Both average temperature and rainfall differed little from year to year during the growing season (particularly, in the months of June, July and August) (Table 1). Over the course of these three months, the most important for seed production because of pollination, fruit setting and fruit growing, the average temperature of these three months was 22.66 ºC, 22.63 ºC and 24.13 ºC, for years 1, 2 and 3, respectively, and total precipitation 24.8 mm., 54.5 mm and 96.3 mm. So, for the three months analysed, year 3 was a little warmer and rainier than years 1 and 2. The long-term averages for temperature and rainfall during June, July and August were, respectively, 21.9 ºC and 25.4 mm, 23.9 ºC and 25.1 mm, and 23.2 ºC and 192 mm, making a three-month average temperature of 23.7 ºC and total precipitation of 242.5 mm (Servei Meteorologic de Catalunya).

The number of days between forage harvest and seed harvest was 126, 183, and 151 for years 1, 2 and 3, respectively, for the NH1 treatment, and 96, 110 and 109 days for the NH2 treatment. The number of days was less in year 1 as it was the seeding year of the crop (seeded on March 15). Furthermore, the seed harvest date also depended on the weather forecast, because in year 1 the seed was harvested earlier (about 7-10 days) to avoid the possible danger of hail at harvest time.

The three-year average seed yield was 476 kg ha^-1, with a moderate annual variation. Seed yield was 448 kg ha^-1 in year 1, 397 kg ha^-1 in year 2 and 583 kg ha^-1 in year 3 (Table 2). On average, the alfalfa harvested only once for forage yielded 500 kg ha^-1 of seed, whereas the two-forage harvest strategy yielded 450 kg ha^-1, with significant differences in two of the three years. However, in year 2, NH2 yielded higher seed production than NH1 (Table 2). The three-year average values ranged from 466 kg ha^-1, for the 20 cm distance, 483 kg ha^-1, for the 40 cm distance, and 476 kg ha^-1, for the 60 cm distance, with significant differences in years 1 and 2 when 60 cm row distances gave the highest seed yields. Row distance affected seed yield significantly in two of the three years, with RD3 giving the highest seed yields. The three-year average seed yields were 469 kg ha^-1 for RD1, 458 kg ha^-1 for RD2 and 522 kg ha^-1 for RD3.

The ‘Ampurdan’ cultivar produced three-year average seed yields of 520 kg ha^-1, compared with 432 kg ha^-1 for ‘Aragon’. However, there were no statistically significant differences in any year, although ‘Ampurdan’ always had higher yield values. There were significant interactions between cultivar and number of harvests in the three years, because in two of the three years, the number of harvest (NH2) of ‘Ampurdan’ yielded proportionally higher than ‘Aragon’, except in one year that in NH2 produced more seed that ‘Ampurdan’.

As commented above, the number of spring harvests and the annual growing season, which can be confused with the stand age of the crop, seem to affect seed production. Temperature and rainfall differed to some degree year to year during the growing season (particularly in the months of June, July and August). Seed set and possibly pollination, therefore, could have been affected by the weather conditions in year 3, despite irrigation.

The overall three-year average forage DM yield was 7,823 kg ha^-1. The corresponding values for NH1 and NH2 were 7,567 kg ha^-1 and 8,078 kg ha^-1, respectively. In year 1, the NH1 treatment gave statistically higher values than the NH2 treatment. In years 2 and 3, the NH2 treatment gave higher values than the NH1 treatment, but without statistical significance (Table 2).

The effect of row distance on forage production was not statistically significant in any year, although, on average, narrow rows (20 cm) showed higher DM yields (8,337 kg ha^-1 compared to 7,844 kg ha^-1 for 40 cm and 7,288 kg ha^-1 for 60 cm). There was also a significant Number of harvests * Row distance interaction. Although average RD1 values were higher than RD2 in any year, the yield decrease slope differed when going from RD1 to RD2 compared to from RD2 to RD3.

Cultivar significantly affected forage yield in years 1 and 2, in which ‘Ampurdan’ outperformed ‘Aragon’, whereas there were no significant differences in year 3. On average, ‘Ampurdan’ yielded 8,260 kg ha^-1 and ‘Aragon’ 7,386 kg ha^-1.

