Adaptability and Phenotypic Stability of Early-Maturing Cowpea (Vigna unguiculata (L.) Walp.) Lines in the Peruvian Amazon

Abstract

This study evaluated the suitability of three floodplain environments near Iquitos for cowpea cultivar recommendations and estimated the adaptability and phenotypic stability of 12 cowpea lines evaluated in 2004. Climatic conditions between 2004 and 2020–2024 were also compared. Three field trials used a randomized complete block design with 12 lines and two replications, assessing grain yield, number of pods per plant, days to flowering, days to 50% maturity, and days to harvest. Combined Analysis of Variance (ANOVA), Duncan’s means test, Genotype + Genotype Environment Interaction (GGE) biplot, and Additive Main Effects and Multiplicative Interactions Interaction (AMMI) analyses revealed that the three sites are contrasting. The Annicchiarico index, GGE biplot, and AMMI analyses identified line CAR 3010 as having superior adaptability and stability. Paired t-tests and Mann–Kendall analyses showed that climatic conditions in 2020–2024 differed significantly from 2004. Therefore, Muyuy, Rafael Belaunde, and San Miguel are suitable locations for testing advanced cowpea lines prior to cultivar recommendation. Line CAR 3010 is recommended for breeding programs in the Peruvian Amazon in response to ongoing climate change. This research addresses a critical knowledge gap as the first study evaluating cowpea adaptability and stability across multiple floodplain locations in the Peruvian Amazon.

1. Introduction

Cowpea (Vigna unguiculata (L.) Walp.) is a legume of significant socioeconomic and nutritional value in semi-arid tropical regions of Africa, Asia, Latin America, and southern Europe, owing to its low production costs, drought tolerance, adaptability to low-fertility soils, and high protein content [1,2]. From a nutritional perspective, dry seeds at ~12.7% moisture contain ~20.2% protein and ~61.2% carbohydrates, with notable concentrations of K, P, and Mg and appreciable levels of B-complex vitamins [3]. These attributes make cowpea a strategic component of food-security policies in areas where alternative protein sources are limited. Despite its global relevance, especially in Nigeria, Niger, Burkina Faso, Brazil, Ghana, and Mali, significant yield disparities persist between countries. In Peru, for example, mean grain yield remains modest (~1.4 t ha⁻¹) while in Qatar and Egypt, mean grain yields are 13.2 t ha⁻¹ and 3.8 t ha⁻¹, respectively [4].

In the Peruvian Amazon, cowpea is grown primarily on floodplains (known as restingas) during the low-water season (approximately July–December). These landscapes are characterized by periodic flooding during December–July and pronounced edaphic heterogeneity. Such environmental conditions significantly constrain productivity and limit the external validity of varietal recommendations developed for non-floodplain or semi-arid agroecosystems. In Loreto, early-maturing (70–75 days) cowpea cultivars with beige seeds and black eyes are most preferred by farmers and consumers over later-maturing varieties (>90 days) because they enable two harvests within the low-water season. Notable examples include cv. Vaina Blanca and cv. CAU 9. These cultivars are primarily utilized for dry grain production, although they are also consumed as fresh pods when harvested at immature stages.

The limited adaptability of available cultivars becomes evident under these challenging conditions. Breeding lines selected under semi-arid or upland conditions frequently exhibit reduced performance in floodplain environments. Genotype × environment (G × E) interactions are pronounced, reducing yield predictability and phenotypic stability across sites and seasons [5]. Even relatively subtle edaphoclimatic differences can reorder genotype rankings in cowpea [6,7,8,9,10,11,12,13,14,15,16,17]. In Amazonian floodplains, where hydrological cycles are intense and soils are alluvial and heterogeneous, these challenges are amplified, explaining the scarcity of broadly adapted cultivars.

Multi-environment trials (METs) are essential for capturing G × E interactions and identifying genotypes that are both productive and stable in restinga environments [5]. Established analytical approaches such as GGE biplot and AMMI analysis, complemented by classical methods (Eberhart–Russell; Lin–Binns; Annicchiarico index) and recent frameworks (REML/BLUP = Restricted Maximum Likelihood/Best Linear Unbiased Prediction), effectively separate main effects from G × E interactions, enabling the identification of genotypes with broad or specific adaptation [11,12,13,14,15,16,17,18,19,20,21,22,23]. Well-designed METs also guide breeding programs’ resource allocation by helping balance selection for yield with early maturity, particularly valuable in floodplain systems, where restricted planting windows increase the importance of timely harvest before flooding.

