Introduction
Chickpea, Cicer arietinum L., is a globally significant legume, ranking as the second most important food legume after dry beans (FAOSTAT, 2025). Its prominence stems from its dual role in human nutrition and sustainable agriculture. Chickpeas are a rich source of protein, making them a vital component of diets in many parts of the world, particularly in regions where animal protein is scarce or expensive. Beyond its nutritional value, chickpea plays a crucial role in agricultural systems due to its ability to fix atmospheric nitrogen, which enhances soil fertility and reduces the need for synthetic fertilisers. This nitrogen-fixing capability makes chickpea an ideal rotational crop in cereal-pulse systems, contributing to the sustainability of farming practices (Rani and Krishna Reference Rani and Krishna2016; Liu et al., Reference Liu, Bandara, Hamel, Knight and Gan2020; Palmero et al., Reference Palmero, Fernandez, Garcia, Haro, Prasad, Salvagiotti and Ciampitti2022).
Chickpea is predominantly cultivated as a cool-season crop, thriving in climates ranging from Mediterranean to subtropical and tropical regions. The 2023, global chickpea production reached approximately 16.5 million t, harvested from nearly 14.1 million ha (FAOSTAT, 2025). Australia, a major player in the global chickpea market, is the largest exporter of desi chickpeas, the crop ecotype widely used in traditional dishes across South Asia. The crop is well-suited to medium-rainfall regions (300–500 mm), where it exhibits slow growth during the cold winter months, followed by accelerated growth in spring as temperatures rise. The area under chickpea cultivation in Australia has expanded significantly, to 1.039 million ha in 2024-25, driven by favourable grain prices and its role as a profitable break crop in cereal rotations. However, despite its economic and agronomic benefits, the average yield of chickpea in Australia remains relatively low at 1.35 t/ha over the last 10 years (2015-16 to 2024-25), primarily due to challenges posed by diseases and abiotic stresses such as drought, frost and heat stress (GRDC, 2011; ABARES, 2024).
Globally, chickpea yields have stagnated at around 1 t/ha, with productivity gains lagging behind those of other winter crops (Joshi and Rao Reference Joshi and Rao2017). Abiotic stresses, particularly drought, are major constraints on chickpea productivity, limiting plant growth, distribution and yield (Garg et al., Reference Garg, Shankar, Thakkar, Kudapa, Krishnamurthy, Mantri, Varshney, Bhatia and Jain2016; Saini et al., Reference Saini, Das, Adhikary, Kumar, Singh, Nayyar and Kumar2022). Drought, exacerbated by climate change, threatens global food security, with chickpea being particularly vulnerable due to its sensitivity to water availability. In Australia, where chickpea is a cornerstone of the pulse industry, the crop’s productivity is highly susceptible to climatic variations, especially during critical growth phases. This vulnerability underscores the urgent need for adaptive strategies to mitigate the impacts of environmental stresses and enhance chickpea yields.
Chickpea’s sensitivity to environmental variability, including temperature, precipitation and other climatic factors, is particularly pronounced during key growth stages. After germination, temperature, photoperiod and soil moisture availability collectively influence the progression through various phenological stages. Among these, flowering is a critical phase, as the environmental conditions during this period and the length of the reproductive phase significantly affect pod formation and final yield (Lake and Sadras Reference Lake and Sadras2014; Peake et al., Reference Peake, Dreccer, Whish and Hochman2020; Graham et al., Reference Graham, Raman, Warren and Anwar2022). Understanding and predicting the complex interactions between environmental conditions, crop growth stages and yield outcomes is essential for improving productivity and resource use efficiency. Temperature, solar radiation and water availability significantly influence chickpea yield and water-use efficiency (WUE) (Sadras and McDonald Reference Sadras and McDonald2011; Siddique et al., Reference Siddique, Johansen, Turner, Jeuffroy, Hashem, Sakar, Gan and Alghamdi2012). These interactions are further modulated by the specific growth stages of the crop, highlighting the need for targeted analyses to identify periods of heightened sensitivity.
Despite significant advancements in agricultural research, there remains a critical knowledge gap in quantifying how combinations of environmental factors and growth stages affect chickpea yield and WUE. Addressing this gap could lead to improved management strategies and adaptive practices, particularly in the face of changing climate conditions. For instance, understanding the impact of cold and heat stress during flowering or the role of soil moisture during podding could inform irrigation schedules, planting dates and varietal selection for mitigating stress and enhancing crop productivity. Therefore, to uncover the most significant drivers of yield and WUE, we split the chickpea growing season phenologically into four intervals: sowing to flowering (sf), flowering to podding (fp), podding to maturity (pm) and critical period (cp) thus isolating the sensitive periods (fp and cp) where the chickpea is prone to pod abortion from sub- and supra-optimal temperatures (Peake et al., Reference Peake, Dreccer, Whish and Hochman2020). In the vegetative phase (sf), the crop accumulates biomass to maximise productivity, while during the seed-filling stage, there is a balance between vegetative and reproductive growth to maximise productivity and WUE.
This study aimed to bridge these knowledge gaps by focusing on four key objectives. Firstly, it sought to quantify the relationships between combinations of environmental factors and chickpea growth stages in relation to grain yield and WUE. Secondly, the study employed correlation analysis and Exclusive LASSO (eLASSO) regression to identify the most influential environmental factors in different growth stages. Thirdly, predictive models were developed to forecast chickpea yield and WUE under varying environmental conditions. Finally, the research aimed to provide a practical tool for farmers and agronomists, enabling informed decision-making to optimise and expand chickpea production.