The correlations coefficients between annual grain and forage production and also with the whole data of the three years, were significant only in year 3.

On average, NH2 produced heavier seeds than NH1 (1.937 g per 1,000 seeds vs. 1.741 g per 1000 seeds), although the difference was only statistically significant in year 1 (Table 2).

Row distance did not significantly affect seed size in any year (Table 2), although an interaction between number of harvests and row distance was detected in years 1 and 2. The average three-year values were 1.769 g per 1,000 seeds for the 20 cm row distance, 1.865 g per 1,000 seeds for the 40 cm rows and 1.882 g per 1000 seeds for the 60 cm rows.

Cultivar affected seed size significantly in years 2 and 3, with heavier ‘Aragon’ seeds. The three-year average seed size was 1.843 g per 1,000 seeds for ‘Aragon’ and 1.836 g per 1000 seeds for ‘Ampurdan’.

The results obtained reveal statistically significant differences in stand density due to number of harvests at trial end after three years of treatment. The number of plants was 78 plants m^-2 for NH1, compared to 129 plants m^-2 for NH2. However, row distance and cultivar had no effect on final stand density.

The seed and forage yields of the harvest treatments were calculated, and their gross income compared. The differences in average seed and forage production of the harvest treatments (seed + forage) were (Table 2): 1 Harvest - 2 Harvests = (500 kg ha^-1 of seed + 7,567 kg ha^-1 of forage DM) - (450 kg ha^-1 of seed + 8,078 kg ha^-1 of forage DM) = 50 kg ha^-1 of seed – 511 kg ha^-1 of forage DM. Therefore, the average differences in gross income between the two harvest treatments (seed + forage) were: 50 kg ha^-1 × (3.5 – 4.0 € kg^-1) – 511 kg ha^-1 × (250 – 270 € Mg-1) = (175 to 200) – (128 to 138) € ha^-1. Taking the means of the price intervals, the difference in gross income is: 187.5 € ha^-1 – 133 € ha^-1 = 54.5 € ha^-1. Gross income is considered to be a more reliable indicator of the success of the crop grown for seed (Pajic & Markovic, 2016). However, when comparing the average economic income of the six possible NH+RD combinations, NH1 + RD1 and NH2 + RD1 gave the best results (3,931 € ha^-1 and 3,903 € ha^-1, respectively) while the NH1 + RD3 combination gave the lowest income (3,494 € ha^-1).

Overall, although the differences between NH1 and NH2 were small, it seems that (using the lowest or highest prices currently paid to farmers of both seed and forage) one harvest prior to letting the alfalfa grow for seed is better than two harvests in terms of economic benefit (gross income).

As earlier described, alfalfa harvested only once for forage yielded on average 500 kg ha^-1 of seed, whereas the two-forage harvest strategy yielded 450 kg ha^-1, with significant differences between treatments in only two of the three years. Average seed yields were 448 kg ha^-1, 397 kg ha^-1 and 584 kg ha^-1 for years 1, 2 and 3, respectively.

Over the study period, temperatures and rainfall in the growing season (mainly in June, July, and August) differed little from year to year (Table 1), and it does not appear that these factors affected the differences in seed set and seed production, except perhaps in year 3. It is known that the development rate is accelerated by increasing temperature (Fick et al., 1988). In year 3, with the highest seed yields, the three months analyzed seem more favorable and were a little warmer and rainier. In year 1, the number of days between forage harvest and seed harvest was 126. This period was shorter than in the other two years because year 1 was the seeding year and because the seed harvest was advanced to avoid hailstorms that had been forecast. This shorter time may have affected seed production. Possible reasons for the average lower seed yields in year 2 include excessive plant stand, growth, and some lodging, as normally happens in that year.

Agronomic precautions may be taken and the quality of the pollination may vary, but some losses of pollinated flowers and pods will always take place due to the competition between seed and vegetative growth. Depending on whether agroclimatic conditions are favorable to vegetative growth, ovule and embryo abortion can range between 30% and 90% of pollinated flowers (Hacquet et al., 2000).

