Two field experiments were conducted in high-yielding environments of central-southern Chile, named (a) the potential yield and (b) genetic progress trials. In addition, a set of historical data from the winter breeding program of the INIA was used to analyze yield progress (Table 1). In the potential yield trials, four elite cultivars and four advanced lines from the INIA winter wheat breeding program were evaluated at seven high-yielding sites located in the southern part of the Mediterranean region of Chile (Santa Rosa, Yungay, and Humán) and the temperate-humid zone (Curacautín, Galvarino, Carillanca, and Mafil) in 2018 and 2019 (Table 1). In the southern part of the Mediterranean zone of Chile, the mean annual temperature range was 12.6–13.3°C, and precipitation was 632–1,139 mm (Supplementary Table 1); precipitation was lower in 2019, particularly at Humán, but the three sites received irrigation in both years. In the temperate-humid zone, the mean annual temperature range was 9.8–11.9°C. Galvarino was the site with the lowest precipitation (679 and 834 mm in 2018 and 2019, respectively) but received irrigation, while Carillanca (925 and 1,400 mm), Curacautin (1,856 and 1,488 mm), and Mafil (1,729 and 1,302 mm) had higher amounts of rainfall (Supplementary Table 1). For the sites that received irrigation, furrow (in Santa Rosa, Yungay, and Humán) and sprinkler (in Galvarino) irrigation of about 50 mm per application was applied three to four times after heading (when required). In all sites, soils were derived from volcanic ashes (Andisol) of low bulk density (∼0.9 g cm^–3), large rooting depth (>100 cm), high water storage capacity and hydraulic conductivity, and high organic (>12%) contents, but also high phosphorus retention (Matus et al., 2014; Fleige et al., 2016). Plots consisted of four rows of 2 m in length and 0.20 m between rows. The seed rate was the equivalent of 200 kg ha^–1. Seeds were disinfected using 250 cc per 100-kg seeds of Real^® Top (BASF; 166.6-g L^–1 thiophanate-methyl, 8.3-g L^–1 pyraclostrobin, and 83.3-g L^–1 triticonazol) and 120 cc per 100-kg seeds of Punto 600 FS (ANASAC Chile; 600-g L^–1 imidacloprid). Fertilizer additions to the plots included 250 kg ha^–1 of triple superphosphate (46% P₂O₅) before sowing and 230 kg ha^–1 of urea (46% N) applied at four leaves (Z1.4 of Zadoks stage; Zadoks et al., 1974), tillering (Z2.4), and first node (Z3.1). In addition, 100 kg ha^–1 of sulpomag (22% K₂O, 18% MgO, and 22% S) and 100 kg ha^–1 of potassium muriate (60% KCL) were applied to tillering. Weeds were controlled with the application of 1 L ha^–1 of Bacara Forte, Bayer (120-g L^–1 flufenacet, 120-g L^–1 flurtamone, and 120-g L^–1 diflufenican) preemergence, and a further application of 4 g ha^–1 of Ally, Dupont (600-g kg^–1 metsulfuron-methyl) and 4 kg ha^–1 of MCPA (750-g L^–1 MCPA-dimethylammonium) post-emergence. The experimental design was a complete block with four replicates.

In the genetic progress trials, 13–14 winter wheat cultivars of pre- and post-green revolution, released in the country from 1965 and 2019, were evaluated at four high-yielding sites: Santa Rosa (in 2018/2019 and 2019/2020) in the Mediterranean climate zone under irrigated conditions, and Carillanca (in 2018/2019 and 2019/2020), Purranque (in 2018/2019), and Mafil (in 2019/2020) in the temperate-humid zone (Table 1). Each plot consisted of 9 rows of 2 m in length and 0.2 m apart, and the seed rate was 180 kg ha^–1 (∼330 plants m^–2). The crop management was similar to the potential yield trials. The plots were sprayed against fungal diseases with 6.25 cc L^–1 of Juwell Top (Basf; 150-g L^–1 fenpropimorph, 125-g L^–1 kresoxim-methyl, and 125-g L^–1 epoxiconazole) and 0.8 L ha^–1 of Priori (Syngenta; 250-g L^–1 azoxystrobin). At Santa Rosa, furrow irrigation was applied around four main developmental stages of the crop (if necessary): tillering (Z2.4), flag leaf just visible (Z3.7), early heading (Z5.0), and medium milk (Z7.5). The experimental design was a complete block with four replicates.

