Two experiments were carried out in a commercial farm of 20-year-old, drip-irrigated mandarin trees ( Dancy tangerine × Clementine ( Citrus clementine Hort. Ex Tanaka × Citrus reticulata Blanco )), grafted onto Cleopatra mandarin ( Citrus reshni Hort. Ex. Tanaka ) trees. The orchard was located in La Palma (Cartagena, SE Spain). The soil with a bulk density of 1.31 g/cm ³ had a clay loam texture (11.45%, 30% and 58.55% sand, silt and clay particle size, respectively) with a medium level of organic matter (1.8%). Trees, spaced at 6 m × 4 m were irrigated by a drip irrigation system with two lines per row of trees. Each tree was irrigated with a total of 6 emitters (4 emitters with 4 L/h and 2 emitters with 2 L/h). The application of each irrigation treatment is explained in detail below. Irrigation water was a mixture of desalinized water with water from the Tajo-Segura Water Transfer System, with an electrical conductivity of 1.42 dS/m and pH of around 8.

Daily meteorological information (temperature, relative humidity and precipitation (T, RH, Prec) was obtained from an automatic weather station (CA52- www.siam.es ) close to the experimental site. The air vapour pressure deficit (VPD) was calculated every day using air temperature and RH data. Crop evapotranspiration (ETc) was weekly determined from the product of reference evapotranspiration (ET ₀ ,), the crop coefficient (kc, between 0.2-0.6) and the correction factor vs the shaded area (kr).

The phenological stages were determined as a function of fruit development (stages I, II and III, representing the typical sigmoid model of growth in citrus crops) ( González-Altozano & Castel, 2003b ). The slow-down of MXDT was used as criterion for dividing stage II of fruit growth into two periods (early and late). Details of the application of both RDI treatments are summarized in Fig 1.

Experiment 1 (2009-2010)

Three irrigation treatments were established: (i) Control treatment (Control) irrigated at 100% ETc throughout the season, (ii) RDI-1 irrigated as the Control, except for the early stage II and from late stage III to harvest (50% of Control) and late stage II (80% of Control) (Fig. 1), and (iii) RDI-2 irrigated as the Control at 100% ETc throughout the season. RDI-2 trees were actually irrigated as Control trees in Experiment 1, but they were termed as RDI-2 trees to maintain consistency with their name in the second experiment. Fruits were stored at 3 °C and 90 ± 5% RH for 50 d followed by a shelf-life period of 3 d at 25 °C and 90 ± 5 % RH.

Experiment 2 (2010-2011)

Three irrigation treatments were established: (i) Control treatment (Control) irrigated at 115% ETc to ensure non-limiting soil water conditions; (ii) RDI-1, irrigated as the Control except for early stage II (50% Control), and (iii) RDI-2, irrigated as the Control except from flowering-fruit set (April) to late stage II (November) at 70% of the Control (Flowering-Fruit set), 50% of Control (stage I), 70% of Control (early stage II), 80% of Control (late stage II) (Fig. 1). Fruits were stored at 5 °C and 90 ± 5% RH for 34 d with or without a subsequent shelf-life period of 3 d at 25 °C and 90 ± 5% RH. Both cold storage experiments ended when the percentages of fruit suffering chilling injury exceeded 10%.

The experimental design of both experiments consisted of completely randomized blocks with three replicates for each treatment. Each replicate consisted of three adjacent tree rows with 6 trees per row. Measurements of plant water status and fruit production were taken in 4 trees of the central row, the other trees serving as borders. Agricultural practices such as pruning, weed control, fertilization and banding were the same for all the trees of the experiment and were carried out by the technical department of the commercial orchard following usual criteria for the area.

The experimental design of both cold storage experiments consisted of 30 fruits per treatment (ten fruits for each replicate). Fruit quality traits - skin color (skin C*), hardness, total solid soluble content, titratable acidity, maturity index and the percentages of peel and juice - were measured at harvest and during storage, as suggested by Conesa et al. (2014) .

Midday stem water potential (Ψ _s , MPa) was mea­sured in 6 mature leaves per treatment (two leaves per replicate) located on the south side, selected from the middle-third of the tree, using a pressure chamber (Soil Moisture Equipment Co., Model 3000) following the recommendations outlined by Hsiao (1990) . In order to estimate the intensity of stress endured by deficit treatments, the water stress integral was calculated from the values of Ψ _s , according to the equation defined by Myers (1988) :

[1]

where t is the number of measurements of Ψ _s , Ψ _i,i+1 is the average Ψ _s for any interval (MPa), Ψ _c is the value of the maximum Ψ _s measured during the season, and n is the number of days in each interval. All values were referred to those of the Control treatment.

