A field screening experiment was conducted over ten years (2013–2022) at the experimental research station of the Department of Agriculture, University of Naples Federico II, located in Sant’Angelo dei Lombardi (40°55’12.0”N, 15°07’12.0”E, 667 m above sea level). This rural site, situated in the foothills of the Apennine Mountain range in the province of Avellino, southern Italy, experiences a warm-summer Mediterranean climate (Csb) as per the Köppen-Geiger classification (Beck et al., 2018). Utilizing 43 years of daily climatic data from the NASA-Power project (https://power.larc.nasa.gov/) covering the period from 1981 to 2023, this region displays an average daily temperature of 14.02°C (± 0.60°C) and an average annual precipitation of 572.36 mm (± 119.38 mm). The soil in this area is predominantly clay loam in texture (clay, sand, and silt: 38.8%, 36.9%, and 24.3%, respectively). The chemical and physical characteristics of the soil at the beginning of the experiment at 0.20 m of soil depth were as follows: pH 8.07, Kjeldahl total nitrogen 0.97 g kg⁻¹, phosphorus pentoxide (P₂O₅: Olsen method) 14.4 mg kg⁻¹, and soil organic matter (SOM) content of 1.16%.

The daily key weather parameters for each wheat growing season at the Sant’Angelo dei Lombardi location, including, precipitation, relative humidity, air temperature, and solar radiation, along with 11 additional climatic variables ( Table 1 ), were obtained from the NASA-Power project (https://power.larc.nasa.gov/). The weather regime, including monthly precipitation (P) and minimum and maximum temperatures (Tmin and Tmax) for the ten growing seasons, is detailed in Supplementary Table S1 . These data are compared against the historical means from 1981 to 2023. To precisely reflect the temporal dynamics of environmental conditions throughout crop development, the crop cycles were segmented into five main phenological stages based on days after sowing (DAS): 0 to 21 days (Emergence), 22 to 110 days (Tillering), 111 to 140 days (Jointing), 141 to 184 days (Heading), and 185 to 217 days (Maturity). These intervals were determined through field observations of various wheat varieties grown in Sant’Angelo dei Lombardi over ten growing seasons. This approach resulted in the generation of 900 combinations of environmental covariates (18 weather parameters) across growing seasons (10 years) and main phenological stages (5 time periods), providing a comprehensive dataset for analyzing climate impacts on wheat development.

Eight-nine varieties were investigated. These varieties were certified by ENSE (Ente Nazionale Sementi Elette) and registered in either the national or community variety registry. Forty-eight varieties tested for less than three years were excluded from the analysis, leaving 41 varieties for genotype-by-environment interaction assessment. Not all varieties were evaluated every year, just 5 varieties were shared between all years ( Figure 1 ). Consequently, the dataset is highly unbalanced, with 45% of the data missing. Comprehensive genealogy records, crop cycles, and breeding institutions for all 41 durum wheat varieties are detailed in Supplementary Table S2 . Each year, a set of 23 ± 3 varieties was evaluated, ensuring that at least five varieties were common across all years ( Figure 1 ). The trials were implemented using an experimental block design (RCBD), wherein varieties were randomized across the plots and replicated three times. The size of the plots was 5 m × 2 m. Sowing density was standardized at 400 grains per square meter for each variety. All varieties were grown under rainfed conditions. Each year, broad beans served as the preceding crop for wheat. Prior to sowing, phosphorus (P) and potassium (K) were applied in the form of superphosphate and potassium sulfate, respectively, at a rate of 100 kg ha⁻¹. The specific amounts of nitrogen applied as urea, as well as the sowing and harvest dates, are detailed in Table 2 .

At harvest, a 1 square meter subplot in the center of each plot was hand-harvested to determine nine phenotypic and quality traits: 1) grain yield (GY), adjusting the moisture to 13% and converting the unit to tons per hectare (t ha⁻¹), 2) plant height (Alt), measured using a graduated scale from ground level to the tip of the tallest spikelet and is expressed in meters (m), 3) Thousand-gain weight (TGW), is expressed in gram (g), 4) number of ears per square meter (Sp, num m⁻²), 5) hectoliter weight (Pe) was determined according to the AACC method 55-10.01 (Cereals and Grains Association, 2024) and reported in kilograms per hectoliter (kg/hL), 6) crude protein concentration (Pro), determined using the Kjeldahl method according to AACC 46-10.01 (Cereals and Grains Association, 2024). The Jones factor was used to convert it to protein (6.25) (Mariotti et al., 2008) and expressed in percentage (%), 7) gluten index value (Glu), analyzed according to AACC 38-12.02 (Cereals and Grains Association, 2024), and expressed in percentage (%), 8) shriveled wheat grain (Cs), referring to grain that is underdeveloped, small, and often misshapen expressed in percentage (%), and 9) bleached wheat grain (Cb), referring to wheat grain that has been exposed to environmental conditions causing them to lose their natural color, resulting in a paler, often whitish appearance, expressed in percentage (%).