Data from 25 advanced lines and cultivars obtained from field experiments conducted at Carillanca from 1957 to 2017 were used for yield progress analysis. The plots consisted of five rows of 2 m in length and 0.20 m between rows. The seed rate was the equivalent of 180 kg ha^–1. The sowing date was in May-June of each year. Crop fertilization and weed control were performed as recommended for each year, but no fungicides or insecticides were used in the trials. The experimental design was a complete block with four replicates. To estimate grain yield, an area of 2 m² was harvested at maturity.

Data on the following were collected: (a) days to heading when 50% of the spikes, which were at stage Z5.9, were determined by regular observations of the plots; (b) plant height (from the soil level up to the ends of the awns); (c) shoot dry weight (Shoot DW) and harvest index (grain yield/Shoot DW) determined from a 1-m row harvested at maturity and placed in a forced-air oven at 60°C for 48 h; (d) the number of spikes per m² was extrapolated from a 1-m row and the number of kernels per spike and the thousand kernels weight were determined in 10 spikes taken at random from the same 1-m row; (e) the number of kernels per m² was calculated by multiplying the number of kernels per spike by the number of spikes per m²; and (f) grain yield by harvesting 1.6 m² (i.e., the whole plot) and 2 m² (i.e., five central rows) for the potential yield and genetic progress trials, respectively.

The leaf area and the specific leaf area were determined on three flag leaves per plot at early grain filling (Z7.0). The leaves were cut, and, immediately, the leaf area was measured with the WinDIAS Leaf Image Analysis System (Delta-T Devices, United Kingdom), and then placed in a forced-air oven at 60°C for 48 h and weighed for specific leaf area determination. Leaf gas exchange parameters, the net assimilation rate (An), and stomatal conductance (gs) were determined in three flag leaves per plot using a portable open system infra-red gas analyzer (CIRAS-3 model, PP Systems, Amesbury, MA, United States) with a 0.250-L min^–1 flow rate, 400 ppm CO₂, and leaf temperature at 25°C. Measurements were made at the heading (Z5.9), anthesis (Z6.9), and late milk (Z7.9) stages between 11:00 and 15:00 on sunny days at a photon flux density of at least 1,500 mmol m^–2 s^–1, using a PLC3 universal leaf cuvette (1.75 cm² of the leaf area). In addition, chlorophyll content was measured using a DUALEX sensor (Dualex Scientific, Force A, France) as a non-destructive measurement.

The normalized difference vegetation index (NDVI) was determined with a handheld spectroradiometer (GreenSeeker, Trimble, United States) in 13–14 genotypes in the genetic progress trial in 2018 and 2019. Measurements were performed on 4–5 occasions from tillering to the grain filling stage for each environment (Z2.7, main stem and seven tillers; Z3.7, flag leaf just visible; Z4.7, flag leaf sheath opening; Z6.9, anthesis complete; and Z7.7, late milk). Measurements were performed at 60 cm above the top of the canopy. The NDVI corresponds to the differences between the reflectance (R) in the near-infrared (760 nm) and red (660 nm) bands and is calculated as (Raun et al., 2001):

In 2019, RGB images of each plot were acquired using a DJI Mavic Pro (DJI, China) unmanned aerial vehicle (UAV) equipped with an RGB 24-mm camera with a 1/2.3″ 12 MP sensor. The flights were carried out 25 m above the ground level using Pix4D software¹ for automated flight control. Three flights were performed at each site at booting (Z4.7), heading (Z5.9), and grain filling (Z7.7, late milk). Orthomosaics were created from RGB images using Agisoft PhotoScan Professional software (Agisoft LLC, Russia), and then individual images from each plot were obtained. The CerealScanner plugin (Shawn Kefauver, University of Barcelona²) of the open-source ImageJ analysis platform FIJI (Fiji is Just ImageJ³) was used to extract the different color parameter information from the images (Hue, Saturation, Intensity, Lightness, a *, b *, u *, v *). Also, other parameters related to the active photosynthetic canopy and senescence, such as green area (GA = pixels with 60° < Hue < 180°); and greener area (GGA = pixels with 80° < Hue < 180°) were extracted (detailed information on these parameters is found in Casadesús et al., 2007; Hunt et al., 2013; Vergara-Díaz et al., 2016; Kefauver et al., 2017).

