Performance of informal market sweet potato cultivars in on-farm trials in South Africa

Abstract Several new sweet potato cultivars have been released in South Africa from 2004 until 2011 with varietal traits suitable for the informal market. This paper reports on results of on-farm trials conducted to evaluate the cultivars in broader agro-ecologies in the country. The trials were conducted at 15 localities across the 2011/12 to 2014/15 planting seasons including six cultivars with varying flesh colours. Data collection included marketable and unmarketable storage root yield classes, dry mass, and taste acceptability. The sites were representative of cool subtropical and warm temperate regions. Statistical analysis included ANOVA, AMMI, Discriminant analysis (DA) and Principal component analysis (PCA). Significant effects were detected for locality*cultivar and region*cultivar interactions for total and marketable yield, and for cultivar only for marketable yield. The factor region was not significant. The mean marketable yield ranged from 13.1 to 19.0 t/ha, being highest for 199062.1 and Ndou, while Bophelo had average stable yield over a broad range of environments. The most prevalent unmarketable class was weevil damage (15.1%). The study demonstrated PCA and DA as quick analytical tools to associate quality traits with cultivars to facilitate cultivar recommendations. Bophelo and Ndou had been sold with success on local informal markets and have large potential to be promoted wider.


Introduction
South Africa annually produces around 60,000 tons of sweet potato (DAFF 2014), a much lower production than most other countries in sub-Saharan Africa (FAOSTAT 2014). Notwithstanding, sweet potato is an important indigenized root crop in South Africa and a popular traditional crop in the northern subtropical regions of the country (Laurie et al. 2015b). The sweet potato industry in South Africa is notably different as there is, on one end, a small number of large commercial mechanized farmers and, on the other end, the crop features prominently in smallholder cropping systems delivering to the informal markets. Smallholder and subsistence farmers easily include sweet potato in their cropping programmes because of its versatility, ease to cultivate and hardiness. Additionally, many people are starting to note the value of orange cultivars as rich sources of provitamin A (Laurie et al. 2015b). This is of great importance since Vitamin A deficiency (VAD) in South Africa is a national public health problem with 43.6% of one-to nine-year-olds being vitamin A deficient (Shisana et al. 2013). The same survey also indicated that food security is at an unfavourable level, with 28.3% of the population at risk of hunger and 26.0% already experience hunger. Based on these alarming statistics, sweet potato is promoted for food and nutrition security. While the estimated informal market of 25,000 -30,000 t for sweet potato brings an opportunity for income generation and enterprise development, there is a large potential for processing sweet potato into various products, such as biscuits, doughnuts, juice and chips that can be processed on kitchen-scale (Laurie et al. 2015b, but may be scaled to commercial size with higher sweet potato production. Since 1995, the ARC-VOP breeding programme has been focussing on high yield, good root quality, adaptability, high dry mass and sweet taste. Furthermore, in 2003, β-carotene became the major breeding goal of the programme. Cultivars released include Ndou and Monate (2003), Impilo (2007) Abstract: Several new sweet potato cultivars have been released in South Africa from 2004 until 2011 with varietal traits suitable for the informal market. This paper reports on results of on-farm trials conducted to evaluate the cultivars in broader agro-ecologies in the country. The trials were conducted at 15 localities across the 2011/12 to 2014/15 planting seasons including six cultivars with varying flesh colours. Data collection included marketable and unmarketable storage root yield classes, dry mass, and taste acceptability. The sites were representative of cool subtropical and warm temperate regions. Statistical analysis included ANOVA, AMMI, Discriminant analysis (DA) and Principal component analysis (PCA). Significant effects were detected for locality*cultivar and region*cultivar interactions for total and marketable yield, and for cultivar only for marketable yield. The factor region was not significant. The mean marketable yield ranged from 13.1 to 19.0 t/ha, being highest for 199062.1 and Ndou, while Bophelo had average stable yield over a broad range of environments. The most prevalent unmarketable class was weevil damage (15.1%). The study demonstrated PCA and DA as quick analytical tools to associate quality traits with cultivars to facilitate cultivar recommendations. Bophelo and Ndou had been sold with success on local informal markets and have large potential to be promoted wider.
(CIP), 199062.1, is being promoted since 2010 (Laurie et al. 2015a,c). Trials at five research stations across 2008/9 to 2010/11 achieved total yield of 29.5 -36.2 t/ha (mean 33.5 t/ha) and 17.3 -24.6 t/ha (mean 20.0 t/ha) for marketable yield. However, the cultivars needed to be tested in a wider range of sites at on-farm conditions to confirm performance and for cultivar recommendations.
Farmers are basically interested in cultivars with consistent superior performance on their own farms, and those specifically adapted to farmer conditions and needs, with a high degree of stability over time Ceccarelli 1989). However, there are specific advantages with varieties that are input-responsive as these may be able to respond to changes in the environment in contrast to stable/non-responsive varieties. Popular multivariate approaches to investigate stability include the Tai stability test (Tai, 1971), Additive Main Effects and Multiplicative Interaction (AMMI) model (Gauch and Zobel 1996) and GGE (genotype main effect plus G x E interaction) biplot model, which simultaneously represents both mean performance and stability (Yan and Kang 2003), and variance components (Tumwegamire et al. 2016). In addition, there is potential in using additional statistical methods (e.g. Principal Component Analysis -PCA) to demonstrate association of variables and cultivars and Discriminant Analysis (DA) to determine the most prominent cultivar characteristics (Rencher, 2002).
Across the 2011/12 to 2014/15, commercially released cultivars for the informal market were evaluated under on-farm conditions and various agro-ecologies to identify those that perform well under farmer conditions and at different sites and, thus, can be widely recommended. This paper reports on the performance of released cultivars in on-farm trials in four provinces, their adaptability and stability, trait association and progress with vine distribution of the cultivars. The secondary purpose of the on-farm trials was to expose farmers to the improved cultivars and thereby extend production.