Figure 2 provides a comprehensive multivariate analysis of weather parameters and wheat traits over several years. The first two principal components (Dims), which have eigenvalues greater than one, account for 88.58% of the total variance after varimax rotation. Dim1 explains 59.24% of the variance, while Dim2 accounts for 29.34% ( Figure 2A ). Dim1 is positively correlated (>0.5) with most major weather parameters except for precipitation (PRECTOT), wind speed (WS), and dew-point temperature (T2MDEW), which are positively correlated with Dim2 ( Supplementary Table S3 ). This analysis highlights significant climatic variability between growing seasons. A positive correlation was identified among minimum temperature (Tmin), fraction of radiation use efficiency (FRUE), vapor pressure deficit (VPD), maximum temperature (Tmax), mean temperature (Tmed), growing degree days (GDD), evapotranspiration (ETP), sunshine duration (N), and radiation at the top of the atmosphere (RTA). Conversely, these climate parameters showed a negative correlation with total precipitation (PRECTOT), precipitation deficit (PETP), and relative humidity (RH) ( Supplementary Table S4 ).

Based on the hierarchical cluster analysis of 17 weather parameters, four distinct groups of growing seasons were identified, each characterized by specific climatic conditions ( Figure 2B ). Group 1 (Y_2021) is marked by high values for most weather parameters, including maximum and minimum temperatures (Tmax, Tmin), average temperature (Tmed), vapor pressure deficit (VPD), evapotranspiration (ETP), fractional radiation use efficiency (FRUE), radiation absorption (RTA), growing degree days (GDD), daylight hours (N), actual duration of sunshine (n), and insolation incident (SIHS), further signifying very warm and dry conditions. Group 2, comprising the seasons Y_2014, Y_2020, Y_2017, Y_2013, and Y_2015, shows lower temperatures and higher precipitation (PRECTOT), relative humidity (RH), and a precipitation deficit (PETP), suggesting a cooler environment with favourable water availability. Group 3 (Y_2022) is distinguished by high values for dew-point temperature (T2MDew), the effect of temperature on radiation use efficiency (FRUE), sunshine duration (N), radiation at the top of the atmosphere (RTA), Growing Degree Day (GDD), minimum and mean temperatures (Tmin, Tmed), and thermal infrared (longwave) radiative flux (DTIRF). These high values indicate a warm environment during this growing season. Finally, Group 4 (Y_2018, Y_2019, Y_2016) is characterized by higher wind speed (WS), evapotranspiration (ETP), sunshine duration (N), and insolation incident (SIHS), along with moderate temperatures, precipitation, and humidity, delineating an intermediate environment.

During the warm years (2021 and 2022), higher vapor pressure deficit (VPD) values were frequently observed, particularly during the heading and maturity stages, with the maximum VPD reaching 2.83 kPa day⁻¹ ( Figure 3A ). High values of maximum temperature (Tmax) ranging from 22.4°C to 35.9°C were evident during almost the entire heading and maturity stages ( Figure 3C ). Both high VPD and Tmax values, causing heat and drought stress, can impair photosynthesis, reduce pollen viability, decrease grain setting, and ultimately impact yield. The precipitation deficit (PETP) data during these years show notable deficits, with values ranging from −13.9 to −9.35 mm day⁻¹ observed on approximately more than 80% of the days during the heading and maturity stages ( Figure 3B ). This exacerbates the effect of both heat and drought stress, and consequently reducing grain yield.