Gas exchange measurements were taken every two weeks between 09.00 and 11.30 h in daylight hours from 6 mature leaves per treatment (2 leaves per replicate) exposed to the sun, as described by Pérez-Pérez et al. (2008a) . Maximum net CO ₂ assimilation rate (A _CO2 , μmol/m ² ·s), maximum stomatal conductance (g _sm , mmol/m ² ·s), and transpiration rate (E _m , mmol/m ² ·s) were measured at a photosynthetic photon flux density (PPFD) ∼ 1200 μmol/m ² ·s above the photosynthesis light saturation intensity for citrus leaves ( Sinclair & Allen, 1982 ). Near constant ambient CO ₂ concentration (Ca ∼ 350 μmol/mol) and leaf temperature (Tleaf ∼ 30 °C) were measured with a portable gas exchange system CIRAS-2 (PP Systems, Hitchin, Hertfordshire, UK). Instantaneous water use efficiency (WUEi) was calculated as the ratio between A _CO2 and E _m (μmol/mmol).

Micrometric TDF were monitored in 6 selected trees (two per replicate) in Control and RDI-1 treatments, using a set of linear variable displacement transducers (LVDT; Solartron Metrology, Bognor Regis, UK, model DF ±2.5 mm, precision ±10 μm) installed on the northern side of trunks and 40 cm above the ground. The transducers were mounted on holders built of aluminum and invar-an alloy comprising 64% Fe and 35% Ni that has minimal thermal expansion. Measurements were recorded by a CR1000X data­logger (Campbell Scientific, Inc., Logan, USA) every 30 s and averaged every 15 min. Several indices were derived from TDF according to Goldhamer & Fereres (2001) : maximum (MXTD) and minimum (MNTD) daily trunk diameter (μm), maximum daily trunk shrinkage (MDS = MXTD MNTD) and trunk daily growth rate (TGR, calculated as the difference between MXTD measurements obtained during two consecutive days).

WUE was calculated as the rate of yield and the total irrigation applied (kg/m ³ ). Fruits were fully harvested in one pick. The commercial picking dates depended on the needs of the commercial farm. In this case, the dates were 22 ^nd February and 3 ^rd March in the first and second experiment, respectively. Trunk perimeter was measured with a tape-measure in 4 trees per replicate to determine trunk cross-section area (TCSA, cm ² ). The effects of RDI treatments on the distribution of photosynthetic resources were also calculated as the ratio between yield/TCSA and yield/ΔTCSA ( González-Altozano & Castel, 2003b ).

Yields components (total yield, number of fruit per tree, fruit weight and fruit diameter) were measured according to Conesa et al. (2014) . An evaluation during flowering was also made weekly in order to obtain the percentage of fruit set (%), fruit drop (%), and the ratios (fruit/branch and fruit/flower) from early March to fruit drop (end of June). The measurements were taken in four fruit-bearing branches, 15-30 cm long per tree and six trees per treatment (2 trees per replicate). Phenological status was determined during the second experiment according to the classification defined by the BBCH scale ( Agustí et al. , 2003 ).

The dynamics of fruit growth was determined weekly from fruit set to harvest with a digital caliper (Mitutoyo, CD-15D) using 45 fruits chosen randomly per treatment (15 fruits per replicate). The relative growth rate (RGR) was determined based on the values of the equatorial diameters at the beginning of the experiment ( Conesa et al. , 2014 ).