Historical datasets from previous breeding programs represent a largely underutilized resource in the Peruvian Amazon. In 2004, the Instituto Nacional de Innovación Agraria (INIA) established trials to evaluate ten early-maturing cowpea lines (CAR series, black-eye type) and two check cultivars across three floodplain sites near Iquitos: Muyuy, Rafael Belaunde, and San Miguel. Although properly designed to assess genotype performance under seasonal flooding conditions, financial limitations prevented data analysis, leaving this dataset unexplored. Retrospective analysis offers strategic value by (i) maximizing return on prior public investment; (ii) establishing comparative baselines for genotype responses under characteristic floodplain hydrology; and (iii) informing parental selection strategies aligned with regional targets [18].

Validation under current climatic conditions is essential. Global temperatures have increased 1.1 °C since 1850–1900 [24]. In cowpea, elevated nighttime temperatures increase respiration and shift energy balance, potentially reducing yields [25], while heat stress affects reproductive processes [26]. Genotypes identified in 2004 must be confirmed under 2020–2024 conditions, when planting windows, rainfall patterns, and flood dynamics may have shifted.

This study aims to (i) determine whether the floodplain sites of Muyuy, Rafael Belaunde, and San Miguel constitute contrasting test environments for evaluating advanced cowpea lines; (ii) assess the adaptability and stability of the twelve early-maturing lines evaluated in 2004 using complementary analytical methods (GGE biplot, AMMI, and Annicchiarico index); and (iii) evaluate the contemporary relevance of the 2004 recommendations under current climatic conditions (2020–2024).

2. Materials and Methods

The experiments were conducted during the 2004 growing season in three floodplain locations in the San Juan Bautista district, Maynas Province, Loreto region, Peru: Muyuy, San Miguel, and San Rafael, located between 6 and 12 km from Iquitos (Figure 1). Meteorological data were obtained from the SENAMHI/San Roque weather station. The study sites experience annual flooding from December to July.

To characterize the edaphic environments, twenty soil cores (0–30 cm depth) were collected and combined to form a single composite sample per environment. Analyses were performed at the INIA-EEA Pucallpa Soil Laboratory (Ucayali, Peru) to determine the following properties: texture (hydrometer method), pH (1:2.5 in water), available phosphorus (modified Olsen method), available potassium (NaHCO₃-EDTA-SUPERFLOC extraction), exchangeable cations (K, Ca, Mg extracted with 1 N KCl), cation exchange capacity (sum of bases plus total acidity, SMP buffer), and organic carbon (Walkley–Black modified method, according to Nelson and Sommers [27]). The results are shown in

Table 1. Physical and chemical characteristics of the soils of the three environments near Iquitos. Loreto, 2004.

Characteristics	Unit	Environments
		San Miguel	Rafael Belaunde	Muyuy
Clay	(%)	34.1	16.1	34.6
Silt	(%)	57.3	23.3	52.6
Sand	(%)	8.6	60.6	12.8
pH H2O		5.7	7.2	5.5
P	mg/kg	18.2	7.5	13.9
H+ + Al3+	cmolc/dm3	0.5	0.3	0.3
K+	cmolc/dm3	0.2	0.1	0.1
Ca2+	cmolc/dm3	9.3	5.1	18.2
Mg2+	cmolc/dm3	2.9	0.5	3.0
ECEC	cmolc/dm3	12.8	6.0	21.6
M	%	3.9	5.0	1.4
Organic Carbon	%	0.9	0.3	0.8
N	%	0.1	0.0	0.1
Bulk density	g/cm3	1.3	1.5	0.0

.

Ten semi-erect, early-maturing lines (CAR series, black-eye type) and two semi-erect, early-maturing control lines of the black-eye type (cv. CAU 9 (TM) and cv. Vaina Blanca (TL)) were evaluated. All materials were obtained from INIA-EEA Vista Florida, Chiclayo, Peru (6°43′38.2″ S, 79°46′52.1″ W) except cv. CAU 9 (TM), which is a local cultivar. The CAR series consists of lines developed through breeding for improved performance in tropical environments.

