively observing identical targets at 10–20 m resolution nominally every 5 days, but due to Yuma’s location, observations were often available at 3-day intervals. Sentinel 2 sensors are multispectral push-broom instruments in sun-synchronous orbits with overpass times in Arizona at ~ 11AM. Because of its still better sampling (5 m spatial and 2 day temporal), Venus data were incorporated when available.

All data sets used were calibrated, multispectral, visible to near infrared surface reflectance images. Multispectral data were used to improve consistency against atmospheric effects and to create vegetation indices. Higher spatial resolution of satellite data was important to resolve lettuce fields without significant field-edge effects. At the Yuma sites, field widths ranged between 174 and 555 m. High temporal frequencies were essential to track rapid lettuce growth and frequent irrigation events.

A majority of the analyzed data (352 scenes) were from Sentinel 2. Orthorectified, surface reflectance L2A, 100 km × 100 km tiles, were downloaded from Copernicus data hub (https://scihub.copernicus.eu/dhus/#/home). From these, NDVI and OSAVI were computed at 10 m resolution, while other red-edge indices were obtained at 20 m. VI values were generated as indicated in Table 5. Sentinel 2 surface reflectance data, L2A, were used for images acquired from mid-2018 and later. For earlier acquisitions where L2A were not available, L1C top of atmosphere data were downloaded and atmospherically corrected using the Sen2Cor program (https://step.esa.int/main/snap-supported-plugins/sen2cor/sen2cor_v2-5-5/).

Additional data provided by Venus data (173 scenes, L2) contributed 5 m resolution data for all indices. Each were atmospherically corrected by CNES via the MAJA program, a combination of the Multi-sensor Atmospheric Correction and Cloud Screening (MACCS) algorithm and ATCOR use estimates of atmospheric thickness as described in Hagolle et al. (2015).

Estimation of lettuce soil evaporation and basal crop coefficients with EC Data

As described in "Evapotranspiration measurements", actual crop evapotranspiration (ET_{c act}) was acquired daily using EC observations at the 12 Yuma lettuce sites in Table 1. At each site, the daily actual single crop coefficient (K_{c act}) was calculated as ET_{c act}/ET_o, where the daily ET_o was provided by the Yuma Valley AZMET station. In this section, additional analyses were developed to estimate the dual crop coefficients (i.e., K_cb and K_e) of daily K_{c act} and, thereby, estimate T_c and E_s components of the daily ET_{c act}. The framework involved modeling a daily soil water balance (SWB) for two depth intervals: a layer representing the lettuce rooting depth (Z_r) and a shallow soil evaporation layer (Z_e) for each of the 12 sites. The simultaneous SWB modeling of the two layers followed methodologies given in FAO56 (Allen et al. 1998).

Soil water depletion (D_r,i) within the rooting zone at the end of each day was estimated from a depletion balance equation (all in mm units):

$${\text{D}}_{{{\text{r}},{\text{i}}}} = {\text{ D}}_{{{\text{r}},{\text{i}} - {1}}} - {\text{ P}}_{{\text{i}}} - {\text{ I}}_{{\text{i}}} + {\text{ ET}}_{{{\text{c}},{\text{i act}}}} + {\text{ DP}}_{{\text{i}}}$$

(2)

where (D_r,i − 1) is the depletion state for the prior day. ET_{c,i act} (actual crop ET from EC), and deep percolation (DP_i) increase depletion amounts, while precipitation (P_i) and irrigation events (I_i) decrease them. The irrigation method, dates, times, and amounts for I_i were available for each field site. Equation 2 does not consider possible water runoff or potential capillary rise from a water table. The P_i were provided by the Yuma Valley AZMET station. Effective P_i included in the two SWBs were cumulative amounts greater than 1.0 mm within a 24-h period. Daily values of total available water (TAW_i) of the daily rooting depth (Z_r,i) were calculated as shown in Eq. 3:

$${\text{TAW}}_{{\text{i}}} = { 1}000{\text{ Z}}_{{{\text{r}},{\text{i}}}} \left( {{\text{FC }} - {\text{ WP}}} \right)$$

(3)

where TAW_i is in mm, Z_r,i in m, and FC and WP are in m³ m⁻³. The values of FC and WP used for each site are provided in Table 3. The limits for D_r,i in Eq. 2 were zero at FC and TAW at WP. Since D_r,i cannot be less than zero on a given day i following irrigation and/or precipitation, an amount for DP_i was computed, when necessary, to balance Eq. 2. Initiation of the SWB model for the analyzed sites was on the day when the dry-planted seed was wetted by sprinkler irrigation. For the dry-seeded lettuce, it was assumed that the D_r,i − 1 in Eq. 2 was at WP. The initial D_r,i in Eq. 2 was the end of the first lettuce sprinkler wet-up day (first day). Description of the lettuce Z_r,i modeling is presented later in this section.

The separate daily SWB of the soil evaporation layer (Eq. 4) was computed for all fields using Z_e = 0.12 m (Allen et al. 1998).

$${\text{D}}_{{{\text{e}},{\text{i}}}} = {\text{ D}}_{{{\text{e}},{\text{i}} - {1}}} - {\text{ P}}_{{\text{i}}} - {\text{ I}}_{{\text{i}}} /{\text{f}}_{{{\text{w}},{\text{i}}}} + {\text{ E}}_{{{\text{s}},{\text{i}}}} /{\text{f}}_{{{\text{ew}},{\text{i}}}} + {\text{ DP}}_{{{\text{e}},{\text{i}}}}$$

(4)

where D_e,i and D_e,i − 1 are the daily cumulative depth of evaporation following wetting of the exposed and wetted fraction of the surface soil at the end of day i and end of day i − 1, respectively, P_i and I_i are as previously defined, E_s,i is daily soil evaporation, and DP_e,i is daily deep percolation loss if soil water content exceeds FC on day i, where all these variables are in units of mm. The f_w,i and f_ew,i are the fraction of the soil surface wetted by P_i and I_i and the exposed and wetted fraction, respectively, on day i. For the lettuce sites, f_w,i was assumed as 1.0 for precipitation events greater than 1.5 mm, sprinkler irrigations, and pre-thinning irrigations, which completely wet the lettuce beds (regardless of irrigation method). For subsequent furrow irrigation events, the f_w,i was assumed to be 0.60, where growers apply water to only cover the furrow and not the beds. The f_ew,i are a minimum function of either f_w,i or (1 − f_c,i), where f_c,i is daily canopy cover fraction. The f_c,i were estimated by the OSAVI data for each site using the OSAVI-based canopy cover fraction relationship derived from drone and satellite observations: f_c,i = 1.33 × OSAVI − 0.312, and presented later in the Results section. The upper limit for D_e,i in Eq. 4 is the TEW, shown for sites in Table 3.