In contrast, cooler years (2013, 2014, 2015, 2017, and 2020) exhibited lower VPD and Tmax values during almost 80% of the heading stage, with values ranging from 0.38 to 0.82 kPa day⁻¹ and 15.5 to 22.4°C, respectively ( Figures 3A, C ). This suggests reduced stress from high temperatures and dry air. These conditions are more conducive to optimal plant development and grain filling. Additionally, in these years, PETP values ranging from −1.83 to 13.4 mm day−¹ were displayed on 13% of days during the heading and maturity stages, indicating a certain availability of water in these two critical phenological stages ( Figure 3B ). Higher relative humidity (RH) observed during the cooler years, particularly at the emergence and tillering stages, showed maximum frequency distributions between 78.7% and 95.8% ( Figure 3D ), which can foster better early growth and establishment.

The likelihood-ratio test (LRT) revealed that only six durum wheat traits (GY, Sp, TGW, Alt, Pe, and Cb) had significant effects of environment, genotype, and genotype by environment interaction effects ( Table 3 ). Additionally, all traits, except gluten, grain protein, and shriveled wheat grain, showed significant genotypic effects ( Table 3 ). The genotype × environment interaction was significant for all traits, demonstrating the varying responses of different varieties to distinct growing seasons ( Table 3 ).

In the analysis of durum wheat traits, it was observed that, with the exception of gluten index, grain protein content, and shriveled wheat grain, the environmental variance (σ_e ²) contributed substantially more to the total phenotypic variance (σ^2P) than any other traits examined ( Figure 4 ). The genotype-by-environment interaction (GEI) variance was stronger on shriveled wheat grain traits, accounting for 50% of the overall phenotypic variation. For gluten content and grain protein, the residual variance (σ_r ²) was more substantial than the other variance components ( Figure 4 ).

The distribution of grain yield for each variety is illustrated in Figure 5 . Wheat grain yield varied significantly across years and varieties, with an average yield of 3.42 ± 1.66 t ha⁻¹ among the 41 tested varieties. The lowest yield was recorded in 2021 at 1.66 t ha⁻¹, while the highest yield was observed in 2014 at 6.09 t ha⁻¹ ( Supplementary Figure S1 ). Grain yield (GY) ranged from 1.09 t ha⁻¹ (variety G75 in 2018) to 7.22 t ha⁻¹ (variety G65 in 2014), with a coefficient of variation (CV) of 48.61%. The coefficient of variation for durum wheat traits between varieties over the ten growing seasons showed maximum variability for traits such as bleached wheat grain (113.07%), shriveled wheat grain (97.8%), gluten content (32.72%), spikes per m² (31.43%), and grain protein (26.15%). Minimum variability was observed for hectoliter weight (6.1%), thousand-grain weight (13.45%), and plant height (14.53%).

The broad-sense heritability (H²) values for all traits were generally low, ranging from 0 for shriveled wheat grain to 0.19 for thousand-grain weight, indicating that most of the variation in these traits is likely due to environmental or other non-genetic factors. Conversely, the estimate heritability of the mean genotype (h²_mg) varied, being low for gluten index, grain protein, and shriveled wheat grain (ranging from 0 to 0.32), moderate for grain yield, ears per square meter, height, and bleached wheat grain (ranging from 0.48 to 0.66), and high for thousand-grain weight and hectoliter weight (ranging from 0.80 to 0.86).

Regarding Piepho heritability (H² Piepho), only thousand-grain weight and ears per square meter showed moderate heritability (ranging from 0.54 to 0.57), while bleached wheat grain exhibited high heritability (0.76), with the remaining traits showing low heritability. For Cullis heritability (H² Cullis), five traits – gluten index, grain protein, height, ears per square meter, and bleached wheat grain – had moderate heritability (ranging from 0.41 to 0.64), with thousand-grain weight showing high heritability (0.72), and the remaining traits displaying low heritability. These differing heritability estimates highlight the complex nature of heritability and the importance of considering multiple perspectives when evaluating genetic contributions to trait variation.

The selection accuracy (AS) exceeded 0.70 for all traits except gluten index, protein content, and shriveled wheat grain. This high selection accuracy indicates reliable identification and selection of superior genotypes for breeding, ensuring effective genetic improvement in these traits.