The experiments were conducted using a randomized complete block design (RCBD) with two replications and twelve lines. Each experimental plot consisted of eight rows, each 9.6 m long, with 0.7 m inter-row spacing and 0.3 m intra-row spacing. Sowing was carried out at all three locations between 25 and 30 May 2004, with three seeds placed per hill at a depth of 3 cm. Two weeks after emergence, plants were manually thinned to two plants per hill.

Before sowing, seeds were treated with Rhyzolex (tolclofos-methyl 75 WP) at 4 g/kg and Orthene (acephate 75 WP) at 5 g/kg to prevent soil-borne pathogens. Additionally, chlorpyrifos 5% (20 mL per 15 L water plus Pegasol^® adjuvant) was applied at 10, 20, and 30 days after sowing as a preventive measure against insect pests. Plots were rainfed throughout the growing season. Manual weed control using hand hoes was performed at 15-day intervals.

Five agro-morphological traits were evaluated:

-

Yield grain (GY): assessed in balance after harvest.

-

Number of pods per plant (NPP): counted on ten randomly selected plants per plot at harvest

-

Days to flowering (DF): days from sowing to the point when 50% of plants had at least one open flower.

-

Days to 50% maturity (D50M): days from sowing until 50% of pods changed color from green to light green.

-

Days to harvest (DH): days from sowing until 95% of pods were fully dry.

Analyses of variance (ANOVA) were performed separately for each environment. The homogeneity of the variances in the three experiments was then assessed according to the criterion proposed by Pimentel-Gomes [28], which states that the ratio of the maximum to the minimum error mean square should be less than seven. Based on this assessment, a combined ANOVA was conducted to evaluate the effects of the interaction between the environment, genotype, and genotype × environment (G × E).

Mean comparisons were carried out using Duncan’s multiple range test at a significance level of p < 0.05. Subsequently, the environmental indices (I_j) and the genotypic confidence indices (W_i) were estimated according to the method of Annicchiarico [27].

The additive linear model used for the combined ANOVA was as follows:

Y_ijk = µ + (B/A) _jk + G_i + A_j + (GA)_ij + e_ijk

(1)

where:

Y_ijk = observation of genotype i in environment j and block k;

µ = overall mean

(B/A)_jk = effect of block k within environment j

G_i = fixed effect of genotype i

A_j = fixed effect of the environment j

(GA)_ij = interaction effect of genotype i and environment j

e_ijk = random error associated with Y_ijk, with normal distribution N(0, σ2)

The environmental index of Annicchiarico [29] was estimated as follows:

I_j = X_j − X₍..₎

(2)

where:

I_j = Index of environment j

X_j = Mean of the environment j

X₍..₎ = Overall mean

The confidence index of Annicchiarico for GY and PNP was estimated as follows:

W_i = Y_(i.) (%) − Z_(1−α) (σ_i)

(3)

where:

W_i = Genotypic confidence index of genotype i

Y_(i.) (%) = Relative mean of genotype i across the evaluated environments

Z_(1−α) = Quantile of the Z distribution at α = 0.95

σ_i = Relative standard deviation of genotype i across the evaluated environments

As early-maturing line recommendation is important in plant breeding, the confidence indices for DF, DPM, and DH were estimated as follows:

W_i = Y_(i.) (%) + Z_(1−α) (σ_i)

(4)

Additionally, to evaluate the environments and lines’ adaptability and stability for GY, both the GGE Biplot and AMMI methods were employed [5,30]. The GGE biplot analysis was performed using a matrix of genotype and environment means, with each value adjusted by subtracting the corresponding environment mean (G_i + (G × E)_ij = Y_ij − E_j). For the AMMI analysis, each value of the genotype-by-environment mean matrix was adjusted according to the following model: G_i × E_j = Y_ij − G_i − E_j + overall mean. Both matrices were subjected to principal component analysis (PCA) based on the covariance matrix.

ANOVA and Duncan’s multiple range test were performed using SAS 9.4 (PROC GLM; SAS Institute Inc., Cary, NC, USA). Annicchiarico analysis was performed in GENES [31]. PCA for the GGE biplot and AMMI was performed in JMP 14 (SAS Institute Inc., Cary, NC, USA) and the Metan package of R software version 4.5.1 [32].

