Multi-trait stability selection drives genetic gains in cowpea [Vigna unguiculata (L.) Walp.] under high-temperature stress

Background

Cowpea (Vigna unguiculata (L.) Walp.) is a crucial food and fodder crop in regions characterised by high temperatures. However, its reproductive stages are particularly sensitive to heat stress, leading to reduced pollen viability, pod set, and yield. Therefore, developing and deploying high-temperature-tolerant cultivars is crucial for maintaining productivity under changing climatic conditions.

Methods

A total of 250 diverse cowpea accessions, along with five checks, were evaluated at Baramati and Jodhpur during the 2022–2023 growing season under both normal and late sowing conditions to simulate high-temperature stress. Growth, yield, and physiological traits, including canopy temperature, were measured. Multi-trait stability indices, including the multi-trait genotype-ideotype distance index (MGIDI), the multi-trait stability index (MTSI), the multi-trait mean performance and stability index (MTMPS), Mean Performance and Stability (MPS) analysis and the factor analysis-based BLUP index (FAI-BLUP), were utilised to identify superior genotypes. Further, genetic parameters were also estimated using mixed linear models.

Results

Late sowing exposed genotypes to temperatures exceeding 35 °C during flowering, inducing severe reproductive stress; only 147 of 250 accessions successfully flowered and set seeds. Traits such as seed yield, pods per plant, and seeds per pod showed high heritability (≥ 60%) and considerable genetic variation, while canopy temperature exhibited lower variability. Grain yield was positively correlated with pods per plant (r = 0.40) and grains per pod (r = 0.36), and negatively correlated with canopy temperature (r = − 0.35). Multi-trait indices identified 29 superior genotypes, with six [G044 (IC488270), G078 (EC240920), G080 (IC488085), G112 (EC240868), G114 (EC243927), G142 (IC402161)] consistently selected across all models. Predicted selection gains were high for grain yield (+ 18.7%), pods per plant (+ 11.2%), and grains per pod (+ 8.5%). Several genotypes [G066 (EC240878), G080 (IC488085), G112 (EC240868)] demonstrated both high yield and stability across environments.

Conclusions

This study reveals substantial genetic variation for high-temperature tolerance in cowpea, highlighting the value of integrating multi-environment phenotyping with multi-trait stability indices for robust selection. The identified genotypes provide immediate candidates for cultivar release and represent valuable genetic resources for breeding programs aiming to develop heat-resilient cowpea varieties.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-026-08597-5.

Highlights

Screened 250 cowpea accessions; 147 reproduced under > 35 °C field heat stress.

Grain yield showed high heritability (71.4%) and a wide range (1.0–52.5 g/plant).

Multi-trait indices identified 29 superior heat-resilient genotypes.

Six genotypes [G044 (IC488270), G078 (EC240920), G080 (IC488085), G112 (EC240868), G114 (EC243927), G142 (IC402161)] were selected across all models owing to multi-trait superiority (high grain yield, 3-4 °C lower canopy temperature under stress).

Predicted selection gains: yield + 18.7%, pods + 11.2%, grains per pod + 8.5%.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-026-08597-5.

Introduction

Cowpea (Vigna unguiculata (L.) Walp.), a C3 legume, plays a critical role in food and nutritional security, particularly in sub-Saharan Africa and parts of Asia. It is valued for its drought resilience, high protein content, and ability to thrive in marginal environments where other legumes struggle to perform [1, 2]. However, cowpea is highly susceptible to heat stress, particularly during reproductive development, resulting in reduced seed set, pod abortion, and yield loss [3, 4].

High temperatures during reproductive stages adversely affect physiological and reproductive functions, including pollen viability, anther dehiscence, and pod formation [5, 6]. Even a 1 °C rise in night temperatures above 16.5 °C can lead to a 13.6% reduction in yield [7–9]. Night temperatures exceeding 20 °C drastically reduce pollen viability, while day temperatures above 35 °C accelerate leaf senescence, impair photosynthesis, and hinder grain filling [10]. Moreover, genotype-specific responses to thermal stress have been reported, with yield losses ranging from 4% to 14% per 1 °C increase in night-time temperature beyond critical thresholds [11, 12].

Breeding heat-tolerant cowpea cultivars has been a long-standing objective. Although a few genotypes, such as BR 17-Gurguéia, have shown tolerance at temperatures up to 38 °C [6, 13], progress remains limited. This is partly due to reliance on controlled-environment screenings, which often fail to replicate the complex interactions and stress dynamics observed under field conditions. Additionally, inconsistent and imprecise phenotyping approaches have hindered the accurate identification of traits associated with heat tolerance [2]. Furthermore, most previous studies have been conducted in greenhouses or controlled environments, which fail to capture genotype × environment interactions (GEI) adequately. Consequently, there is a critical need for robust, field-based evaluations that integrate physiological, biochemical, and agronomic traits to identify high-performing, stable cultivars in multi-environment trials that experience variable temperatures during the cropping season.

Multi-environment trials (METs) are essential for evaluating genotypic performance across variable temperature regimes and quantifying GEI effects [14, 15]. Understanding GEI is crucial for improving selection accuracy and achieving greater genetic gain in breeding programs [16]. The stability of genotypic performance across environments also enhances adaptability and broadens the target production area.

Various statistical tools have been used to analyse GEI and stability, including ANOVA, regression-based models [17, 18], and multivariate methods such as principal component analysis (PCA), additive main effects and multiplicative interaction (AMMI), and genotype plus genotype-by-environment (GGE) biplots [19, 20]. However, these methods typically emphasise single-trait analysis, often focusing on grain yield while neglecting other correlated traits that influence overall performance. On the other hand, multi-trait selection approaches offer a more comprehensive assessment by incorporating both primary and secondary traits without introducing multicollinearity. Traditional indices such as the Smith-Hazel index [21, 22] are limited in METs due to multicollinearity and biased estimates. Recently developed multi-trait stability indices, including the multi-trait genotype-ideotype distance index (MGIDI), multi-trait stability index (MTSI), multi-trait mean performance and stability index (MTMPS), and factor analysis-based BLUP index (FAI-BLUP), address these limitations and allow for the identification of genotypes with both high yield and stable performance [23–26].

The yield potential of cowpea genotypes under high-temperature conditions is known to be influenced by several secondary traits. Hence, we hypothesise that “selecting thermostable genotypes using multiple indices improves the genetic gain with stable genotypes suitable for high temperature conditions”. The specific objectives of this study were to: (i) assess phenotypic variability, heritability, and inter-trait associations under high temperature stress; (ii) quantify the effects of GEI on key agronomic and physiological traits; and (iii) compare the efficacy of four multi-trait stability indices (MGIDI, MTSI, MTMPS, and FAI-BLUP) in identifying stable and high-performing genotypes.

Materials and methods

Site description

Screening trials were conducted at two locations representing distinct agro-climatic zones: the ICAR–National Institute of Abiotic Stress Management (NIASM), Baramati, Maharashtra, and the ICAR–National Bureau of Plant Genetic Resources, Regional Station (NBPGR, RS), Jodhpur, Rajasthan. Detailed geographical, climatic, and soil characteristics of these sites are provided in Table 1.

Table 1.