Daily coefficients of basal (K_cb,i), soil evaporation (K_e,i), and water stress (K_s,i) were derived based on the daily actual ET_c (ET_{c,i act}), daily ETo (ET_o,i), estimated Z_r,i and soil water properties (Eq. 5):

$$\left( {{\text{K}}_{{{\text{cb}},{\text{i}}}} {\text{K}}_{{{\text{s}},{\text{i}}}} + {\text{ K}}_{{{\text{e}},{\text{i}}}} } \right){\text{ ET}}_{{{\text{o}},{\text{i}}}} = {\text{ET}}_{{{\text{c}},{\text{i act}}}} \approx {\text{ K}}_{{{\text{c}},{\text{i act}}}} {\text{K}}_{{{\text{s}},{\text{i}}}} {\text{ET}}_{{{\text{o}},{\text{i}}}}$$

(5)

where ET_{c,i act} and ET_o,i are in mm and the coefficients are dimensionless. The K_{c,i act} is the daily actual single K_c. The daily K_s was calculated as:

$${\text{K}}_{{{\text{s}},{\text{i}}}} = \left( {{\text{TAW}}_{{\text{i}}} - {\text{ D}}_{{{\text{r}},{\text{i}}}} } \right)/\left( {\left( {{1 }{-}{\text{ p}}} \right){\text{ TAW}}_{{\text{i}}} } \right) = \, \left( {{\text{TAW}}_{{\text{i}}} - {\text{ D}}_{{{\text{r}},{\text{i}}}} } \right)/\left( {{\text{TAW}}_{{\text{i}}} - {\text{ RAW}}_{{\text{i}}} } \right)$$

(6)

where K_s,i, TAW_i, and D_r,i are as previously defined; p is the fraction of TAW_i that can be extracted from Z_r before suffering water stress (i.e., K_s less than 1.0) and RAW_i is the daily readily available soil water in Z_r. For D_r,i > RAW_i, K_s,i is less than 1.0. The p value for lettuce was set to 0.40 for ET_{c act} = 5.0 mm/day, and adjusted daily for atmospheric demand, per FAO56 Table 22 and footnote 2. The K_s,i were not computed until the f_c,i for the given lettuce site was above 0.15, i.e., the K_s,i were set to 1.0 before f_c,i > 0.15.

Assessment of K_s,i during the lettuce growing seasons was dependent on the estimation of Z_r,i. A lettuce model for lettuce Z_r,i based on cumulative growing degree days (CGDD; ℃ day) by Thorup-Kristensen (2006) was used for this estimation. For each site, an initial minimum Z_r value of 0.12 m at planting was increased at a rate of 1.2 mm per ℃ day starting after 200 ℃ days were accumulated (Thorup-Kristensen 2006). A maximum lettuce Z_r,i of 0.50 m for modeling water stress was suggested in FAO56 and by others, where that depth corresponds with the approximated depth of primary lettuce root mass (Jackson and Stivers 1993). However, assuming a maximum Z_r,i of 0.50 m in the SWB calculations (Eq. 2) generated calculated D_r,i values that far exceeded RAW_i on numerous days for all but one site (YID19-20a). Consequently, assuming 0.50-m or even 0.70-m as the maximum Z_r resulted in K_s,i values substantially below 1.0 on numerous days for 11 of the 12 sites. Given that the commercial Yuma lettuce sites were managed to minimize water stress for attaining high yield quality, estimated TAW values were considered reasonable, and ET_{c act} from EC was representative. This means that assessment of K_s,i should be based on a deeper maximum Z_r,i, such that the calculated water stress was minimal. Detailed analyses of the 11 sites showed that a maximum Z_r,i of 0.90 m was a reasonable assumption, resulting in K_s,i less than 1.0 on only a few days for some fields. The differences in the calculated RAW_i and resulting K_s,i as affected by assumed maximum Z_r of 0.50, 0.70, and 0.90 m are illustrated for site YID17c (Fig. 3a–c, respectively). Note in Fig. 3 that the K_s,i were assumed as 1.0 until estimated f_c,i was greater than 0.15, which occurred after CGDD exceeded 730 ℃-days for this site. The Thorup-Kristensen (2006) model, supported by tap root measurements beyond 1.0 m, predicted a lettuce Z_r,i of 0.90 m at CGDD of about 800–900 ℃-days for the Yuma sites, which coincided with the occurrence of maximum f_c,i for each site. Gallardo et al. (1996) also reported measured tap roots as deep as 0.90 m, 63 days after planting, for well-watered romaine and crisp head lettuce in California, while Erie et al. (1982) showed lettuce depleted soil water to a depth of 0.92 m in studies in Arizona.

Fig. 3

Daily values of the water stress coefficient (K_s), depletion (D_r), readily available water (RAW), irrigation (I), and precipitation (P) as a function of cumulative growing degree day (CGDD) assuming maximum root depths of 0.50 (a), 0.70 m (b), and 0.90 m (c) for site YID17c in Yuma

Besides the parameters used in Eq. 4, derivation of daily K_e and K_cb values included REW and h. The estimated REW varied between 9 and 10 mm based on soil type (Table 3). Prior to lettuce h measurements, it was assumed that the h was zero for the first 5–8 days after the first wet-up irrigation. Subsequently, measured h data were interpolated on a daily basis for each site.

The daily K_e calculations started on the first day of wet-up and were governed by Eqs. 7 and 8, which follow the FAO56 (Allen et al. 1998) equations:

$${\text{K}}_{{{\text{e}},{\text{i}}}} = {\text{ min }}\left[ {{\text{K}}_{{{\text{r}},{\text{i}}}} \left( {{\text{K}}_{{{\text{cmax}},{\text{i}}}} {-}{\text{ K}}_{{{\text{cb}},{\text{i}}}} } \right),{\text{ f}}_{{{\text{ew}},{\text{i}}}} \times {\text{ K}}_{{{\text{cmax}},{\text{i}}}} } \right]$$

(7)

where K_r,i is the daily evaporation reduction coefficient as calculated in Eq. 8:

$${\text{K}}_{{{\text{r}},{\text{i}}}} = \, \left( {{\text{TEW}} - {\text{ D}}_{{{\text{e}},{\text{i}} - {1}}} } \right)/\left( {{\text{TEW}} - {\text{REW}}} \right)$$

(8)

where K_r,i = 1.0 during stage 1 drying (energy limiting), i.e., when D_e,i − 1 ≤ REW, K_r,i less than 1.0 during stage 2 drying (falling rate), and K_r,i is zero when D_e,i − 1 = TEW. In FAO56, K_cmax,i is the theoretical upper limit of the combined daily K_cb and K_e coefficients that can occur following I_i or P_i and is calculated as a function of the daily values of mean u₂, RH_min, and h. In this analysis, the K_cmax,i used was the greater of either K_{c,i act} or the K_cmax,i calculated by the FAO56 equation (Eq. 72), which ensured that the sum of K_cb,i plus K_e,i was equal to K_{c,i act}.

For each of the 12 lettuce site data sets, iterations of the daily SWB for Z_e and Z_r were made to derive the K_e,i and K_cb,i. A first iteration required initial estimates of daily K_cb so that K_e,i by Eq. 7 could be solved. In the initial iteration, the daily K_cb,i were calculated as a function of estimated f_c,i by the relation derived by Johnson and Trout (2012) for lettuce, where K_cb,i = 0.209 + 1.08 f_c,i − 0.07 f_c,i². Using these initial K_cb,i values in Eq. 7 to solve for K_e,i and then, calculating resultant K_cb,i (as K_cb,i = K_{c act,i} − K_e,i) gave a new set of K_cb,i estimates. The second iteration used the new estimates of K_cb,i in Eq. 7 resulting in another set of estimated K_cb,i. These iterations continued for each site until the mean absolute percent difference (MAPD; Li et al. 2005) between the entire set of estimated K_cb,i and those used in solving in Eq. 7 was less than 5.0%. The number of iterations needed varied from four to six.