To assess whether environmental conditions in 2004 were comparable to those in 2020–2024, we retrieved daily meteorological data (June to August) from the San Roque SENAMHI weather station (Iquitos, Peru). Paired t-tests compared 2004 versus 2020–2024 for daily mean, minimum, and maximum temperature; sunshine hours; precipitation; and relative humidity, using Statistica 64 (StatSoft, Inc., Tulsa, OK, USA).

Additionally, the Mann–Kendall test was applied to time series data (June to August) spanning 2004–2024 to assess long-term climate trends. Missing values in the climate time series were imputed using linear interpolation. Prior to trend analysis, serial autocorrelation was evaluated using the Ljung–Box test to verify the independence assumption required for standard trend detection methods. Where autocorrelation was detected, pre-whitening was implemented following the methodology of Zhang [33], utilizing the ‘zyp’ package [34] in R statistical software [35]. The modified Mann–Kendall test was subsequently applied to detect monotonic trends while accounting for serial dependence in the dataset. These statistical analyses were conducted using the ‘trend’ package [36] in the R statistical environment.

3. Results

3.1. Environmental Effects

In this study, three floodplain environments near Iquitos were analyzed. ANOVA results indicated that the environment significantly affected all traits (

Table 2. ANOVA for agronomic characteristics of cowpea (V. unguiculata) lines in three environments of the Peruvian Amazon: Muyuy, San Rafael, and San Miguel. Loreto, 2004.

		Mean Square
FV	GL	Yield Grain (kg/ha)	Number of Pods per Plant	Days to Flowering	Days to 50% Maturity	Days to Harvest
Environment (E)	2	1,564,786.7 **	1109.9 **	1.7 **	30.5 **	27.7 **
Block/E	3	24,993.0 ns	73.9 *	1.4 *	4.0 *	18.0 **
Genotype (G)	11	8971.4 ns	34.0 ns	9.9 **	25.0 **	10.1 **
G × E	22	8490.9 ns	15.2 ns	0.8 *	3.6 **	2.2 *
Error	33	22,705.7	17.7	0.4	1.1	1.1
Total	71
Mean		815.38	33.25	43.19	62.32	69.47
CV (%)		18.82	12.66	1.39	1.66	1.50

Legend: *, **, and ns indicate statistical significance at p ≤ 0.05 and p ≤ 0.01, and no significant difference, respectively.

). According to Duncan’s means test, Muyuy presented the highest GY, with reductions of 14% in San Miguel and 48% in Rafael Belaunde compared to Muyuy (

Table 3. Duncan’s multiple range test for three environments near Iquitos where cowpea (V. unguiculata) was grown. Loreto, 2004.

Environment	Yield Grain (kg/ha)	Number of Pods per Plant	Days to Flowering	Days to 50% Maturity	Days to Harvest
Muyuy	1027.8 a	40.2 a	43.5 a	63.3 a	70.7 a
Rafael Belaunde	532.1 c	26.6 c	43.1 b	62.6 b	69.2 b
San Miguel	886.3 b	33.0 b	43.0 b	61.1 c	68.6 b

Legend: Within each column, means followed by different letters differ significantly at p ≤ 0.05.

). Similarly, the NPP was highest in Muyuy, but was 18% lower in San Miguel and 34% lower in Rafael. Belaunde. The GGE biplot and AMMI biplot analysis for GY showed that Muyuy, San Miguel, and Rafael Belaunde are contrasting environments (Figure 2). These results suggest that the three environments provide different conditions and should all be included in cowpea grain yield trials.

For DF, DPM, and DH, the mean comparisons (Table 3) also showed higher values in Muyuy. Although the differences among locations were statistically significant (p < 0.05), the maximum difference between sites for these traits was only two days, indicating that these lines exhibit early maturity characteristics.

On the other hand, analyses according to Annicchiarico [29] showed that for GY and NPP, Muyuy was identified as the favorable environment, while Rafael Belaunde was unfavorable (

Table 4. Environmental indices (I_j) for three sites near Iquitos, where 12 cowpea (V. unguiculata) lines were cultivated.