Details of the experimental site utilised for screening and identification of high-temperature-tolerant cowpea genotypes

Environments	Year	Sowing Conditions	Test location	Soil properties
ENV1	2022	Normal sown	NIASM, Baramati (18.1592° N, 74.5007° E)	Soil Texture: sandy clay, pH: slightly alkaline pH of 8.0, EC: 0.26 dS m⁻¹, Soil NPK status: 176, 25, and 148 kg ha⁻¹
ENV2	2022	Late sown
ENV3	2023	Normal sown
ENV4	2023	Late sown
ENV5	2023	Normal sown	NBPGR, RS, Jodhpur (26°14′45.823 N, 73°0′1.223″ E)	Soil Texture: sandy loam, pH: 7.95, EC: 0.24 dS m⁻¹, Soil NPK: 120, 47, and 573.75 kg ha⁻¹
ENV6	2023	Late sown

Plant materials

A total of 250 cowpea accessions were selected from the core collection maintained at the National Genebank, NBPGR, New Delhi. These comprised both indigenous (IC) and exotic (EC) germplasm, representing a wide range of genetic and geographical diversity. Additionally, five check varieties, viz., DC15, Pant Lobia 3, Pant Lobia 4, GC3, and RC101, were included. Notably, Pant Lobia 3 and Pant Lobia 4 are recommended for summer cultivation due to their known tolerance to high-temperature stress. Detailed accession information is provided in Supplementary Table 1.

Experimental design and sowing

The screening trials were conducted during the 2022 and 2023 cropping seasons. At NIASM, Baramati (hereafter BMT), trials were carried out in both years (2022 and 2023), whereas at NBPGR, RS, Jodhpur (hereafter JDP), screening was performed only in 2023. An alpha lattice design with two replications was used at both locations. Each germplasm accession was manually sown in a single row of 2 m in length, with a row-to-row spacing of 45 cm and a plant-to-plant spacing of 15 cm. Standard agronomic practices were followed to ensure optimal crop growth, and need-based irrigation was provided to mitigate any moisture stress. To expose plants to different temperature regimes during reproductive development, two sowing dates were adopted at both locations: normal sowing (NS) during the first week of February and late sowing (LS) during the first week of March, approximately 30 days after NS. The LS treatment was specifically designed to impose high-temperature stress during the reproductive phase of development, enabling the identification of thermotolerant genotypes.

Trait measurements

Growth and morphological traits such as plant height (PH; cm), number of primary branches per plant (NBP), number of secondary branches per plant (NSB), pods per plant (PP), pod length (PL; cm), seeds per pod (GPP), and grain yield per plant (GY; g) were measured in five randomly selected plants in each genotype from each replication and both NS and LS separately and then mean data of NS and LS was used for data analysis.

Canopy temperature was measured using an Infrared thermal Camera (Vario CAM hr. inspect 575, Jenoptic, Germany) with a spatial resolution of 760 × 570 pixels at 30 days after sowing (pre-flowering; CT1) and 60 days after sowing (post-flowering; CT2).

Weather parameters

Daily maximum (Tmax) and minimum temperatures (Tmin) were recorded at both experimental locations to characterise the environments during the crop growth period. Weather parameters were monitored from February 1st to June 30th for each year, corresponding to the sowing and harvesting periods under both NS and LS conditions. The data were compiled according to standard meteorological weeks and are presented in Supplementary Fig. 1.

Data analysis

Variance component analysis

Genotypic performance across environments was evaluated using a linear mixed-effects model implemented in the R software (version4.51.1) using metan package. The mixed-effects ANOVA model used in this study was expressed as:

where:

Yijk = observed trait value of the ith genotype in the jth environment and the kth replication.

µ = overall mean.

Ej = effect of the jth environment.

Rk(Ej) = effect of the kth replication nested within the jth environment.

Gi = effect of the ith genotype.

(GE)ij = genotype × environment interaction effectεijk = residual error associated with the observation.

In this model, environment (E) and replication within environment [R(E)] were treated as fixed effects, whereas genotype (G) and genotype × environment interaction (GE) were treated as random effects.

Variance components were estimated using the restricted maximum likelihood (REML) method, and best linear unbiased predictions (BLUPs) of genotypic effects were obtained by solving the mixed model equations:

where:

y = vector of phenotypic observations.

β = vector of fixed effects (environment and replication within environment).

υ = vector of random effects (genotype and genotype × environment interaction).

X = design matrix relating observations to fixed effects.

Z = design matrix relating observations to random effects.

ε = vector of residual errors.

Residuals were assumed to be independently and normally distributed with mean zero and variance σ²ε. Analyses followed the framework described by [27].

Genetic parameters in the multi-environment trial (MET) were estimated following the procedures described by [27, 28]. Broad-sense heritability on a plot basis (h²_g) was estimated as:

The symbols c represent genotypic, genotype-by-environment interaction, and residual variances, respectively.

Broad-sense heritability of genotype means across environments (h²_gm) was calculated as

Where 𝑒 is the number of environments and b is the number of blocks per environment.

Accuracy of selection (AC) was estimated as

The proportion of phenotypic variance explained by GEI(Ir²) was calculated as

And the genotype–environment variance ratio (r_ge) was estimated as

Coefficients of variation (CVg and CVr)

The genotypic (CVg) and residual (CVr) coefficients of variation were computed as

Where µ is the grand mean.

Multi-trait selection and performance–stability indices

Multi-trait-based indices were employed to identify superior genotypes by simultaneously considering multiple agronomic and physiological traits. These indices were used to support selection for high mean performance and, stability across environments.

Multi-trait Genotype–Ideotype Distance Index (MGIDI)

The multi-trait genotype–ideotype distance index (MGIDI) was computed following the procedure described by [26]. MGIDI is a multi-trait selection index based on BLUP-derived mean performance and is primarily designed to identify genotypes with desirable overall trait profiles. Although MGIDI does not explicitly quantify phenotypic stability, it can indirectly support stability-oriented selection when traits associated with adaptability and consistent performance across environments are included. In this study, MGIDI was therefore used as a complementary tool to identify genotypes combining favourable multi-trait performance across environments, rather than as a direct stability statistic. Trait values were first rescaled to a 0–100 range to ensure comparability, with 100 assigned to the most desirable value and 0 to the least desirable value. For traits in which higher values were favourable (plant height, number of primary and secondary branches, pods per plant, seeds per pod and grain yield), the maximum value was set to 100. For traits in which lower values were favourable (canopy temperature), the minimum value was set to 100. The rescaled data were subjected to factor analysis with varimax rotation to group correlated traits into latent factors and reduce dimensionality. An ideotype was defined as a hypothetical genotype with a score of 100 for all rescaled traits, representing the breeding target. The MGIDI for each genotype was calculated as the Euclidean distance between its factor scores and those of the ideotype, with lower MGIDI values indicating genotypes closer to the desired ideotype.

Where MGIDIi is the multi-trait genotype-ideotype distance index for the ith genotype; Yij is the score of the ith genotype in the jth factor (i = 1, 2, …, g; j = 1, 2, …, f), being g and f the number of genotypes and factors, respectively, and Yj is the jth score of the ideotype. The genotype with the lowest MGIDI is then closer to the ideotype and should therefore exhibit the desired values for all the analysed traits.