Materials and methods
Study area and agronomy
The workflow for this investigation – a meta-analysis of nineteen environmental variables across four growth stages of chickpeas – is summarised in Figure S1. Data were collected from 29 field experiments at 10 sites between 2013 and 2019 (with multiple sowing times per year at some sites) in South Australia (SA), Victoria (VIC), New South Wales (NSW) and Queensland (QLD), representing diverse agroclimatic conditions in Australia’s chickpea regions (see Table 1 and Figure 1). Detailed information on years, sowing times, cultivars, agronomic practices and experimental designs is available in prior studies (Anwar et al., Reference Anwar, Luckett, Chauhan, Ip, Maphosa, Simpson, Warren, Raman, Richards, Pengilley, Hobson and Graham2022; Chauhan et al., Reference Chauhan, Anwar, Richards, Lake, Sadras, Luckett, Raman, Krosch and Graham2023). The experiments followed the National Variety Trials (NVT) protocols, including guidelines for crop establishment and pest control (NVT Online, 2025). Although procedures were generally consistent, variations in row spacing, sowing depth, plant density and harvested area were noted. Specific details can be found in Table 1 of Anwar et al. (Reference Anwar, Luckett, Chauhan, Ip, Maphosa, Simpson, Warren, Raman, Richards, Pengilley, Hobson and Graham2022).
Table 1. Weather data spanning 75 years (1950–2024) for 10 experimental sites (Fig. 1), summarised for both long-term (LT) and short-term (ST) periods (years with experiments). Data covers the growing season (April 1–October 31) and was sourced from SILO (https://longpaddock.qld.gov.au/silo/; Jeffrey et al., Reference Jeffrey, Carter, Moodie and Beswick2001)
Figure 1. The ten field experimental sites (black dot) in southeastern Australia used in this study.
Phenological stages (flowering and podding) were recorded when 50% of the plants exhibited at least one open flower or visible pod. Yield was measured at physiological maturity via machine harvest and expressed in kg/ha. Daily weather data, including temperature, rainfall and solar radiation, were primarily sourced from the SILO database (https://legacy.longpaddock.qld.gov.au/silo/about.html; Jeffrey et al., Reference Jeffrey, Carter, Moodie and Beswick2001), with some locations using on-site climate data. Soil properties for each experimental site are detailed in Anwar et al. (Reference Anwar, Luckett, Chauhan, Ip, Maphosa, Simpson, Warren, Raman, Richards, Pengilley, Hobson and Graham2022) and Chauhan et al. (Reference Chauhan, Anwar, Richards, Lake, Sadras, Luckett, Raman, Krosch and Graham2023).
Growing periods and weather indices
The growth stages of chickpea – emergence, flowering, pod set and physiological maturity – are crucial for grain yield and water use efficiency. Abiotic stresses like heat and drought during these stages can significantly affect crop performance (Soltani and Sinclair, Reference Soltani and Sinclair2011; Devasirvatham and Tan, Reference Devasirvatham and Tan2018; Bicard et al., Reference Bicard, Faucon, Pedas, Vequaud, Pin, Elmerich and Lange2025). This study defines four key crop growth periods (Lake and Sadras, Reference Lake and Sadras2014; Bicard et al., Reference Bicard, Faucon, Pedas, Vequaud, Pin, Elmerich and Lange2025):
a. SF (Sowing to Flowering): The period from sowing to the onset of flowering.
b. FP (Flowering to Pod): The interval between flowering and pod formation.
c. PM (Pod to Maturity): The period from pod formation to maturity.
d. CP (Critical Period): Defined as 300°Cd before flowering to 500°Cd after flowering, based on the thermal time concept.
These growth periods may overlap, requiring careful interpretation. The growth periods were calculated for each genotype based on observed phenology. To assess environmental conditions during these periods, 19 weather indices were calculated, capturing temperature, water availability and solar radiation aspects that influence chickpea growth and yield. Key indices include:
1) H30: Number of days with maximum temperatures ≥ 30°C, indicating heat stress.
2) H35: Number of days with maximum temperatures ≥ 35°C, representing extreme heat stress.
3) F: Frost frequency, measured as the number of days with minimum temperatures ≤ 0°C.
4) sumTT: Cumulative thermal time (°Cd), calculated as the sum of daily temperatures above a base temperature.
5) SE: Cumulative soil evaporation (mm), representing water loss from the soil surface.
6) PET: Cumulative potential evapotranspiration (mm), estimated using the Priestley-Taylor method.
7) SDR: Accumulated water deficit ratio, calculated as the ratio of water supply to water demand.
8) RAIN: Cumulative rainfall (mm) during the growth period.
9) ET: Cumulative crop water use (mm), including both transpiration and soil evaporation.
10) T: Cumulative crop transpiration (mm), excluding soil evaporation.
11) meanVPD: Mean vapour pressure deficit (kPa), indicating atmospheric demand for water.
12) meanPTQ: Mean photothermal quotient (MJ/m²/°Cd), calculated as the ratio of photosynthetically active radiation (PAR) to mean temperature.
13) meanPTQvpd: Mean photothermal quotient corrected by vapour pressure deficit (MJ/m²/°Cd/kPa).
14) meanPAR: Mean photosynthetically active radiation (MJ/m²/day), derived by multiplying global solar radiation by 0.47 (Pinker and Laszlo, 1992).
