(PDF) A spatiotemporal decision support protocol based on thermal imagery for variable rate drip irrigation of a peach orchard

Irrigation Science https://doi.org/10.1007/s00271-022-00830-x ORIGINAL PAPER A spatiotemporal decision support protocol based on thermal imagery for variable rate drip irrigation of a peach orchard L. Katz1,2,3,4   · A. Ben‑Gal4 · M. I. Litaor3,5 · A. Naor3 · M. Peres6 · A. Peeters7 · V. Alchanatis1 · Y. Cohen1 Received: 15 March 2022 / Accepted: 26 September 2022 © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022 Abstract Precision irrigation can affect orchard water status and water productivity (WP). It is hypothesized that crop water status- based irrigation at the subfield scale can maintain tree water status according to targets, thereby increasing WP. Our objectives were to define a spatiotemporal decision support protocol for variable rate drip irrigation (SDSP-VRDI) in a well-watered peach orchard and to evaluate protocol efficiency on a subfield scale. Research was initiated during 2017 in a uniformly irri- gated commercial peach orchard. In 2018, half the orchard was converted to SDSP-VRDI utilizing a model developed to study the relationship between stem water potential (SWP) and thermal image-based crop water stress index (CWSI). In 2019, the orchard's south subplot continued to be irrigated uniformly while its north subplot was managed according to SDSP-VRDI during the primary stage of fruit growth and the period of peak irrigation (stage III). The SDSP-VRDI included seven steps including calculation of the CWSI per management cell (MC) using thermal imagery. The CWSI was used to estimate SWP that was compared to a specified target range driving irrigation applied per MC based on FAO-56. The target range was reached in most MCs by applying MC-specific irrigation. Some specific MCs responded well to higher amounts of irrigation while others did not, as evident from relative yield, WP, and water cost efficiency data. Management downscaling from field to subfield scale appears to be beneficial and could advance precision irrigation management of complex orchard systems. Abbreviations NRCI Normalized relative comparison index CWSI Crop water stress index PI Performance index MC Management cell SDSP Spatiotemporal decision support protocol NDVI Normalized difference vegetation index SWP Stem water potential (MPa) UI Uniform irrigation VRDI Variable rate drip irrigation * L. Katz WP Water productivity (Mg ­m−3)

1 Institute of Agricultural Engineering, Agricultural Research Organization, Volcani Center, 50250 Rishon‑LeZion, Israel Introduction 2 Department of Soil and Water Sciences, The Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew Precision irrigation strives to maximize crop yield and University of Jerusalem, P.O. Box 12, 7610001 Rehovot, quality while streamlining irrigation inputs over space and Israel time in spatially and temporally variable fields. The vari- 3 Department of Precision Agriculture, MIGAL Galilee ability can be manifested in various parameters including Research Institute, P.O.B. 831, 11016 Kiryat Shmona, Israel crop water status, canopy size, yield, and soil properties. 4 Environmental Physics and Irrigation, Agricultural Research The irrigation management of such variability is possible Organization, Gilat Research Center, 85280 M.P. Negev, through detailed acquisition of the relevant field parameters Israel and the use of sophisticated data analysis. 5 Department of Environmental Sciences, Tel Hai College, 1220800 Upper Galilee, Israel 6 Precision irrigation approaches Northern R&D Center, MIGAL Galilee Research Institute, P.O.B. 831, 11016 Kiryat Shmona, Israel Precision water management approaches can be catego- 7 TerraVision Lab, P.O. Box 225, rized based on the calculation of plant water requirement 8499000 Midreshet Ben‑Gurion, Israel 13 Vol.:(0123456789) Irrigation Science and according to scale of application. The classic irrigation approach following FAO-56 (Allen et al. 1998) has been widely implemented for field crops and fruit trees (Steduto et al. 2012) for estimating crop evapotranspiration (­ ETc) (mm) and is a function of environmental (meteorological) and plant-based factors: ETc = ET0 ∗ Kc [mm], (1) where ­ET0 (mm) is the evapotranspiration of a reference crop calculated by the Penman–Monteith equation (Monteith 1965) which takes into account local meteorological param- eters including air temperature, vapor pressure, radiation, and wind speed; crop coefficient (Kc) is a factor for a specific crop and growth stage traditionally determined empirically or estimated using remotely sensed satellite imagery of crop canopy size (Rozenstein et al. 2018; Beeri et al. 2020). The ­ETc is the total amount of water expected to be consumed by soil evaporation and plant transpiration. Assuming no need for additional factors due to crop stress or leaching, the applied irrigation (I), therefore, replenishes 100% of the ­ETc to the plant root zone. Two central precision irrigation approaches include: 1. Irrigation adjustment is implemented using crop-specific Kc values, i.e., I = ETc. 2. The irrigation amount aims at maintaining a certain Fig. 1  Field (a) and subfield (b) scale irrigation management and periodic crop water status target and is adjusted by a application strategies incorporating crop water status thresholds. crop water status coefficient (α) based on the difference Applied irrigation (I) is calculated by I = ­ET0 * Kc * α, incorporat- ing field-scale reference evapotranspiration ­ (ET0), crop coefficient between a pre-defined water status target and the actual (Kc) and crop water status coefficient (α). Both Kc and α can be deter- measured crop water status. The α increases or decreases mined at the field and subfield scales. The more stressed the plants depending on the deviation from the pre-defined target: (darker red), the higher the α and the amount of applied irrigation (larger valve) (color figure online) I = ET 0 ∗ Kc ∗ 𝛼[mm]. (2) the field-scale approach as it takes into account the spatial These approaches can be implemented at the field and variability of within-field crop water status for the applica- subfield scale if the adjustment parameters (Kc and α) and tion of specific irrigation per management zone. It assumes technical irrigation control are available at the relevant scale. that irrigation management downscaling from the field to The crop water status target approach can be implemented subfield scale will enable each management zone to reach with a single α on the field scale (Fig. 1a) or with a specific α the defined target water status, thus increasing the WP (the per management zone delineation (subfield scale) (Fig. 1b). amount of irrigation necessary to produce a unit of yield) of The field-scale crop water status target is an improvement each management zone and of the entire field. Although the over the classic approach (Eq. 1) as the irrigation decision subfield-scale crop water status approach is an improvement incorporates the adjustment of irrigation according to the over the field-scale approach, it requires detailed water status actual crop water status coefficient (Eq. 2), albeit using only assessments to calculate α per management zone; however, a limited number of plant water status measurements for the use of point measurements (e.g., SWP) is impractical. the entire field. A number of research groups worldwide The introduction of remote sensing imagery unlocked conducted extensive research to determine SWP irrigation the opportunity to monitor crop indices with high spatial thresholds (Naor 2006; Ben-Gal et al. 2021). In commer- and temporal resolution enabling subfield irrigation man- cial orchards that monitor SWP, the irrigation is adjusted by agement. Sensors can be mounted on a variety of platforms trial and error according to the average SWP of a number of (ground, unmanned aerial vehicle (UAV), plane, satel- measured trees to achieve the target value (e.g., Naor et al. lite) and acquire reflected (visible and near infrared range) 2006). The subfield-scale approach is an improvement over or emitted (thermal infrared range) spectra at specific 13 Irrigation Science wavelengths (Khanal et al. 2017). Variable rate irrigation dependent on vapor pressure deficit (VPD) (Idso et al. 1981; decisions have been made using vegetation indices (Sanchez Jackson et al. 1981). Tdry has been calculated with an empiri- et al; 2017; Shi et al. 2019), thermal imagery-based crop cal method (Meron et al. 2010; Gonzalez-Dugo et al. 2013), water stress index (CWSI) (Meron et al. 2010; Gonzalez- and Twet has been determined by