To assess the strengths and weaknesses of each genotype, the contribution of each factor to the MGIDI was calculated as:

Here, D_iⱼ represents the distance between the ith genotype and the ideotype in relation to the jth factor. Higher values of indicate greater contribution of the corresponding factor to the overall MGIDI, reflecting relative weakness, whereas lower values indicate strength for the associated traits.

Multi-trait index based on factor analysis and genotype–ideotype distance (FAI-BLUP)

After defining the ideotype, the spatial probability of each genotype was determined based on its distance from the ideotype. The probability that the ith genotype resembles the ideotype was calculated as:

Where 𝑃𝑖 is the probability associated with the ith genotype, 𝑑𝑖 is the standardised Euclidean distance between the ith genotype and the ideotype, and n is the total number of genotypes. Genotypes with higher probabilities are considered closer to the ideotype and therefore more desirable [23].

Multi-Trait Stability Index (MTSI)

The multi-trait stability index (MTSI) was developed using factor analysis as described by [24]. Unlike MGIDI, which is based solely on BLUP-derived mean performance, MTSI integrates both performance and stability by incorporating the weighted average of absolute scores from the singular value decomposition of the BLUP × environment interaction matrix (WAASBY) into the genotype × trait matrix used for factor analysis.

Genotype rankings were obtained by calculating the Euclidean distance between each genotype’s factor scores and those of the ideotype:

Where is the multi-trait stability index for the ith genotype; F_iⱼ denotes the score of the ith genotype on the jth factor, while Fⱼ denotes the corresponding score of the ideotype on the same factor. Lower MTSI values indicate genotypes with superior mean performance and greater stability across environments. When the stability component is excluded, the MTSI reduces to the MGIDI.

Multi-Trait Mean Performance and Stability Index (MTMPS)

The multi-trait mean performance and stability index (MTMPS) was computed following the framework of [26]. In this study, MTMPS used Wricke’s ecovalence (Wi) as the stability measure instead of WAASBY, thereby distinguishing it from the MTSI. This index simultaneously accounts for average performance and phenotypic stability across environments.

Mean Performance and Stability (MPS) analysis

Mean performance and stability of genotypes across environments were evaluated using the Mean Performance and Stability (MPS) approach. Genotypic mean performance was estimated using BLUPs derived from the mixed-effects model fitted across environments. These BLUPs represent the expected genotypic performance after accounting for environmental and GEI effects. Genotypic stability was quantified using the weighted average of absolute scores from the singular value decomposition of the BLUP-based genotype × environment interaction matrix (WAASB). The WAASB index summarises the contribution of genotype × environment interaction across interaction principal components, with lower values indicating greater stability across environments. The MPS model jointly considers BLUP-based mean performance and WAASB-based stability to identify genotypes that combine high productivity with consistent performance across environments. Genotypes with high mean BLUP values and low WAASB scores were considered superior in terms of both performance and stability. All analyses were conducted using the metan R package.

Selection differential and genetic gain under selection

Genotypes were selected across environments based on MGIDI, MTSI, MTMPS, and FAI-BLUP rankings using a selection intensity of approximately 15%. The selection differential (SD) was calculated as the difference between the mean of selected genotypes X̄s and the overall population mean X̄o. Selection differential expressed as a percentage was calculated as:

Genetic gain under selection (ΔG) was estimated as

Furthermore, genetic gain expressed as a percentage was calculated as :

Where is the heritability of genotype means across environments.

GT and GYT biplot analyses

To evaluate genotype performance across multiple traits, both Genotype × Trait (GT) and Genotype × Yield × Trait (GYT) biplots were constructed. In the GT biplot, mean trait values across replications were subjected to principal component analysis (PCA) to visualise inter-trait correlations and identify superior genotypes. In the GYT biplot, grain yield was multiplicatively combined with individual trait values, and the resulting GYT matrix was analysed using PCA.

Results

Day–night temperature patterns and associated atmospheric conditions

Across locations and years, nighttime temperatures during the cowpea reproductive phase were consistently elevated and showed limited temporal variability (supplementary Fig. 1). Minimum temperatures during flowering and early pod development generally ranged from 21 to 26 °C at Baramati and 22–28 °C at Jodhpur, with only a few weeks recording minimum temperatures below 20 °C. These warm night-time conditions coincided with periods of high daytime temperatures.

Daytime maximum temperatures during flowering frequently exceeded 35 °C across all environments, particularly under the late-sown regime. At Baramati, maximum temperatures during the reproductive phase commonly ranged from 38 to 40 °C, while at Jodhpur values regularly exceeded 40 °C and reached 44–45 °C in several meteorological weeks. Consequently, both daytime (> 35 °C) and nighttime (> 20 °C) heat-stress thresholds were exceeded simultaneously during flowering across locations and years. Further, relative humidity exhibited moderate week-to-week variation but generally declined during periods of peak temperature, particularly during the reproductive stage. These conditions coincided with low and erratic rainfall, with only isolated rainfall events recorded during late flowering to maturity at both sites. No sustained rainfall occurred during the early reproductive phase, indicating limited alleviation of thermal stress through atmospheric or soil moisture conditions.

Under these environmental conditions, a substantial proportion of genotypes failed to flower or set seed. Of the 255 cowpea accessions evaluated, only 147 (57.6%) completed reproductive development across environments, while the remaining 108 (42.4%) failed to reach reproductive maturity. Flowering and seed set were therefore treated as binary outcomes reflecting genotype performance in environments characterised by concurrent high daytime and nighttime temperatures.

Mean performance of cowpea genotypes

The mean performance of the cowpea genotypes varied substantially across locations, years, and sowing times (Table 2; Fig. 1). Plant height (PH) was consistently higher under normal sowing (NS) than under late sowing (LS). At BMT, PH under NS remained stable across years, averaging 157.90 ± 3.68 cm in 2022 and 157.48 ± 3.76 cm in 2023. In contrast, LS conditions led to approximately 14% reductions, with heights of 136.27 ± 3.14 cm and 136.20 ± 3.20 cm, respectively. At JDP, PH was generally lower, with NS plants averaging 80.92 ± 2.43 cm and LS plants showing a 13.3% reduction to 70.06 ± 1.96 cm. Similar trends were observed in branching traits. The number of primary branches per plant (NPB) showed slight differences between sowing times at BMT, with LS plants having slightly more branches (5.26 ± 0.08) than NS plants (5.00 ± 0.10 in 2022 and 4.76 ± 0.10 in 2023). In contrast, at JDP, NS plants had about 6.2% more primary branches than LS plants (4.41 ± 0.07 vs. 4.02 ± 0.06). Secondary branching (NSB) was consistently higher under NS at BMT, averaging around 10.2 branches compared to approximately 9.6 under LS, indicating a 5–6% reduction under heat stress. At JDP, secondary branches were nearly half of those observed at BMT, with LS causing a further 7.7% decrease relative to NS.

Table 2.