Sites
On-farm trials were conducted in four provinces; nine locations in Kwa Zulu-Natal, four in Gauteng, one in the Eastern Cape province and one in Limpopo province. The experiments were conducted across the 2011/12 to 2014/15 planting seasons, one trial per site. The location information is presented in Table 1 and were representative of two regions, namely cool subtropical and warm temperate.

Planting material and trial maintenance
Six cultivars, as indicated in Table 2, were tested in a RCBD with three replicates at all 15 locations. Sweet potato cuttings were planted 30 cm apart in-row and 1 m apart intra-row. Plot size varied from 30 to 40 plants/ plot depending on availability of cuttings. All plants were used for data collection. Cuttings were produced in a field multiplication block at the ARC-VOP which was established with source material from the sweet potato diseasetested scheme. Fertilizer mix (500 kg/ha 2:3:4 N,P,K) was broadcast before planting due to the low nutrient status generally found in resource-poor farmer sites. Fertilizer application was, however, half of the recommended dosage (Niederwieser, 2004) since farmers seldom afford to apply more. Supplementary irrigation was applied during the growing season for five to six months at each site (Table 1).

Data collection
Data collection included rating of storage root defects such as veins, constrictions, grooves (where on a 5-point ordinal scale, 1=very bad, 5=very few), marketable storage root yield (good quality roots between 100 and 1200 g) and unmarketable storage root yield classes (cracked, insect damaged, long irregular, <100 g, >1200 g, mechanical damaged, rotten). Two medium storage roots (250 to 500 g) were sampled for determination of dry mass (dried in oven at 70°C for 48 hours). Taste acceptability of 3 to 4 cooked storage roots was determined with small informal panels, using a rating scale: 1=excellent to 5=very bad.

Statistical analysis
A combined analysis of variance was performed using SAS 9.4 (SAS Institute Inc., Cary, North Carolina) with sources: cultivar, location and region and their interaction (John and Quenouille 1977). The Additive Main Effects and the Multiplicative Interaction (AMMI) model was used to analyse genotype by environment interaction using Gen Stat Release 15.2 (VSN International Ltd., Hemel, Hempstead) for cultivar * location. "Year" was not a variable because the trials were not repeated per location (Gauch et al. 2008). Principal component analysis was performed to demonstrate association of variables (Loadings) and cultivars (Scores) with PCA Factor scores. A discriminant analysis (using XLSTAT 2013 (Addinsoft, Paris, France) was done to determine the most prominent cultivar characteristics (Rencher 2002).   3 Results