Organic acids in the first experiment were extracted according to Obando-Ulloa et al. (2009) . After squeezing, the juice was filtered through four layers of cheesecloth and then centrifuged for 15 min at 12,000 rpm and 4 °C. The supernatant was immediately treated with physical (Oasis HLB 6 cc 500 mg LP Waters Co., Milford, MA, USA) and chemical (Millex-HV PVDF Durapore 0.45 μm, 13 mm Millipore, Billerica, MA, USA) filters. All the samples were thermostated at 5 °C in an auto-sampler prior to analysis using a photodiode array (PDA) detector (model 2996; Waters) connected to an HPLC (model Alliance 2695; Waters). The juice was diluted 1:10 for quantification of citric, ascorbic and succinic acids using a Rezex Roa- Organic acid H + (8%) 300 × 7.80 mm column (Phenomenex Inc., Torrance, CA, USA) working at room temperature, with a guard column KD-0138-03B. The mobile phase was 0.005 N H ₂ SO ₄ with a constant flow of 0.5 mL/min. Malic acid was diluted 1:10 and quantified at room temperature using another column (LUNA C18 5 μm, 250 × 4.6 mm; Phenomenex Inc.) and a water/methanol/trifluroacetic solution (97.7/2.2/0.1 v/v/v) at a constant flow of 0.7 mL/min. The glutamic acid was quantified at 1:100 dilution according to the method described by Obando-Ulloa et al. (2009) . All acids were quantified using a 10 μL volume for injections, UV absorbance readings at 210 nm and calibration curves were prepared with external standards of organic acids (Sigma-Aldrich, Spain) in the linear range of concentrations (usually 0 to 1000 mg/L).

A fully randomized two-way analysis of variance (ANOVA) using treatment (Control, RDI-1 and RDI-2) and cold storage time as factors was performed; or a two-way ANOVA using treatment and shelf-life as factors. Data analysis was carried out using Statgraphics Plus for Windows vers. 5.1 (Manugistics, Inc., Rockville MD, USA). Post hoc pairwise comparison between all means was performed by Duncan´s multiple range test at p <0.05. For elucidating the possible effect of RDI treatments on variability of the production measurements at harvest, an additional one-way ANOVA of the variable residuals ² previously obtained from each ANOVA model reported above was evaluated ( Romero & Zúñica, 2005 ). Linear and nonlinear regressions between plant water stress indicators were fitted using Sigma Plot 2000 (Systat, Richmond, CA, USA).

The average water applied in the Control treatment was around 630 mm during the experimental period and the reductions in water applied in RDI-1 compared with the Control were 17% and 21% during the first and second experiment, respectively (Table 1). Of note was the reduction of 39% in water consumption in RDI-2 compared with the Control in the second season (Table 1). As can be observed, the first season was very wet (annual rainfall 678 mm) and the annual ET ₀ was 1183 mm (Table 1). In the second season, the climatology was characterized by low temperatures, occasionally below 0 °C during the winter, and the annual ET ₀ and total rainfall registered were 1202 mm and 273 mm, respectively (Table 1). The highest VPD values were registered in early stage II of fruit growth - about 3.56 kPa and 3.29 kPa in the first and second experiment, respectively (Fig. 2A). ETc was 580.3 mm and 576.5 mm during the first and second experiment, respectively.

During the studied seasons, a similar pattern of Ψ _s was observed in both RDI strategies (Fig. 2E), reflecting the irrigation volume applied and the evapotranspirative demand. The Control treatment showed Ψ _s values close to -1.0 MPa during both experiments (ranging between -1.3 MPa in early stage II and -0.5 MPa during stage I). As regards RDI-1 during the first season, this treatment registered Ψ _s values below -2.0 MPa at the end of early stage II, while maximum differences of Ψ _s with respect to the Control (around 1.1 MPa) were registered in September (late stage II). In the second experiment, Ψ _s values in RDI-2 were lower than the corresponding Control values until the second-half of late stage II of fruit growth, and the greatest differences in Ψ _s compared with the Control (Δ _Ψ ∼ 0.8 MPa) were reached in August (early stage II). Coinciding with higher VPD values (Fig. 2A), RDI-1 also showed maximum differences of about 0.5 MPa with respect to the Control in early stage II. Related to this, the maximum values of the water stress integral (S _Ψ ) were registered in early stage II in both RDI treatments (Fig. 3). RDI-1 presented an accumulated S _Ψ of 106.4 MPa*day during the first experiment, whereas the accumulated S _Ψ in RDI-2 was 27% higher than in RDI-1 during the second experiment (Fig. 3B).

Of note was the sharp fall in the Ψ _s trend of all treatments during both experiments in stage III (coinciding with the winter months), reaching less than -1.8 MPa. A strong dependence was established between the Ψ _s values of the Control (correctly irrigated trees) and the minimum temperature during the previous night [Ψ _s,md =0.465 ln (Tmin) - 2.347; r ² = 0.72, ( p <0.01)] (Fig. 4).