The intercepted PAR was evaluated at the same stages of the NDVI by using a 1-m long probe with 64 PAR sensors and a BF5 reference PAR sensor (SunScan canopy analyzer); three determinations at the bottom of each plot were taken by placing the 1-m sensor in parallel with the crop row between 11:00 and 16:00 h. Beer’s law was used to estimate the leaf area index:

and then equation was transformed to:

where LAI is the leaf area index, Io is the incident PAR at the top of the canopy, I is the transmitted PAR at the bottom of the canopy, K is the canopy extinction coefficient (calculated using the Campbell equation; more information is available in the SunScan manual⁴), and I/Io is the light transmittance (from heading, it included transmittance through both active and senescenced leaves). The fraction of PAR intercepted (fi) was measured as (Tao et al., 2018):

Later, the fi by each cultivar was plotted against accumulated thermal time (Tb = 0°C) from sowing to maturity at Santa Rosa and Carillanca, using measurements performed in 2018 and 2019.

The daily incoming radiation (MJ m^–2) during 2018 and 2019 was obtained from weather stations located at Santa Rosa and Carillanca; the incident PAR (PARi) was considered to represent half of the total incident radiation. The daily intercepted PAR was calculated as [fi × PARi], and then the accumulated intercepted PAR (IPAR) was calculated from crop emergence to heading and from heading to maturity. RUE (g MJ^–1) for each cultivar was estimated as Shoot DW/IPAR.

Differences among genotypes (G) and environments (E) were determined through ANOVA using Statgraphics Centurion Version 18.1.12 (Statgraphics Centurion, 2018). In the ANOVAs, genotypes and sites were considered fixed factors and replicate random factors. For the potential yield trials, regression analyses were performed between the grain yield BLUE value (best linear unbiased estimator) of each cultivar and the grain yield BLUE value of cultivars at each site and year (environmental index; Finlay and Wilkinson, 1963). The sites’ phenotypic correlation (Pearson) was determined using the BLUE values for each trait of the genetic progress trials. Also, regression analyses of the year of cultivar release and grain yield with the traits were performed using BLUE values for each site in the 2 years of evolution. The BLUE values were calculated by a site and across all sites using META-R (Multi Environment Trail Analysis with R) for Windows (META-R, 2016) created by the Centro Internaciónal de Mejoramiento deMaíz y Trigo (CIMMYT). It is a suite of R scripts (R Core Team, 2018) linked by a graphical user interface (GUI) designed in Java language. Figures were created by MS Excel (Microsoft Corporation, 2019).

The shoot dry weight (Shoot DW) across the seven sites between latitudes 36°31′ S and 39°90 S and 2 years (2018 and 2019) was between 25.64 (cv. Rocky) and 29.04 Mg ha^–1 (Line 26) (Figure 1A), but the average of all genotypes reached 36.35 Mg ha^–1 at one site (Curacautin) in 2018 (Figure 1B). The grain yield across the seven sites and 2 years were between 12.34 (Kiron) and 13.85 Mg ha^–1 (Chevignon), and the average of all genotypes was 17.90 and 18.58 Mg ha^–1 at Galvarino in 2018 and 2019, respectively (Figures 1C,D). The harvest index exhibited significant (p < 0.001) differences among genotypes and sites in both years, but not for the genotype x site interaction (data not shown). The harvest index across the sites and years was between 0.36 (cv. Kiron) and 0.44 (cv. Chevignon), and the average of all genotypes was the highest (0.47) at Human in 2018 (Figures 1E,F).

