2.1. Locations
The field experiments were situated at Ihosy (Ihorombe Region, Madagascar, 2014–2017) and Batchenga (Centre Region, Cameroon, 2013–2016), which have different edapho-climatic characteristics (
Table 1). Batchenga had a humid climate with mean annual rainfall of 1375 mm and mean temperature of 29.6 °C during the trial period of 3 years. Ihosy had a dry-subhumid climate with mean annual rainfall of 635 mm and mean temperature of 24.9 °C during the trial period of 3 years. Detailed weather data is provided in
Tables S1 and S2. Typically, the abundance of inflorescences follows a unimodal pattern at Ihosy and a bi-modal pattern at Batchenga; at least 50% of the jatropha plants have inflorescences from October to March at Ihosy, whereas at least 50% of the jatropha plants have inflorescences from April to July and again from September to December at Batchenga. Batchenga has a sandy loam soil and Ihosy has a loamy sand soil. Both locations were characterized by low pH values, as well as low (Ihosy) and very low (Batchenga) amounts of available phosphorus in the soil according to Food and Agriculture Organization (FAO) soil fertility classification [
13]. Locally available mineral fertilizers and plant protectants were applied according to location and plant development specific requirements. Weed management was performed mechanically and with herbicides.
2.2. Agronomy
In Ihosy, seeds were sown in perforated polybags of 1.5 liters, which contained substrate composed of 1/3 local topsoil, 1/3 sand, and 1/3 cow dung. Seedlings were raised in a nursery under shade for 60 days and under plain sunlight for rustification for additional 14 days prior to transplanting to the field.
In Batchenga, seeds were sown in polybags, which contained substrate composed of 2/3 local topsoil and 1/3 organic manure. Seedlings were raised in a nursery under shade for 114 days and a rustification period without shade of 14 days prior to transplanting to the field. At nurseries at both locations, watering was done complementing rainfall in order to keep the substrate moist. Seedlings were transplanted in pits of 30 cm × 30 cm × 30 cm and fertilized with 30 g/pit of NPK 11/22/16 in Ihosy and 30 g/pit of NPK 20/10/10 in Batchenga.
The direct sowing variant was tested only at Ihosy. The soil was plowed and seeds were sown in pits of 30 cm × 30 cm × 30 cm, containing local topsoil, 1200 g of cow dung and 30 g of NPK 11/22/16. Two seeds were sown at a depth of 2 cm and the weaker seedling in each pit was rogued after 74 days. Direct sowing in the field was done one week after sowing of seeds in the nursery with the onset of rains.
Plants were arranged in spacings of 1.5 m × 4 m and 2.0 m × 4 m corresponding to planting densities of 1667 and 1250 plants/ha, respectively, at both locations.
The jatropha genotypes for this study were selected from the breeding program of JatroSolutions (JatroSolutions GmbH, Stuttgart, Germany). At Batchenga, the plant material under study was half sib progeny descending from jatropha varieties JSPE101, JSPE102, and JSPE002 and experimental jatropha varieties JSPE10d and JSPE10x, and, at Ihosy the plant material under study was half sib progeny descending from jatropha varieties JSPE101, JSPE102, and JSPE002 and experimental jatropha varieties JSPE10d and JSPE00x. In previous field experiments, JSPE101 and JSPE102 had shown high and stable yields across years and locations. JSPE002 and JSPE00x produce seeds lacking phorbol esters and are considered edible jatropha variants. JSPE101, JSPE102, and JSPE002 were registered in the Paraguayan national register of protected cultivars in 2018 [
16].
2.4. Experimental Design and Statistical Analysis
The experiment at Ihosy was laid out as a strip-split-plot experiment with two replications. Plant spacings and crop establishment methods represented the main strip plots and genotypes were assigned to sub plots within the spacings-establishment method combinations.
Analysis of variances for the field experiment at Ihosy was carried out on each annual data set using the following model
where
µ is the general mean,
si is the effect of the spacing,
mj is the effect of the establishing method,
gk is the genotype effect,
smij,
sgik,
mgjk, and
smgijk are the corresponding interaction effects among the main effects,
rl is the effect of the replication,
sril is the main plot error for the spacing,
mrjl is the main plot error for the establishing method,
smrijl is the interaction error and
εijkl is the residual error term.
The experiment at Batchenga was laid out as a split-plot experiment with three replications. The different spacings represented the main plots and genotypes were assigned to sub plots.
Analysis of variances for the field experiment at Batchenga was carried out on each annual data set using the following model
where
µ is the general mean,
si is the effect of the spacing,
gk is the genotype effect,
sgik is the interaction effect among the main effects,
rl is the effect of the replication,
sril is the main plot error for the spacing and
εikl is the residual error term. For the annual analyses per location, the ANOVA directive of Genstat 19 (VSN International, Hemel Hempstead, UK) [
17] was used and multiple comparisons of means were carried out after Bonferroni adjustment of the significance level.
For the combined analyses of plant spacings and genotypes in both field experiments across years and locations, the following mixed model was used
where
µ is the general mean,
ylmn is the fixed effect of the environment comprising location and year,
si is the fixed effect of the spacing,
gk is the fixed genotype effect and
gsik,
sylimn,
gylkmn, and
sgylikmn are the corresponding interaction effects among the main effects,
ryllmn is the random effect of the replication per environment,
srylilmn is the random main plot error for the spacing per environment and
εiklmn is the random residual error term. Due to the repeated measurements nature in the analysis of perennial crops, it is recommended to account for serial correlation of measurements taken on the same observational unit [
18]. Therefore, we applied a first order autoregressive (AR1) model structure on the residual error term
εiklmn. The combined analysis across environments was carried out using the VCOMPONENTS and REML directives in Genstat 19.