Descriptive statistics for agronomic and physiological traits recorded in screening trials conducted at BMT and JDR under normal and late sowing conditions in 2022 and 2023

Trait	Year	2022		2023		2023
	Treatment	NS	LS	NS	LS	NS	LS
	Locations	BMT		BMT		JDP
Plant Height (cm)	Mean ± SE	157.90 ± 3.68	136.27 ± 3.14	157.48 ± 3.76	136.20 ± 3.20	80.92 ± 2.43	70.06 ± 1.96
	Min.	40.00	32.00	40.01	16.00	22.92	24.67
	Max.	376.00	288.00	376.00	288.00	220.83	198.53
	SD	63.17	53.81	64.52	54.80	41.61	33.63
Number of primary branches per plant	Mean	5.00 ± 0.10	5.26 ± 0.08	4.76 ± 0.10	5.26 ± 0.08	4.41 ± 0.07	4.02 ± 0.06
	Min.	1.00	2.00	1.00	2.00	2.00	1.00
	Max.	10.00	11.00	10.00	11.00	9.00	6.83
	SD	1.71	1.31	1.65	1.35	1.12	1.00
Number of Secondary branches per plant	Mean	10.23 ± 0.08	9.59 ± 0.31	10.19 ± 0.26	9.62 ± 0.32	5.99 ± 0.13	5.50 ± 0.12
	Min.	2.00	1.00	2.00	1.00	1.50	1.09
	Max.	25.00	30.00	25.00	30.00	18.26	16.38
	SD	4.28	5.39	4.44	5.54	2.26	2.11
Pods per Plant	Mean	21.87 ± 0.83	19.97 ± 0.79	21.45 ± 0.83	19.95 ± 0.79	15.17 ± 0.18	13.21 ± 0.21
	Min.	1.27	2.00	1.27	2.00	10.06	0.00
	Max.	71.72	70.00	71.72	70.00	30.62	25.81
	SD	14.20	13.48	14.15	13.54	3.05	3.55
Pod length (cm)	Mean	14.81 ± 0.14	13.67 ± 0.13	14.74 ± 0.13	13.71 ± 0.13	25.83 ± 0.45	19.44 ± 0.55
	Min.	8.00	5.89	7.50	8.00	10.00	1.64
	Max.	22.05	22.33	22.05	22.33	49.56	45.87
	SD	2.33	2.24	2.27	2.16	7.71	9.44
Grains per pod	Mean	12.46 ± 0.17	11.91 ± 0.15	12.37 ± 0.17	11.89 ± 0.15	13.35 ± 0.19	11.36 ± 0.22
	Min.	5.00	4.67	5.00	4.67	5.00	2.75
	Max.	22.33	18.00	22.33	18.00	34.18	33.00
	SD	2.89	2.57	2.87	2.59	3.23	3.70
Grain yield per plant (g)	Mean	19.04 ± 0.67	13.77 ± 0.56	20.72 ± 0.62	15.04 ± 0.56	24.92 ± 0.44	13.23 ± 0.49
	Min.	2.10	1.49	5.51	2.14	8.35	1.02
	Max.	52.75	46.87	50.39	46.15	49.90	36.02
	SD	11.41	9.66	10.59	9.66	7.49	8.46
Canopy Temperature 1 (O C)	Mean	30.35 ± 0.12	32.11 ± 0.09	25.10 ± 0.08	28.63 ± 0.13	29.73 ± 0.10	31.47 ± 0.09
	Min.	24.50	27.32	20.87	24.01	25.20	26.85
	Max.	34.00	36.56	30.17	37.65	33.15	36.37
	SD	1.77	1.62	1.49	2.24	1.72	1.59
Canopy Temperature 2 (O C)	Mean	30.83 ± 0.12	30.80 ± 0.13	25.35 ± 0.08	28.30 ± 0.10	29.96 ± 0.10	32.14 ± 0.12
	Min.	20.90	24.69	22.02	23.53	24.98	27.01
	Max.	38.63	41.2	31.19	38.40	34.25	46.81
	SD	2.03	2.15	1.39	1.78	1.78	2.00

Fig. 1.

Boxplots showing the distribution of morpho-physiological traits under normal and late sowing conditions at Baramati and Jodhpur during the 2022 and 2023 growing seasons

The number of pods per plant (PP) was significantly influenced by sowing time. At BMT, NS plants recorded 21.87 ± 0.83 and 21.45 ± 0.83 pods per plant in 2022 and 2023, respectively, whereas LS plants showed an approximate 9% reduction, averaging around 19.95 pods. At JDP, pod number was lower overall, with NS plants producing 15.17 ± 0.18 pods compared to 13.21 ± 0.21 under LS, reflecting a 13% decline. Pod length (PL) also exhibited substantial site-specific variation. At JDP, pods were significantly longer under NS (up to 25.83 ± 0.45 cm in 2023), compared to 19.44 ± 0.55 cm under LS—a reduction of about 33%. In contrast, pod length at BMT remained relatively stable across treatments and years, averaging around 14 cm. Grain number per pod (GPP) was moderately higher under NS at both locations, with LS plots showing 4–7% fewer grains per pod, underscoring the negative impact of elevated temperatures on reproductive output.

Grain yield per plant (GY) was significantly higher under NS across all years and sites. At BMT, NS treatments produced yields of 19.04 ± 0.67 g in 2022 and 20.72 ± 0.62 g in 2023, while LS treatments yielded 13.77 ± 0.56 g and 15.04 ± 0.56 g, representing an average reduction of 28%. The disparity was even more pronounced at JDP in 2023, where NS yields reached 24.92 ± 0.44 g, nearly double (88% higher) the LS yield of 13.23 ± 0.49 g, underscoring the more substantial impact of heat stress at this site. Canopy temperature (CT) measurements further validated the imposition of stress. At BMT, canopy temperatures under NS ranged from 25.10 ± 0.08 °C to 25.35 ± 0.08 °C, while under LS they increased to 38.63 ± 0.13 °C and 38.40 ± 0.10 °C, representing an approximate 12% increase. At JDP, canopy temperatures were generally higher, with NS averaging around 30.3 °C and LS reaching 32.1 °C—a 6% increase under LS conditions. These results confirm that late sowing effectively imposed high-temperature stress during critical reproductive stages, leading to measurable physiological and agronomic changes in cowpea genotypes.

Combined analysis of variance for growth, yield and canopy temperature traits of cowpea genotypes

Combined ANOVA revealed that the mean sum of squares (MSS) due to genotypes, management, and location was highly significant (p < 0.01) for all traits (Table 3). In contrast, the effect of years was non-significant for PH, NSB, PL and GPP. Two-way and three-way interactions involving genotypes (Gen), sowing time (Mgt) and locations (Loc) were significant for all traits. Interactions involving Year with Gen and Mgt were significant for NPB, GY, CT1 and CT2 but non-significant for the remaining traits.

Table 3.