Environmental influences on alfalfa seed yield have also been reported under many other growing conditions, including in France (Bolaños-Aguilar et al., 2002), Australia (Downes, 2002), the Czech Republic (Simon, 1997), China (Zhang et al., 2008), Italy (Iannucci et al., 2002) and Spain (Delgado et al., 2013).

The Ebro Valley climate, with its warm, sunny summer days and scant rainfall in summer (rainfall may affect pollination by insects), favours alfalfa seed production, with our experiment conducted under flood irrigation conditions. However, our seed yields were much lower than the 1200-2110 kg ha^-1 reported in California and Washington state (Rincker et al., 1988; Mueller, 2007). The reasons for the large differences between our seed production conditions and those reported in the USA potentially include their longer growing seasons and use of pollinators. We did not analyse the number of pollinators because commercial fields are relatively small in our region and are normally surrounded by fruit trees (almond, apple, peach, etc.), which host many pollinating insects. According to Hacquet et al. (2000), the best way to promote pollination is to conserve local fauna, although pollinating insects have been deliberately incorporated in other countries (Palmer & Donovan, 1980; Rincker et al., 1988; Mueller, 2007). Such diverse strategies and management conditions could explain the variations in seed yields seen in our Mediterranean environment and the USA.

As row distance affected seed production in two of the three years in our study, it seems appropriate to use, in our conditions, wide rows (60 cm) for seed production, although this would tend to lead in particular to lower biomass yields. Similar results were reported by Delgado et al. (2013) in the Ebro valley, with an average four-year seed yield of 429 kg ha^-1. Our average seed yield of 476 kg ha^-1 was higher than the 367 kg ha^-1 reported in a previous study comparing the row distance effect on seed production in quite similar conditions, but with only one harvest date in spring (Chocarro & Lloveras, 2015). The main difference between the latter study and the present one was use of a sprinkler as opposed to flooded irrigation. It is known that sprinkler irrigation sometimes can interfere with pollination when plants are at the blooming stage (Rincker et al., 1988).

The row distance recommended for alfalfa seed production varies greatly between and in different countries, probably due to climatic conditions, irrigation practices, different varieties, and previous field experience. A distance of at least 35 cm is recommended by the French Association of Seed Producers (Hacquet et al., 2000), whereas 76-102 cm is commonly used for commercial seed production in California (Mueller, 2007), and 50 cm is considered the most appropriate distance in Lebanon (Abu-Shakra et al., 1977).

In our studies, the ‘Ampurdan’ cultivar produced a three-year average seed yield of 520 kg ha^-1 compared with 432 kg ha^-1 for the ‘Aragon’ cultivar. However, there were no statistically significant differences in any year, although ‘Ampurdan’ yield was always higher.

Seed production is cultivar dependent (Abu-Shakra et al., 1977; Rincker et al., 1988; Bolaños et al., 2000). In our experiments, ‘Ampurdan’ produced yields that were significantly higher than those with ‘Aragon’. These differences were, however, smaller than those obtained in previous experiments involving only one spring harvest: 437 kg ha^-1, for ‘Ampurdan’ and 298 kg ha^-1 for ‘Aragon’ (Chocarro & Lloveras, 2015).

Our seed yields were normal by European standards: 210-570 kg ha^-1 in France (Hacquet et al., 2000), 440 kg ha^-1 in Italy (Iannucci et al., 2002), 150-220 kg ha^-1 in the Czech Republic and 155-299 kg ha^-1 in Hungary (Bócsa & Pummer, 1997). The differences in yield within Europe probably reflect both climatic and varietal diversity.

The three-year average forage DM yield was 7,823 kg ha^-1, with the NH1 treatment yielding an annual average of 7,567 kg ha^-1 and the NH2 treatment 8,078 kg ha^-1. A statistically significant difference was observed in year 1 when the NH1 treatment had a higher forage DM yield than the NH2 treatment, but not in years 2 and 3 when the NH2 presented higher values than NH1 (Table 2).

The forage DM yields of NH1 and NH2 were very similar, although a little higher for NH2, possible because one more harvest (and more growing time) was given for forage. As a trade-off, the plants used for seed in NH1 had, on average, 153 days to produce seed (from forage harvest to seed harvest), whereas in the NH2 treatment only 105 days elapsed from the second harvest to seed harvest. As previously mentioned, the alfalfa of the NH1 treatment yielded on average 500 kg ha^-1 of seed, whereas the NH2 yielded 450 kg ha^-1, with significant differences between treatments only in two of the three years.