15) DL: Cumulative day length (hours:minutes).
16) RADN: Cumulative solar radiation (MJ/m²/day).
17) MINT: Mean daily minimum temperature (°C).
18) MAXT: Mean daily maximum temperature (°C).
19) meanPZT: Mean temperature corrected for photoperiod (°C).
Vapour pressure deficit (VPD) was calculated from saturated and actual vapour pressure (Zeleke et al., Reference Zeleke, Anwar, Emebiri and Luckett2023), and the photothermal quotient (PTQ) was derived from cumulative photosynthetically active radiation (PAR) over mean temperature (Fischer, Reference Fischer1985; Soltani and Sinclair, Reference Soltani and Sinclair2011). The PTQ adjusted for VPD (PTQvpd) was calculated by dividing PTQ by mean VPD. The photoperiod-adjusted temperature (meanPZT) was determined using the approach from Gallagher et al. (Reference Gallagher, Biscoe, Dennis-Jones, Wright and Wynn-Williams1983) and Verghis et al. (Reference Verghis, McKenzie and Hill1999). The saturated vapour pressure and actual vapour pressure was estimated for maximum (Tmax) and minimum (Tmin) temperatures using the equations provided by Dreccer et al. (Reference Dreccer, Fainges, Whish, Ogbonnaya and Sadras2018) and Jeffrey et al. (Reference Jeffrey, Carter, Moodie and Beswick2001). Equations to compute VPD, PTQ, PTQvpd and meanPZT, respectively, are given in the Supplementary material.
Soil water balance
Soil water balance is crucial for crop growth and yield as it determines the availability of water for plants. Key components include rainfall, evapotranspiration (ET), runoff and drainage (Unkovich et al., Reference Unkovich, Baldock and Farquharson2018; Unkovich et al., Reference Unkovich, McBeath, Moodie and Macdonald2023). This study calculates the soil water balance using temperature, rainfall and simulated soil water content at sowing and harvest, following He and Wang (Reference He and Wang2019).
We used the Agricultural Production Systems sIMulator (APSIM) to simulate soil water dynamics. APSIM is a validated model for crop growth, water cycling and nutrient dynamics (Holzworth et al., Reference Holzworth, Huth, deVoil, Zurcher, Herrmann, McLean, Chenu, van Oosterom, Snow, Murphy, Moore, Brown, Whish, Verrall, Fainges, Bell, Peake, Poulton, Hochman, Thorburn, Gaydon, Dalgliesh, Rodriguez, Cox, Chapman, Doherty, Teixeira, Sharp, Cichota, Vogeler, Li, Wang, Hammer, Robertson, Dimes, Whitbread, Hunt, van Rees, McClelland, Carberry, Hargreaves, MacLeod, McDonald, Harsdorf, Wedgwood and Keating2014; APSIM, 2023). The Soil Water module was parameterised using previous research data (Liu et al., Reference Liu, Anwar, O’Leary and Conyers2014; Zeleke and Nendel, Reference Zeleke and Nendel2019; Wang et al., Reference Wang, Liu, Asseng, Macadam and Yu2017; Xing et al., Reference Xing, Liu, Li, Wang, Anwar, Crean, Lines-Kelly and Yu2017; Anwar et al., Reference Anwar, Luckett, Chauhan, Ip, Maphosa, Simpson, Warren, Raman, Richards, Pengilley, Hobson and Graham2022). To estimate the initial soil water content for the 2013–2019 experimental period, we ran APSIM from January 1, 2000, assuming soil water content was equal to the lower limit of plant-available water (LL15) (He and Wang, Reference He and Wang2019).
Total water use (WU) for crops, measured as ET, was determined by the difference between starting soil moisture at planting and final soil moisture at harvest, plus total irrigation and rainfall during the growing season. Water use efficiency (WUE) is defined as crop yield (Y, kg/ha) per unit of water lost through ET (mm). Transpiration (T), which excludes soil evaporation, was calculated similarly (Yang et al., Reference Yang, Liu, Anwar, O’Leary, Macadam and Yang2016). Equations to compute ET, T and WUE using ET and T, respectively, are given in the Supplementary material.
APSIM also calculates daily potential evapotranspiration (PET) using the Priestley-Taylor method, based on the relationship between crop yield and ET (Paredes et al., Reference Paredes, Rodrigues, Alves and Pereira2014; Trout and DeJonge, Reference Trout and DeJonge2017; Akumaga and Alderman, Reference Akumaga and Alderman2019).
Water supply-demand ratio (SDR)
The APSIM model computes a water-deficit index (Chapman et al., 19193; Chenu et al., Reference Chenu, Cooper, Hammer, Mathews, Dreccer and Chapman2011), also known as the ‘water supply’ and ‘water demand’ ratio, which indicates how well the water extractable by a crop’s roots (water supply) meets the crop’s potential transpiration needs (water demand). The water supply is calculated for each soil layer with roots, depending on root growth and soil properties. Water demand is estimated daily based on crop growth and atmospheric conditions. The water supply-demand ratio (SDR) ranges from 0 to 1, reflecting water stress levels in plants. An SDR of 1 indicates no stress, while a lower value indicates stress. The water supply-demand ratio (SDR) is the ratio between water supply and water demand, bounded between 0 and 1, which indicates if the plant is water-stressed. Equations 8 and 9 in the Supplementary material describe how SDR and water deficiency was calculated.