Site	Mean	Ij	Class
Grain yield—GY (kg ha−1)
San Miguel	886.25	70.87	Favorable
Rafael Belaunde	532.08	−283.29	Unfavorable
Muyuy	1027.79	212.41	Favorable
Number of pods per plant—NPP
San Miguel	32.96	−0.29	Unfavorable
Rafael Belaunde	26.60	−6.65	Unfavorable
Muyuy	40.19	6.94	Favorable
Days to flowering—DF †
San Miguel	43.00	−0.19	Favorable
Rafael Belaunde	43.08	−0.11	Favorable
Muyuy	43.50	0.31	Unfavorable
Days to 50% maturity—D50M †
San Miguel	61.08	−1.24	Favorable
Rafael Belaunde	62.58	0.26	Unfavorable
Muyuy	63.29	0.97	Unfavorable
Days to harvest—DH †
San Miguel	68.58	−0.89	Favorable
Rafael Belaunde	69.17	−0.31	Favorable
Muyuy	70.67	1.19	Unfavorable

^† Consider early development as a favorable environment.

). However, for DF, DPM, and DH, San Miguel was identified as the favorable environment for earliness, whereas Muyuy was the least favorable.

3.2. Adaptability and Stability Analyses

The G × E interaction was not significant for GY and NPP, but was significant for DF, DPM, and DH (Table 2). This indicates that the G × E interaction did not affect GY, and selection can therefore be based on means of the the genotype. However, DF, DPM, and DH require further analysis of stability and adaptability.

Duncan’s mean test did not detect significant differences for GY and NPP, but it detected significant differences for DF, DPM and DH (

Table 5. Duncan’s means tests (p < 0.05) of cowpea (V. unguiculata) lines evaluated in three environments near Iquitos, Peru.

Lines	Yield Grain (kg ha−1) †	Number of Pods per Plant	Days to Flowering	Days to 50% Maturity	Days to Harvest
CAR 3002	780.6	35.2	42.5 de	62.3 c	69.8 a
CAR 3003	753.4	36.2	40.7 f	60.5 e	68.0 b
CAR 3004	778.6	34.0	42.2 e	61.0 de	70.3 a
CAR 3005	825.5	35.3	45.0 a	65.0 ab	70.0 a
CAR 3006	799.0	31.9	42.0 e	64.3 b	68.0 b
CAR 3009	845.7	31.3	45.0 a	62.0 cd	70.3 a
CAR 3010	897.5	34.2	43.3 c	61.0 de	70.3 a
CAR 3013	852.7	33.2	44.2 b	64.7 ab	70.2 a
CAR 3014	799.2	27.4	44.3 ab	65.7 a	70.2 a
CAR 3015	804.0	31.9	43.0 cd	60.0 e	70.2 a
cv. CAU 9 (TM)	817.6	34.9	43.2 cd	60.8 de	66.3 c
cv. Vaina Blanca (TL)	830.4	33.5	43.0 cd	60.5 e	70.0 a

^† Values within a column followed by different letters differ significantly.

). The lines with the highest GY values were CAR 3010, CAR 3013, and CAR 3009 (Table 5).

However, for GY, Annicchiarico confidence index (Wi), GGE biplot, and AMMI analysis, all identified line CAR 3010 as having high adaptability and stability. The confidence index W_i was 103.54% (

Table 6. The confidence index (W_i) estimated by the method of Annicchiarico [29] for grain yield (GY), number of pods per plant (PNP), days to flowering (DF), days to 50% maturity (D50M), and days to harvest (DH) in cowpea (V. unguiculata).

		Wi (%) †
N°	Genotype	GY	NPP	DF *	D50M *	DH *
1	CAR 3002	81.16	100.56	100.69	102.57	104.66
2	CAR 3003	90.61	84.57	97.58	99.03	99.70
3	CAR 3004	67.98	86.25	101.41	100.16	102.57
4	CAR 3005	81.59	98.41	105.23	107.45	103.33
5	CAR 3006	90.68	87.96	98.23	105.68	100.27
6	CAR 3009	95.71	88.22	105.23	99.03	102.57
7	CAR 3010	103.54	101.86	101.52	102.49	102.57
8	CAR 3013	89.40	91.62	106.02	100.94	103.83
9	CAR 3014	85.37	53.19	106.46	111.59	102.92
10	CAR 3015	79.55	86.97	100.57	107.80	102.92
11	cv. CAU 9 (TM)	92.80	100.80	100.12	99.18	98.63
12	cv. Vaina Blanca (TL)	95.82	91.61	100.69	100.14	103.33

^† alpha = 0.05, z = 1.645. * W_i was estimated by adapting the Annicchiarico index as described in Formula (4). Under this adaptation, lines with W_i values less than 100% were considered suitable for earliness.