employing empirical, theo- Dugo et al. 2013; Bahat et al. 2019), or a combination of retical, and statistical methods (Cohen et al. 2017; Meron continuous point measurement of water status and remote et al. 2010) or by calculating the average temperature of the sensing vegetation indices (Osroosh et al. 2018; Bonfante coolest 5–10% of canopy pixels of each individual thermal et al. 2019). For example, vineyards were divided into yield image (Gonzalez-Dugo et al. 2013; Rud et al. 2014; Cohen potential groups based on normalized difference vegetation et al. 2017). The calculation of Tcanopy requires the removal index (NDVI) values and irrigated according to SWP values of non-canopy and “mixed pixels” (combinations of canopy, in each group to reach a target value (Bellvert et al. 2020). In soil, weeds, and shade) for which various techniques have other cases, vineyards were divided into management zones been suggested. Meron et al. (2010) proposed using the cold- according to NDVI: zones with low NDVI values received est 33% of canopy pixels for the calculation assuming that more irrigation and high NDVI values received less (Nadav they are unaffected by non-canopy material (Meron et al. and Schweitzser 2017; Sanchez et al. 2017). In some cases, 2003). Park et al. (2017) used a feature-extraction algorithm Kc was assumed uniform for the entire field (Bellvert et al. that detected peach tree edges in an orthomosaic image and 2016a; McCarthy et al. 2010; Katz et al. 2022); while in dilated these boundaries. Recently, Osroosh et al. (2018) other cases, both Kc and 𝛼 were calculated for each manage- developed a novel algorithm for the overlay and masking ment zone (Sanchez et al. 2017; Bahat et al. 2019). Field out of soil pixels and shaded apple leaves from a thermal division into management zones has been implemented on image using an RGB image. Thermal imagery-based CWSI the basis of spatial variability of field properties including was employed to determine the spatial variability of plant yield (Leroux et al. 2018), soil electrical conductivity (Per- water status in cotton (Cohen et al. 2005, 2017), grain sor- alta and Costa 2013), NDVI and other vegetation indices that ghum (O’Shaughnessy et al. 2012), grapevines (Bellvert are indicators of crop yield potential (Bellvert et al. 2020). et al. 2014; Bahat et al. 2021), olives (Agam et al. 2013; A field can also be divided arbitrarily into grid-based man- Egea et al. 2017), apples (Osroosh et al. 2016) and peaches agement cells (MC) irrespective of field spatial variability (Gonzalez-Dugo et al. 2013; Bellvert et al. 2016b), using as demonstrated in center pivot sprinkler (McCarthy et al. a variety of platforms such as UAV, aircraft, satellite or 2010) and drip-irrigated fields (Nadav and Schweitzser ground-based measurements, with various spatial resolu- 2017; Sanchez et al. 2017; Bahat et al. 2019; Katz et al. tions (Ishimwe et al. 2014). 2022). There are only a few examples of actual implementa- tion of CWSI for on-farm irrigation management and fewer descriptions of detailed protocols and decision support algo- Integration of thermal remote sensing imagery rithms. Bellvert et al. (2020) proposed a comprehensive spa- in irrigation management tial decision support system for vineyards that incorporates controlled water stress to increase wine quality. A protocol Thermal image-derived CWSI is an indirect measurement of based on thermal remote sensing for peach was briefly out- crop water status used in research and in commercial agri- lined in Katz et al. (2022) whose focus was introducing a culture. The absolute canopy temperature is a function of novel normalized relative comparison index (NRCI) com- stomata opening and subsequent crop transpiration and it is paring uniform application to variable rate application at the affected by meteorological factors including ambient tem- field scale. The present paper constitutes a completion and perature, vapor pressure, wind speed, and radiation (Jones deepening of the ideas outlined in Katz et al. (2022), and 1999). To compare thermal images and eliminate the need to concentrates on the detailed implementation and evaluation measure the meteorological parameters, normalization of the of the VRDI protocol at the subfield scale. It was hypoth- canopy temperature via the CWSI was proposed as a proxy esized that crop water status-based irrigation at the subfield of crop water status (Idso et al. 1981; Jackson et al. 1981): scale can maintain tree water status according to a target Tcanopy − Twet goal, thereby increasing WP. The main objective of this CWSI = , (3) study was to present a novel spatiotemporal decision sup- Tdry − Twet port protocol for variable rate drip irrigation (SDSP-VRDI) where Tcanopy is the temperature of the canopy, Twet is the in a spatially variable peach orchard. A secondary objective temperature of a fully-transpiring canopy and Tdry is the tem- was to quantify and categorize the response of peach trees perature of a non-transpiring (stressed) canopy. The CWSI in terms of water status, yield and WP to varying irrigation ranges from 0 to 1 where higher values indicate higher water amounts at the subfield scale and to discuss ways to improve stress. The difference between Tcanopy (°C) and Tair (°C) is the proposed protocol. 13 Irrigation Science Materials and methods 2017 to 2019 in a 4  ha commercial late-harvest peach (Prunus persica cv. 1881) orchard located near Mishmar The spatiotemporal decision support system Hayarden, Israel (33.01° N; 35.60° E). Elevation of the protocol for variable rate drip irrigation site ranges from 171 to 188 m above sea level, the average slope is 5% to the northwest and within the orchard the The SDSP-VRDI consists of seven steps (Fig. 2): (1) field slope ranges from 0 to 11.3%. The total precipitation for division into subfield units, (2) thermal imaging campaigns, years 2017 and 2019 was 228 and 603 mm, respectively, (3) thermal image processing, (4) calculation of crop water and was limited to the winter months, from October to stress index (CWSI), (5) calculation of SWP (SWPc) per April. The orchard was planted in 2007 with spacing of MC using a pre-defined linear model linking measured stem 2.6 and 5 m between trees and rows, respectively. Peach water potential (SWP) to CWSI, (6) determination of a tar- orchards are typically irrigated in excess to ensure mini- get SWP for each growth stage and (7) precision irrigation mum stress. The experiment was conducted during stage decision-making for each MC. In this case, the irrigation III of fruit development when about 100% and sometimes amounts were calculated using Eq. (2). The implementa- more of E ­ T0 is returned to the orchard, representing most tion of the SDSP-VRDI was demonstrated in a commercial of the annual irrigation. peach orchard. Step 1: Division of the field into subplots Research area and experiment setup The objective of this step was to divide the field into subfield The SDSP-VRDI was implemented and evaluated based units for data collection and VRDI implementation at the on a field experiment conducted over three seasons, from subfield scale. Each subfield unit had individual irrigation control for VRDI implementation. Fig. 2  Schematic flow chart of a spatiotemporal decision support of SWPc using a linear regression model (SWP vs. CWSI), (6) deter- protocol for variable rate drip irrigation by management cells (MC): mination of SWP target curve and operational range and (7) irrigation (1) sub-division of field into MCs, (2) thermal image campaign, (3) (I) decision-making per MC using I = ­ET0 * Kc * α, incorporating the thermal image processing—removal of soil and mixed pixels, (4) cal- reference evapotranspiration (­ ET0), crop coefficient (Kc), and the crop culation of crop water stress index (CWSI) per MC, (5) calculation water status coefficient (α) (color figure online) 13 Irrigation Science Fig. 3  Field sub-division. a The Mishmar Hayarden orchard was irrigated by uniform irrigation (UI) in 2017. b During 2018– 2019, differential drip irrigation (VRDI) was implemented in the north subplot (orange squares), while UI was applied in the south subplot. Black dashed squares differentiate cells for data collection within UI management. Red dots indicate measurement trees (color figure online) General division During the base year (2017), uniform Step 2: Thermal image campaigns irrigation (UI) was implemented in the entire orchard (Fig. 3a). During the subsequent years (2018–2019), the The objective of step 2 was to conduct a thermal image cam- orchard was divided into north and south subplots managed paign to produce a