Crop growth stages (Allen et al. 1998) were assessed after final estimation of K_cb,i at each lettuce site. Length of the initial growth stage, corresponding to the interval up to 5 true leaves, was estimated as the number of days before the estimated K_cb,i were consistently greater than 0.20, which was then the start of the development stage. The end of development stage was estimated as the day when the estimated f_c,i reached 75% of the maximum lettuce f_c. This event is approximately 30 true leaves for romaine and heads > 10 cm for iceberg and corresponds approximately to the beginning of stage 4 as presented by Jenni and Bourgeois (2008). From that day on, the mid-season continued until the day before the lettuce was harvested. A late season was not defined for fresh lettuce (Pereira et al. 2021). For each site, average K_cb values were calculated from all data within the initial and mid-season growth stages and used to construct FAO56 segmented K_cb curves. The average K_cb values for the initial and mid-season stages were interpolated daily to provide the development stage diagonal line of the K_cb curve. To facilitate comparison of the derived lettuce crop coefficients at Yuma to the literature, the local average K_c and K_cb values for the mid-season stage for each site were adjusted to the standard climate conditions proposed in FAO56, where RH_min = 45% and u₂ = 2 m s⁻¹. The climate adjustments were made using the transfer equations provided in Pereira et al. (2021), which require the h at mid-season. The h values used in the adjustments were those measured at mid-season for each site.

The average K_{c act} over estimated growth stages was compared to an estimated K_c (K_{c est}) calculated by the soil depletion balance over the growth stage lengths for each site:

$${\text{K}}_{{\text{c est}}} = \left[ {{\text{D}}_{{{\text{r}},{2}}} - {\text{ D}}_{{{\text{r}},{1}}} + \, \sum \left( {{\text{P}}_{{\text{i}}} + {\text{ I}}_{{\text{i}}} } \right) \, - \, \sum {\text{DP}}_{{\text{i}}} } \right]/\sum {\text{ET}}_{{0,{\text{i}}}} = \left( {\Delta {\text{D}}_{{\text{r}}} + {\text{ TWA }}{-}{\text{ DP}}} \right)/{\text{ET}}_{{\text{o}}}$$

(9)

where D_r,1 and D_r,2 (∆D_r) are the depletion values at the beginning and end of the growth stage, respectively, ∑(P_i + I_i) is the summation of precipitation and irrigation over the stage, i.e., the total water applied (TWA) over the stage, ∑DP_i is the summation of DP loss over the stage, and ∑ET_0,i is summation of ET_o over the stage, where units are in mm. Linear regression of K_{c est} versus K_{c act} over each growth stage was performed. Performance indicators for K_{c est} accuracy were the coefficient of variation (r²) and the root mean squared error (RMSE). Similarly, linear regression of average K_{c act} versus the average estimated K_e was computed over growth stages to evaluate the variation in average K_e at sites, as explained by K_{c act}.

Estimation of total ET_c with heat units

Weather-based heat units are useful for estimating h, the number of days to maturity, and expected total ET_c. We evaluated GDD and EDD heat units accumulated between the day of planting (DOP) to the day of first harvest. The two heat unit metrics were computed from the AZMET Yuma Valley station. To estimate total ET_c, a combined empirical and climatological approach was used. First, the observed cumulative GDD and EDD values were computed for the three growth stage transition times: end of the initial, end of development, and date of first harvest. Second, these cumulative heat unit values were combined with AZMET ET_o data observed 2003–2021, along with average K_c values (reported in Sect. "Results"), to obtain daily K_c and subsequently total ET_c for DOP values ranging between 1 September and 31 December.

Results

Evapotranspiration from iceberg and romaine lettuce

Daily ET_c, ET_o, and water inputs from all 12 sites are shown in Fig. 4. DOP was between 13 September and 19 November (days of year 256–323 respectively), though at several sites initial seed wet-up by sprinkler occurred 1–2 days later. Planting and harvest days corresponded closely to EC deployments delineated in black symbols (Fig. 4). Day of first harvest was from 14 November to 5 March in the subsequent year. There were 83 ± 14 days to maturity (DTM). DTM increased with DOP: approximately 65 days for early September planting, to 107 days for 19 November planting. However, neither DTM nor DOP were significantly correlated to total ET_c: r² values were 0.29 and 0.15, respectively. ET_c values were distinguished by an initial period with high day-to-day variability followed by a subsequent period with lower ET_c variability. The initial period, approximately 20–30 days, exhibited spikes, 6–8 mm/day, and were due to evaporation from bare soil during sprinkler and furrow irrigation events. In the subsequent period, encompassing development and mature stages, ET_c tended to track ET_o. Thus, daily ET_c values did not correspondingly increase with plant development as would be expected for other crops such as wheat or cotton, but they followed the seasonal ET_o trend, which decreased towards solstice (DOY 355), and then increased afterwards.

Fig. 4

EC observed daily ET_c (black), reference evapotranspiration ET_o (red), sprinkler events (blue), furrow irrigation events (orange), rain (green) (colour figure online)

Cumulative ET_c ranged between 242 and 314 mm when considering all sites (Table 6). Average cumulative ET_c was 283 ± 34 mm for romaine (N = 4), and 276 ± 25 mm for iceberg (N = 8), suggesting that romaine and iceberg crop water requirements were not much different in this study, although site sample sizes were unequal. Cumulative TWA (irrigation plus precipitation) was about 21% higher for iceberg than romaine. Consumptive use fraction of TWA (Pereira et al. 2012) averaged 82%. With the exception of YID17a, a site that was sprinkler irrigated for the entire season, sites partitioned seasonal irrigation applications similarly, with ≈35% of water applied by sprinkler (120 mm) and the remainder by furrow (228 mm). TWA to lettuce varied from 257 mm (YID17a) to 508 mm (YID19-20a). On average sites planted in 2016 and 2017 received about 100 mm less TWA than those planted in 2018 and 2019. The 2018 and 2019 sites had more precipitation and generally heavier furrow irrigation applications than those in 2016 and 2017. Precipitation constituted ≈3% of TWA for all sites but accounted for 5–11% of TWA for 2018 and 2019 sites, and less than 1% of TWA for sites in 2016 and 2017.