Exclusive least absolute shrinkage and selection operator (LASSO)
The technique ‘least absolute shrinkage and selection operator’ (LASSO) introduced by Tibshirani (Reference Tibshirani1996) is a variable selection method that ‘shrinks’ some of the regression coefficients to zero during the estimation procedure and aims to retain only the essential features, leading to a more interpretable model. Recent applications of LASSO in agricultural studies include Anwar et al. (Reference Anwar, Emebiri, Ip, Luckett, Chauhan and Zeleke2024) and Heilemann et al. (Reference Heilemann, Klassert, Samaniego, Thober, Marx, Boeing, Klauer and Gawel2024). Since its introduction, LASSO has been extended in various ways to handle different data structures, including situations where the predictor variables can be divided into several groups. For example, a group of indicator variables are often used to represent a multi-level categorical variable collectively. In genomic analysis, for example, several genes may be treated as a group if they belong to the same pathway. Group LASSO (Bakin, Reference Bakin1999; Yuan and Lin, Reference Yuan and Lin2006) is used to select the most important groups of variables without focusing on the selection of individual variables. If one is interested in selecting both the important groups and the important variables within the groups, a bi-level selection method such as those proposed by Huang et al. (Reference Huang, Ma, Xie and Zhang2009) and Breheny and Huang (Reference Breheny and Huang2009) is needed. A review of group LASSO and bi-level selection methods can be found in Huang et al. (Reference Huang, Breheny and Ma2012). Since both group LASSO and bi-level selection methods aim to select the most representative groups of variables, some variable groups may be missing in the final model, which may not be desirable in some circumstances. The method of exclusive LASSO (eLASSO), recently introduced by Campbell and Allen (Reference Campbell and Allen2017), selects at least one variable from each group, thus maintaining all variable groups in the final model. Campbell and Allen (Reference Campbell and Allen2017) reported that eLASSO tends to select the correct number of variables even when the explanatory variables within and across groups are correlated. This property is essential since multicollinearity often exists in agricultural studies. Like all other LASSO-based methods, eLASSO seeks to solve a constrained optimisation problem. The objective function involves a penalty term (λ) that governs the number of non-zero coefficients. Under eLASSO, with a sufficiently large λ, there will be exactly one non-zero coefficient per group (that is, only one explanatory variable per group will remain), representing the most parsimonious model possible under such a method. During the shrinking process, biases are inevitably introduced to the regression coefficients. Therefore, the predictive power of eLASSO models may not be as high as that of other models. Yet, eLASSO models offer a unique way for users to identify the most important explanatory variable(s) within and across variable groups.
In this study, the growth periods were considered the variable groups and the 19 weather indices were considered as the explanatory variables. All explanatory variables were standardised by subtracting the mean and further dividing by the standard deviation before analysis. Two eLASSO models were developed for each of the response variables:
Model 1: Selected by minimising the Bayesian Information Criterion (BIC), balancing interpretability and goodness of fit (Schwarz, Reference Schwarz1978).
Model 2: Selected using the minimum penalty parameter (λ) that retained one variable from each group, prioritising simplicity and identifying key factors.
The chief intention of including Model 2 was to identify the single most dominant variable per growth stage under the eLASSO constraint, rather than for prediction. The performance of Model 2 also provides a contrast to the more complex but better-performing BIC-selected Model 1.
eLASSO model performance evaluation
We compared the observed and the eLASSO-modelled chickpea yield, water use efficiency based on transpiration (WUE_T) and water use efficiency based on evapotranspiration (WUE_ET) with least square linear regressions, including the coefficient of determination (R2, the amount of variation explained by the model). We tested the goodness of fit by calculating the Root Mean Square Error (RMSE) (Piñeiro et al., Reference Piñeiro, Perelman, Guerschman and Paruelo2008) using Equation 10 given in Supplementary material. Additionally, we expressed the normalised root means square error (NRMSE) as precision parameters given in Equation 11 and the Willmott index (Willmott, Reference Willmott1982) using Equation 12 (Supplementary material).
The analyses were carried out using the ExclusiveLasso package in R (Weylandt et al., Reference Weylandt, Campbell and Allen2018). We used R version 4.4.2 (R Core Team, 2024) and the following R packages: correlationfunnel v. 0.2.0 (Dancho, Reference Dancho2020), emmeans v. 1.10.5 (Lenth, Reference Lenth2024), ExclusiveLasso v. 0.0 (Weylandt et al., Reference Weylandt, Campbell and Allen2018), ggpmisc v. 0.6.1 (Aphalo, Reference Aphalo2024), ggpubr v. 0.6.0 (Kassambara, Reference Kassambara2023), glmnet v. 4.1.8 (Friedman et al., Reference Friedman, Hastie and Tibshirani2010; Simon et al., Reference Simon, Friedman, Hastie and Tibshirani2011; Tay et al., Reference Tay, Narasimhan and Hastie2023), grateful v. 0.2.10 (Rodriguez-Sanchez and Jackson, Reference Rodriguez-Sanchez and Jackson2023), gt v. 0.11.1 (Iannone et al., Reference Iannone, Cheng, Schloerke, Hughes, Lauer, Seo, Brevoort and Roy2024), janitor v. 2.2.0 (Firke, Reference Firke2023), MESS v. 0.5.12 (Ekstrøm, Reference Ekstrøm2023), naniar v. 1.1.0 (Tierney and Cook, Reference Tierney and Cook2023), patchwork v. 1.3.0 (Pedersen, Reference Pedersen2024), qs v. 0.27.2 (Ching, Reference Ching2024), remotes v. 2.5.0 (Csárdi et al., Reference Csárdi, Hester, Wickham, Chang, Morgan and Tenenbaum2024), reshape2 v. 1.4.4 (Wickham Reference Wickham2007), tidyverse v. 2.0.0 (Wickham et al., Reference Wickham, Averick, Bryan, Chang, McGowan, François, Grolemund, Hayes, Henry, Hester and Kuhn2019), running in RStudio v. 2024.9.1.394 (Posit team, 2024).