high-resolution thermal image mosaic. by VRDI and UI, respectively (Fig. 3b). The entire field was Bi-weekly thermal imaging campaigns were conducted divided into 22 MCs with dimensions of 35 m × 35 m due during growth stage III of seasons 2019 (21 July 2019–15 to technical restrictions. Six healthy trees of representative Aug 2019). Each campaign employed a sensitive (± 2 °C) canopy size were selected in each MC for measurements of uncooled FLIR SC655 ­(FLIR® Systems, Inc., Billerica, MA, various parameters such as SWP and crop yield. The 11 MCs USA). The thermal camera was mounted on a six-engine in the north subplot served as the base for irrigation during drone (Datamap Group, Bnei Brak, Israel). The flight height VRDI implementation. During the 2018 season, an experi- was 100 m, producing images with a 7 cm spatial resolution. ment was conducted to quantify the relationship between Mosaics were created using ThermCam software ­(FLIR® CWSI and SWP. Additionally, VRDI was implemented in Systems, Inc., Billerica, MA, USA) and Pix4D mapper the north subplot based on SWP measurements, but did not software (Pix4D, Prilly, Switzerland). Campaigns were utilize thermal imagery for decision-making. conducted on cloudless days lasting approximately 30 min Irrigation UI decision-making occurred weekly. Irriga- between 12:30 and 15:15 concurrently with the SWP tion amounts were calculated based on Eq. (2). ­ET0 was measurements. acquired from Gadot meteorological station (33.03° N; 35.62° E) and Kc values from the Agricultural Extension Step 3: Thermal image processing Service of Israel where in growth stage III the recommended Kc was 1.1. The UI subplot irrigation was via a single lateral Thermal image mosaics were processed to remove all non- per tree row with 2.3 L ­h−1 integral pressure compensating canopy pixels from the original thermal images of the whole Uniram drippers (NetafimTM, Tel Aviv, Israel) every 0.5 m. orchard prior to calculation of CWSI. The temperature histo- VRDI irrigation decisions were made twice a week. The gram of the whole orchard was analyzed and the lowest third VRDI irrigation was delivered by a pair of laterals, one on of temperature pixels weas extracted and retained (Fig. 4a). each side of the tree, with 1.6 L ­h−1 integral pressure com- The highest two-thirds were removed from the images, pensation Uniram drippers (NetafimTM, Tel Aviv, Israel) including the soil pixels and most of the "mixed pixels" every 0.5 m. A computerized system (Dream2, Talgil Ltd., (weeds, shaded soil or a combination) (Fig. 4b). Lastly, the Kiriyat Motzkin, Israel) was used to control irrigation and morphologic erosion technique was incorporated to remove fertilization delivery to each of the MCs 1–11 of the VRDI. any of the remaining "mixed pixels" around the canopy (Dag 13 Irrigation Science Fig. 4  Thermal image processing: a the original thermal image, b primary separation of canopy (from soil and mixed) pixels by retaining the lowest third of temperature pixels, c geographical erosion of two pixels for additional removal of mixed pixels (orange oval) (color figure online) et al. 2015) using reclassification and shrink tools (with two Step 5: Computation of SWP with a linear regression model pixels) (ArcGIS Pro software, ESRI, Redlands, CA, USA) (Fig. 4c). The remaining pixels were classified as canopy During this step, the CWSI value per MC was used to calcu- pixels. late the SWP value per MC. Conventional irrigation thresh- olds are specified in terms of SWP (MPa), therefore it was Step 4: Calculation of CWSI per MC necessary to convert the CWSI values to SWP to determine the required amount of irrigation. The relationship between The purpose of this step was to use the thermal image of SWP and CWSI was established experimentally. canopy-only pixels in order to calculate the CWSI per MC. Establishment of the relationship between SWP and CWSI The CWSI was calculated as follows: the average of the The model was developed based on an experiment that took coolest 33% of the "canopy pixels" (Tcanopy) (Meron et al. place during growth stage III of season 2018 (Fig. 5a). Three 2010) of each MC and the average of the coolest 5% (Twet) experimental cells (11 m × 33 m) were designated in the (Rud et al. 2014; Cohen et al. 2017) of the "whole plot" northeast corner of the orchard, and three irrigation levels canopy pixels (R Studio software, Boston, MA, USA). Tdry were applied (0, 100, or 150% ­ET0) throughout stage III to was calculated empirically as Tair + 2° (Gonzalez-Dugo et al. create a range of soil water contents and respective plant 2013). water status. The 0% ­ET0 cell irrigation valve was closed at the onset of stage III (28 June 2018). Irrigation calculations for the 100 and 150% ­ET0 cells were made bi-weekly based on the meteorological station data and the irrigation was applied daily. 13 Irrigation Science Fig. 5  Computation of calculated stem water potential (SWPc) using water stress index (CWSI) calculated for each tree (Tcanopy = low- a linear regression model. a Thermal image acquired on 05/08/18 est 33% of canopy pixels, Twet  = lowest 5% of canopy pixels, during stage III of experimental plots irrigated with 0% (red), 100% and Tdry = Tair + 2 °C). b Linear regression model of SWP and CWSI (green), and 150% (blue) of reference evapotranspiration (­ ET0). Stem (color figure online) water potential (SWP) measured on 15 trees (gray circles) and crop A campaign including SWP measurements and thermal The optimal water status range was designated as ± 10% the imaging took place on 05 August 2018 during stage III. Five target SWP value. representative trees per irrigation treatment were designated as measurement trees. Plant water status was evaluated by Step 7: Irrigation decision‑making measuring SWP using a Scholander-type pressure chamber (Arimad, MRC Ltd., Holon, Israel). Two shaded leaves were In the final step the crop water status coefficient (α, Eq. 2) covered with an aluminum foil zip-lock bag 1.5 h before was calculated empirically for each MC based on the differ- the measurement (Naor et al. 2006). Measurements were ence between the SWP value (calculated from CWSI) and performed between the hours of 12:30–15:15. The meas- the target SWP value for each growth stage. The α value urements were averaged per tree. CWSI was calculated per increased or decreased when SWP was, respectively, lower measurement tree using the method described in step 4. A or higher than the optimal water status limits. This calcu- linear regression model was constructed using the SWP lation was repeated after each thermal imagery campaign, and CWSI values measured in each of the 15 trees in the and served as the basis for the next iteration of the above experiment: SWP = − 1.2229*CWSI − 1.1663; R2 = 0.916, process (Fig. 2). p < 0.0001 (Fig. 5b). The SWP of each MC was calculated using this model. (Fig. 5b). Protocol evaluation at the subfield scale Step 6: Determination of SWP target curve and optimal operational range The SDSP-VRDI was evaluated at the subfield scale using a variety of parameters: applied irrigation, plant water status The purpose of this step was to define an operational tar- in comparison to a target value, a spatiotemporal analysis get SWP values for irrigation scheduling and additionally, of CWSI during stage III, relative SWP at the beginning an operational range, inside which irrigation adjustment in and end of stage III, relative total irrigation, yield, WP, and not required. Each crop species and region have a differ- irrigation cost analysis. ent target value per growth stage determined empirically. In Applied irrigation Descriptive statistics were calculated this experiment, a single target value was chosen per growth for applied irrigation data. Also, the paired t test was com- stage: − 0.75, − 1.3, − 1.3, and − 1.8 MPa for growth stages puted on daily applied irrigation data between the VRDI and I, II, III, and post-harvest, respectively, roughly based on the UI subplots and compared with the recommended amount SWP of an optimal irrigation treatment (Shimshowitz 2018). 