Table 6 Cumulative ET_c and total water applied (TWA), partitioned to sprinkler irrigation, furrow irrigation, and precipitation for 12 lettuce sites at Yuma, AZ

Flux footprint results

To evaluate the existence and possible impact on the ET_c data due to advection from beyond field boundaries, two-dimensional flux footprints were estimated for all sites. The estimation used the algorithm and R script developed by Kljun et al. (2004), local weather data from AZMET, crop heights, and EC instrument heights. The climatology option was selected and roughness height to Monin–Obukhov ratios (z_m /L) less than − 15.5 were excluded. Most of the flux footprints lay within the experimental field (Fig. 5), however, there were three instances: YCWUA17-18a, YCWUA17-18b and YID19-20a, where flux footprints significantly extended into adjacent fields. To estimate the impact upon total ET_c from these cases, ET_c for the core and adjacent fields were modeled using NDVI, crop coefficients, and local weather data. NDVI time-series were obtained from analyses of Sentinel 2 surface reflectance (L2A). NDVI served to partition growth stages, while tabulated FAO56 crop coefficients for lettuce were used, where only K_{c INI}, K_{c MID} (respectively 0.7 and 1.0) were relevant. Neighboring crops were unknown but assumed to have lettuce crop coefficients.

Fig. 5

Seasonal flux footprint contours for lettuce, colors denote cumulative percentage of total flux: 50% (red), 80% (yellow), and 90% (blue). Plot experimental sites outlined in red rectangles. Background gray scale is NDVI derived from Planet satellite data with acquisition dates indicated in white (colour figure online)

The modeled ET_c influence was computed using weighted averages of flux footprint fractions inside and outside of the EC sites. In all three instances where the flux footprint was a concern, the modeled influence of neighboring fields was a seasonal reduction in observed ET_c. This suggests that tabulated cumulative ET_c might be underestimated as follows: YCWUA17-18a by 41 mm, YCWUA17-18b by 108 mm, and YID19-20a by 83 mm. In comparison, modeling of adjacency effect at the other sites showed minor biases ranging between 1 and 6 mm.

Assessment of EC closure

Gap-filled EC data were assessed for energy balance closure at daily time steps (Table 7) using daily cumulative radiative and advective fluxes (Anderson and Wang 2014). Based on linear regression, mean closure was strong for 8 sites, with mean slope coefficient (b1) of 0.80. Two sites, YID17a and YID17b could not be assessed because of data gaps. Site NGIDD19-20 had anomalously high closure error and could be possibly due to flux input from the adjacent field to the north, but unfortunately documentation is unavailable to assess that possibility.

Table 7 EC closure statistics, coefficient of determination (r²) and slope (b1), performed at daily time steps. IRGA failures precluded closure assessment at YID17a and YID17b

Plant heights

Iceberg and romaine heights increased linearly by heat units, with effective degree days (EDD, Wurr et al. 1988) providing better estimates than GDD (Table 8). For the Yuma sites maximum h were ≈ 0.30 m, reached about 71 days after planting (DAP) and when 1012 GDD and 672 EDD were accumulated. Using all measured h from iceberg and romaine (no height data available from YID16), DAP = 23.8 + 191 × h (m), with r² = 0.61, and RMSE = 13 days, which means that h at emergence is about 24 DAP, while maximum h is about 81 DAP.

Table 8 Estimation coefficients for iceberg and romaine heights (h) vs. growing-degree-days (GDD) and effective-degree-days (EDD), where h = b0 + b1 × [GDD or EDD]

Fractional cover

Fractional vegetation cover (f_c) vs. vegetation indices (VI) were assessed using 2 romaine and 5 iceberg sites between fall 2019 and winter 2021. Fractional cover was continuously tracked for romaine and episodically for iceberg. Results from one field, ‘S9’, are shown in Fig. 6. Drone and satellite observations spanned 40 days from planting to near maturity. Observed cover ranged from 0 to 0.55 for S9 with other sites reaching 0.75 cover. No index exhibited saturation for the range of observed cover. VI types showed smooth patterns and would require second or third order polynomials to obtain accurate f_c predictors for all conditions. However, all VI types exhibited an early season, low-cover inflection at 0.05 cover, beyond which a good linear fit was possible. Adopting this approach for simplicity, whereby inflection points were chosen for the lower limit, each VI/cover relation was linearly modeled (red lines in Fig. 6). Results show that all indices predict f_c with high correlation (Table 9). However, two of them, and notably NDVI, ranked relatively low with r² of 0.92. The top ranked indices were red-edge CIRE, REIP, and TCARI. Comparisons between these suggest that TCARI could have better linearity at high cover, while REIP is non-linear below 0.10 f_c. Thus, one might choose TCARI as a preferred fractional cover estimator. However, the fourth ranked index was a non-red-edge index, OSAVI, and its correlation was marginally different from TCARI. Considering ease of use, accessibility, linearity, and accuracy, OSAVI was chosen to estimate f_c: it is almost as accurate as the top indices in Table 9 and it is more widely accessible because it uses conventional red and NIR bands.

Fig. 6

Romaine cover development and fractional cover (f_c) vs. VI for 11 indices. For cover development (upper left sub-figure) drone observations are in black, daily interpolations in green. For the VI displays (all other sub-figures), interpolated and 5-day smoothed values are in black. Linear models (red) are fit to values following a subjectively chosen early season inflection (colour figure online)

Table 9 Ranked linear regression statistics for romaine fractional cover vs. vegetation index

Measurement and modeling crop coefficients

Values of daily K_{c act} and estimated daily K_e and K_cb as a function of CGDD are illustrated for four of the Yuma lettuce field sites in Fig. 7a, b (YID17b and YCWUA17-18a, initial wet-up dates DOY 264 and 317, respectively) and Fig. 7c, d (YID18-19a and YCWUA19b, initial wet-up dates DOY 282 and DOY 270, respectively). The initial CGDD on the first sprinkler irrigation wet-up dates varied from ≈ 17 to 21 ℃-days for the sites. The estimated number of days for sites from wet-up through the end of initial growth stages averaged 23.5 ± 5.1 days (coefficient of variation, CV, of 22%), while the CGDD of the initial stage for sites averaged of 398 ± 50 ℃-days and CV of 13%. From wet-up through the estimated end of development stages the sites averaged 51 ± 8.2 days and CV = 16%, whereas CGDD for sites over that period were much more alike (787 ± 37 ℃-days and CV = 5%), as depicted in Fig. 7a–d. From wet-up through harvest date, sites averaged 83 ± 10.2 days and CV = 13% and CGDD averaged 1146 ± 88 ℃-days and CV = 8%. The four sites in Fig. 7 had different planting dates and diverse day lengths for growth stages and total season. Using CGDD as the temporal scale shows that growth cycles for the sites were more uniform at that scale than day lengths.