Combined Analysis of Variance (ANOVA) for morphological and physiological traits of cowpea genotypes across management practices, locations, and years

Source of variation	DF	Mean Sum of Squares
		PH	NPB	NSB	PP	PL	GPP	GY	CT1	CT2
Gen	146	19,913**	11.39**	121**	946**	77**	39.1**	702**	15**	13**
Mgt (NS vs. LS)	1	141,670**	6.63**	141**	1405**	3593**	446.2**	25,124**	2420**	1278**
Loc	1	2,002,403**	285.99**	6806**	17,198**	27,693**	15**	1463**	6419**	6614**
Year	1	17	4.41**	0	14**	0	1	639**	115**	17**
rep	1	590	0.11	0	0	13**	35.3	54**	1	0
Gen: Mgt	146	1266**	1.28**	15**	77**	42**	8.6**	75**	4**	3**
Gen: Loc	146	10,895**	7.44**	62**	494**	127**	19.1**	219**	9**	9**
Mgt: loc	1	11,013**	58.88**	1	7	2752**	215.4**	3779**	311**	346**
Gen: Year	146	42	0.29**	1	3	1	0.1	46**	2**	2**
Mgt: Year	1	9	4.17**	0	12**	1	0.3	13**	0	364**
Gen: Mgt: Loc	146	880**	1.06**	7**	45**	83**	8.8**	64**	4**	4**
Gen: Mgt: Year	146	18	0.24**	0	3	0	0.1	11**	3**	10**
Residual	881	154	0.22	2	3	2	5.4	0	0	0

Legend: SV Source of Variation, DF Degrees of Freedom, PH Plant Height (cm), NPB Number of Primary Branches, NSB Number of Secondary Branches, PP Number of Pods per Plant, PL Pod Length (cm), GPP Number of Grains per Pod, GY Grain Yield (g/plant), CT1 Canopy Temperature at Mid-Flowering (°C), CT2 Canopy Temperature at Mid-Podding (°C)

Genetic parameters, heritability, and selection potential of growth and high temperature adaptation traits in cowpea across environments

Broad-sense heritability (H²) estimates ranged from 31.02 to 71.39%. The highest value was for grain yield per plant (71.39%), followed by grains per pod (67.06%), pods per plant (60.46%), pod length (61.56%) and number of primary branches (49.79%) (Table 4). Moderate heritability was found for NSB (44.23%), PH (33.35%), CT2 (32.56%), and CT1 (31.02%). The genotypic coefficient of variation (CVg) was highest for PL (39.45%), followed by PP (37.11%), GY (36.34%), and NSB (24.81%). In contrast, CT1 (11.26%), CT2 (13.07%), GPP (12.1%), and NPB (14.36%) showed moderate variability (Table 4). The ratio of genotypic to residual coefficient of variation (CVg/CVr) was above 1.0 for PP (2.45), GY (1.83), PH (1.53), and NSB (1.51). Selection accuracy was highest for GY (84.49%), GPP (81.89%), and PP (77.75%). The proportion of total variation due to genotype × environment interaction (R²ge) was highest for CT2 (2.619), CT1 (2.581), and PL (2.273). The stability of trait expression in experimental genotypes across different environments, measured by r_ge, was highest for GPP (0.820), GY (0.755), and PP (0.625).

Table 4.

Likelihood ratio test and genetic parameters analysis of cowpea genotypes for high temperature tolerance adaptive traits

Trait	LRTg	LRTge	LRTgt	s2p	Heritability	R2ge	R2gt	h2mg	Accuracy	rge	CVg	CVr	CVg_CVr
PH	8.55	2415.778	273.235	1659.47	33.35	1.809	0.103	33.35	57.75	0.395	19.11	12.48	1.53
NPB	26.482	1016.132	125.711	0.949	49.79	1.17	0.157	49.79	70.56	0.546	14.36	12.95	1.1
NSB	19.274	1738.431	425.109	10.1	44.23	1.303	0.214	44.23	66.5	0.503	24.81	16.48	1.51
PP	50.394	2067.054	553.977	78.835	60.46	0.942	0.146	60.46	77.75	0.625	37.11	15.17	2.4
PL	51.56	2156	542.6	73.49	61.56	1.273	0.303	61.56	61.58	0.56	39.45	21.94	1.8
GPP	55.923	45.531	16.695	3.259	67.06	0.327	0.169	67.06	81.89	0.82	12.1	18.87	0.6
GY	96.382	916.409	315.941	58.525	71.39	0.539	0.177	71.39	84.49	0.755	36.34	19.88	1.8
CT1	21.02	1533.14	102.939	2.13	31.02	2.581	0.183	25.65	49.45	0.39	11.26	13.75	0.8
CT2	20.89	917.236	1.322	1.933	32.56	2.619	0.033	26.45	51.23	0.35	13.07	15.42	0.8

Where, (LRT - The Likelihood Ratio Test for the random effects; Phenotypic variance the phenotypic variance; Heritability the broad-sense heritability BLUP basis; R2ge the coefficient of determination of the interaction effects; h²mg the heritability on the mean basis; Accuracy the selective accuracy; rge the genotype-environment correlation; CVg the genotypic coefficient of variation; CVr the residual coefficient of variation; CV ratio the ratio between genotypic and residual coefficient of variation)

*** significant at P < 0.001, LRT significance test is conducted against the Chi-square value

Genetic correlations among the growth, yield and canopy temperature traits of cowpea

Grain yield per plant (GY) exhibited strong positive correlations with PP (r = 0.40), GPP (r = 0.36), and PL (r = 0.34), and a negative correlation with CT1 (r = − 0.35) and CT2 (r = − 0.31) (Fig. 2). Canopy temperatures at the two stages, CT1 and CT2, were strongly and positively correlated (r = 0.71, p < 0.001), reflecting a consistent thermal response across the two growth stages. Among the morphological traits, NPB and NSB were positively correlated (r = 0.55, p < 0.001). PH showed moderate positive correlations with NPB (r = 0.38) and NSB (r = 0.26), but it had negative associations with traits related to yield. Notably, PL had strong negative correlations with CT1 (r = − 0.57) and CT2 (r = − 0.54), indicating that pod development is susceptible to heat stress.

Fig. 2.

Scatterplot matrix showing pairwise relationships, frequency distributions, and Pearson’s correlation coefficients among morpho-physiological, and yield traits in cowpea. * = p < 0.05 and *** = p < 0.001

Identification of superior genotypes using multi-trait indices

A total of 147 cowpea genotypes were evaluated under high-temperature stress. Multi-trait selection was performed using MGIDI, FAI-BLUP, MTSI, and MTMPS indices, applying a selection intensity of 15%. MGIDI identified 29 genotypes with superior multi-trait performance, including G067 (EC240924), G066 (EC240878), G142 (IC402161), G044 (IC488270), G065 (EC240989-A), G111 (IC433448), and others (Table 5; Fig. 3). Similarly, FAI-BLUP selected 22 genotypes, including G119 (IC472252), G112 (EC240868), G132 (EC724905), G044 (IC488270), and G004 (EC240926) and others. Further, MTSI selected 22 genotypes (Fig. 4), including IC472254, IC400155, EC243927, EC240868, and IC418505. MTMPS also selected 22 genotypes, including G120 (EC240868), G106 (IC400155), G086 (IC418505), G120 (IC472254), and G080 (IC488085).

Table 5.