The linear regression between grain yield and the environmental index indicated that the highest regression coefficient (Finlay–Wilkinson slope) was for cv. Chevignon (b = 1.16) and the lowest was for cv. Kiron (b = 0.88) (Figure 2A). Grain yield had a positive and significant correlation with shoot DW (r² = 0.47, p < 0.01; Figure 2B). Weak but positive significant correlations were detected for grain yield with number of spikes per m² (r² = 0.27, p < 0.01; Figure 2C), and number of kernels per spike (r² = 0.17, p < 0.01; Figure 2D). Moreover, medium significant positive correlations were detected between plant height with both grain yield (r² = 0.38, p < 0.01; Figure 2E) and Shoot DW (r² = 0.49, p < 0.01; Figure 2F).

The average grain yield of the 25 cultivars and advanced lines evaluated in southern Chile (at INIA Carillanca) between 1959 and 2017 increased from 2.7 Mg ha^–1 in 1959 to 12.9 Mg ha^–1 in 2017 (Figure 3); the regression analysis indicates that the rate of increase in grain yield between 1959 and 2017 was 128.8 kg ha^–1 per year. However, analysis of the last 10 years showed much higher progress in grain yield (of 637 kg ha^–1 per year) and stagnation of grain yield during the last 5 years at an average of 12.45 Mg ha^–1.

In this study, the potential yield achieved by winter wheat cultivars in the temperate-humid zone of southern Chile was close to the 20.50 Mg ha^–1 (Figure 2B) that some authors (Parry et al., 2011; Reynolds et al., 2011; Ray et al., 2012; Hawkesford et al., 2013) have proposed as a target for high-yielding areas. The new cv. Chevignon had the highest yield and regression coefficient (Finlay and Wilkinson slope), indicating greater adaptability (Figure 2A). The prolonged growing period from sowing to the heading of about 180–190 days, a grain filling period of 60–70 days, mild temperatures in December-January, ample water availability (Supplementary Table 1), and favorable soil conditions explain this high potential yield. Indeed, the average photothermal coefficient (Q = solar radiation/mean temperature) during grain filling (15 November-15 January) ranged from 1.40 MJ m^–2 °C^–1 at Santa Rosa to 1.78 MJ m^–2 °C^–1 at Galvarino.

The historical yield progress observed at Carillanca (a city in the region of Araucania) between 1959 and 2017 resulted from genetic progress and improvements in agronomic practices (Figure 3). The interannual variability detected in grain yield can be explained by changes in environmental conditions, especially precipitation, and the incidence of fungal diseases (e.g., rusts and others since the trials were not sprayed). The highest yield (∼12.90 Mg ha^–1) was attained in the last 5 years (2013–2017), evidencing the high potential yield of recent cultivars and advanced lines. However, wheat yields attained by farmers are usually lower than the potential yield exhibited by the best-adapted cultivars grown in an experimental field with optimum crop management. Therefore, a yield gap can be estimated by the difference between the potential yield and the average farm yield at a regional level (Lobell et al., 2009; Fischer and Edmeades, 2010). The average yield of the whole Araucania region from 1980 to 1989 was 2.10 Mg ha^–1, and between 2010 and 2017 was 5.70 Mg ha^–1 (Oficina de Estudios y Políticas Agrarias [ODEPA], 2020). Compared to the yields obtained at Carillanca for the same period, the yield gap has increased from 3.80 Mg ha^–1 in the 80s to 5.70 Mg ha^–1 after 2010. The yield potential of wheat in two high-yielding countries, the United Kingdom and New Zealand, estimated using the Sirius crop model, was between 15.00–19.00 Mg ha^–1 and 15.60–19.50 Mg ha^–1 under water-limited and irrigated conditions, respectively (Senapati and Semenov, 2019). The yield gap was estimated at 4.00–6.00 Mg ha^–1 in both countries, similar to the gap we found between the potential yield and average yield in the Araucania region in southern Chile.