Fig. 7

Daily values of actual single (K_c) and estimated basal (K_cb) crop coefficients, estimated soil evaporation (K_e), coefficients, and fitted FAO56 K_cb curve, along with estimated daily cover fraction (f_c), irrigation and precipitation events at YID17b (a), YCWUA17-18a (b), YID18-19a (c), and YCWUA 19b (d) sites at Yuma, Arizona

During the initial growth stage, the K_{c act} are generally highest following the near daily establishment sprinkler irrigations at the beginning of the seasons (Fig. 7). The K_{c act} then declined for following periods when no irrigation was applied and increased again following the irrigation prior to lettuce thinning, where those irrigation applications were given at a CGDD of ≈ 285, 270, 270, and 375 ℃-days for sites (Fig. 7a–d, respectively). The variations in estimated daily K_e during the initial stages corresponded to the irrigation application periods (high K_e) and the non-irrigated periods (low K_e). Estimated f_c greater than zero for the sites was not observed until CGDD was ≈ 600, 490, 550, and 570 ℃-days for sites (Fig. 7a–d, respectively). During initial stages, the estimated daily K_cb values for the sites fluctuated from slightly above zero to values 0.20 or higher for several days prior to the thinning irrigations. The higher K_cb at those times occurred as the computed E_s and, thus, estimated K_e became very small, while K_{c act} values were generally 0.45 and higher for those days, forcing the higher estimated K_cb. Following the thinning irrigations, the K_cb values again dropped below 0.20 for a period of a few days, which varied by site. The initial stages ended when daily K_cb were consistently above 0.20 for the remainder of the growing seasons. The averages of K_{c act} and estimated K_cb in the initial stage for the four sites are tabulated in Table 10.

Table 10 Growth stage lengths in days, cumulative growing degree days (CGDD), and cumulative effective growing degree days (CEDD), with averages of estimated basal (K_cb) and actual single (K_c) crop coefficients and cumulative amounts of actual crop evapotranspiration (ET_c), estimated crop transpiration (T_c), and estimated soil evaporation (E_s) during the growth stages for 12 Yuma lettuce sites

During the development stages, before subsequent irrigations were applied at sites, the estimated K_e declined rapidly such that the estimated daily K_cb became closer to the K_{c act} (Fig. 7). When the irrigations were then applied during the development stages, the K_{c act} increased markedly for several days, as did the K_e estimates. In general, the estimated K_cb increased with CGDD during the development stage, though slight dips in K_cb occurred immediately following the irrigations at sites, reflecting the balance between estimated K_e and K_{c act}.

During the mid-season growth stages, when estimated f_c was highest and estimated K_e was low, the K_{c act} at the two YID sites were seen to occasionally fluctuate ~  ± 0.50 to 0.60 over periods of about 4–8 days, e.g., sites YID17b (Fig. 7a) around 830–890 ℃-days and YID18a-19 (Fig. 7c) around 950–1035 ℃-days. The YID sites were closer to the NGIDD than Yuma Valley AZMET station than for sites in Fig. 7b, d, which were much closer to Yuma Valley station, though those sites occasionally had high K_{c act} fluctuations over several days during the mid-season. While using NGIDD ET_o instead of Yuma Valley ET_o generally reduced day to day K_{c act} variation at YID sites, the NGIDD station ET_o also resulted in K_{c act} values that were theoretically inconsistent, e.g., 1.8–2.2. Using Yuma Valley ET_o for all sites provided a more reasonable upper limit constraint for K_{c act} values (less than 1.80). As seen in Fig. 7, estimated K_cb became close in magnitude to K_{c act} during mid-season for sites. Like prior stages when f_c was smaller, there was also an association of increased K_{c act} values following the irrigation and precipitation events. The estimated f_c declined rapidly towards the end of the mid-season stage for YID17b and YCWUA17-18a (Fig. 7a, b), respectively. The decline in late-season f_c indicates that parts of these fields were being harvested, though not at the EC location, where measurements continued until the EC location was harvested. Thus, as the field-wide f_c declined at those sites, the estimated K_e increased and the differences between daily K_{c act} and K_cb became larger. In contrast, sites YID18-19a (Fig. 7c) and YCWUA19b (Fig. 7d) had much smaller changes in estimated f_c at the end of the mid-season stages, which indicated limited harvesting in the fields had begun prior to the harvests at the EC location. Accordingly, late-season K_e were not increased appreciably at those sites.

Estimated growth stage lengths in days, CGDD, and CEDD and the average K_cb and actual K_c values for the growth stages are given for all 12 Yuma lettuce sites in Table 10. Wet-up dates for the sites varied from DOY 256 to 324 (Sep. 13 to Nov. 20). Wet-up through harvest varied from 61 to 106 days. The estimated initial growth stage lengths from wet-up varied from 18 to 37 days for sites and were generally shorter for sites having wet-up before DOY 288 i.e., Oct. 15 (Table 10). However, the duration of the initial stages based on CGDD and CEDD was more uniform than day lengths, where the average CGDD and CEDD for sites was 399 and 265 ℃-days, respectively, with coefficients of variation (CV) of 10% compared to a CV of 24% based on day lengths. The estimated day lengths from wet-up through development stages were variable (39–77 days; average of 52 days and CV of 24%) and generally increased from earlier to later wet-up dates. Variation of sites through development stage were small (CV less than 6%), where average CGDD and CEDD values were 783 and 520 ℃-days, respectively. Estimated mid-season day lengths varied for sites (21–41 days) and averaged 30 ± 6.3 days and CV of 21%. While mid-season lengths were generally shorter for earlier wet-up dates, the trend was inconsistent for the later planting dates. Mid-season lengths were likely also influenced by farm manager harvest decisions. There was also more variation in CGDD and CEDD duration for sites over mid-season stages than over the initial and development stages, where the mid-season CV for CGDD and CEDD were 21 and 17%, respectively, compared to CV between 10 and 12% for the earlier growth stages. For the 12 sites, total season lengths averaged 82 ± 14 days with a CV of 18%. The average CGDD and CEDD for total season were 1133 ± 87 and 754 ± 48 ℃-days, respectively, (Table 10) with CV of 8 and 6%, respectively. Thus, the heat unit results indicate consistency in season durations for sites despite differing day lengths of up to 45 days.

The average K_{c act} for the initial growth stage for sites was variable, from 0.67 to 1.14, and averaged 0.90 ± 0.13 for the 12 sites (Table 10). The lowest initial K_{c act} was at YCWUA17-18b having the latest wet-up date (DOY 324) and longest initial growth stage length in days, while the highest was at YID19-20a, which was also a late wet-up date (DOY 313).

During the initial lettuce stages, growers at the 12 sites applied variable amounts of cumulative irrigation, i.e., 190–220 mm at four sites and 85–170 mm at the others. YID19-20a was the only site that had significant precipitation (32 mm) during the initial stage in addition to 160 mm of cumulative irrigation. Cumulative ET_o during initial stages varied from 75 mm (YID19-20a) to 134 mm (YCWUA19b). According to the daily SWB of Z_r for the sites (Eq. 2), irrigation amounts during initial stages often resulted in DP, where estimated cumulative DP varied from 3 mm (YID17b) to 115 mm (YID18-19a) and averaged 75 mm for all sites. The SWB for sites also indicate that the change in estimated D_r from the beginning to the end of the initial stage was variable for fields (− 5.0 to 19 mm), where small or negative changes in D_r were generally at sites where higher amounts of irrigation were applied at the thinning irrigation, e.g., the site in Fig. 7c. The greatest increase in D_r over the initial stage (19 mm) was calculated for YID17b, where a much lighter thinning irrigation was applied (Fig. 7a). The K_{c est}, based on the SWB for sites, varied from 0.65 to 1.11 for the initial stage. Regression indicated that K_{c est} accounted for 97% of the variability in average K_{c act} for the 12 sites, where the RMSE was 0.03.