13 Irrigation Science of irrigation using JMP statistical software (JMP Inc., Cary, pattern classification per bin. The frequency (number of pix- NC, USA). els) and relative frequency per pattern were calculated for Thermal image sensitivity Descriptive statistics of CWSI both north and south subplots and for each MC, enabling the canopy pixels per MC were calculated, including mean, analysis of pattern variance for each of these areas. median, standard deviation, minimum, maximum, skew- Yield and water productivity Total fruit weight was ness and kurtosis. summed per measurement tree and averaged per MC from Measured SWP SWP was measured at the subfield scale both VRDI and UI subplots. Water productivity (WP) is the throughout the orchard to evaluate the response of each MC efficiency by which an orchard produces a crop defined by to VRDI (north subplot in 2019) and UI (2017 and south peach yield as a function of input irrigation and calculated subplot in 2019) management. During 2017, SWP was meas- per MC: ured on six trees per MC (all 22 MCs) on four days dur- ing stage III of 2017 (9 July, 20 July, 6 Aug, and 29 Aug). WP = [Yield]/[Irrigation] [Mg m−3 ], (5) During stage III of 2019, between three and four trees were where Yield units are Mg h­ a−1 and Irrigation is the cumula- measured per MC in the VRDI subplot and three trees per tive seasonal amount of applied irrigation in ­m3 ­ha−1. MC bi-weekly in the UI subplot. SWP measurements were Relative parameter calculation Values of SWP and irri- averaged for each MC. The method of SWP measurement is gation, were calculated per MC relative to the respective described in the SDSP-VRDI step 5. subplot average for a specific year: SWP Performance Index A performance index (PI) was calculated to determine the relative change between a meas- Pi ured SWP value and the respective target value at each point Relative parameter = Avg ⋅ Pj , (6) in time (McCarthy et al. 2010): where Pi is the value of the specific parameter of interest at SWPtarget − SWPi MC i (1–22) in respective subplot j (VRDI or UI). The cal- PI = , (4) SWPtarget culation enables the comparison between MCs in a specific subplot for a specific season and additionally the comparison where ­SWPtarget is the defined target value, S ­ WPi is the of a specific MC between seasons. measured SWP value at a specific time. A PI value of zero Normalized relative comparison index The normal- indicated that the measured SWP value was identical to ized relative comparison index (NRCI) (Katz et al. 2022) the target value. The PI was calculated in order to simplify compares variable rate application (VRA) relative to the future analysis when the target SWP may change. base-year uniform application (UA) data and calculates the Emerging hot spot analysis The emerging hot spot analy- improvement or impairment of specific parameters of inter- sis tool (ArcGIS Pro software, ESRI, Redlands, CA, USA) est due to VRA management: was used to identify spatiotemporal trends of hot and cold spots in a specific area. Hot spots are defined as high val- ⎛ VRAti ⎞ ues (of any parameter) surrounded by high values and cold ⎜ UAt ⎟ Normalized relative comparison index(NRCI) = ⎜ VRA∗i ⎟ , spots are defined as low values surrounding by other low ⎜ t0 ⎟ values. In this analysis, we analyzed raster CWSI data. The ⎝ UAt0 ⎠ mp emerging hot spot analysis tool was used here to evaluate (7) spatiotemporal trends of orchard CWSI and to compare them where the ratios of VRA/UA plots are compared between a to the SWP PI of each MC in the orchard during stage III base year (t0) and an additional season (ti), for any measured incorporating ten thermal images acquired during stage III parameter (mp). VRA* is the future-VRA subplot. Addition- of season 2019. A 3-D space–time cube was constructed ally, a subfield-scale NRCI (herein MC-scale NRCI) was cal- with raster CWSI layers of the orchard that were stacked by culated for yield and WP performance parameters in order date. The X and Y axes were the map coordinates and the Z to determine the relative change in performance of a specific axis of the cube was time. Each CWSI raster layer was pre- cell under VRA management: processed and included only canopy pixels. The Getis-Ord VRA Gi* statistic was calculated per bin (pixel) and statistically ⎛ MCi t i ⎞ significant hot and cold spots of CWSI within the orchard ⎜ UAti ⎟ MC-scale NRCI = ⎜ VRA∗ ⎟ , (8) canopy were identified. The hot and cold spot trends were ⎜ MCi t 0 ⎟ evaluated per bin over time using the Mann–Kendall trend ⎝ UAt0 ⎠m p test. The tool then combined these two statistics and clas- sified each bin into one of 17 categories (Online Resources where ­MCi refers to the specific MC that was under VRA Table S1). The tool output is a trend z-score, p-value, and management at ti. 13 Irrigation Science Irrigation cost analysis The cost of water per ton of due to a malfunctioning valve. The plant status coefficient fruit ($ ­ton−1) (Eq. 9) was calculated to determine the cost mirrored the trends of the applied irrigation. Throughout the effectiveness of the SDSP-VRDI in terms of irrigation and experiment, α was equal to or greater than one indicating to determine the differences in cost per MC under VRDI that the orchard water status was suboptimal and therefore, management: more irrigation was necessary to reach the defined target.α values ranged from 0.6 on 25 July 2019 to 1.6 on 04 Aug Water price ∗ Water amount [ ] Cost of water = $ton−1 , (9) 2019 and 15 Aug 2019. Yield The price of water was calculated per year with data from CWSI variability the Israel Water Authority, found via the Israeli Central Bureau of Statistics (http://​www.​cbs.​gov.​il/). Descriptive Differences in the CWSI were evident at the MC level statistics were computed for each subplot, including aver- throughout the experiment. In response to this variability, age and standard deviation. a range of differential irrigation amounts were applied to bring each MC to the pre-defined target water status. Figure 6 shows the canopy temperature maps and CWSI distribution within three selected management cells rep- Results resenting a range of daily applied irrigation amounts at the beginning (25 July 2019) and end (15 Aug 2019) of Applied irrigation the experiment. The planned irrigation for MC 9 was not fully executed on 25 July 2019 due to a malfunctioning Table 1 lists the applied irrigation in the VRDI and UI sub- valve, resulting in a shift of its distribution to higher tem- plots during stage III of 2019 and the Kc * α values per MC peratures and, therefore, higher CWSI relative to the other 1–11. The average daily applied irrigation with SDSP-VRDI two cells. The applied irrigation amounts for MC 4 were was greater than that of the UI subplot (excluding 8 and 12 relatively high during most of the experiment; however, Aug 2019). Additionally, on most of the days, the irriga- the target water status was not reached, indicating that tion of the VRDI subplot was significantly higher than the MC 4 did not respond to irrigation as was expected. The recommended amount (­ ET0 * Kc) (p < 0.05). The applied higher CWSI values in MC 4 may suggest that moisture irrigation for the UI subplot was also higher than the rec- availability was not the only reason for tree water stress. ommended amount. The range of applied water per irriga- Additionally, there were noticeably warm pixels sur- tion decision for MC 1–11 was relatively stable; however, rounding the tree canopy indicating that the quality of it was larger on 25 July 2019 and 15 Aug 2019. An outlier canopy extraction of MC 4 was of lower quality than the of 3.9 mm ­day−1 was recorded on 25 July 2019 for MC 9, other cells and may have affected the CWSI. The water Table 1  Applied irrigation (mm ­day−1), crop water status coefficient VRDI) and the reference evapotranspiration ­(ET0) during stage III of (α) for management cells (MC) 1–11 managed by the spatiotempo- 2019 (color figure online) ral decision support protocol for variable rate drip irrigation (SDSP- Applied irrigation per MC is compared to the recommended ­(ET0 * Kc) value per day: higher (orange) and lower (blue). The recommended Kc for the duration of the experiment was 1.1 and the α was equal to 1 for the uniformly irrigated (UI) subplot 13 Irrigation Science Fig. 6  Example of thermal image variability of management cells (MC) 2, 4, and 9 on 25 July 2019 and 15 Aug 2019. a Thermal images of can- opy pixels. b Pixel frequency of CWSI following removal of soil and mixed pixels. c Descriptive statistics of the pixel data (color figure online) stress in MC 9 was due to the malfunctioning valve and Peach water status relative to recommended target was reduced by the end of the experiment (15 Aug 2019) curve demonstrating that SDSP-VRDI was effective in reaching a target crop water status in specific cells. Additionally, The spatiotemporal response of peach water status to applied the range of CWSI values of all 11 management cells irrigation per MC was examined during 2017 (Fig. 7) and decreased from 0.16 to 0.9 at the beginning and end of 2019 (Fig. 8), enabling the evaluation of the SDSP-VRDI the experiment, indicating that the SDSP-VRDI was