Results
Weather data
Significant variations in ambient temperature and in-season rainfall were observed across ten diverse locations in the south-eastern Australian cropping belt (Table 1, Figure 1 and Supplementary Table S5). Mean maximum temperatures ranged from 15.8°C to 22.7°C, with Narrabri recording the highest, while Wagga Wagga experienced the lowest. In contrast, minimum temperatures varied from 1.3°C to 9.4°C. Rainfall patterns also displayed considerable differences. Breeza, the driest site, received only 71 mm of in-season rainfall. In contrast, Wagga Wagga, the wettest, recorded up to 300 mm. Notably, all experimental sites experienced substantially lower growing-season rainfall during the experimental years compared to their long-term averages. For instance, Breeza’s rainfall during the experiment was 80.7 mm, significantly lower than its 303.7 mm long-term average. While minimum temperatures remained relatively consistent between short-term experimental periods and long-term averages, with slight increases of over 1°C at Trangie and Yanco, maximum temperatures showed a clear upward trend in the short term, exceeding long-term means across all sites. Trangie, for example, experienced a temperature increase of more than 2°C. Furthermore, the frequency of frost days was notably higher during the experimental years, particularly at Breeza and Wagga Wagga. These findings collectively suggest that the experiments were conducted during an unusually warm and dry period in the region, characterised by reduced cloud cover, which likely contributed to the increased frequency of frost days. Such climatic conditions could potentially impact chickpea flowering and podding.
Statistics of response variables
The summary descriptive statistics for chickpea yield and water use efficiency (WUE) based on two different measures: transpiration (WUE_T) and evapotranspiration (WUE_ET) are given in Table 2. These data are derived from 438 observations across ten field experimental sites in southeastern Australia. Yields varied considerably, ranging from 113 to 4128 kg/ha, with an average of 1725 kg/ha. WUE_T ranged from 0.85 to 30.8 kg grain/ha/mm, averaging 11.7 kg grain/ha/mm. WUE_ET showed similar variability, ranging from 0.42 to 16.0 kg grain/ha/mm, with an average of 6.41 kg grain/ha/mm. Standard deviations indicate substantial variability in the dataset.
Table 2. Descriptive summary statistics for chickpea yield and water use efficiency (WUE) based on two different measures: transpiration (WUE_T) and evapotranspiration (WUE_ET). These data are derived from 438 observations across 29 experiments at ten field experimental sites in southeastern Australia with multiple genotypes per experiment. Some experiments had several sowing times to generate different growing conditions (Anwar et al., Reference Anwar, Luckett, Chauhan, Ip, Maphosa, Simpson, Warren, Raman, Richards, Pengilley, Hobson and Graham2022). *sd = standard deviation, min = minimum and max = maximum
Correlation analysis of predictor variables
The Pearson correlation coefficients between nineteen weather indices and chickpea yield, water use efficiency based on transpiration (WUE_T), and water use efficiency based on evapotranspiration (WUE_ET) are presented in Table 3 and S1–S3. The results revealed significant relationships between weather indices and chickpea performance, with varying degrees of influence across different growth stages. For instance, during the sowing to flowering (SF) period (Table S1), frost frequency (F) showed negative correlations with yield (r = −0.33, p < 0.001), WUE_T (r = −0.27, p < 0.001) and WUE_ET (r = −0.39, p < 0.001). Sum of daily thermal time (sumTT) and accumulated daily water deficit (SDR) were negatively correlated with yield and WUE. Mean photosynthetically active radiation (meanPAR) and daily mean minimum temperature (MINT) showed positive correlations with yield and WUE. In the flowering to pod (FP) period (Table S2), heat stress indices (H30, H35) had negative correlations with yield (r = −0.13, p < 0.01; r = −0.30, p < 0.001) and WUE. Frost frequency (F) continued to affect yield and WUE negatively. Mean vapour pressure deficit (meanVPD) correlated negatively with yield and WUE. The photothermal quotient corrected by VPD (meanPTQvpd) showed positive correlations with WUE_T and WUE_ET.
In the pod to maturity (pm) period (Table S3), sumTT and MINT continued to have positive correlations with yield (r = 0.44, p < 0.001; r = 0.25, p < 0.001) and WUE. Frost frequency (F) had a negative impact but was slightly lower than in earlier periods. Mean VPD negatively correlated with WUE_T and WUE_ET.
During the critical period (cp), heat stress indices (H30, H35) negatively correlated with yield (r = −0.17, p < 0.001; r = −0.35, p < 0.001) and WUE (Table 3). Frost frequency (F) remained strongly negative across all measures. Mean PAR and MINT had positive correlations with yield and WUE.