The average estimated K_cb during the initial stage varied from 0.13 to 0.30 and averaged 0.20 ± 0.05 for all sites. The difference between average K_{c act} and estimated K_cb for the initial stage of sites in Table 10 is the average K_e, which averaged 0.70 for all sites. The initial estimated K_e was highest at YID19-20a (0.91) and lowest at YID17a and YCWUA17-18b (both 0.54). Generally, the lower average K_{c act}, the lower the average estimated K_e during the initial stage. Regression indicated the average K_{c act} explained 85% of the variation in estimated average K_e. Other factors, such as irrigation frequency during the initial stage and variable TEW values (Table 3), would also affect average K_e.

The average K_{c act} for individual sites for the development stage was also highly variable, from 0.76 to 1.25 with an average of 0.96 ± 0.14 for all sites (Table 10). In development stages, growers at the 12 sites applied either 0, 1, 2, or 3 irrigations, where cumulative TWA was from 32 to 203 mm and averaged 93 mm for all sites. Cumulative ET_o varied were from 66 to 111 mm (96 mm average), and estimated DP was from zero to 80 mm, where positive DP occurred at only five sites during development stages. The estimated ∆D_r over the development stage varied from − 16 to 51 mm. The two highest average K_{c act} during the development stage occurred at the NGIDD19-20 (1.25) and YID19-20a (1.13) sites. The NGIDD19-20 site (highest K_{c act}) was the only site not irrigated during the development stage but had three precipitation events totaling 32 mm. It also had the lowest cumulative ET_o, no DP, and a ∆D_r of 46 mm. At that site, K_{c est} was 1.20 and highest among all sites. Conversely, site YID19-20a had the highest TWA (3 irrigations plus some small precipitation), highest ET_o, the most estimated DP, and a ∆D_r of near zero during its development stage. At that site, K_{c est} was 1.10 or second highest for sites. For the other 10 sites, one or two irrigations were applied during development stages and precipitation was minimal. The three sites that had two irrigations during the development stage had slightly higher K_{c est} (0.90) than the seven sites that received only one irrigation (0.84). These differences were similar to average K_{c act} for the sites, which were 0.98 and 0.87 for two versus one irrigation. Considering the development stages for all sites, K_{c est} explained 97% of the variation of average K_{c act} with RMSE of 0.03.

The average K_cb for all sites over the development stage varied from 0.54 at YCWUA17-18b to 0.92 at NGIDD19-20 and the average for all sites was 0.70 ± 0.10. The difference between K_{c act} and estimated K_cb for all sites in Table 10 indicates K_e averaged about 0.26 during the development stage. The highest K_e were at the YID19-20a (0.41) and YID17a (0.38) sites, where the latter site was the only field site that was sprinkler irrigated during the development stage. Regression indicated that average K_{c act} explained only 47% of the variation of K_e with RMSE of 0.06. However, K_e variation was also affected by differences in f_c and the number of irrigations applied at sites during the development stages, which would help explain the K_e variation among sites.

The average mid-season K_{c act} varied from 1.02 to 1.37 for the 12 sites and the 12-site average was 1.19 ± 0.11 (Table 10). During mid-season, growers applied either 1, 2, or 3 irrigations, where cumulative TWA was from 35 to 196 mm and averaged 98 mm for all sites. Several fields planted in 2018 and 2019 had a few precipitation events during mid-season that totaled less than 20 mm. Cumulative ET_o varied were from 59 to 114 mm (87 mm average) and estimated DP was from zero to 69 mm, where positive DP occurred at only four sites during mid-season. The estimated ∆D_r over the mid-season stage varied from − 32 to 59 mm. For mid-season, 92% of the variation in average K_{c act} was explained by K_{c est} with RMSE of 0.03.

The average mid-season K_cb varied from 0.88 to 1.13 for sites and the 12-site average was 1.01 ± 0.11 (Table 10). The variation of average mid-season K_cb values for sites did not show trends due to the different wet-up dates or length of the mid-season stages. However, there appeared to be a trend of higher mid-season K_cb (and K_{c act}) for the Romaine sites than at the Iceberg sites. Though sample sizes are too small to generalize, it is worth noting that estimated f_c averaged 0.71 for the four Romaine sites at mid-season, which was 18% greater than mid-season f_c averaged for the eight Iceberg sites, respectively. The average K_e for the 12 sites for mid-season was about 0.18 but varied for sites from 0.07 to 0.31. The difference in average K_e for sites was not explained by differences in K_{c act}. However, sites with lower average K_e (less than 0.18) were generally those having higher f_c (0.65–0.75) and earlier termination of seasonal irrigations prior to harvest. The average estimated K_cb values at mid-season after adjusting to the standard climate of FAO56 were from 0.84 to 1.09 for sites and averaged 0.97 for all sites. The adjusted average mid-season K_{c act} values varied from 0.99 to 1.33 for sites with an average of 1.15 for all sites. Thus, climate-adjusted mid-season coefficients were about 0.04 lower than the locally derived values at the Yuma sites.

The root zone SWB computations indicated water stress (K_s < 1.0) occurred on several days for three field sites. Water stress was computed for sites YID17c (4 days in mid-season), YID17d (8 days in mid-season), NGIDD19-20 (5 days in mid-season). On days when K_s < 1.0 was computed, the K_{c act} (and estimated K_cb) were adjusted according to Eq. 5. The estimation of water stress for the most extreme case (YID17d) increased the estimated K_cb at mid-season by only 3.7% (i.e., from that if water stress was not considered).

Cumulative ET_{c act} and the partitioned cumulative T_c and E_s estimates are shown for each growth stage for the 12 sites (Table 10). The cumulative ET_{c act} for growth stages match the calculated TWA-DP + ∆D_r from first to last day of stage and that for the total season determined in the SWB for each site. The average cumulative ET_{c act} during the initial stage for all sites was 93 ± 14 mm, where eight of the 12 sites had ET_{c act} less than 93 mm. Higher cumulative ET_{c act} at NGIDD19-20, YCWUA18d, and YCWUA19b corresponded to higher amounts of cumulative TWA-DP (108–119 mm) during the initial stage than for other sites. The estimated cumulative E_s was also greatest at those sites (82–96 mm). In contrast, the lowest cumulative ET_{c act} in the initial stage, i.e., at YCWUA17-18b and YID18-19a (≈75 mm) had cumulative TWA-DP from 58 to 78 mm and lower cumulative E_s. Table 10 shows that the average estimated cumulative T_c for the sites during the initial stage was quite small (19.4 mm) compared to average estimated cumulative E_s (74 mm). Given the differences in irrigation and climate for the 12 lettuce sites, the ratio of cumulative E_s to cumulative ET_{c act} during the initial stage was generally consistent, averaging 0.79 ± 0.04 for all sites.