Selection of cowpea genotypes for high temperature stress tolerance based on different Multi-Trait Stability Approaches

Models	Selected Genotypes
MGIDI	G004 (EC240926), G022 (IC402159), G028 (EC240829), G031 (IC488077), G032 (IC402161), G036 (EC243938), G039 (EC241048), G042 (EC244138), G044 (IC488270), G047 (IC488185), G049 (IC488264), G063 (EC240966-A) G065 (EC240989-A), G066 (EC240878), G067 (EC240924), G074 (IC402103), G078 (EC240920), G079 (IC410043), G080 (IC488085), G085 (IC420660), G086 (IC418505), G088 (IC488146), G105 (IC402182), G111 (IC433448), G112 (EC240868), G114 (EC243927), G119 (IC472252), G142 (IC402161), G147 (GC3)
FAI_BLUP	G001 (EC240930), G004 (EC240926), G006 (EC240884), G011 (EC107151), G012 (EC107176), G039 (EC241048) G044 (IC488270), G054 (EC240875), G078 (EC240920), G080 (IC488085), G091 (IC421917), G095 (IC569092) G111 (IC433448), G112 (EC240868), G114 (EC243927), G119 (IC472252), G120 (IC472254), G132 (EC724905) G136 (EC528421), G142 (IC402161), G145 (PL4), G147 (GC3)
MTSI	G001 (EC240930), G013 (EC107173), G014 (EC117720), G018 (EC240824), G044 (IC488270), G047 (IC488185) G049 (IC488264), G078 (EC240920), G080 (IC488085), G086 (IC418505), G091 (IC421917), G097 (IC596961) G106 (IC400155), G108 (IC402101), G112 (EC240868), G114 (EC243927), G115 (IC471948), G120 (IC472254) G130 (EC724547), G132 (EC724905), G142 (IC402161), G143 (RC101)
MTMPS	G001 (EC240930), G014 (EC117720), G018 (EC240824), G044 (IC488270), G047 (IC488185), G049 (IC488264) G078 (EC240920), G079 (IC410043), G080 (IC488085), G086 (IC418505), G091 (IC421917), G097 (IC596961) G106 (IC400155), G108 (IC402101), G112 (EC240868), G113 (IC421881), G114 (EC243927), G120 (IC472254) G124 (EC724364), G130 (EC724547), G132 (EC724905), G142 (IC402161)

Fig. 3.

Selection of Cowpea Genotypes Based on Multi-Trait Genotype-Ideotype Distance Index (left side), The circular plot displays MGIDI values for 147 cowpea genotypes, with selected genotypes (superior multi-trait performance) in red and non-selected in grey. The red circle indicates the selection threshold. Strengths and weaknesses view (Right side) of the selected genotypes based on the Multi-trait Genotype–Ideotype Distance Index. The radar chart displays the contribution of each factor (FA1–FA4) to the MGIDI value for the selected genotypes. The center of the chart represents the lowest MGIDI contribution (i.e., closer proximity to the ideotype and superior performance for traits within that factor), whereas values toward the outer edge indicate a greater distance from the ideotype (less desirable performance)

Fig. 4.

Selection of Cowpea Genotypes Using Factor Analysis- Best Linear Unbiased Predictions (FA-BLUP) (Left) and Multi-Trait Stability Index (MTSI) (Right). Circular plots showing 147 cowpea genotypes ranked by FA-BLUP and MTSI values, respectively. Selected genotypes are highlighted in red, non-selected genotypes in black, and the red circles indicate the selection thresholds

Coincidence analysis (Table 6) revealed the highest overlap between MTSI and MTMPS (19 genotypes, CI = 83.96%), while MGIDI and MTSI shared only nine genotypes (CI = 14.09%). Six genotypes, viz., [G044 (IC488270), G078 (EC240920), G080 (IC488085), G112 (EC240868), G114 (EC243927), G142 (IC402161)], were consistently selected by all four indices, indicating robust multi-trait performance and stability across environments.

Table 6.

Common cowpea genotypes between the multi-trait stability models based on the coincidence index

Model 1	Model 2	CI_Percent	Common Count	Common Genotypes	Common Genotypes across all models
MGIDI	FAI-BLUP	22.68	11	G004 (EC240926), G039 (EC241048), G044 (IC488270), G078 (EC240920), G080 (IC488085), G111 (IC433448), G112 (EC240868), G114 (EC243927), G119 (IC472252), G142 (IC402161), GC3	G044 (IC488270), G078 (EC240920), G080 (IC488085), G112 (EC240868), G114 (EC243927), G142 (IC402161)
MGIDI	MTSI	14.09	9	G044 (IC488270), G047 (IC488185), G049 (IC488264), G078 (EC240920), G080 (IC488085), G086 (IC418505), G112 (EC240868), G114 (EC243927), G086 (IC418505)
MGIDI	MTMPS	18.38	10	G044 (IC488270), G047 (IC488185), G049 (IC488264), G078 (EC240920), G079 (IC410043), G080 (IC488085), G086 (IC418505), G112 (EC240868), G114 (EC243927), G142 (IC402161)
FAI-BLUP	MTSI	35.85	10	G001 (EC240930), G044 (IC488270), G078 (EC240920), G080 (IC488085), G091 (IC421917), G112 (EC240868), G114 (EC243927), G120 (IC472254), G132 (EC724905), G142 (IC402161)
FAI-BLUP	MTMPS	35.85	10	G001 (EC240930), G044 (IC488270), G078 (EC240920), G080 (IC488085), G091 (IC421917), G112 (EC240868), G114 (EC243927), G120 (IC472254), G132 (EC724905), G142 (IC402161)
MTSI	MTMPS	83.96	19	G001 (EC240930), G014 (EC117720), G018 (EC240824), G044 (IC488270), G047 (IC488185), G049 (IC488264), G078 (EC240920), G080 (IC488085), G086 (IC418505), G091 (IC421917), G142 (IC596961), G106 (IC400155), G108 (IC402101), G112 (EC240868), G114 (EC243927), G120 (IC472254), G130 (EC724547), G132 (EC724905), G142 (IC402161)

Predicted selection differential and genetic gain

The predicted selection differentials (ΔS%) and genetic gains (ΔG%) across all measured traits are summarised in Table 7. Positive ΔS values were recorded for all traits except plant height (PH: − 2.4%). Grain yield per plant (GY) showed the highest ΔS (24.3%), reflecting substantial improvement potential. The mean genetic gain across traits was 5.97%, with GY showing the highest gain (18.7%) and PH again showing a slight reduction (–1.7%). The WAASBY index, representing overall stability, had a mean ΔS of 8.81%, highlighting the effectiveness of combined performance and stability selection.

Table 7.

Multi-Environment Selection Gain Estimates for Mean Performance via MGIDI

Traits	Factor	Xo	Xs	SD	SD%	h2	SG	SG%	Indicator
CT1	FA 1	29.6	29.9	0.4	1.2	0.5	0.2	0.6	decrease
CT2	FA 1	29.6	29.9	0.3	1.1	0.5	0.2	0.5	decrease
PP	FA 2	18.6	21.5	2.9	15.5	0.7	2.1	11.2	increase
GY	FA 2	17.8	22.1	4.3	24.3	0.8	3.3	18.7	increase
NPB	FA 3	4.8	5.3	0.5	10.0	0.6	0.3	6.4	increase
NSB	FA 3	8.5	9.9	1.4	16.5	0.7	1.0	11.8	increase
PH	FA 4	123.0	120.0	-3.0	-2.4	0.7	-2.1	-1.7	increase
PL	FA 4	17.0	17.8	0.8	4.6	0.1	0.1	0.6	increase
GPP	FA 4	12.2	13.3	1.0	8.5	0.7	0.7	5.6	increase

Xo mean for WAASBY index of the original population, Xs mean for WAASBY index of the selected genotypes, SD and SD perc, The selection differential and selection differential in percentage, respectively; SG and SG%, The selection gains and selection gains in percentage, respectively

The strengths and weakness view

Radar plots were used to visualise the contribution of each factor (FA1–FA4) to MGIDI scores (Fig. 3). Lower factor contributions indicate proximity to the ideotype: Genotypes G067 (EC240924), G066 (EC240878), G142 (IC402161), and G044 (IC488270) contributed minimally across all factors, indicating desirable multi-trait profiles. Genotypes G047 (IC488185), G080 (IC488085), and G028 (EC240829) showed higher contributions in specific factors, reflecting relative weaknesses. FA4 accounted for the largest variability, followed by FA3 and FA2, highlighting key traits driving differences among genotypes.