The genetic gain determination is useful to evaluate the progress of crop breeding programs. Studies on yield progress conducted worldwide have reported increases between 0.44 and 1.3% per year (Battenfield et al., 2013); in general, higher yield increases are reported when older and taller cultivars (from before the green revolution) are included in the analysis. In our study, the range of genetic gain in grain yield was 70.20 kg ha^–1 y^–1 (0.49%), representing about 55% of the yield progress observed at Carillanca (Figures 3, 4D). Other studies in winter wheat have reported lower or higher genetic gains: 11.00 kg ha^–1 y^–1 (0.40%) for semi-dwarf cultivars in low-yielding environments in Grain Plains, United States (Battenfield et al., 2013); 58.00 kg ha^–1 y^–1 (1.37%) under irrigation in Turkey for cultivars released between 1963 and 2004 (Gummadov et al., 2015); 57.50 kg ha^–1 y^–1 (0.70%) in Henan Province, China, for cultivars released after the 1950s (Gao et al., 2017); 47.40 kg ha^–1 y^–1 (0.72%) in the Hebei Province in China for cultivars released between 1964 and 2007 (Yao et al., 2019); and 60.40 kg ha^–1 y^–1 (0.72%) under irrigation and 47.50 kg ha^–1 y^–1 (0.66%) under rain-fed conditions in England, for cultivars released between 1964 and 2009 (Foulkes et al., 2016). In spring wheat cultivars, the genetic gain in a high-yielding environment (under irrigation) in the Mediterranean zone of Chile has been 43.50 kg ha^–1 y^–1 (0.51%) for bread wheat (del Pozo et al., 2014) and 72.80 kg ha^–1 y^–1 (0.73%) in durum wheat (del Pozo et al., 2019).

The genetic gains in wheat grain yield since the 1960s (i.e., green revolution) have been achieved by introducing the semi-dwarf genes Rht1 and Rht2, leading to greater partitioning of the shoot biomass to spikes and grains (Shearman et al., 2005; Trethowan et al., 2007; Gummadov et al., 2015). Thus, many agronomic traits have been attributed to these gains, including increased harvest index, thousand kernel weight, number of kernels per spike, Shoot DW, number of spikes per m², and ear length, and decreased plant height (Calderini et al., 1995; Elazab et al., 2021).

Indeed, introducing semi-dwarfing genes in the winter wheat breeding program of INIA in the 1960s reduced plant size from more than 130 cm (cv. Druchamp) to 90–100 in modern cultivars (Table 2). The reduced stem size allows new genotypes to utilize assimilates more efficiently to grow tillers than early genotypes and receive higher N application rates (Rife et al., 2019).

Although the positive correlation was detected between grain yield and Shoot DW in 2019 (Table 2), no genetic gains were reported for Shoot DW (Table 2; Figure 4C). Moreover, the increase in grain yield was associated negatively with the number of spikes per m² (Table 4), and the plant height showed negative gains with the year of cultivar release (Figure 4B). Something similar has been reported for bread wheat varieties released in Argentina between 1918 and 2011 (Lo Valvo et al., 2018), and for winter wheat varieties released in Northern China between 1960 and 2000 (Zhou et al., 2007), but other studies have reported increases in the aboveground biomass with the year of cultivar release (e.g., Xiao et al., 2012; Gao et al., 2017).

The yield progress of winter wheat cultivars released after the 1960s was associated with increases in the number of kernels per spike and the harvest index (Table 2; Figures 4E,F). Indeed, comparing the oldest cv. Druchamp (1965) with the modern cv. Kiron (2017), the increases in the number of kernels per spike and Harvest index were, on average, 43 and 75%, respectively. Increases in the number of kernels per spike after the 1960s have also been reported by other authors (Zheng et al., 2011; Xiao et al., 2012). In our study, genetic gain in the harvest index was observed until 1993, and no further increase has been observed after (Figure 4E). In spring wheat cultivars released by CIMMYT between 1966 and 2009 in the irrigated, high-potential environment of northwest Mexico, there was no increase in harvest index (Aisawi et al., 2015), but other studies performed in high-yielding environments have reported continuing increases in harvest index since the 1960s (Xiao et al., 2012; Lo Valvo et al., 2018; Yao et al., 2019). Modern cultivars are now in the range of 0.45–0.5 (Shearman et al., 2005; Zhou et al., 2007), but further increases are still possible to the theoretical limit of ∼0.6 calculated by Austin et al. (1980). The highest values of the harvest index (0.41) were obtained in cvs. Maxwell (2012) in 2018 and 2019 and Kumpa (2002) in 2018 (Table 2). Reynolds et al. (2011) pointed out that the increase in harvest index might be achieved by optimizing the partitioning of assimilates to different plant organs and increasing spike fertility.