The correlation of genotype–environment interaction effects (rge) varied in magnitude, being low for gluten index, grain protein, and thousand-grain weight (ranging from 0.08 to 0.39), moderate for grain yield, spikes per square meter, height, and ears per square meter (ranging from 0.42 to 0.53), and high for shriveled wheat grain and bleached wheat grain (ranging from 0.75 to 0.84). The genotypic coefficient of variation (CVg) exhibited a wide range, from 0% for Cs to 22.7% for Cb, demonstrating significant genetic diversity among the varieties. Conversely, the residual coefficient of variation (CVr) ranged from 1.91% for ears per square meter (Pe) to 31.2% for shriveled wheat grain (Cs). Additionally, the ratio between the genotypic and residual coefficients of variation (CVratio) was below 1 for all traits except for bleached wheat grain (Cb).

Due to the data being unbalanced, Sampaio Filho et al. (2023) recommend evaluating performance and genotypic stability using factorial analysis, and they suggest focusing on the deviations from regression (S² _di), the coefficient of determination (R²), and the root mean square error of the regression (RMSE). This approach is preferred over the originally proposed WAASB index method. According to Table 4 , three factor components (FA) were identified, explaining 68.61%, 63.87%, and 69.55% of the total variance for S² _di, R², and RMSE, respectively, in durum wheat traits.

In our factor analysis, we evaluated the communality (h²ⁱ) and uniqueness (u²ⁱ) of each variable to understand how well they are represented by the common factors. Communality indicates the proportion of a variable’s variance explained by these factors, while uniqueness shows the variance not explained by them. The average h²ⁱ values were 0.69 for S² _di, 0.64 for R², and 0.70 for RMSE. For S² _di, all durum wheat traits had h²ⁱ values greater than 0.5 except for TGW and Cb. For R² and RMSE, all traits had h²ⁱ values above 0.5 except for Alt and TGW. This indicates that most traits are well represented by the common factors in our analysis.

The nine durum wheat traits were grouped into three factors (FA) for each stability parameter. For S² _di, the first factor (FA1) showed a strong positive correlation with important quality traits, such as the gluten index (0.92) and protein content (0.90), and a moderate correlation with bleached wheat grain (0.61). The second factor (FA2) included traits related to plant productivity, like grain yield (0.94) and ears per m² (0.84), and some quality traits such as thousand-grain weight (0.50) and hectoliter weight (0.48). The third factor (FA3) had a strong negative correlation with the quality trait shriveled wheat grain (−0.88) and a significant positive correlation with the morphological trait of plant height (0.70).

For R² and RMSE parameters, FA1 was associated with quality traits, including protein content (−0.92), gluten index (−0.91), and thousand-grain weight (−0.59) for R², and gluten index (−0.92), protein content (−0.89), bleached wheat grain (−0.61), and thousand-grain weight (−0.50) for RMSE. The second factor (FA2) grouped productive traits (grain yield (−0.80) and ears per m² (−0.83)) and the morphological trait of plant height (0.62) for R², and productive traits (grain yield (−0.86) and ears per m² (−0.91)) for RMSE. Finally, FA3 showed a high correlation with quality traits (bleached wheat grain (0.67) and shriveled wheat grain (−0.80)) for R², and with the morphological trait plant height (−0.58), and quality traits such as shriveled wheat grain (0.83) and protein content (0.72) for RMSE.

Figure 6 illustrates Pearson’s correlation coefficients between various agronomic and quality traits. These correlations illustrate the relationships among traits using the mean performance and stability (MPS) index, which was established based on three stability parameters (S² _di, R², and RMSE). Approximately 25% of the correlations were significant for both R² and RMSE, while 30.56% were significant for S² _di. Traits show varying degrees of correlation depending on the stability parameter considered. Notably, protein content and gluten index exhibited a highly positive correlation (>0.88) across all stability parameters. Additionally, grain yield and spike per square meter (Sp) demonstrated a moderate positive correlation, with coefficients ranging from 0.52 to 0.70. There was a low positive correlation between shriveled wheat grain (Cs) and hectoliter weigh (Pe), with values ranging from 0.33 to 0.43. Conversely, plant height showed a low negative correlation with Cs for S² _di (−0.36), Pe for RMSE (−0.33), and Sp for R² (−0.34). Thousand-grain weight (TGW) exhibited low positive correlations with several traits: grain yield (0.40), Pe (0.37), and gluten index (0.35) for S² _di. For RMSE, TGW was positively correlated with Pe (0.46), and for R², it correlated positively with protein content (0.39) and gluten index (0.35). Pe also had low positive correlations with grain yield (0.35) and protein content (0.38) for R², with protein content (0.35) for RMSE, and with grain yield (0.38), Sp (0.38), and protein content (0.44) for S² _di. Additionally, a low but significant positive correlation between bleached wheat grain (Cb) and both protein content and gluten index (0.32 and 0.40, respectively) was observed for the RMSE parameter.