), the only value exceeding 100%. Similarly, the GGE biplot positioned CAR 3010 close to the abscissa of the Average Environment Axis (AEA) (Figure 2A, blue lines) and near the ideal genotype (Figure 2A, red circles), indicating strong adaptability and stability. The AMMI analysis further supported these findings, showing that CAR 3010 had a principal component 1 (PC1) score close to zero and high GY (Figure 3), confirming the previous results. Together, these results demonstrate the wide adaptability and phenotypic stability of CAR 3010 across diverse environments. Therefore, CAR 3010 could be recommended for cultivation in the floodplains of the Peruvian Amazon, provided that the current climate remains comparable to that of 2004.

For NPP, the lines CAR 3010, cv. CAU 9 (TM), and CAR 3002 had W_i values greater than 100% (Table 6, W_i estimated by Formula (3)). Once again, line CAR 3010 exhibited the highest index (W_i = 101.86%), indicating that this line has adaptability and stability for this trait.

When selecting early-maturing lines for recommendation, Wi values below 100% are preferred for phenological traits (DF, DPM, and DH), as lower Wi values indicate shorter cycles with greater stability across environments (Table 6, W_i estimated by Formula (4)). For DF, the lines with W_i < 100% were CAR 3003 and CAR 3006; for DPM, CAR 3003, CAR 3009, and cv. CAU 9 (TM); and for DH, CAR 3003 and cv. CAU 9 (TM) (Table 6). Thus, CAR 3003 uniquely exhibited Wi < 100% for all three traits, demonstrating superior stable earliness.

However, although CAR 3010 exhibited high adaptability and stability for GY and NPP, it had Wi values greater than 100% for DF, DPM, and DH, indicating a lack of adaptability and stability for earliness.

3.3. Comparative Analysis of Climate Data and Trends

Paired t-tests were performed to compare daily climatic variables between identical calendar dates from June to August in 2004 and the 2020–2024 period. The analyses revealed significant differences (p < 0.01) for mean, maximum, and minimum temperatures, precipitation, and relative humidity (

Table 7. Paired t-test for daily climatic variables between June and August 2004 and the period 2020–2024 in Iquitos, Peru. Source: SENAMHI San Roque.

Variable	n	Mean (2004)	Mean (2020–2024)	Difference	t	p-Value
Mean Temperature (°C)	90	26.39	26.96	0.57	−3.42	0.0009
Max. Temperature (°C)	90	30.96	32.18	1.21	−4.98	<0.0001
Min. Temperature (°C)	90	21.96	22.71	0.75	−6.66	<0.0001
Sunshine Hours	92	3.72	3.96	0.25	−0.73	0.4699
Precipitation (mm)	92	3.12	5.38	2.26	−2.71	0.0081
Relative Humidity (%)	90	93.63	87.03	−6.60	20.85	<0.0001

).

Among the variables analyzed, only the daily sunshine hours did not differ significantly between 2004 and 2020–2024. This indicates that the observed temperature increase is not attributable to changes in daily sunshine hours. Therefore, these results across six climatic variables demonstrate that the recent climate is warmer and drier, yet with more precipitation.

The Mann–Kendall test was applied to assess trends in six climatic variables over the 2004–2024 period (

Table 8. Mann–Kendall trend test results for daily climatic variables during the low water season (June-August) from 2004–2024 in Iquitos, Peru. Source: SENAMHI San Roque.

Variable	Interpolated Data	Effective Number	S	τ	p-Value	Trend per Year
Mean Temperature (°C)	1	21.0	10	0.048	0.7858	0.003
Max. Temperature (°C)	0	21.0	30	0.143	0.3812	0.028
Min. Temperature (°C)	0	10.4	106	0.505	0.0001	0.055
Sunshine Hours	8	21.0	−118	−0.562	0.0004	−0.045
Precipitation (mm)	0	21.0	−2	−0.010	0.9759	−0.001
Relative Humidity (%)	1	21.0	32	0.152	0.3492	0.069

). The results revealed a statistically significant increasing trend in minimum temperature at a rate of 0.055 °C year⁻¹. Conversely, daily sunshine hours exhibited a significant declining trend of 0.045 h year⁻¹. These findings provide evidence of climatic shifts in the Peruvian Amazon region.