Although significant genotypic effects were found for thousand kernels weight, with cvs. Melifen (1974) and Tukan (1993) presented the lowest and highest thousand kernels weight, the correlation with the year of cultivar release was not significant (Table 4). Nevertheless, the correlation between grain yield and thousand kernels weight for all genotypes across sites was positive and significant in both years (Table 4). Similar studies in winter wheat have reported no changes in thousand kernels weight with the year of release (Xiao et al., 2012), but others have reported a positive relationship (Zhou et al., 2007; Gao et al., 2017; Yao et al., 2019).

We divided the genetic progress data into two different periods (1965–1993) and (1993–2019) to understand why no genetic gains were detected in the harvest index after 1993 (Supplementary Table 3). The results were interesting as the genetic gains from 1965 to 1993 were due to increases in harvest index and thousand kernels weight, and decreases in Shoot DW, days to heading, and plant height, while the genetic gains from 1993 to 2019 were due to increased kernels number per m² and decreases in thousand kernels weight and plant height.

Overall, the results of genetic progress trials confirmed that the main cause for genetic gains detected in grain yield is the increase in the partitioning of biomass toward reproductive organs, without a significant role for increases in shoot biomass production (Austin et al., 1980; Deckerd et al., 1985; Siddique et al., 1989; Slafer et al., 1990; Calderini et al., 1995; Royo et al., 2007; Tshikunde et al., 2019).

During grain filling, leaves of the canopy bottom are senescenced, while the flag leaf and the penultimate leaf remain green for a longer time. Thus, the flag leaf is the main photosynthetic organ to support assimilates to the grain, besides the non-foliar tissues (i.e., spikes) and redistribution of assimilates stored in the stem (Simkin et al., 2020; Elazab et al., 2021; Tambussi et al., 2021).

The current study showed significant genotypic differences in An and gs, but there was no significant correlation with the year of cultivar release (Table 5). Positive and significant correlations were found between grain yield of genotypes across environments and An (r = 0.55; p < 0.01) and gs (r = 0.64, p < 0.01) in 2019, but not in 2018 (Table 5). Since yield and biomass accumulation are determined by the integration of various metabolic processes, where measurement of the photosynthesis in the leaf represents a small area at a specific moment (Gutiérrez-RodrıìGuez et al., 2000; Faralli and Lawson, 2020), this might be the reason why we did not always find high correlations between An and grain yield as in the 2018 example. However, there are many examples where improvement in the photosynthetic capacity led to increases in grain yield in wheat cultivars, which could be an option for further improvement of yields in high-yielding environments. For example, significant genetic gains were reported in An at the heading stage (Sun et al., 2014) and, also, after anthesis with 0.70–6.80% per year for facultative wheat released between 1981 and 2008 in Henan Province (Zheng et al., 2011); 0.47% per year for winter wheat cultivars used between 1969 and 2006 in Shandong Province (Xiao et al., 2012); 0.21% per year in ten cultivars released between 1940 and 2009 in Brazil (Beche et al., 2014); and 0.18% per year for winter wheat released between 1964 and 2007 in Hebei Province (Yao et al., 2019).