In the first experiment, the g _sm and the A _CO2 /g _sm ratios were sensitive to the deficit applied in RDI-1 during late stage II and late stage III to harvest, respectively. In the second experiment, the lowest values of A _CO2 (∼ 4.0 μmol/m ² ·s) and g _sm (∼ 50 mmol/m ² ·s) obtained in RDI-2 during early stage II (summer) also underlined the effects of water restrictions on the physiological response of this cultivar (Fig. 5). However, the deficit imposed during stage I did not significantly influence the leaf gas exchange parameters (Fig. 5). The year effect was only significant at p <0.05 in the g _sm of the second experiment, which is consistent with the longer deficit applied in RDI-2 (Fig. 5D).

The most pronounced differences in TGR between Control and RDI-1 (around 30 μm/d) occurred when the MDS of the trunks in RDI-1 was 70 μm greater than in the Control (Fig. 2C). At that moment, the difference in Δ _Ψ was around 0.5 MPa (Fig. 2E). Despite the deficit applied in early stage II, TGR values in RDI-1 were significantly higher than in the Control in August as a result of summer rainfall events (Fig. 2A). Consistent with this, a compensatory increase in the MDXT of RDI-1 was observed, reaching similar values to the Control (Fig. 2D). Moreover, a good linear relationship was found between MDS and Ψ _s,md [MDS= 1.39 Ψ _s,md + 3.02; r ² = 0.88 ( p <0.001)] during the study (Fig. 4). Reflecting the maximum differences in Ψ _s observed in the late stage II of the first experiment (Fig. 2E), fruit RGR in RDI-1 slowed down and the mean equatorial diameter fell by 10% compared with the Control value (Fig. 2B). However, the abundant rainfall registered in September (335 mm) promoted a higher fruit RGR in RDI-1 and, as a result, there was a compensatory effect on the equatorial diameter (Fig. 2B). In the second experiment, the maximum differences in Ψ _s compared with the Control were reached in early stage II (around 0.5MP and 0.8MPa in RDI-1 and RDI-2, respectively), but had no negative effect on fruit growth (Fig. 2B). However, the moderate deficit applied (Δ _Ψ ∼ 0.5MPa) in RDI-2 during late stage II, promoted a significant decrease in the rate of fruit growth in this treatment (Fig. 2B). Consequently, the fruit diameter of RDI-2 at harvest was significantly lower (4.5%) than in the Control (Table 2).

Non-significant differences in the means or variability of yield, fruit weight and fruit diameter were found among the three treatments assayed during both experiments (Tables 2 and 3). By contrast, the interaction treatment × season for fruit weight was significant (Table 3), and related to the significant increase in the number of fruits of RDI-2 trees in the second experiment compared with the rest of the combinations. The percentage of fruit set in the RDI-1 treatment was lower than in the Control (11 ± 5 and 40 ± 8%, respectively), but RDI-2 produced similar results to the Control (38 ± 5%) (Table S1 [suppl.]). A logarithmic relationship was found between total yield (Y) and fruit set percentage (FS): [Y=66.98·ln (FS) -242.7; r ² =0.93; p <0.001].

On the other hand, neither the irrigation treatment nor the experiment affected the distribution of resources (yield: TCSA ratio and the increment of TCSA) during this study (Table S2 [suppl.]). Nevertheless, during the second experiment the more severe RDI (RDI-2) resulted in significantly higher WUE values than were observed in the Control or RDI-1 treatments (6.44 vs 3.86 or 3.53).

In the first experiment, the fruit composition and skin color at harvest were affected by RDI but differences were not very pronounced for RDI-1, and any changes in TSS were significant (Fig. 6; Table 2 and 4). At harvest, RDI-2 showed the more vivid fruit (higher C* values) and RDI-1 the dullest (slightly lower C* values) (Fig. 6).

In the second experiment, RDI-2 TSS levels at harvest were above the corresponding Control levels (Table 2). Also, the statistics for the second experiment indicated a significant effect of temperature (T) as regards TA and hardness, together with a decrease in TA and a peak of hardness at the end of storage (Fig. 6). The interpretation of the former effect is that RDI-2 fruit were more acidic and softer than Control fruit irrespective of the storage condition (Fig. 6 and Table 4). In contrast, RDI-1 showed the highest peel percentage but only in the first experiment (Table 2).

During cold storage in the first experiment, the decrease in TA followed the same pattern as ascorbic acid, and the opposite to glutamic acid (Figs. 6A and 7), but did not follow the trend of citric acid (mean values of 26.6 ± 4 g/L throughout the experiment). Malic and succinic acids were not significantly affected at any time during the study period (means of 332.3 ± 21 and 550.1 ± 66, mg/L, respectively).