4. Discussion

This study represents the first evaluation of cowpea line adaptability and phenotypic stability in the floodplain environments of the Peruvian Amazon. Although INIA conducted cowpea research between 2000 and 2007, the breeding program was subsequently discontinued. With the reactivation of this program in 2025, new strategies must be adopted to address contemporary agricultural challenges in the region.

The results demonstrated that the three floodplain environments near Iquitos represented contrasting test locations for the five traits evaluated. Line CAR 3010 exhibited superior adaptability and stability for grain yield (GY) based on the Annicchiarico index, GGE biplot, and AMMI analysis. Notably, climatic conditions near Iquitos in 2020–2024 differed significantly from those in 2004, with Mann–Kendall trend analysis revealing increasing minimum temperatures and decreasing sunshine hours over the two-decade period. These climatic shifts underscore the importance of re-evaluating historical variety recommendations under current environmental conditions.

4.1. Importance of Muyuy, Rafael Belaunde, and San Miguel

Evaluating genotypes in diverse and representative environments is critical for identifying broadly adapted lines [5]. In this study, environmental effects significantly influenced all five traits, indicating substantial heterogeneity among the environments near Iquitos. Although the three soils are classified as floodplain and are separated by up to 16.5 km, they differ in their physical and chemical properties (Table 1).

Several statistical approaches, including ANOVA, Duncan’s means tests, GGE biplot, and AMMI biplot, showed marked environmental differences among the three sites, with Muyuy and Rafael Belaunde identified as the most contrasting environments. These variations are primarily attributed to the different edaphic properties described in Table 1. Muyuy exhibited a higher silt fraction (52.6%) and lower sand content (12.8%), which contributed to better water retention. Furthermore, Muyuy presented the highest Ca²⁺ (18.2 cmolc/dm³) and Mg²⁺ (3.0 cmolc/dm³) concentrations, indicating greater base saturation and nutrient availability. Organic carbon (0.8%) and total ECEC (21.6 cmolc/dm³) were also higher in Muyuy, enhancing soil fertility. These results corroborate previous studies indicating that elevated levels of exchangeable Ca²⁺, Mg²⁺, and organic matter are essential for optimizing legume productivity in acidic tropical soils [37].

San Miguel and Rafael Belaunde showed marked differences in soil properties that influenced cowpea performance (Table 1). San Miguel’s silty clay texture, higher fertility, and greater organic carbon and nutrient content supported intermediate yields. In contrast, Rafael Belaunde’s sandy loam texture, low nutrient reserves, and minimal organic matter limited water and nutrient retention, resulting in the lowest productivity. Additionally, at Rafael Belaunde, a soil pH of 7.2 may have reduced the availability of micronutrients, including B, Cu, Fe, Mn, and Zn [38]. Collectively, the contrasting soil characteristics and productivity levels observed across Muyuy, Rafael Belaunde, and San Miguel confirm that these sites constitute suitable test environments for conducting adaptability and stability trials in cowpea and other short-cycle crops.

4.2. G × E Interaction and Adaptability and Stability Analysis

The G × E interaction was significant for DF, DPM, and DH, which confirms the environmental heterogeneity of the region, as observed in other tropical settings [11,14,39]. In particular, phenological traits were more sensitive to interaction, while GY and NPP were relatively stable, a pattern consistent with earlier findings in cowpea [40].

The Annicchiarico confidence index (W_i), GGE biplot, and AMMI were powerful tools for identifying adaptability and stability of the cowpea lines. The CAR 3010 line exhibited the highest W_i for both GY (897.5 kg/ha) and NPP (34.2), indicating wide adaptability in all environments. This aligns with previous studies that emphasize the importance of environmental factors in influencing cowpea yield and the need for robust statistical methods to account for genotype-environment interactions [11,14,20]

In contrast, Wi values below 100% for DF, DPM, and DH indicate lower relative means and standard deviations, which are desirable when selecting early-maturing lines. In this regard, lines CAR 3003 and cv. CAU 9 (TM) were noteworthy. CAR 3003 showed Wi values below 100% for all three phenological traits, while cv. CAU 9 (TM) had the lowest W_i for DH, suggesting that this local cultivar maintains a consistently short cycle (66 days to harvest).