An and gs correlated positively in both years (r = 0.35, p < 0.05 in 2018; and r = 0.40, p < 0.01 in 2019; data not shown). However, the reducing trends in the flag leaf area and the specific leaf area in 2019 with the year of cultivar release and the opposite response of chlorophyll content (Table 5 and Figure 5D) supported the idea that the positive association between grain yield and An in 2019 was not only due to higher gs but also due to a higher photosynthetic capacity as well. The lower specific leaf area results in a higher assimilation rate due to increased photosynthetic machinery (i.e., chlorophyll content) per unit leaf area (Sharma-Natu and Ghildiyal, 2005). Furthermore, the positive correlation of chlorophyll content with the year of release (Table 5 and Figure 5D) may reflect more delayed senescence (stay-green) of modern cultivars. Yao et al. (2019) found an increase of 0.17% in chlorophyll content (determined as SPAD index) with the year of cultivar release, and other authors found similarly significant genetic changes in chlorophyll content (Xiao et al., 2012; Balota et al., 2017). Previous studies reported positive and significant correlations between chlorophyll content and grain yield, and Shoot DW and number of kernels per spike, indicating that improvements in chlorophyll content would benefit the genetic gain in grain yield (Yao et al., 2019).

The negative genetic gains of the fraction of intercepted PAR (fi) by the crop (Figure 5B) could be due to the more erectophile leaf habit (vertical leaf angle) of modern cultivars compared to older ones (Sadras et al., 2016), which increases the efficiency of intercepting radiation (Siddique et al., 1989). In dense canopies with fi > 0.8 (or leaf area index > 3), as in the case of our study (Figures 5A,B), the vertical leaf angles will increase photosynthetic activity (Loomis and Williams, 1969; Stöckle and Kemanian, 2009). The horizontal angle of the leaves increases the extinction coefficient, leading to shading to the bottom part of the canopy, which leads to drying of the bottom leaves, and thus reductions in photosynthetic activity (Thorne, 1971; Borojevic and Kraljevic-Balalic, 1984; Tan et al., 2020). In contrast, the vertical leaf angle facilitates sunlight penetration to the bottom of the canopy, thus enhancing photosynthesis activity by allowing greater light access to a larger proportion of the canopy (Loomis and Williams, 1969; Stöckle and Kemanian, 2009). Such an erectophile leaf habit seemed to be a vital trait for enhancing photosynthetic activity and extending the grain yield performance in wheat.

Like the fi, negative genetic gains were detected for the NDVI with the year of cultivar release (Figure 5C). Such a negative association could be due to the nature of the NDVI formulation, which uses the NIR reflectance in its formulation. There are many factors that may cause artefactual decreases in the measured NIR reflectance signal (Gitelson et al., 2002; Elazab et al., 2015, 2016) such as: (1) the canopy architecture, where canopies with erectophile leaves generally scatter more radiation into lower leaf layers than planophile canopies, and, thus, more radiation is trapped within the canopy, increasing the ability of the plant to absorb more light, and, thereby, the NIR reflectance decreases (Gitelson et al., 2002; Elazab et al., 2016); (2) increasing soil moisture content causes a decrease in NIR reflectance of the soil, which may lead to a decrease in the NIR reflectance from the total canopy (Kanemasu, 1974; Huete et al., 1985; Gitelson et al., 2002; Elazab et al., 2016); and (3) transformation of the color of wheat lower leaves from green to brown due to chlorophyll loss (i.e., senescence), which starts from the booting stage, and, thus, the active reflecting leaf layer area decreases, and the consequent NIR reflectance decreases (Leamer et al., 1980; Gitelson et al., 2002).

The Shoot DW in the present study did not show any genetic gains with the year of cultivar release (Figure 4C). Both the Shoot DW determinants, the accumulated intercepted PAR (IPAR), and the RUE showed no genetic gains with the year of cultivar release (r² = 0.03, p > 0.05 for IPAR from emergence to maturity, figure not shown; r² = 0.06, p > 0.05 for RUE, Figure 6B). Moreover, the IPAR and RUE did not show significant correlations with Shoot DW (r² = 0.006, p > 0.05 for IPAR; r² = 0.28, p > 0.05 for RUE; figures not shown). These results are in agreement with the studies of Deckerd et al. (1985) and Slafer et al. (1990), who suggested that the Shoot DW did not show any trend with the year of cultivar release because either the IPAR or the RUE changed in opposite directions and/or both traits have no association with Shoot DW. The correlations between photosynthetic capacity and grain yield might not necessarily be straightforward for modern wheat genotypes, which are reported to be source limited (Reynolds et al., 2000a). This source limitation is because the amount of CO₂ fixed for plant growth is reduced by physiological processes, such as the respiration of assimilates during the dark period, as well as the loss of carbon from root exudates, senescence, and other processes (Amthor, 1989; Reynolds et al., 2000b).