Table 3. Pearson correlation coefficients between nineteen weather indices and chickpea yield, water use efficiency based on transpiration (WUE_T), and water use efficiency based on evapotranspiration (WUE_ET) during the critical period (cp). n = 438; cp = 300 °Cd before flowering to 500 °Cd after flowering (Lake and Sadras, Reference Lake and Sadras2014); Sig= significance levels: *** = p < 0.001; ** = p < 0.01; * = p < 0.05; and ns = not significant
Abbreviations: H30 = heat stress, number of days with maximum temperatures > = 30°C; H35 = number of days with maximum temperatures >= 35°C; F = frost frequency, number of days with minimum temperatures < = 0°C; sumTT = sum of daily thermal time (°Cd); SE = sum of daily soil evaporation (mm); PET = sum of daily potential evapotranspiration (mm); SDR = accumulated daily water deficit; RAIN = sum of daily rainfall (mm); ET = sum of daily crop water use (mm); T = sum of daily crop transpiration (mm); meanVPD = mean vapour pressure deficit (kPa); meanPTQ = mean of photothermal quotient (MJ/m2/°Cd); meanPTQvpd = mean photothermal quotient corrected by VPD (MJ/m2/°Cd); meanPAR = mean of photosynthetically active radiation (MJ/m2/d); DL = sum of day length (hours); RADN =sum of solar radiation (MJ/m2/d); MINT= daily mean minimum temperature (°C); MAXT = daily mean maximum temperature (°C); meanPZT = daily mean temperature corrected for photoperiod (°C).
These results indicate significant weather influences on chickpea yield and water use efficiency across growth periods. Heat stress, frost and water deficit negatively affect chickpea performance, whereas adequate light and moderate temperatures contribute positively.
Correlation funnels of explanatory variables
The correlation funnels (Figure 2) illustrate the most influential explanatory variables (climate index plus period) for each of the three response variables. For grain yield (kg/ha), of the 10 most influential features, nine are positively correlated with yield while only one is strongly negative (‘F_cp’ = number of frost days in the ‘critical period’) (Figure 2A). Of the positive features, six were from the podding-to-maturity (‘pm’) period; the other three were from the sowing-to-flowering (‘sf’) period. Four of the influential features involved evapotranspiration (‘ET_pm’ and ‘ET_sf’) or transpiration alone (‘T_pm’ and ‘T_sf’). The sum of thermal time for podding to maturity (‘sumTT_pm’) was the only purely temperature-derived feature present in this top 10. The final two were the sum of PZT in the same periods (‘sumPZT_pm’) and ‘SE_pm’. Another notable negatively correlated point was ‘SDR_sf’ (soil moisture supply/demand ratio for the sowing to flowering period). Many other interesting observations can be made; for example, while high heat stress at the critical period (‘H35_cp’) and flowering to podding (‘H35_fp’) were significantly negatively correlated with yield, the incidence of warm days from podding to maturity (‘H30_pm’) is positively correlated with yield. We can speculate about the mechanisms for these observations.
Figure 2. Correlation funnels showing the relationships between response (yield (A), water use efficiency based on transpiration (WUE_T) (B) and based on evapotranspiration (WUE_ET) (C)) and explanatory variables (see text for explanation) across four growth stages (sf =sowing to flower, fp =flower to pod, pm =pod to maturity and cp =critical period). The critical t-test values for these Pearson correlation coefficients (n = 438) are: +/- 0.094 (for 0.05 > p > 0.01); +/- 0.123 (for 0.01 > p > 0.001); and +/- 0.157 (p < 0.001; see Table S1 - S3).
When seasonal water-use efficiency for transpiration from sowing to maturity (‘wue_T’) was examined as the response variable (Figure 2B) it produced a somewhat dissimilar set of most-influential features: in the top 10, four were positively and six were negatively correlated. Low temperatures (‘MINT_cp’ and ‘MINT_fp’) were strongly positive while actual frost effects (‘F_cp’, ‘F_sf’, ‘F_fp’) featured strongly on the negative side. The sum of vapour-pressure deficit for the critical period (‘sumVPD_cp’) and for sowing to flowering (‘sumVPD_sf’) were strongly negatively correlated with ‘wue_T’, along with soil moisture supply/demand ratio in the sowing to flowering period (‘SDR_sf’). Finally, ‘meanPZT_sf’ and ‘mean_PTQvpd_fp’ made up the positive correlations.
The third response variable (seasonal water-use efficiency for evapotranspiration for sowing to maturity = ‘wue_ET’) only exhibited one highly positively correlated feature in this top group (Figure 2C), namely ‘MINT_cp’ (mean daily minimum temperature in the critical period). The other top influencers (features) were all negative: five in the sowing-to-flowering period, plus frost incidence in all three periods involving flowering time (‘F_cp’, ‘F_sf’, ‘F_fp’). Surprisingly, accumulated day-length in the sowing-to-flowering period (‘DL_sf’) was strongly negatively correlated with ‘wue_ET’, in addition to ‘SDR_sf’, ‘sumVPD_sf’, ‘sumPTQ_sf’ and ‘sumPTQvpd_sf’.
Exclusive LASSO for Variable Selection
The estimated coefficients based on eLASSO Model 2 for chickpea yield, WUE_T and WUE_ET are presented in Figure 3. Under this model’s selection criterion, only one explanatory variable was retained for each growing period. Figure 3 (left panel) reveals that the key factors affecting yield are ET during pm, MINT during fp, SDR during sf and F during cp. Yield increases by approximately 5 kg/ha for every standard deviation increase in ET during these periods, while F during fp and cp and SDR during sf show negative effects on yield.