Yield components
Significant effects were detected for location*cultivar and region*cultivar for total and marketable storage root yield, and for marketable yield only cultivar (Table 3). Unmarketable classes insect damage (%) and cracked (%) showed significant effects for locality, region and cultivar, and their interactions. The marketable yield (Table 4) of cultivars 199062.1 and Ndou were significantly higher than that of all the other cultivars, and the mean marketable yield over 15 locations was 15.8 t/ha. The mean total yield ranged from 24.8 to 28.5 t/ha (mean 26.8 t/ha). Cultivar 199062.1 had significantly less insectdamaged roots than all other cultivars (Table 4). Ndou and 199062.1 had lower percentage cracked roots. The average marketable root weight was 242.2 g, which was significantly higher in cultivar 199062.1 than in other cultivars; therefore, indicative of larger storage roots. Ndou was shown to be high in dry mass with a good taste rating (Table 4). The most prevalent unmarketable class was insect-damaged roots (mean of 15.1%), predominantly caused by sweet potato weevil ( Table 5). The mean marketable percentage was 59.1% (data not shown).

Genotype by environment interaction
The AMMI biplot of the first PCA versus marketable storage root yield explained 52.39% of the genotype by environment interaction (Fig. 2a). AMMI indicated a stable (PC1 score between 1.5 and -1.5) and high marketable yield for Ndou, while Monate had also stable although average marketable yield. The highest yielding cultivar 199062.1 as well as Mvuvhelo had unstable marketable yields; PC1 scores surpassing 2.5 and -2.5, respectively. The PC1 score of orange-fleshed cultivar Bophelo was just inside the cut-off line of being stable. This can be an advantage as the cultivar may have the potential to adapt to changes in conditions. Ndou performed well at environment sites Ulundi 2 (CEZ), Pongola (PON), Vryheid  In terms of total storage root yield (Fig. 2b), the AMMI biplot of the first principal component versus total yield explained 58.45% of the genotype by environment interaction. The AMMI analysis indicated stable total yield performance from Ndou and Monate (placed close to a zero PC1 score), while the other four cultivars had unstable performance (PC1 scores surpassing 2.5 and -2.5). Impilo cultivar performed well at environment sites SYA and Soshanguve (KUT). Monate and Ndou performed particularly well at MON, while cultivars 199062.1 and Bophelo performed better at MLA.
The AMMI graph insect (mostly weevil) damage percentage presented in Figure 3a revealed that the first principal component explained 59.02% of the genotype by environment interaction. The highest insect (mostly weevil) damage % was detected for Impilo. The cultivar was stable with regards to this negative trait (PC1 score -1.25). Monate, Mvuvhelo and Bophelo had medium high insect damage but unstable in this regard, while Ndou was stable, lying close to the zero line. The lowest weevil damage was observed for 199062.1 cultivar. A megaenvironment (group of similar environments) was seen for KUT, Winterveld (WIN), SIM, MJI, MON, CEZ, DUM and MVU associating with Ndou and 199062.1 cultivars.
The AMMI biplot of the first principal component versus cracked roots percentage explained 58.45% of the genotype by environment interaction (Fig. 3b). Bophelo, Mvuvhelo and Impilo had high mean cracked root percentage. Impilo and Ndou was stable (PC1 score between 1.5 and -1.5) and the other four cultivars were unstable. Environments reacting similarly included WIN, MAP, KUT, MVU, DUM, PON, NEB, MON and BUL forming a mega-environment with Ndou cultivar, which had the lowest cracked percentage. Monate and 199062.1 cultivars were associated with MJI, whereas Bophelo cultivar associated with SYA and CEZ sites.

Cultivar association and predominant traits of cultivars
The discriminant analysis divided the cultivars into three groups (Fig. 4a), of which the grouping was mostly based on veins (%), dry mass% and average root mass. The three cream-fleshed, dry cultivars Ndou, Monate and Mvuvhelo grouped together in the top right quadrant, whereas the orange-fleshed cultivars Bophelo and Impilo are in the bottom right quadrant medium dry mass percentages). The yellow-orange cultivar 199062.1 was separated from all other cultivars and tended to produce many storage roots with the defect veins.
The results of the PCA biplot showed that the characteristics curved (%), veins, damaged (%), grooves and dry mass (%), those displayed furthest away from the centre, were most important to distinguish the cultivars (Fig. 4b). The PCA further demonstrates the association of variables (loadings) and cultivars (scores) with PCA factor scores in terms of quality traits of cultivars. Figure  4b shows that Ndou has high % marketable roots and high dry mass; therefore, it is a cultivar with good root quality. However, Ndou may also show some grooves on its roots. Monate and Mvuvhelo cultivars, despite their good yielding ability, tend to have high root-infestation by weevil and some rotten roots. The 199062.1 cultivar produced a high total yield but the large-sized roots showed several defects such as veins, curved and damaged roots. Orange-fleshed cultivars bear more unmarketable small roots (<100g), and tend to have cracked, rat-damaged and insect-infested roots.