The negative association detected between IPAR calculated from sowing to heading and the year of cultivar release (Figure 6A) is mainly due to the longer growing period required to reach heading in older cultivars (Figure 4A), as positive associations have been detected between DH and IPAR from sowing to heading (r = 0.86; p < 0.0001, figure not shown). The positive genetic gains detected for IPAR calculated from heading to maturity (Figure 6A) with the year of cultivar release are mainly due to the extended period from heading to maturity, which increased the total amount of incident radiation, and not due to enhanced fi, which showed negative correlations with the year of cultivar release (see Figure 5B). The extended flag leaf duration may partially play a role in the positive genetic gains detected for IPAR with the year of cultivar release. After heading, most of the dry weight of the wheat grains is mainly derived from CO₂ taken up by parts of the canopy located above the flag leaf node (Thorne, 1965). Moreover, previous studies have reported that enhanced flag leaf duration (Fischer and Kohn, 1966; Simón, 1999; Foulkes et al., 2016) and ear photosynthesis (Maydup et al., 2010, 2012, 2014; Merah and Monneveux, 2015; Merah et al., 2018; Elazab et al., 2021) are correlated with grain yield. The flag leaf chlorophyll content from heading until grain filling (Table 5 and Figure 5D) showed a weak but significant association with the year of cultivar release. Furthermore, Elazab et al. (2021) reported a role for ear photosynthesis during grain filling for the same set of genotypes.

Like both the fi and NDVI, most RGB-derived vegetation indices (except Hue, a*, and u*) showed negative correlations with the year of release and grain yield, especially at the booting stage, followed by the heading stage (Supplementary Table 2 and Figure 7). Casadesús et al. (2007); Casadesús and Villegas (2014), Vergara-Díaz et al. (2016); Kefauver et al. (2017), and Fernandez-Gallego et al. (2019) using the same digital RGB-derived vegetation indices have shown opposite correlations (i.e., positive correlations) with grain yield in wheat and maize at different growth stages (tillering, booting, heading, anthesis, and grain filling). A positive correlation between these RGB-derived vegetation indices and grain yield is expected when a positive correlation is detected between Shoot DW and grain yield (Casadesús et al., 2007), but this was not fully accomplished in our study as discussed before.

The three locations showed that the booting stage is the best phenological stage for detecting variability in grain yield (Supplementary Table 2). The weak correlations after the booting stage could be due to several disturbance factors, such as the appearance of productive organs of lower chlorophyll content (i.e., spikes) (Tambussi et al., 2007), the previously mentioned erectophile architecture of the canopy, and the senescence of bottom canopy leaves (starting from booting), which may decrease the ability of the RGB-derived vegetation indices to capture the characteristics of the green biomass after the booting stage.

In general, explaining the grain yield and shoot biomass relationship using the spectral and/or digital RGB-derived vegetation indices is not always straightforward (Casadesús et al., 2007). The vegetation index response could vary in stressed environments according to stress timing and severity (Bort et al., 2005), such as in Mediterranean regions where terminal water and/or heat stress during the grain-filling period are common (López-Castañeda and Richards, 1994; del Pozo et al., 2016; Royo et al., 2018). Thus, low values and even negative associations of the vegetation index, especially during grain filling, could indicate a shorter crop cycle or an avoidance mechanism from the stress conditions (Bort et al., 2005; Casadesús et al., 2007). Overall, the studied spectral and RGB-derived vegetation indices confirmed a minor role for Shoot DW production in the genetic gains detected in grain yield, as confirmed previously by the yield components (Tables 2, 4; Figure 4).