Figure 3. Estimated Exclusive LASSO coefficients based on Model 2 (see text) for chickpea yield (A, Left), water-use efficiency based on transpiration (WUE_T, B_middle) and evapotranspiration (WUE_ET, C_right) in relation to climatic indices across four growing periods (sf = sowing to flower, fp = flower to pod, pm = pod to maturity and cp = critical period).
For WUE_T, Figure 3B (middle panel) shows that mean PZT during sf and frost during other periods (pm, fp and cp) are the key influencing factors. Frost exhibited the most significant effects on WUE_T across these periods, with values ranging from -0.04 to -0.09. In contrast, mean PZT during sf demonstrated a positive effect on WUE_T, with an expected increase of 0.06 for every standard deviation increase in mean PZT. For WUE_ET, the key factors identified include RAND during pm, SDR during sf and frost during fp and cp periods. Among these, only RAND during pm showed a positive effect (approximately 0.03), while all other factors exhibited negative effects ranging between −0.05 and −0.095.
Figure 4 displays the estimated non-zero eLASSO coefficients based on Model 1 for chickpea yield, WUE_T and WUE_ET. For better visualisation of the effects of the explanatory variables, similar enlarged plots showing the complete list of explanatory variables, including those with a coefficient being shrunk to zero, are displayed in Figures S2–S4. This model minimised BIC without restricting the number of explanatory variables. The left panel indicates that ET, SE and mean PTQ during pm and sf periods had the strongest positive effects on yield, while F during cp and mean PTQvpd during sf had the most detrimental effects. For WUE_T (Figure 4, middle panel), SE during sf showed the most positive effects, while F during cp and RAIN during pm had the most negative impacts. Similar patterns were observed for WUE_ET, except that SE during sf had less influence than mean PTQvpd. Detailed eLASSO coefficients based on Model 1 are provided in Table S4.
Figure 4. Estimated Exclusive LASSO coefficients based on Model 1 (see text) for chickpea yield (A, Left), water-use efficiency based on transpiration (WUE_T, B_middle) and evapotranspiration (WUE_ET, C_right) in relation to climatic indices across four growing periods (sf = sowing to flower, fp = flower to pod, pm = pod to maturity and cp = critical period).
Predictions of response variables from eLASSO
Figure 5 compares observed and predicted values of yield and water-use efficiency (WUE) in chickpea using two eLASSO models: one based on BIC (Model 1) and the other on the minimum penalty parameter λ (Model 2). Panels A–F in Fig. 5 show predictions versus observations for yield, WUE based on evapotranspiration (WUE_ET) and WUE based on transpiration (WUE_T) for both models.
Figure 5. Comparison of observed yield (A and B), water-use efficiency based on transpiration (WUE_T) (C and D) and water-use efficiency based on evapotranspiration (WUE_ET) (E and F). The left panel A, C and E are predictions by exclusive LASSO Model 1 (minimising the Bayesian Information Criterion), and right panel B, D and F are based Model 2 (Minimum Penalty Parameter (λ)). The text inserts show the fitted regression equation, the coefficient of determination of the linear relationship (r2), the root mean square error (RMSE) and the normalised root mean square error (NRMSE), and Willmott’s index (d). LR = linear regression (solid black line); 1:1 = the line for a theoretical one-to-one fit (dashed black line).
Each graph displays scatter plots of observed against predicted values, with a black dashed 1:1 line representing perfect prediction and a black solid line indicating the linear regression fit (LR). The insets include key statistical metrics such as the coefficient of determination (r²), root mean square error (RMSE), normalised RMSE (NRMSE) and Willmott’s index (d).
For each of the three response variables (yield, WUE_T and WUE_ET) we observed a very similar result within the two eLASSO models (BIC-based and λ-based models) but a large difference between them when plotting observed versus predicted values. The λ-based model fits (see Figure 5B, D and F) had a tighter distribution around the regression lines but much lower R2s than the BIC-based model fits (Figure 5A, C and E). This was also reflected in their higher RMSE and NRMSE values. However, these λ-based model fits were poorly predictive of the response variables because the model retained only one explanatory variable in each growth stage; hence, the low explanatory and predictive powers reflected by the almost horizontal spread of the points. Predictions of this type would be of little or no utility in the field for predicting crop water use and yield.
In contrast, the BIC-based model fits were much more predictive with much higher R2 values (all > 0.7) but with a greater scatter about the regression lines (which is also reflected in the lower RMSE and NRMSE statistics). The BIC-based model also fits and estimates a sensible origin (i.e., (0,0)) much more closely for all response variables (actual values not given), so these models perform better across a typical real-world range of yield and water-use scenarios. The λ-based model fits are almost useless for predictive purposes except for the fact that the LASSO procedure uncovered these ‘best-fit’ solutions. Examination of the range of the predicted values from λ-based model fits illustrates their poor predictive value. For example, predicted λ-based model yields (Figure 5B) range from about 1400 to 2100 kg/ha, whereas the observed yields range from 100 kg/ha to over 4000.