Discussion
In the present study, an overall mean total yield of 15.8 t/ ha was measured for the six sweet potato cultivars in the on-farm trials at 15 sites. A mean total storage root yield of 15.3 t/ha was observed for 10 sweet potato cultivars evaluated by Tumwegamire et al. (2016). In a study by Andrade et al. (2016), cultivar 199062.1 achieved a total yield of 15.3 t/ha over a three-year period at Umbeluzi, Mozambique. The mean yield observed in the present study is slightly higher than the mean marketable yield of 13.5 t/ha achieved by informal market cultivars grown by four farmers in Limpopo Province in sweet potato enterprise establishment projects. The results of this study are comparable to farmers' yield and present representative sites for selection of cultivars for this level of farmers. However, the factor region was not significant for total yield nor for marketable storage root yield (Table 3). Thus, the differences in cultivar performance need to be explained in follow-up experiments by other factors; possibly water application, management, and soil conditions.
An alternative way of evaluating genotype by environment interaction is by using variance components as demonstrated by Tumwegamire et al. (2016). Furthermore, Yan et al. (2015) developed a function to calculate the optimum number of locations. Results of the present study showed wide variations for locations for both total and marketable root yields (Figs. 2a and 2b). Cracked and insect-damaged percentage (predominantly caused by weevil) showed pertinently similar performance at various locations (Figs. 3a and 3b). These traits formed mega-environments and, consequently, require fewer environment sites to make recommendations. The significant environment effects on root yield are consistent with previous studies (Tumwegamire et al. 2016, Mcharo andNdolo 2013). As expected, cultivar Ndou performed well under farmer conditions over several sites; thus, it findings need to be further validated using multi-location and multi-season data to confirm repeatability and for future recommendation of the cultivars. Principle component analysis and discriminative analysis were found to be quick analytical tools to associate quality traits with cultivars to ease cultivar recommendations in conjunction with ranking.
The present study indicated that a mean marketable yield of 15.8 t/ha was achieved in on-farm trials, and that cultivar 199062.1, followed by Ndou had the highest marketable yields. The best orange-fleshed cultivar in terms of uptake by farmers was Bophelo, with slight advantage with regards to yield and yield stability, dry mass (%) and taste over Impilo cultivar. Ndou also produced stable high total as well as marketable yield over 15 environment sites. The most prevalent unmarketable root yield class was weevil infestation. Veins, dry mass and mean marketable root mass were the main variables that made cultivars distinguishable. The improved sweet potato cultivars produce sustainable yields and, therefore, have a large commercial potential to generate income for producers to sell in the informal markets. This study is also significant because it confirmed the cultivar Bophelo as the best orange-fleshed cultivar to address vitamin A deficiency in South Africa.

Acknowledgements:
The authors thank the ARC-VOP sweet potato team, Department of Agriculture collaboration with on-farm trials and funding by ARC for on-farm trials and the National Department of Rural Development for vine and production enterprise project.
can be widely recommended for farmer use. Furthermore, a performance index can be calculated to recommend cultivars per site using various traits and ranking each as equally important if data from two years of evaluation per location is available (Lehmann and D'Abrera 2006).
A nursery foundation block established at ARC-VOP from disease-tested planting material of the evaluated cultivars, during February to April 2016 disseminated vines to nine vine grower enterprises in five provinces. In total 2.1 million cuttings from 10 nurseries were disseminated to farmers (3868 bags of vines), whereas a total of 3885 bags were disseminated in 2014/15 ( Table 6). The amount of vine dissemination in 2015/16 was negligible due to severe drought in those years. The area planted during 2016/17 season was 86 ha, of which 59% were orangefleshed sweet potato -almost all from Bophelo cultivar. The fresh produce was delivered to the informal markets in the various areas. Bophelo was the dominant orangefleshed cultivar, while Ndou topped the cream-fleshed informal market cultivars ( Table 6). The dissemination of vines is a continuation of projects that were initiated in 2010 and 2013 (Laurie et al. 2015b).

Conclusions
Over the past three planting seasons, there has been a considerable increase in availability of vines and fresh produce of the informal market cultivars. The cultivars have a large potential for income generation. The present