able irrigation management at the subfield (MC) scale. SWP was to respond to CWSI variability. measured on four dates during stage III of season 2017 dur- ing which UI was implemented for both north and south plots (Fig. 7). The SWP PI (Eq. 4) was above the upper threshold (plants not stressed) for 75% of the measurements. During the final measurement day (29 Aug 2017), the PI 13 Irrigation Science Fig. 7  Performance index (PI) (primary y-axis, purple line) and (blue line) and the lower threshold is PI = -− 0.1 (red line). The crop applied irrigation per day (mm  ­day−1) (secondary y-axis, blue col- coefficient (Kc), reference evapotranspiration ­(ET0) (mm) and rec- umns) (blue columns) during stage III of 2017 in management cells ommended irrigation amount (Kc * ­ET0) (mm) are listed in the table (MC) 1–22. Uniform irrigation (UI) was implemented in the whole insert (delineated in blue) (color figure online) plot. The target is PI = 0 (black line), the upper threshold is PI = 0.1 Fig. 8  Performance index (PI) (primary y-axis, purple line) and drip irrigation—SDSP-VRDI) and MCs 12–22 (uniform irrigation— applied irrigation per day of decision-making (mm ­day−1) (secondary UI). The PI target is PI = 0 (black line), the upper threshold is PI = 0.1 y-axis, blue columns) during stage III of 2019 for management cells (blue line) and the lower threshold is PI = − 0.1 (red line) (color fig- (MC) 1–11 (spatiotemporal decision support protocol for variable rate ure online) was within the operational water status range for 15 MCs. During the 2019 experiment, 68% of the measure- This suggests, that for most of growth stage III excess irriga- ments under VRDI management were within the target tion was applied to the orchard. The PI of the south (future range (Fig. 8). The average amount of applied irrigation UI) subplot mirrored the general trend of the north (future for all cells and days was 7.7  mm  ­day−1. However, the VRDI) MCs. amounts of irrigation ranged between 5.7 mm ­day−1 (MC 13 Irrigation Science 6) and 9.3  mm  ­day−1 (MC 10), indicating that various different pattern categories was more even in the VRDI MCs responded differently to irrigation. For example, high subplot. This indicates that the spatiotemporal variability amounts of irrigation were applied to both MC 4 and MC of CWSI was lower in the VRDI subplot as compared to the 10. However, the PI of MC 4 was below the lower threshold UI subplot, perhaps as a result of the precision management (wet) for most of the experiment (five days) while the PI implementation. The relative frequency of MC 2 patterns of MC 10 remained within the target range for most of the indicates a large area of statistically significant cold spots experiment (6 days). MC 2 remained within the target range for most of the experiment's duration (intensifying cold spot) during the experiment despite reduced applied irrigation where indeed low CWSI values were recorded and the trees amounts. The different responses of MCs to applied irriga- were non-stressed, indicating that an adequate amount of tion could be the result of underlying soil and plant-based water was available. Conversely, MC 9 CWSI values were parameters. The UI subplot followed a similar pattern to higher than the MC 2 value and sometimes the trees were many of the VRDI subplot MCs. The average applied irriga- stressed and therefore, almost all of MC 9 pixels fell into a tion (7.4 mm ­day−1) was similar to that of the VRDI subplot. different category, indicating that the pixels were statisti- cally significant cold spots during the last time-step (thermal Spatiotemporal analysis of within‑season crop image) but previously were hot spots (oscillating cold spot) water stress index (Online Resources Table S1). MC 4 were characterized by relatively high CWSI and areas of stressed trees within the The emerging hot spot analysis was executed to analyze both MC and the emerging hot spot analysis indicated that MC 4 spatial and temporal trends of CWSI canopy pixels during was composed of pixels representing both statistically sig- stage III (Fig. 9). The pattern variance (calculated by rela- nificant cold and hot spots during the experiment. This sug- tive frequency per pattern) of the VRDI subplot (3.64 × gests that MC 4 was perhaps not optimally irrigated, with ­10–2) was lower than that of the UI subplot (5.92 × 1­ 0–2) soil limiting factors likely contributing to the stress. The demonstrating that the overall distribution of pixels in the emerging hot spot patterns of MC 2, 4, and 9 are similar to Fig. 9  Emerging hot spot analysis based on all CWSI canopy pixels from ten thermal images during stage III of season 2019. The relative fre- quency per pattern for management cells (MC) 2, 4, and 9 is shown (color figure online) 13 Irrigation Science Fig. 10  Map of relative stem water potential (SWP) for management cells (MC) 1–22 for the beginning and end of stage III during seasons 2017 and 2019. Uniform irrigation (UI) was implemented for the whole plot during 2017 and in the south plot during 2019. The spatiotemporal decision support protocol for variable rate drip irrigation (SDSP-VRDI) was implemented in the north sub- plot during 2019 (color figure online) the SWP PI trends of each individual MC, indicating that and the north (future VRDI) subplot of 2017. At the subfield this analysis can possibly give insight into spatiotemporal scale, the relative SWP values in the VRDI subplot differed water status patterns of the orchard (Fig. 8). between the beginning and end of stage III in 2019. By the end of stage III, MC 4 substantially increased relative to Stem water potential the other MCs. Contrarily, the values in MC 9, 10, and 11 decreased and were similar to most of the VRDI subplot The SWP per MC relative to the subplot (relative SWP, values, as a result of the VRDI management. Eq. 6) was examined at the beginning and end of stage III during 2017 and 2019 to determine the efficiency of the Relative irrigation, yield and water productivity SDSP-VRDI protocol at the subfield scale (Fig. 10). During 2017, at the beginning of stage III, the range of relative SWP Relative irrigation, yield and WP were examined to determine values was higher in the south (future UI) subplot (0.27), the efficiency of the SDSP-VRDI protocol at the subfield scale. while by the end of stage III, the range was higher in the A 32% difference in relative irrigation was recorded between north (future VRDI) subplot (0.34). However, during 2019, the lowest (MC 1) and highest (MC 4) amounts of irrigation at the beginning of stage III, the range of relative SWP val- (Fig. 11). To determine how the SDSP-VRDI affected yield ues was slightly higher in the VRDI (in comparison to the and WP relatively between the different MCs, the parameters' UI) subplot. By the end of stage III, this trend was reversed: performance in each MC in 2019 were compared to their per- the range of relative SWP values in the VRDI subplot had formance in 2017, prior to the start of the protocol (MC-scale decreased substantially as compared to both the UI subplot NRCI, Eq. 8). In each cell, the MC-scale NRCI of the yield 13 Irrigation Science was most likely due to the different number of fruit on trees following thinning (62% more fruit on trees in the VRDI plot in comparison to the UI plot in 2019). This may have been due to slightly later fruit thinning in the VRDI plot. The thinning instructions were to leave 15 cm between individual fruit and to remove only up to four peaches per branch. The MC-scale NRCI values were compared to the field-scale NRCI value and grouped accordingly, adding depth to the field-scale insights. The WP NRCI values of MCs 1, 2 9, 10, and 11 were higher than the field-scale value, while MCs 4 and 8 values were substantially lower. The large amounts of irrigation applied to MC 4 and 8 did not result in higher yield, and MC 10 and 11 responded differently to large amounts of irrigation in terms of the yield increase. The response of crop yield to the annual irrigation in both Fig. 11  Map of relative irrigation management cells (MC) 1–11 dur- ing season 2019 implemented by spatiotemporal decision support VRDI and UI management cells shows an increase in crop protocol for variable rate drip irrigation (SDSP-VRDI). Average irri- yield with increasing seasonal applied irrigation (Fig. 13). gation is 734 mm (color figure online) The irrigation of both north (future VRDI) and south (UI) subplots was higher in 2017 than in succeeding years. Addi- tionally, in 2017, the orchard was irrigated 130% of the sea- sonal recommended amount, most likely as a result of higher crop load and general