The 31 occurrence points in Japan of S. edule occurred in the Cfa climate (temperate, no dry season, and hot summer). As for soil types, Acrisol (60.9%), Andosol (17.4%), Cambisol (13%), Fluvisol (4.35%) and Gleysol (4.35%) were observed. The distribution patterns were inferred by layers bio1, bio3, bio4, bio7, bio8, bio12, bio14, bio15, bio17, and elevation.
The seven distribution models along with the assembly showed optimal values in terms of their AUC and TSS. The Maxent model obtained the highest AUC value (0.93), while the GLM obtained the best TSS value (0.84) (Table 2). The SVM model obtained the largest land area for suitability ≥ 0.5 with 100,394.4 km2 (Table 2), while GLM obtained 175.5 km2. The Maxent models and the assembly reported similar territorial extensions. Regarding the Jackknife test, the greatest contribution was obtained by the bioclimatic variables related to temperature: bio1, bio3, bio4, and bio5 (Table 3). The CTA model assigned the same percentage of contribution for each one of the variables.
Table 2
Statistical parameters for the evaluation of the models and area with suitability > 0.5
Model | AUC | Sensibility | Specificity | TSS | Suitability ≥ 0.5 |
GLM | 0.91 | 0.95 | 0.89 | 0.84 | 175.5 km2 |
GBM | 0.78 | 0.78 | 0.78 | 0.56 | 48,259.6 km2 |
CTA | 0.85 | 0.71 | 0.99 | 0.70 | 65,768.7 km2 |
RF | 0.78 | 0.78 | 0.78 | 0.56 | 85,950.9 km2 |
SVM | 0.78 | 0.78 | 0.78 | 0.56 | 100,394.4 km2 |
MAXENT | 0.93 | 0.78 | 0.80 | 0.58 | 17,139.1 km2 |
ANN | 0.60 | 0.50 | 0.64 | 0.14 | 61,536.4 km2 |
Ensemble | 0.82 | 0.82 | 0.82 | 0.64 | 13,636.5 km2 |
Table 3
Contribution (%) of the bioclimatic variables to the distribution models using the Jackknife test
Model | Bio1 | Bio3 | Bio4 | Bio7 | Bio8 | Bio12 | Bio14 | Bio15 | Bio17 | Elevation |
GLM | 8.19 | 27.86 | 28.65 | 27.51 | 0 | 1.41 | 1.89 | 2.04 | 0.48 | 1.95 |
GBM | 12.24 | 0.42 | 79.08 | 1.77 | 1.89 | 0.95 | 1.93 | 1.08 | 0.21 | 0.43 |
CTA | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 | 10 |
RF | 26.54 | 2.58 | 26.27 | 10.7 | 10.71 | 2.72 | 1.91 | 7.7 | 5.53 | 5.34 |
SVM | 10.6 | 7.37 | 5.71 | 8.44 | 1.84 | 25.2 | 12.09 | 12.66 | 11.8 | 4.29 |
MAXENT | 40.5 | 5.05 | 5.85 | 5.98 | 3.7 | 20.3 | 2.19 | 2.5 | 3.78 | 10.15 |
ANN | 19.28 | 9.68 | 11.05 | 18.66 | 9.68 | 0 | 12.49 | 9.59 | 9.57 | 0 |
Ensemble | 15.9 | 11.31 | 19.47 | 15.02 | 4.93 | 8 | 5.89 | 7.7 | 4.72 | 7.07 |
All seven models and their ensemble exhibited suitability areas in the eastern coastal region of Japan, mainly in the Kantō, Chūbu, Kinki, Shikoku, Chūgoku, and Kyūshū prefectures (Fig. 1). In Fig. 2A the GLM showed 0.25 suitability for the Hokkaido region; however, in the rest of the models, except for ANN (Fig. 2G) there was no suitability for this region. The GBM (Fig. 2B), ACT (Fig. 2C), RF (Fig. 2D), and MVS (Fig. 2E) models showed zones of suitability ≥ 0.5 for Ibaraki, Chiba, Saitama, Tokyo, Kanagawa, Yamanashi, Shizuoka, Aichi, Mie, Nara, Wakayama prefectures and the prefectures of Shikoku and Kyūshū. The Maxent models maps (Fig. 2F) and the ensemble (Fig. 2H) exhibited similarities on their suitability areas and their ideal areas ≥ 0.5 are lower than the GBM, ACT, RF, ANN, and MVS models. The ANN model (Fig. 2G) was the only one that showed high suitability zones for the western coastal areas of Japan, specifically in Yamagata, Niigata, Toyama, Ishikawa, and Fukui prefectures, the ANN model also showed high suitability in Sapporo.
Figure 2 Distribution models of S. edule in Japan. A: GLM. B: GBM. C: CTA. D: RF. E: SVM. F: Maxent. G: ANN. H: Assembly
At the varietal group level, each one has different environmental preferences. Variables with significant differences were selected, leaving Bio13, Bio3 and Bio4 out of the analysis. Therefore, due to temperature differences in Japan, where the winter is long and low temperatures, as well as very humid summers, therefore the relative weight assigned to minimum temperatures were more relevant than humidity.
The combination of the variables was carried out with a medium risk, giving all theweight of importance to the minimum temperatures and their ranges of variation. Due to the above, a reduction was obtained in the aptitude surfaces with general data. The most adaptive regional trends are southern and southeastern Japan, with persistent warm temperatures in early autumn. The following table shows a hypothetical approximation of the adaptation of the varietal groups, all of them in the southern part of the country.
Table 4 shows the potential area of distribution with adequate characteristics for five varieties of S. edule, considering 19 bioclimatic variables in consensus with González-Santos et al. (2016). The variety with the greatest distribution potential is albus dulcis in the middle and southern part of Japan, followed by nigrum minor, nigrum spinosum, virens levis and nigrum xalapensis with the lowest value (Table 4, Fig. 3).
Table 4
Potential distribution area with suitable agroecological conditions for five varieties of Sechium edule in Japan
var. nigrum xalapensis | var. virens levis | var. nigrum spinosum | var. nigrum minor | var. albus dulcis |
Surface (km2) |
2,225 | 13,959 | 32,681 | 39,820 | 51,737 |
Figure 3 Distribution models of five edible varieties of chayote (Sechium edule) for Japan