The ranking of the 41 durum wheat varieties based on the Multi-Trait Index (MTMPS) analysis, with a 15% selection intensity, is illustrated in Figure 7 . Among the 41 varieties evaluated, six were selected for each of the three regression variance parameters (S² _di, R², and RMSE). Regarding the R² parameter, the varieties with lower MTMPS values were Core (G18), Furio Camillo (G31), Antalis (G4), Bering (G9), Dylan (G24), Gibraltar (G32), and Anco Marzio (G3) ( Figure 7A ). For the RMSE parameter, Core (G18), Aureo (G7), Svevo (G76), Antalis (G4), Bering (G9), and Marco Aurelio (G49) were selected ( Figure 7B ). In the S² _di parameter, the varieties with lower MTMPS values were Core (G18), Aureo (G7), Marco Aurelio (G49), Antalis (G4), Furio Camillo (G31), Claudio (G14), and Svevo (G76) ( Figure 7C ). These six varieties selected across three regression parameters demonstrated satisfactory mean grain yield performance. Specifically, their yields ranged from 3.41 t/ha in R², to 3.91 t/ha in RMSE, and 3.61 t/ha in S² _di. Additionally, these varieties exhibited better stability in grain yield across different growing seasons, highlighting their resilience and consistency under varying environmental conditions.

The multi-trait mean performance and stability index (MTMPS) proved highly effective in selecting traits with selection gain (SG) for all durum wheat traits, achieving a success rate of 77.77% (7 out of 9 traits) based on RMSE and R², and 88.89% (8 out of 9 traits) based on S² _di, with a few exceptions. Notably, the trait Cb showed reductions of 4.95%, 10.17%, and 13.55% for S² _di, RMSE, and R², respectively, indicating its favorable desirability as a wheat trait in this context. Specifically, the trait Cb exhibited reductions of 4.95%, 10.17%, and 13.55% for S² _di, RMSE, and R², respectively, which is desirable wheat trait in this context. Additionally, Cs showed a slight reduction of 0.04% for S² _di. Conversely, the trait Alt experienced increased gains of 0.33%, 1.21%, and 1.50% for RMSE, S² _di, and R², respectively, which are not suitable for mechanized harvesting. For the parameters R² and RMSE, Cs increased by 1.92% and 4.46%, respectively, which negatively impacts durum wheat quality. Regarding S² _di, the selection gain percentages for traits with desirable values ranged from 0.70% for Pe to 3.65% for TGW ( Table 5 ). Similarly, the MTMPS index gains for the R² and RMSE parameters followed a comparable pattern, with increases in selection gains for six desirable wheat traits. For R², the gains ranged from 1.28% for Pe to 6.08% for Sp, while for RMSE, the gains ranged from 0.19% for Pe to 3.67% for Glu ( Table 5 ).

Figure 8 illustrates the strengths and weaknesses of selected durum wheat varieties identified over ten consecutive growing seasons. Based on the R² parameter, the “G31” variety exhibits strengths in FA1, indicating high performance in quality wheat traits such as protein (Pro), gluten (Glu), thousand-grain weight (TGW), and plant height (Pe). The variety “G4” shows strengths associated with FA2, which includes grain yield (GY), ears per m² (Sp), and plant height (Alt), while the variety “G9” excels in FA3, encompassing traits like shriveled wheat grain (Cs) and bleached wheat grain (Cb) ( Figure 8A ). For the RMSE parameter, FA1, which primarily relates to Pro, Glu, TGW, and Cb, significantly influences the deviation of the “G49” variety from the ideotype, suggesting areas for improvement in these traits. Meanwhile, the “G4” variety demonstrates strengths in both FA2 (GY and Sp) and FA3 (Alt, Pe, and Cs), indicating a well-rounded performance across these factors ( Figure 8B ). Regarding the S² _di parameter, the “G18” variety performs very well in FA1, showing strong traits in protein (Pro), gluten (Glu), and bleached wheat grain (Cb). The “G14” variety shows robustness in FA2, associated with GY, Sp, and Pe. On the other hand, the “G31” variety presents a positive gain in traits related to Alt and Cs ( Figure 8C ).