As the lettuce season progressed to later growth stages, cumulative E_s was diminished and cumulative T_c increased at all sites (Table 10). For the development stage, cumulative ET_{c act} averaged 87 mm, cumulative T_c averaged 62 mm, and cumulative E_s averaged 25 mm for the sites. The ratio of cumulative E_s to ET_{c act} was variable among sites, 0.19–0.40, and averaged 0.29 ± 0.07. Sites with the higher ratios (0.37–0.40) were irrigated two or three times during the development stage, including YID17a, YCWUA17-18b, and YID19-20a. The cumulative ET_{c act} at mid-season stages varied from 70 to 132 mm for sites and averaged 98 ± 18 mm for all sites. Sites with ET_{c act} > 100 mm were those having mid-season lengths > 30 days and two to three irrigations plus some precipitation. The ratio of cumulative T_c to ET_{c act} at mid-season varied from 0.75 to 0.93 and averaged 0.84 for all sites. The ratio of cumulative E_s to ET_{c act} at mid-season varied from 0.07 to 0.25 and averaged 0.15 ± 0.05. The cumulative E_s /ET_{c act} ratio at mid-season was highly correlated with estimated K_e at mid-season (r of 0.96, p < 0.01). For the total lettuce season, estimated E_s was a large contribution of ET_{c act} and constituted between 37 and 46% of ET_{c act} with an average of 41% for all sites. However, on average 65% of the cumulative E_s occurred in the initial growth stage, which coincided with the frequent establishment and subsequent thinning irrigations under bare soil conditions.

Application of heat unit modeling

In "Plant heights" heat units were used to estimate h, but they also provide a way to estimate FAO56 crop growth stages, and from there a way to model total ET_c vs. day of planting. Assessing cumulative heat units vs. crop growth stages as defined in FAO56 results in cumulative heat units at the three transitions (Table 10): end of INI, end of DEV, and end of season (Begin of Harvest). Maximal units were 1133 and 754 for CGDD and CEDD respectively. By comparison, Wurr et al. (1988) reported lower values for their Wellesbourne experiments: ~ 900 CGDD and 589 CEDD. These differences highlight that heat unit thresholds need to be developed locally. At Yuma the variability of CGDD and CEDD units were nearly the same, with the coefficient of variation ranging 0.06–0.10. Using the Table 10 CEDD transition values along with AZMET weather data for 2003–2021 shows the dependence of growth stage lengths (L) upon day of planting (Fig. 8). Shown are mean and standard deviations of L value vs. DOP. Progressively later planting days increases the duration of the INI stage from ~ 20 days (22 September) to over 30 days by DOY 310 (6 November). At still later planting days, the maximum INI of 45 days is reached on 1 December, while the length of DEV begins to decrease. Total season length varies from less than 60 days when lettuce is planted early September to more than 100 days when planted in November.

Fig. 8

Climatological estimation of lettuce growth stages: L INI, L DEV, and L Total, vs. day of planting (DOP) at Yuma, Arizona. Black lines denote mean stage lengths estimates obtained from EDD for DOP ranging September 1 to December 31 while blue dots denote values as reported in Table 10. Light blue shading indicates ± 1 standard deviation around mean DOP values (colour figure online)

Combining growth stage results shown in Fig. 8 with mean single K_c values from Table 10 allows simulation of total ET_c vs. day of planting (Fig. 9). The simulations capture much of the observed ET_c variability (red symbols) and show that minimal total ET_c (~ 280 mm) is obtained for an early October planting date. For earlier or later plantings, expected total ET_c would be 10–20% greater, approximately 310–340 mm. Mid-season stress events at YID17c, YID17d, and NGIDD19-20 (circled red symbols) did not affect total ET_c.

Fig. 9

Total ET_c vs. DOP for Yuma Valley lettuce sites. EDD-based modeled ET_c values for weather observed at Yuma Valley AZMET range ~ 280 to 340 mm (black), with averaged ET_c values shown in blue. EC-based ET_c values indicated by red symbols. Circled red symbols denote instances for mid-season stress at sites YID17c, YID17d, and NGIDD19-20 (colour figure online)

Discussion

Primary results for iceberg and romaine lettuce at the 12 Yuma sites are summarized in Table 11. Average seasonal total water was 278 ± 24 mm, a value that represents an ET_c increase of 29% from the 216 mm value reported for lettuce in Mesa, AZ by Erie et al. (1982). However, it is important to note that the Erie et al. data show minimal ET_c (25 mm) for the first 30 days of the season. At the Yuma sites, cumulative ET_c during the initial growth stage (or about the first 20–30 days of the season for sites) averaged 93 ± 14 mm, which was predominately soil evaporation, arising from frequent sprinkler irrigations applied at the Yuma sites. While environmental differences between the locations may be important, the Yuma region is generally warmer in the fall-winter season than Mesa, it is possible Erie et al. (1982) did not use sprinkler irrigation for establishment or did not measure ET_c until after the lettuce crops were established. Specific details on the lettuce experimental methodologies used in the 1960’s Mesa, AZ field studies could not be found. Measured cumulative ET_c for the development stage through harvest (28–63 DAP) was reported to be approximately 150 mm for spring-seeded lettuce grown under sprinklers in Salinas, CA (Gallardo et al. 1996). Cumulative ET_c for development stage through harvest (31–110 DAP) computed for the Mesa, AZ data would be about 190 mm. These data would be comparable with the cumulative ET_c at Yuma, which averaged 185 mm from development through harvest (25–82 DAP). The fall-planted cycles in Arizona were longer than the cycle shown in Gallardo et al. (1996).

Table 11 Summary of lettuce water use and growth attributes for 12 Yuma lettuce sites

In recent years, progress has been made in Yuma to better control furrow irrigation applications for improved efficiencies, where long run furrows are replaced by shorter runs, field slopes are laser or gps-leveled, and furrows are compacted. For six of the Yuma sites, seasonal ET_c closely matched the total water applied indicating those sites were managed efficiently. On the other hand, several sites had poor irrigation efficiency, where TWA was about 150–200 mm more than seasonal ET_c. However, at those sites, managers applied furrow irrigations shortly before harvest, whereas more efficient sites did not. Presumably, the late irrigations were applied to preserve lettuce quality. Such sites illustrate a potential need to provide Yuma lettuce growers with updated information on crop water requirements so that in-season irrigations can be more efficiently scheduled.

The lengths of lettuce seasons, as well as the estimated initial, development, and mid-season stages were variable for sites, where differences were primarily related to date of lettuce wet-up. The FAO56 paper (Allen et al. 1998) gives an indicative initial growth stage length of 25 days for lettuce planted Oct./Nov. in arid regions. The six sites that were wet-up in Oct./Nov. in the present study averaged 27 days for the initial stage, which is close to the FAO56 estimate. Average initial stage length was 21 days for sites with wet-up prior to Oct. 1. This is similar to the 20 days for an indicative initial stage for lettuce planted in April in a Mediterranean climate (Allen et al. 1998). Although not specifically addressed by Gallardo et al. (1996) in lettuce studies in Salinas, CA, also similar to a Mediterranean climate, initial stage length was likely in the 25-day range for spring-planted lettuce. Bryla et al. (2010) reported a 24-day initial stage length for iceberg lettuce in SJV, CA experiments planted in late summer. Both California studies also suggest that lettuce cover is negligible until after the initial stage ends, as at Yuma. Modeling with CEDD indicates that the initial stage duration depends on day of planting, and ranges between 19 and 44 days. However, the duration of initial stages at the 12 study sites in thermal units of CGGD and CEDD were more uniform than number of days, where the thermal unit variations were less than half the variation in day lengths.