practice of over-irrigation. During 2019, the range of yield values in the VRDI subplot was substantially smaller than during 2017 under UI manage- ment. Higher yields were recorded in each of the MCs of the VRDI subplot in comparison to the UI subplot (excluding MC 4) in 2019. The seasonal applied irrigation in 2019 was higher than the seasonal recommended amount during this year: 117 and 122% in the VRDI and UI subplots, respec- tively. In 2018, both yield and applied irrigation for most MCs in both the VRDI and UI subplots were particularly low (Katz et al. 2022). Irrigation cost analysis The cost of water per ton fruit in 2019 in the VRDI MCs (Fig. 14) varied, from $101 t­on−1 (MC 1) to $182 t­on−1 (MC 4) (81%). In comparison to 2017, a decrease in cost of water was recorded in all VRDI MCs with the exception of MC 4 and MC 8. On the field scale, a 16% decrease in average water cost was calculated in the 2019 VRDI sub- plot in comparison to 2017, while an increase of 20% was recorded between 2017 and 2019 in the UI subplot. This Fig. 12  Map illustration of yield and water productivity (WP) per- formance in each management cell (MC 1–11) under spatiotemporal shows that the SDSP-VRDI was more cost-effective (lower decision support protocol for variable rate drip irrigation (SDSP- cost of water) than UI in terms of irrigation. Also, within the VRDI) management in 2019, normalized to its performance in 2017, VRDI subplot, the cost of water per ton of harvested fruit prior to the application of variable rate irrigation (subfield MC-scale was substantially higher in MC 4 than the average, indicating normalized relative comparison index (NRCI)). The field-scale NRCI values are shown for comparison (Katz et  al. 2022) (color figure that water could possibly be reduced in the future. online) and WP where higher than one (Fig. 12), indicating, improve- ment relative to the base year. The relative increase in yield 13 Irrigation Science Fig. 13  Crop-water production function for the peach orchard. Average yield (ton h­ a−1) response to seasonal applied irrigation (mm) per manage- ment cell (MC) (large dots) in the uniform irrigation (UI) and variable rate drip irrigation (VRDI) subplots in seasons 2017–2019 (R2 = 0.7718). VRDI management accomplished using stem water potential (2018) and thermal imagery (2019). 2018 data from (Katz et al. 2022). Seasonal recom- mended irrigation (reference evapotranspiration ­(ET0) * crop coefficient (Kc)) per season was calculated from the Gadot mete- orological station (table insert) (color figure online) Fig. 14  Cost of water per ton of fruit ($ t­on−1) in management cells (MC) 1–22 during seasons 2017 and 2019. Uniform irriga- tion (UI) was implemented for the whole plot during 2017 and in the south plot during 2019. The spatiotemporal decision support protocol for variable rate drip irrigation (SDSP- VRDI) was implemented in the north subplot during 2019 Discussion SWP) at the subfield scale to determine if tree water status can be maintained by precision irrigation and the effect of This paper presents a novel thermal image-based SDSP- productivity at the subfield scale. Currently, the most reli- VRDI protocol, including detailed steps for implementation able and common measure of orchard water status is SWP, and evaluation at the subfield scale of a commercial peach a point measurement with a limited potential for applica- orchard. The paper is a continuation of Katz et al. (2022) tion, which is impractical for detailed water status assess- that first introduced the idea of a field-scale comparison of ment required for variable rate irrigation decision-making. uniform application to variable rate application, and presents The introduction of thermal imagery allows high-resolution the following new ideas: (1) a detailed description of the mapping of CWSI, which is well-correlated with SWP, and SDSP-VRDI protocol based on the assumption that a target thus can serve as a basis of the decision-making process tree water status can be maintained in each MC by adjusting for subfield-scale variable rate irrigation. In the protocol the irrigation, (2) a comprehensive explanation of two cen- described in this paper, the water status of individual MCs tral steps within the protocol, namely, the thermal imagery is determined by the CWSI using a local linear calibration processing tools and the development of a linear regression model. model establishing the connection between SWP and CWSI, Protocol efficiency was analyzed and evaluated with and (3) analysis of relevant parameters (e.g., yield, WP and parameters including thermal image sensitivity and SWP 13 Irrigation Science throughout stage III, and yield, WP and water cost per MC well (higher relative yield and WP) to higher amounts for seasons 2017 and 2019: of irrigation (MC 10 and 11) while others did not (MC 4). Underperforming areas may stem from the soil and 1. The SDSP-VRDI was able to identify and respond to plant-based limiting factors presented above (#3). MC CWSI variability within the well-watered orchard. 4, both unresponsive to irrigation and underperform- CWSI was found to be sensitive to differences in plant ing, was highlighted using this protocol. The range of water status, similar to other reports (Park et al. 2017; performance response, emphasizes the need to regroup Gonzalez-Dugo et al. 2013). This research builds on the current MCs in order to improve the WP. the current literature, demonstrating that in addition to 5. The SDSP-VRDI managed subplot was more cost effec- identifing a range of CWSI values within the orchard, tive overall than the UI subplot in terms of cost of water the CWSI variability can be managed at the subfield per ton of fruit harvested (Fig. 14). However, in specific scale. The SDSP-VRDI was able to reduce the CWSI areas within the subplot (MC 4 and 8) cost effectiveness variability. was reduced (higher water cost and lower WP). 2. The SDSP-VRDI demonstrated the ability to manage and maintain the actual SWP within the SWP target range The data from the current research highlights the com- throughout almost the entire stage III. During 2019, plexity of the orchard system and indicates that SDSP-VRDI 68% of the irrigation decisions (n = 88; 8 dates × 11 management can benefit the orchard at both the field and cells) were able to bring the SWP PI to the target range subfield scale. On the field scale, a clear correlation between (Fig. 8), while under the UI management in 2017 only crop yield and applied irrigation in the orchard was found 25% of the SWP measurements were found to be in the during seasons 2017 and 2019 (Fig. 13). This corresponds target range (Fig. 7). During 2017, the base year, the to literature that shows a clear and primarily linear corre- orchard was distinctly above the upper threshold, indi- lation in crop-water production functions of various crops cating over-irrigation and opportunity for improvement. including peach (Steduto et al. 2012). Both the crop yield 3. Various MCs responded differently to irrigation empha- and applied irrigation were higher during 2017 as compared sizing the need for VRDI within the orchard during stage with 2019. The higher yields were most likely the result of a III, for example the daily applied irrigation in MC 2 greater number of fruit per tree. Additionally, the seasonal and MC 6 was an average of 6.5 and 5.7 mm, which is recommended irrigation ­(ET0 * Kc) was higher in 2017. Dur- much lower than MC 8, MC 10, and MC 11 (8.3, 9.3 and ing 2017, the grower irrigated 130% of the seasonal recom- 8.8 mm, respectively) (Table 1). Despite this range of mended amount with UI management, while during 2019 applied irrigation, the target goal was met. In contrast, the irrigation was only 117% of the seasonal recommended a large amount of daily irrigation was applied to MC 4 amount with SDSP-VRDI management (Fig. 13). This indi- (8.7 mm) but SWP remained out of the target range. MC cates that the SDSP-VRDI is of value but can be further 4 represent an area in the orchard with trees that did not improved. respond to increasing irrigation, raising the possibility Orchard complexity on the subfield scale is evident in that water stress in these trees was due to other factors both the spatial and temporal analysis of relative SWP, yield, and not low moisture availability. Soil aeration, pests and WP during stage III and between seasons, the MC-scale and disease, and high crop load (number of fruits per NRCI results, and the emerging hot spot analysis of CWSI. tree) may decrease the water absorption capacity and The relatively low values of WP NRCI for MC 4 and 8 indi- cause