There was also a significant interaction between Number of harvests * Row distance. Although average forage yield values for RD1 were higher, RD2 outperformed RD1 in the NH1 treatment.

Cultivar significantly affected forage yield in years 1 and 2, when ‘Ampurdan’ performed better than ‘Aragon’, but there were no significant differences in year 3. On average, ‘Ampurdan’ yielded 8,260 kg ha^-1 and ‘Aragon’ 7,388 kg ha^-1.

The annual average forage DM yields of 7,823 kg ha^-1 observed in our trial for seed production, with 2-3 forage harvests per year, were much lower than the 15-20 Mg DM ha^-1 yields produced when alfalfa was harvested 6-7 times per year, only for forage, under normal conditions (Lloveras et al., 2012).

Our study sought to maximize alfalfa seed yields while penalising forage production as little as possible. In our case, the plots that received two forage harvests before the crop was left for seed production yielded 511 kg ha^-1 DM of forage more per year than plots that had only one harvest (6.7 %), whereas the corresponding alfalfa seed yields fell by 50 kg ha^-1 per year (10%). The correlations coefficients between annual grain and forage production, show that only in one year there were significant correlations between seed and forage production.

Many forage breeders expect a negative correlation between dry matter production and seed yield. In alfalfa, a significant positive phenotypic correlation between seed and forage yield was observed by Melton (1969), while Heinrichs (1965) and Bolaños-Aguilar et al. (2000) found no relationship between seed yield and dry matter yield.

On average, the NH2 treatment produced heavier seeds than NH1 (1.937 g vs. 1.741 g per 1000 seeds). However, the difference between number of harvests was only statistically significant in year 1, possibly because the seed components (number of pods and seeds per pod) were lower and reduced competition. Our seed size generally fell within the normal range reported in the literature. In Italy, Iannucci et al. (2002) recorded a seed weight of 1.74-2.27 mg compared to our 1.741-1.937 mg. In the Lebanon, Abu-Shakra et al. (1977) recorded a seed weight of 2.0 mg and, in France, Bolaños-Aguilar et al. (2000) reported a weight of 1.87-2.14 mg, which was slightly higher than our mean value of 1.86 mg. The variation in yield components (in our case the P1000) reported in the literature probably reflects the responses of different cultivars to different irrigation and harvest management practices (Abu-Shakra et al., 1977; Martinello & Emilio, 1997).

Plant mortality over time is a normal behavior of alfalfa due to root and crown competition, disease conditions, etc. As stated by Volenec et al. (1987), the density of the plant stands declined with time with loss of plants being greatest in stands established using high rates of seeding. Similar results were obtained by Lloveras et al. (2008) in irrigated Mediterranean conditions. In their field seed rate studies, 10 kg ha^-1 of seed produced about 340 plants m^-2. In the present study about 600-800 plants m^-2 emerged, and only 78 and 129 plants m^-2 were left after the three years of the study period for the NH1 and NH2 treatments, respectively.

At present, alfalfa seed has a price of about 3.5-4.0 € kg^-1, whereas the price of forage depends on its quality and may be around 250-270 € Mg-1 (Jaume Areny, Aldahra Europe, pers. commun.). Considering these values and comparing the costs and revenues of the two harvest treatments (seed + forage), the increases in revenue to be obtained from selling either the extra seed or forage were 175-200 € ha^-1 for the 50 kg ha^-1 of extra seed in the NH1 treatment vs. 147-158 € ha^-1 for the 511 kg ha^-1 of extra forage for the NH2 treatment. This gives a difference in gross income in favour of NH1 of about 54.5 € ha^-1.

Although the differences in economic benefit (gross income) between the NH1 and NH2 treatments are relatively small, on the basis of current prices it seems be better to have only one harvest prior to letting the alfalfa grow for seed and that maximising seed yield maximizes gross income. However, this will always depend on variations in price as well as the quality of the seed and forage.