Comparison of stability-based indices (MTSI vs. MTMPS)

MTSI and MTMPS both integrate performance and stability; however, they use different stability metrics. MTSI uses WAASBY, which captures the overall magnitude of genotype × environment interaction. However, MTMPS incorporates Wricke’s ecovalence (Wi), which is more sensitive to crossover interactions and genotype-specific instability. Using both indices provides complementary perspectives on stability, improving confidence in identifying genotypes with both high performance and consistent expression. Coincidence analysis showed substantial overlap between these indices, confirming their effectiveness and complementarity.

Multi-environment trial performance (MPS Model)

The mean performance and stability of cowpea genotypes across environments were evaluated using the Mean Performance and Stability (MPS) approach based on BLUP-estimated mean grain yield and the WAASB stability index (Fig. 5). The presence of substantial genotype × environment interaction was reflected by the wide dispersion of genotypes along the stability axis, indicating differential responses to environmental conditions. In the MPS plot, BLUP-based mean grain yield is displayed on the x-axis, while the WAASB stability index is shown on the y-axis, with lower WAASB values indicating greater stability. The vertical and horizontal reference lines represent the overall mean performance and mean stability, respectively, dividing the plot into four quadrants. Genotypes positioned in the bottom-right quadrant, characterised by above-average grain yield and below-average WAASB values, were identified as high-performing and stable across environments. These included G066 (EC240878), G067 (EC240924), G078 (EC240920), G080 (IC488085), G085 (IC420660), G086 (IC418505), G088 (IC488146), G105 (IC402182), G111 (IC433448), G112 (EC240868), G114 (EC243927), G119 (IC472252), G142 (IC402161), and GC3. In contrast, genotypes such as G013 (IC488077), G036 (EC243938), and G091 (IC421917) were in regions associated with higher WAASB values, indicating greater yield instability across environments. The MPS results were largely consistent with the outcomes of the multi-trait selection indices (MGIDI, MTSI, MTMPS, and FAI-BLUP), reinforcing the robustness of the identified superior genotypes and confirming their suitability for selection under multi-environment conditions.

Fig. 5.

Mean performance and stability (MPS) plot showing BLUP-based mean grain yield plotted against the WAASB stability index for genotypes evaluated across multiple environments. The vertical and horizontal dashed lines represent the overall mean performance and stability, respectively. Genotypes located in the bottom-right quadrant combine high yield with greater stability across environments

Genotype–Trait (GT) and Genotype–Yield × Trait (GYT) biplot analysis

The GT and GYT biplots were employed to visualise genotype performance across multiple traits and to provide complementary insights to the multi-trait selection indices (Fig. 6). The GT biplot summarises trait correlations and facilitates the identification of genotypes with desirable trait combinations. In the present study, the GT biplot accounted for 47.28% of the total variation (PC1 = 26.94% and PC2 = 20.34%). Although this proportion is moderate, it reflects the inclusion of multiple traits with distinct biological functions and weak-to-moderate inter-trait correlations, which tend to distribute variation across higher-order principal components. Consequently, the GT biplot was used as an exploratory tool to elucidate genotype–trait associations rather than as a standalone criterion for selection. Despite the moderate explained variance, the GT biplot revealed biologically meaningful patterns, including the positive association of yield-related traits (GY, PP, GPP, and PL) with superior genotypes such as G066 (EC240878), G067 (EC240924), G111 (IC433448), G142 (IC402161), and G022 (IC402159), as well as the negative association between canopy temperature traits (CT1 and CT2) and yield, highlighting potential physiological trade-offs. In contrast, the GYT biplot integrates grain yield with individual trait values, thereby directly incorporating yield contribution into the multivariate analysis. The GYT biplot explained a substantially higher proportion of the total variation (78.3%), with PC1 and PC2 accounting for 60.3% and 17.97%, respectively. This enhanced explanatory power improved interpretability and enabled apparent discrimination among genotypes based on overall multi-trait performance. Genotypes such as G142 (IC402161), G004 (EC240926), and G022 (IC402159) were positioned favourably with respect to key yield-related combinations (YPL, YGPP, and YPP), supporting their identification as superior lines. Overall, while the GT biplot was effective for assessing trait correlations and trade-offs, the GYT biplot provided a more comprehensive and yield-focused evaluation of genotype performance. The complementary use of both approaches strengthened the interpretation of genotype responses and reinforced the selection decisions derived from MGIDI, FAI-BLUP, MTSI, and MTMPS.

Fig. 6.

Genotype-by-yield-Trait (GYT) Biplot and Genotype-by-Trait (GT) Biplot of Cowpea Genotypes and Traits Based on Principal Component Analysis. Left: The GYT biplot shows the relationship between cowpea genotypes (blue dots) and multiple traits (green vectors), with the first two principal components capturing most of the variation. Right: The GT biplot illustrates the association between 147 cowpea genotypes (blue dots) and measured traits (green diamonds), explained by the first two principal components

Discussion

This study aimed to identify heat-tolerant cowpea genotypes with stable performance across environments characterised by high ambient temperatures during critical phenological stages. To achieve this, 255 genetically diverse cowpea genotypes including checks were evaluated under terminal heat stress imposed through late sowing across two contrasting agro-ecological locations over two consecutive years. To our knowledge, this represents one of the most comprehensive multi-environment trials (METs) in cowpea, integrating the Weighted Average of Absolute Scores from the BLUP matrix (WAASB) with multiple multi-trait selection and stability indices including MGIDI, MTSI, MTMPS, and FAI-BLUP for the assessment of genotype × environment × management (G×E×M) interactions under field conditions. This integrative framework overcomes the limitations of single-trait selection and controlled-environment phenotyping by enabling a multidimensional, field-relevant evaluation of genotype performance and stability [28, 29].

Heat stress is a major constraint to cowpea productivity in tropical and subtropical regions, where both daytime and nighttime temperatures increasingly exceed optimal thresholds [30, 31]. Reproductive development in cowpea is susceptible to temperature extremes. Elevated daytime temperatures impair pollen viability, fertilisation, and pod set, while elevated nighttime temperatures increase respiratory carbon losses and reduce carbon-use efficiency during grain filling [2, 6, 13, 31–33]. In this study, late sowing effectively synchronised flowering and grain filling with periods of sustained high temperature, enabling the evaluation of genotype responses to both day- and night-time heat stress under field conditions that incorporate diurnal temperature variation, soil heterogeneity, and root–soil interactions.