plant stress (Orcutt and Nilsen 2000). To cope cate that the (low) yield did not necessarily justify the high with such conditions in SDSP-VRDI there is a need amounts of applied irrigation. Conversely, relatively less to define an index to identify areas that are unable to irrigation was applied in MC 1 and 2 and high WP NRCI reach the SWP target within a defined time period. The values were attained. This suggests that the SWP thresholds reasons for low moisture absorption capacity will need should be revisited and perhaps slightly adjusted. The origi- to be determined in these MCs. The emerging hot spot nal number and size of the MCs was determined arbitrarily analysis reinforces the finding that MC 4 behaved dif- due to technical restrictions. The performance of the VRDI ferently than other MCs, and it might contribute in the subplot enable us to suggest a reorganization of the irrigation future in diagnosing such portions of an orchard where system and reduce the number of irrigation operations from water stress is not due to limitations in moisture avail- eleven to three based on their WP values: high WP potential ability (Fig. 9). (MC 1, 2, 9, 10, and 11), medium WP (MC 3, 5, 6, and 7) 4. SDSP-VRDI management resulted in a range of rela- and low WP (MC 4 and 8) (Fig. 15). This will permit setting tive yield and WP values (MC-scale NRCI) (Fig. 12) a different SWP target range for each group that will mini- despite the fact that most MCs were able to reach the mize unnecessary over-irrigation and could improve both the target SWP range. Specific areas were found to respond subfield and field-scale water productivity. 13 Irrigation Science Plant vigor can be as estimated retroactively using NDVI or additional vegetation indices. In the context of the rec- ommended orchard reorganization, upper and lower SWP thresholds should be revisited per future irrigation group to increase the WP. Also, an index needs to be defined that identifies MCs that are unable to reach the SWP target range (e.g., MC 4) within a specific time period. Here, additional parameters affecting the orchard water status will need to be examined (e.g., soil aeration) in order to prevent excessive irrigation in an MC and low WP. The protocol could also be improved by incorporating vegeta- tion indices such as NDVI to estimate K c in field crops (Rozenstein et al. 2018; Pelta et al. 2021). The estimation of Kc per MC in the orchard could further improve irriga- tion decisions. Lastly, a future long-term study incorporat- ing the protocol would enable a broad cost–benefit analysis of the system (including fixed equipment costs and vari- Fig. 15  Division of management cells 1–11 of Mishmar Hayarden able input costs) to evaluate economic feasibility of VRDI. peach orchard into future irrigation groups for improved SDSP-VRDI management (color figure online) The proposed SDSP-VRDI protocol is an important step Conclusion forward in the implementation of thermal imagery-based precision irrigation management. Despite clear benefits, A spatiotemporal decision support protocol for variable both challenges and limitations have been identified and rate drip irrigation (SDSP-VRDI) for a well-watered should be addressed to improve the current protocol: (1) stage III peach orchard was presented and the protocol the current thermal image processing (step 3) was not able efficiency evaluated at the subfield scale. The SDSP-VRDI to adequately extract pure canopy pixels in MC 4, leaving incorporated high-resolution thermal imagery to calculate “warmer” mixed pixels in the processed thermal image. This the CWSI per MC. Using a SWP-CWSI linear regression may have affected the CWSI and been a factor in the sub- model, calculated SWP was then compared to a defined sequent high irrigation amounts. It is important to fine-tune target range. A crop status coefficient was determined and the current process of canopy extraction in order to make the applied irrigation per MC calculated. Evaluation of more precise future irrigation decisions; (2) irrigation deci- the protocol indicated that CWSI was sufficiently sensi- sions were made twice a week, a high number of decisions tive to changes to differences in plant water status and by both research and field standards. However, the precise therefore can be used for peach irrigation decision-mak- amount of water necessary to produce a desired change in ing. The crop water status target approach at the subfield SWP is still unknown. In order to quickly reach and remain scale was able to maintain the water status within the tar- within a target SWP range, it is necessary to first apply get range throughout the experiment thus decreasing the larger irrigation adjustments (higher α) and then correct orchard water status variability. However, MCs responded accordingly with smaller irrigation modifications (smaller differently to irrigation. Differences in yield, WP and sub- α); (3) in order for the protocol to be sufficiently robust for sequent cost effectiveness of each MC were highlighted commercial decision-making, the constructed SWP-CWSI using the protocol. Management downscaling from field model should be examined and adapted for different varie- to subfield scale appears to be beneficial and may pos- ties in various environments; (4) the SDSP-VRDI is pres- sibly improve profits for specific areas in a well-watered ently implemented manually, however, automation of the orchard. This research provides a basis for the fine-tuning protocol can be incorporated into a decision support system of decision support systems to advance precision irrigation relatively easily. management. The protocol can be improved with future research. Supplementary Information  The online version contains supplemen- First, it is important to evaluate the proposed orchard divi- tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00271-0​ 22-0​ 0830-x. sion and reorganization of the VRDI subplot by analyz- ing spatial orchard parameters including soil depth and Acknowledgements  The authors would like to thank the grower, texture, canopy area, tree circumference and actual yield. Shlomo Cohen, for collaborating and allowing the research to be con- ducted in his peach orchard; Reshef Elmakais, Tomer Hagai, Shai Levi, 13 Irrigation Science Suliman Farhat, Omer Levi, Ishai Gilad, Ohad Masad, and Shlomi Kfir Bonfante A, Monaco E, Manna P, De Mascellis R, Basile A, Buo- for field measurements and technical support; Datamap company for nanno M, Cantilena G, Esposito A, Tedeschi A, De Michele C, imagery acquisition and pre-processing. Belfiore O, Catapano I, Ludeno G, Salinas K, Brook A (2019) LCIS DSS—an irrigation supporting system for water use effi- Funding  This research is a part of The “Eugene Kendel” Project for ciency improvement in precision agriculture: a maize case study. Development of Precision Drip Irrigation funded via the Ministry of Agric Syst 176(May):102646. https://d​ oi.o​ rg/1​ 0.1​ 016/j.a​ gsy.2​ 019.​ Agriculture and Rural Development in Israel (Grant no. 20-12-0030). 102646 The project has also received funding from the European Union’s Hori- Cohen Y, Alchanatis V, Meron M, Saranga Y, Tsipris J (2005) Estima- zon 2020 research and innovation program under Project SHui, Grant tion of leaf water potential by thermal imagery and spatial analy- agreement no. 773903. sis. J Exp Bot. https://​doi.​org/​10.​1093/​jxb/​eri174 Cohen Y, Alchanatis V, Saranga Y, Rosenberg O, Sela E (2017) Declarations  Mapping water status based on aerial thermal imagery: com- parison of methodologies for upscaling from a single leaf to Conflict of interest  The authors declare that they have no conflict of commercial fields. Precis Agric 18:801–822. https://​doi.​org/​10.​ interest. 1007/​s11119-​016-​9484-3 Dag A, Alchanatis V, Zipori I, Sprinstin M, Cohen A, Maravi T, Naor A (2015) Automated detection of malfunctions in drip-irrigation systems using thermal remote sensing in vineyards and olive References orchards. In: Proceedings of the 10th European conference on precision agriculture, pp 12–23 Agam N, Cohen Y, Berni JAJ, Alchanatis V, Kool D, Dag A, Yermi- Egea G, Padilla-Díaz CM, Martinez-Guanter J, Fernández JE, Pérez- yahu U, Ben-Gal A (2013) An insight to the performance of crop Ruiz M (2017) Assessing a crop water stress index derived from water stress index for olive trees. Agric Water Manag 118:79–86. aerial thermal imaging and infrared thermometry in super-high https://​doi.​org/​10.​1016/j.​agwat.​2012.​12.​004 density olive orchards. Agric Water Manag. https://​doi.​org/​10.​ Allen RG, Pereira LS, Raes D, Smith M (1998) Crop evapotranspira- 1016/j.​agwat.​2017.​03.