Estimated development length for Oct./Nov. wet-up at Yuma averaged about 33 days, which is consistent with the available literature. However, for the six Yuma sites wet-up before Oct. 1 (Sep. 13–28), average development length was only 23 days. The shorter development length for earlier wet-ups corresponded to accumulations of GDD and EDD that were similar to those with longer development day lengths. Typical length of the development stage for lettuce is 35 days per FAO56 (Allen et al. 1998) for the arid region (Oct./Nov. planting), which was also 35 days in the Bryla et al. (2010) study. In a similar way for the initial stage length results, CEDD modeling shows that length of development strongly depends upon day of planting and at Yuma can range between 19 and 43 days. However, considering the period from wet-up through the end of the development stage, the variations in CGDD and CEDD for sites were four times smaller than the variation in the number of days for that period. The mid-season length averaged about 26 and 34 days for early and later wet-up sites at Yuma, respectively. Thus, average mid-season lengths for Yuma sites were 4 days shorter and 4 days longer than FAO56 for early and later wet-ups, respectively, but otherwise were consistent with FAO56. Total season lengths for the Yuma sites averaged 82 days with a coefficient of variation of 18%. As with the individual growth stages, the consistency of site total season in terms of thermal units was much tighter than days, where the CVs for CGDD and CEDD were only 8 and 6%, respectively. The study also found that date of maturity can be estimated from either conventional sinusoidal heat units or the enhanced heat unit model which incorporates solar radiation. Results were similar where GDD estimated maturity within 1 ± 8 days and EDD < 1 ± 6 days. Based on climatological data, EDD appears to be a better growth stage estimator for late-in-year plantings.

In this Yuma study, we found average K_{c act} for initial and mid-season stages were 0.90 and 1.19, respectively, while they were 0.20 and 1.01 for K_cb, respectively. Corresponding K_e averages were 0.70 and 0.18, respectively. The FAO56 standard K_c for the initial stage of lettuce is given as 0.70 (Allen et al. 1998), which is lower than site-wide average and also lower than most of the sites in this study. However, as pointed out in FAO56, the initial K_c in lettuce and other vegetables is highly dependent on irrigation wetting frequency, infiltration depth, and potential evaporation rates. Frequent sprinkler irrigations may increase K_c to 1.10.

Updated standardized FAO56 mid-season K_cb and K_c that considered lettuce studies conducted after FAO56 was published are 0.95 and 1.0, respectively (Pereira et al. 2021). However, the fall-winter lettuce climates in Yuma are much drier than the FAO56 standard climate, where average RH_min at Yuma was typically less than 20% at mid-season. The locally derived mid-season K_cb and K_c averages at the Yuma sites, after adjustment to the FAO56 standard climate, averaged 0.97 and 1.15, respectively. Thus, the average adjusted K_cb was only 0.02 higher than FAO56 standardized lettuce K_cb given in Pereia et al. (2021). However, the average adjusted K_c at Yuma was 0.15 higher than the standardized FAO56 K_c. This indicates that soil evaporation was a larger contribution of the actual ET_c at the Yuma sites during mid-season than generally assumed. Higher mid-season soil evaporation might occur at Yuma due to more frequent irrigation events than applied in other lettuce studies (e.g., Bryla et al. 2010), or when drip irrigation was used (e.g., Oliveira et al. 2005; Fernandez-Pacheco et al 2014). Nevertheless, estimated soil evaporation at mid-season for Yuma sites was relatively low, ranging 7–24 mm. In the lettuce sprinkler irrigation study in Salinas, Gallardo et al. (1996) did not provide specific mid-season crop coefficient values but did evaluate the ratios of T_c and E_s to total ET_c for various lettuce growing stages, including head formation and maturity. These events approximately correspond to the beginning and end of the FAO56 lettuce mid-season stage. Their E_s/ET_c ratio at maturity was 0.18, which was slightly greater than the average ratio for all Yuma sites at mid-season (0.15). However, their E_s/ET_c at head formation was much higher, 0.33, suggesting a higher soil evaporation component of ET_c at mid-season than for the Yuma sites. Gallardo et al. (1996) also presented the E_s/ET_c ratios for other lettuce growth stages, which averaged about 0.79 for rosette stages, where the rosette stages would be comparable to the FAO56 lettuce development stage. Although actual E_s/ET_c ratios were not provided for lettuce stages encompassing the initial growth stage, Gallardo et al. (1996) inferred this ratio would be near 1.0. Other data on the contribution of soil evaporation in irrigated lettuce systems are limited. For the Yuma lettuce systems in this study, soil evaporation was estimated at about 40% of total ET_c.

Multiple vegetation indices were analyzed for fractional cover and all were potentially useful. Considering accuracy and accessibility, OSAVI was preferred: its accuracy for cover estimation over lettuce was marginally less than red-edge indices—CIRE, REIP, and TCARI—while its input bands, red and NIR, were much more widely available for high spatial and temporal resolution sensors. The slope and intercept coefficients for fractional cover estimation using NDVI were 1.05 and − 0.315 respectively, significantly different from Johnson and Trout (2012) who reported 1.26 and − 0.18, r² = 0.96. Application of those coefficients would have over-estimated cover by ~ 0.20. However, neither of those relations were used in this study: the NDVI cover regression at Yuma was less accurate (r² = 0.917) than OSAVI (r² = 0.986).

Conclusion

A study was conducted to evaluate water used by lettuce grown in the Yuma Valley, Southwestern Arizona, USA. Lettuce is a crucial food crop for North America and is produced under irrigation from late summer through early winter using Colorado River supplies. Despite its importance, water requirements, crop coefficients, and heat unit relationships, have not been updated for decades. To remediate these shortfalls, assessments of crop water use were conducted over 12 production scale fields from fall 2016 to winter 2020. Water requirements were evaluated using eddy covariance (EC) measurements of crop evapotranspiration, ET_c. The EC data were processed using standardized corrections, including spike removal, gap filling, energy balance closure analysis, and flux footprint assessment. Average lettuce ET_c was found to be 278 ± 24 mm per year. To improve modeling water requirements under the FA056 methodology, crop coefficients were then evaluated. Plant heights and drone-calibrated fractional vegetation cover data were collected and applied to model the daily soil water balance, soil evaporation, and crop transpiration. Data from these activities enabled an updating of FAO56 single and dual crop coefficients. The updates increased K_c values compared with those reported in FAO56. Based on measured ET_c and irrigation applications, the average consumptive use fraction was 82%, where 38% of irrigation amounts were provided by sprinkler. To improve the ability to forecast lettuce water requirements, heat unit relationships with growth stages were analyzed using two metrics, one based on air temperature alone (the degree-day metric GDD) and another based on air temperature and incoming solar radiations (the enhance metric, EDD). EDD was found to be a better estimator for days to maturity, with accuracy to within 6 days, and also a better way to forecast seasonal ET_c from local weather and day of planting. The water use and crop coefficient findings from this study are the first in years and will be useful for farmers, irrigators and water planners.