4.3. Climate Change in the Amazon Region

Climatic comparisons between 2004 and 2020–2024 revealed significant warming trends, increased precipitation, and reductions in relative humidity (Table 7). These changes are consistent with IPCC projections [24] and highlight the urgent need to develop climate-resilient cowpea cultivars [9] for environments characterized by higher temperatures (mean = 27 °C, min = 22.7 °C, max = 31.2 °C), greater precipitation (5.38 mm day⁻¹), and lower relative humidity (87%). Consequently, the GY of CAR 3010 under these new environmental conditions remains uncertain.

Additional multi-year trials conducted with CAR 3010 and three other lines from June to August 2005–2007 in San Miguel and Muyuy demonstrated superior GY for CAR 3010, ranging from 1089 to 1263 kg ha⁻¹ (personal communication, Rodrigo Gonzales). However, the climate during 2005–2007 also differed significantly from that observed in 2020–2024 (

Table 9. Paired t-test for daily climatic variables between June and August 2005–2007 and 2020–2024 in Iquitos, Peru. Source: SENAMHI San Roque.

Variable	n	Mean (2005–2007)	Mean (2020–2024)	Difference	t	p-Value
Mean Temperature (°C)	92	27.07	26.96	−0.11	0.98	0.3277
Max. Temperature (°C)	92	32.04	32.15	0.12	−0.88	0.3817
Min. Temperature (°C)	92	22.01	22.72	0.71	−8.73	<0.0001
Sunshine Hours	92	5.38	3.96	−1.41	5.80	<0.0001
Precipitation (mm)	92	4.26	5.38	1.12	−1.44	0.1538
Relative Humidity (%)	92	87.23	87.04	−0.19	0.66	0.5131

). In particular, the minimum temperature during 2020–2024 increased by 0.7 °C, while the sunshine hours decreased by 1.4 h compared to previous years. Similarly, the Mann–Kendall trend analysis revealed increasing minimum temperatures at 0.55 °C decade⁻¹ and declining sunshine hours at 0.45 h decade⁻¹ over the 2004–2024 period.

These climatic shifts have important implications for cowpea production, as elevated minimum temperatures, indicative of higher night temperatures, have been shown to accelerate respiration, increase energy expenditure, and ultimately reduce GY in cowpea [25]. Furthermore, day/night temperature regimes of 33/30 °C induced male sterility in cowpea [26], suggesting that the observed warming trend may increasingly challenge reproductive success in this crop.

Therefore, CAR 3010 should be incorporated into genetic improvement programs in the Peruvian Amazon in all three locations and evaluated for at least two consecutive years in the context of ongoing climate change.

4.4. Limitations and Future Research

In this study, Duncan’s mean test did not detect significant differences in GY or NPP among the lines. These results could potentially be improved by increasing the number of blocks and reducing the size of the experimental plot. In the present experiment, each plot consisted of 256 plants (spacing 0.7 × 0.3 m) with two blocks and produced a coefficient of variation (CV) of 18.8%, indicating high experimental accuracy. In similar studies with cowpea’s pure lines, however, it has been observed that plots with 16 plants (spacing 0.5 × 1 m) and four blocks resulted in a CV of 31% [18], or that plots with 10 plants (spacing 0.7 × 0.4 m) and three blocks produced a CV of 33% [13]. These observations suggest that experimental plot size can be reduced to ≥16 plants and the number of blocks increased to ≥4 to allow the evaluation of a greater number of genotypes within the same total area.

In addition, alternative methodologies can be utilized to assess adaptability and stability. Specifically, this study found that the Annicchiarico method, the GGE biplot, and AMMI effectively differentiated CAR 3010, likely due to their efficient use of genotype means across environments and consideration of genotype variation within the context of all environments. More research is needed to evaluate the effectiveness of these and other approaches in the Peruvian Amazon in different years [15,20].

Furthermore, it is recommended that CAR 3010 be evaluated for additional traits, including grain quality and disease resistance, in order to gain a more comprehensive perspective of its potential for local breeding programs.

5. Conclusions

Breeding strategies for cowpea and other short-cycle crops in the floodplains of the Peruvian Amazon should include a core testing network comprising Muyuy, Rafael Belaunde, and San Miguel, as these sites represent contrasting edaphic conditions.

Line CAR 3010 is recommended for inclusion in cowpea breeding programs in the Peruvian Amazon due to its superior adaptability and stability for grain yield across contrasting floodplain environments. As climatic conditions in the region continue to shift toward warmer nights and reduced sunshine hours, sustained investment in cowpea breeding programs is critical for developing climate-adapted varieties and ensuring long-term food security.