​030 tion—guidelines for computing crop water requirements—FAO Gonzalez-Dugo V, Zarco-Tejada P, Nicolás E, Nortes PA, Alarcón Irrigation and drainage paper 56 JJ, Intrigliolo DS, Fereres E (2013) Using high resolution UAV Bahat I, Netzer Y, Ben-Gal A, Grünzweig JM, Peeters A, Cohen Y thermal imagery to assess the variability in the water status of (2019) Comparison of water potential and yield parameters under five fruit tree species within a commercial orchard. Precis Agric uniform and variable rate drip irrigation in a cabernet sauvignon 14(6):660–678. https://​doi.​org/​10.​1007/​s11119-​013-​9322-9 vineyard. Precis Agric 19:125–131. https://​doi.​org/​10.​3920/​978-​ Idso SB, Jackson RD, Pinter PJ, Reginato RJ, Hatfield JL (1981) 90-​8686-​888-9_​14 Normalizing the stress-degree-day parameter for environmental Bahat I, Netzer Y, Grünzweig JM, Alchanatis V, Peeters A, Goldshtein variability. Agric Meteorol 24:45–55 E, Ohana-Levi N, Ben-Gal A, Cohen Y (2021) In-season interac- Ishimwe R, Abutaleb K, Ahmed F (2014) Applications of ther- tions between vine vigor, water status and wine quality in terrain- mal imaging in agriculture—a review. Adv Remote Sens based management-zones in a ‘cabernet sauvignon’ vineyard. 03(03):128–140. https://​doi.​org/​10.​4236/​ars.​2014.​33011 Remote Sens. https://​doi.​org/​10.​3390/​rs130​91636 Jackson RD, Idso SB, Reginato RJ, Pinter PJ (1981) Canopy tem- Beeri O, Netzer Y, Munitz S, Mintz DF, Pelta R, Shilo T, Horesh perature as a crop water stress indicator. Water Resour Res 17(4) A, Mey-tal S (2020) Kc and LAI estimations using optical and Jones HG (1999) Use of infrared thermometry for estimation of sto- SAR remote sensing imagery for vineyards plots. Remote Sens matal conductance as a possible aid to irrigation scheduling. 12(21):1–15. https://​doi.​org/​10.​3390/​rs122​13478 Agric for Meteorol 95:139–149 Bellvert J, Marsal J, Girona J, Zarco-Tejada PJ (2014) Seasonal evolu- Katz L, Ben-Gal A, Litaor MI, Naor A, Peres M, Bahat I, Netzer tion of crop water stress index in grapevine varieties determined Y, Peeters A, Alchanatis V, Cohen Y (2022) Spatiotemporal with high-resolution remote sensing thermal imagery. Irrig Sci normalized ratio methodology to evaluate the impact of field- 33(2):81–93. https://​doi.​org/​10.​1007/​s00271-​014-​0456-y scale variable rate application. Precision Agric. https://​doi.​org/​ Bellvert J, Zarco-Tejada PJ, Marsal J, Girona J, González-Dugo V, 10.​1007/​s11119-​022-​09877-4 Fereres E, Bellvert J (2016a) Vineyard irrigation scheduling based Khanal S, Fulton J, Shearer S (2017) An overview of current and on airborne thermal imagery and water potential thresholds. Aust potential applications of thermal remote sensing in precision J Grape Wine Res 22(2):307–315. https://​doi.​org/​10.​1111/​ajgw.​ agriculture. Comput Electron Agric 139:22–32. https://​doi.​org/​ 12173 10.​1016/j.​compag.​2017.​05.​001 Bellvert J, Marsal J, Girona J, Gonzalez-Dugo V, Fereres E, Ustin SL, Leroux C, Jones H, Taylor J, Clenet A, Tisseyre B (2018) A zone- Zarco-Tejada PJ (2016b) Airborne thermal imagery to detect the based approach for processing and interpreting variability seasonal evolution of crop water status in peach, nectarine and in multi-temporal yield data sets. Comput Electron Agric Saturn peach orchards. Remote Sens 8(1):1–17. https://​doi.​org/​ 148(November 2017):299–308. https://​doi.​org/​10.​1016/j.​com- 10.​3390/​rs801​0039 pag.​2018.​03.​029 Bellvert J, Mata M, Vallverdú X, Paris C, Marsal J (2020) Optimiz- McCarthy AC, Hancock NH, Raine SR (2010) VARIwise: a general- ing precision irrigation of a vineyard to improve water use effi- purpose adaptive control simulation framework for spatially and ciency and profitability by using a decision-oriented vine water temporally varied irrigation at sub-field scale. Comput Electron consumption model. Precis Agric. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 1 007/​ Agric 70(1):117–128. https://​doi.​org/​10.​1016/j.​compag.​2009.​09.​ s11119-​020-​09718-2 011 Ben-Gal A, Ron Y, Yermiyahu U, Zipori I, Naoum S, Dag A (2021) Meron M, Tsipris J, Charitt D (2003) Remote mapping of crop water Evaluation of regulated deficit irrigation strategies for oil olives: status to assess spatial variability of crop stress. Precision agri- a case study for two modern Israeli cultivars. Agric Water Manag culture. In: Proceedings of the Fourth European conference on 245(April 2020):106577. https://​doi.​org/​10.​1016/j.​agwat.​2020.​ precision agriculture, pp 405–410 106577 Meron M, Tsipris J, Orlov V, Alchanatis V, Cohen Y (2010) Crop water stress mapping for site-specific irrigation by thermal imagery and 13 Irrigation Science artificial reference surfaces. Precis Agric 11(2):148–162. https://​ management. Comput Electron Agric 99:218–226. https://​doi.​ doi.​org/​10.​1007/​s11119-​009-​9153-x org/​10.​1016/j.​compag.​2013.​09.​014 Monteith JL (1965) Evaporation and environment. Symp Soc Exp Biol Rozenstein O, Haymann N, Kaplan G, Tanny J (2018) Estimating cot- 19:205–234 ton water consumption using a time series of Sentinel-2 imagery. Nadav I, Schweitzer A (2017) VRDI—variable rate drip irrigation in Agric Water Manag 207(May):44–52. https://​doi.​org/​10.​1016/j.​ vineyards. Adv Anim Biosci. https://d​ oi.o​ rg/1​ 0.1​ 017/S ​ 20404​ 7001​ agwat.​2018.​05.​017 70005​04 Rud R, Cohen Y, Alchanatis V, Levi A, Brikman R, Shenderey C, Naor A (2006) Irrigation scheduling and evaluation of tree water status Heuer B, Markovitch T, Dar Z, Rosen C, Mulla D, Nigon T (2014) in deciduous orchards. Hortic Rev. https://​doi.​org/​10.​1002/​97804​ Crop water stress index derived from multi-year ground and aerial 70767​986.​ch3 thermal images as an indicator of potato water status. Precis Agric Naor A, Gal Y, Peres M (2006) The inherent variability of water stress 15(3):273–289. https://​doi.​org/​10.​1007/​s11119-​014-​9351-z indicators in apple, nectarine and pear orchards, and the validity Sanchez LA, Sams B, Alsina MM, Hinds N, Klein LJ, Dokoozlian N of a leaf-selection procedure for water potential measurements. (2017) Improving vineyard water use efficiency and yield with Irrig Sci 24:129–135. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00271-0​ 05-0​ 016-6 variable rate irrigation in California. In: Proceedings of the 11th O’Shaughnessy SA, Evett SR, Colaizzi PD, Howell TA (2012) A crop European conference on precision agriculture, vol 8, no 2, pp water stress index and time threshold for automatic irrigation 574–577. https://​doi.​org/​10.​1017/​s2040​47001​70007​72 scheduling of grain sorghum. Agric Water Manag 107:122–132. Shi X, Han W, Zhao T, Tang J (2019) Decision support system for vari- https://​doi.​org/​10.​1016/j.​agwat.​2012.​01.​018 able rate irrigation based on UAV multispectral remote sensing. Orcutt DM, Nilsen ET (2000) Physiology of plants under stress: soil Sensors (Switzerland) 19(13) and biotic factors, vol 2. Wiley, New York Shimshowitz E (2018) The effect of irrigation and crop load on crop Osroosh Y, Peters RT, Campbell CS, Zhang Q (2016) Comparison of yield anf fruit size distribution in nectarine Cv. Arctic mist. Tel irrigation automation algorithms for drip-irrigated apple trees. Hai Academic College, Qiryat Shemona Comput Electron Agric 128:87–99. https://​doi.​org/​10.​1016/j.​ Steduto P, Hsiao TC, Fereres E, Raes D (2012) Crop yield response to compag.​2016.​08.​013 water, vol 1028. Food and Agriculture Organization of the United Osroosh Y, Khot LR, Peters RT (2018) Economical thermal-RGB Nations, Rome imaging system for monitoring agricultural crops. Comput Elec- tron Agric 147(August 2017):34–43. https://​doi.​org/​10.​1016/j.​ Publisher's Note Springer Nature remains neutral with regard to compag.​2018.​02.​018 jurisdictional claims in published maps and institutional affiliations. Park S, Ryu D, Fuentes S, Chung H, Hernández-Montes E, O’Connell M (2017) Adaptive estimation of crop water stress in nectar- Springer Nature or its licensor holds exclusive rights to this article under ine and peach orchards using high-resolution imagery from an a publishing agreement with the author(s) or other rightsholder(s); unmanned aerial vehicle (UAV). Remote Sens. https://​doi.​org/​10.​ author self-archiving of the accepted manuscript version of this article 3390/​rs908​0828 is solely governed by the terms of such publishing agreement and Pelta R, Beeri O, Shilo T, Mey-Tal S (2021) Estimating Kc on a plot applicable law. level using remote sensing and artificial intelligence. Precis Agric 21:25–39 Peralta NR, Costa JL (2013) Delineation of management zones with soil apparent electrical conductivity to improve nutrient 13