During reproductive development, maximum daytime temperatures frequently exceeded 35 °C and reached 42–45 °C across locations and years, well beyond thresholds known to impair reproductive success in cowpea [2, 31]. Night temperatures remained consistently elevated, likely intensifying heat injury by increasing respiratory demand and reducing assimilate availability for reproductive sinks [33]. The combined exposure to high day and night temperatures, therefore, constituted a biologically relevant and agronomically realistic terminal heat-stress scenario. Although other atmospheric factors, such as radiation load and vapour pressure deficit, often co-occur with high temperatures in field environments, soil moisture was adequately maintained, and precipitation was monitored, minimising confounding drought effects. The strong negative association between canopy temperature and grain yield further supports the view that temperature is the primary driver of yield reduction in this study.

A major biological outcome of terminal heat stress was the substantial reduction in the number of genotypes progressing from planting to successful flowering and seed set. More than 40% of the initially evaluated genotypes failed to reach reproductive maturity under late-sown conditions, revealing pronounced variation in reproductive resilience. This failure was not due to pre-selection or experimental exclusion, but rather likely reflects the interacting effects of intrinsic heat sensitivity, phenological mismatch, and photoperiod responsiveness under a fixed sowing date. High night temperatures combined with long day length can suppress floral bud initiation, delay flowering, and induce floral bud abortion [5]. Elevated night temperatures during floral development can also disrupt microsporogenesis, resulting in reduced pollen viability, poor pod set, and fewer seeds per pod [5, 7]. Survival through flowering and seed set, therefore, integrates multiple physiological and developmental processes and represents a critical, though often overlooked, component of heat tolerance in cowpea.

Restricting subsequent analyses to the 147 genotypes that flowered and set seed may introduce selection bias, potentially inflating apparent heat tolerance among surviving genotypes. Accordingly, tolerance estimates should be interpreted as conditional on reproductive survival under heat stress. Importantly, reproductive failure itself constitutes a biologically meaningful indicator of heat susceptibility and should be explicitly considered in breeding strategies. Together, reproductive survival and post-flowering performance capture complementary dimensions of heat tolerance that are essential for realistic genotype evaluation.

Among genotypes that completed reproduction, terminal heat stress caused substantial reductions in growth, physiological, and yield-related traits. Decreases in plant height, branching, pod set, and grain yield were consistent with previous reports in cowpea and other legumes [2, 12, 31]. Canopy temperature increased markedly under stress, providing physiological evidence of impaired heat avoidance. Although physiological characterisation was limited mainly to canopy temperature, this reflects a deliberate balance between physiological depth and scalability in large, field-based multi-environment trials. Canopy temperature is a high-throughput, non-destructive proxy trait that integrates stomatal conductance, transpiration cooling, canopy architecture, and root water uptake. It is therefore interpreted as an integrative indicator of whole-plant heat avoidance rather than a direct mechanistic explanation of tolerance [34].

The magnitude of the yield reduction observed at Jodhpur (~ 88%) reflects the severity of the environment rather than an anomaly in the experimental design. While high temperature was the primary imposed stress, the severity of yield loss likely resulted from compounded effects of heat with site-specific factors such as high vapour pressure deficit, intense radiation, and coarse-textured soils. From a breeding perspective, Jodhpur represents an extreme screening environment rather than a typical target population environment (TPE). Such environments are valuable for eliminating highly susceptible genotypes and identifying exceptional resilience, but should not be interpreted as representative of farmer-managed conditions. Genotypes that perform consistently across both Jodhpur and Baramati, therefore, demonstrate broad adaptation and robustness across a gradient of heat-stress intensity.

Correlation analyses showed that pods per plant, grains per pod, and pod length were positively associated with grain yield. In contrast, canopy temperature at both measurement stages was negatively associated with yield, reinforcing the importance of canopy cooling as a heat avoidance mechanism [2, 31, 35]. These relationships support the use of canopy temperature as a practical physiological indicator in heat tolerance breeding when combined with reproductive and yield traits.

The comparative use of multiple multi-trait indices enabled a nuanced interpretation of genotype performance and stability. MGIDI identified genotypes based on proximity to an ideotype but does not explicitly penalise environmental instability, which explains its lower concordance with stability-oriented indices. In contrast, MTSI and MTMPS explicitly integrate stability by combining BLUP-based performance with WAASB and Wricke’s ecovalence, respectively. FAI-BLUP represents an intermediate approach. The partial divergence among indices, therefore, reflects biologically meaningful trade-offs between maximising mean performance and ensuring stability across environments [24–28]. From a practical breeding perspective, MGIDI may be most useful for ideotype-driven selection under relatively uniform conditions or early breeding stages, whereas MTSI and MTMPS are more appropriate for advanced selection and cultivar recommendation under climate variability [24–28].

Six genotypes, viz., G044 (IC488270), G078 (EC240920), G080 (IC488085), G112 (EC240868), G114 (EC243927), G142 (IC402161) were consistently identified across all indices, underscoring their robust performance and stability. These genotypes combined higher pod number and seed set with cooler canopy temperatures (3–4 °C lower than checks PL3 and PL4), resulting in grain yields approximately 60–65% higher than the population mean under heat stress. Their stability reflects relative performance across stress gradients rather than absolute yield stability under extreme conditions (Figs. 7 and 8).

Fig. 7.

Performance of selected genotypes along with checks for yield trait across different locations and management conditions. Environemts are E1:NS_BMT22, E2:LS_BMT22, E3: NS_BMT23, E4:LS_BMT23, E5: NS_JDP23, E6:LS_JDP23

Fig. 8.

Performance of selected genotypes along with checks for canopy temperature trait at pre-flowering (CT1) and post-flowering (CT2) across different locations and management conditions. Environemts are E1:NS_BMT22, E2:LS_BMT22, E3: NS_BMT23, E4:LS_BMT23, E5: NS_JDP23, E6:LS_JDP23

Predicted genetic gains based on index selection indicate substantial improvement potential. At a 15% selection intensity, relative gains of approximately 18–20% translated into absolute yield advantages of around 3–5 g plant⁻¹ under late-sown conditions. Although realised gains under farmer-managed systems are likely to be lower due to environmental and management variability, the consistent superiority of selected genotypes over standard checks suggests meaningful agronomic benefits in heat-prone regions [36].

In conclusion, while individual multi-trait indices are not novel per se, their comparative application under field-imposed terminal heat stress in cowpea represents a significant methodological contribution. The identification of consensus genotypes across multiple indices is particularly impactful, providing breeders with high-confidence candidates for further evaluation and deployment. More broadly, the analytical framework presented here is transferable to other legumes and stress-prone cropping systems. Future studies should expand environmental coverage and integrate deeper physiological phenotyping and genomic tools such as GWAS and QTL mapping to elucidate the genetic architecture of heat tolerance further and accelerate the development of climate-resilient cowpea cultivars.

Authors’ contributions

B.P.S. conceptualised the study and wrote the original manuscript. M.B. and R.B. conducted the research and collected the data. V.S.M. and K.T. provided seed materials and conducted experiments at the JDP location. D.S., B.K.M., H.C.B., S.H.B., R.S., and S.R.K. contributed to manuscript editing. A.C. critically revised the manuscript and suggested the formal analysis.

Funding

This research was supported by ICAR–NICRA under the project “Phenotyping of Pulses for Enhanced Tolerance to Drought and Heat (OXX01737).”

Data availability

The data generated during this research are included in the manuscript. Additional datasets are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.