Discussion
The need to improve resource-use efficiency drives agricultural innovation (Sadras and McDonald, Reference Sadras and McDonald2011; Lorite et al., Reference Lorite, Castilla, Cabezas, Alza, Santos, Porras, Gabaldón-Leal, Muñoz-Marchal and Sillero2023; Dreccer et al., Reference Dreccer, Whish, Delahunty, Richards, Graham, Power, Clancy, Rodriguez, De Voil and Munford2024); understanding the environmental factors influencing crop performance becomes increasingly critical, particularly in the context of climate change and its impact on crop yields (Challinor et al., Reference Challinor, Watson, Lobell, Howden, Smith and Chhetri2014). This study comprehensively analyses the key environmental drivers, including abiotic stress factors, influencing chickpea yield and water-use efficiency (WUE) under Australian field conditions. By integrating advanced statistical modelling with extensive field data, the research identifies temperature extremes, water availability and solar radiation as critical factors affecting chickpea growth and productivity. The study’s use of eLASSO regression (Tibshirani, Reference Tibshirani1996; Campbell and Allen, Reference Campbell and Allen2017) to identify these key drivers represents a significant methodological advancement. Unlike traditional LASSO, which may exclude entire groups of variables, eLASSO ensures that at least one variable from each group is retained, providing a more comprehensive understanding of the factors influencing crop performance. This approach is particularly useful in agricultural studies, where multiple interrelated variables often influence crop outcomes (Kumar et al., Reference Kumar, Attri and Singh2019). The strong predictive performance of the models developed in this study (R² > 0.7 for yield and WUE predictions, Figs. 5A, C and E) demonstrates the potential of eLASSO to identify key drivers of crop performance under varying environmental conditions. Despite its advantages, eLasso regression has a notable drawback. Since the algorithm must retain at least one explanatory variable from each group, some of the retained variables may not be important at all. Caution must be applied while interpreting the model, especially when only one variable is selected in any group.
The findings described here offer novel insights into the complex interactions between environmental variables and chickpea performance, which can inform improved agronomic practices and breeding strategies to enhance crop resilience in the face of climate variability (Vadez et al., Reference Vadez, Berger, Warkentin, Asseng, Ratnakumar, Rao, Gaur, Munier-Jolain, Larmure, Voisin and Sharma2012; Garg et al., Reference Garg, Shankar, Thakkar, Kudapa, Krishnamurthy, Mantri, Varshney, Bhatia and Jain2016; Vadez et al., Reference Vadez, Grondin, Chenu, Henry, Laplaze, Millet and Carminati2024). This methodology could also be applied to other crops to better understand the complex interactions between environmental factors and crop growth, ultimately leading to more targeted and effective management strategies.
Chickpea yield is highly sensitive to environmental fluctuations, with evapotranspiration (ET) and frost frequency emerging as major determinants (Singh et al., Singh et al., Reference Singh, Malhotra, Halila, Knights and Verma1993; Berger et al., Reference Berger, Ali, Basu, Chaudhary, Chaturvedi, Deshmukh, Dharmaraj, Dwivedi, Gangadhar, Gaur and Kumar2006). The positive correlation between ET during podding to maturity (pm) including sowing to flowering (sf) and yield (Figure 2) underscores the importance of water availability during these stages for biomass accumulation and grain filling. Chickpea is a drought-avoidant crop through plasticity in phenology and employs a deep root system to access subsoil moisture when it cannot escape drought. However, water availability during the reproductive phase is particularly critical, as it supports both pod retention and seed filling (Singh et al., Reference Singh, Sharma, Varshney, Sharma and Singh2008; Comas et al., Reference Comas, Becker, Cruz, Byrne and Dierig2013; Waqas et al., Reference Waqas, Azhar, Rana, Arif, Atif and Wani2019). Inadequate moisture supply during this period can lead to increased flower abortion and reduced seed set, ultimately lowering yield potential (Peake et al., Reference Peake, Dreccer, Whish and Hochman2020). Similar findings have been reported in other crops such as wheat and barley, where post-anthesis water availability significantly enhances grain yield by maintaining assimilate supply to the developing grain (Samarah, Reference Samarah2005; Passioura, Reference Passioura2006; Foulkes et al., Reference Foulkes, Sylvester-Bradley, Weightman and Snape2007). The physiological need for sustained photosynthesis and assimilate translocation to seeds during these stages aligns with ET’s observed importance in chickpea’s final growth phases (Sadras and McDonald, Reference Sadras and McDonald2011; Dreccer et al., Reference Dreccer, Fainges, Whish, Ogbonnaya and Sadras2018).
Conversely, frost during flowering to podding (fp) and the critical period (cp) had a strong negative impact on yield (Table 3; Figures 3 and 4), highlighting chickpea’s vulnerability to low temperatures during reproductive stages (Lake and Sadras, Reference Lake and Sadras2014; GRDC, 2016). Unlike cereals such as wheat and barley, which have a degree of frost tolerance due to protective floral structures and cold acclimation mechanisms, chickpea lacks sufficient protective adaptations (Croser et al., Reference Croser, Clarke, Siddique and Khan2003; Barlow et al., Reference Barlow, Christy, O’Leary, Riffkin and Nuttall2015; Peake et al., Reference Peake, Dreccer, Whish and Hochman2020). Frost exposure during flowering causes sterility by disrupting pollen viability, while frost at early podding stages can lead to pod abortion, ultimately reducing the number of harvestable seeds (Clarke and Siddique, Reference Clarke and Siddique2004; Chauhan and Ryan, Reference Chauhan and Ryan2020). Additionally, frost-induced cellular damage can lead to reduced photosynthetic efficiency and lower carbon assimilation rates, further compounding yield losses (Allen and Ort, Reference Allen and Ort2001; Sage and Kubien, Reference Sage and Kubien2007). Frost damage at flowering can lead to flower abortion and reduced pod set, similar to its effects in canola (Brassica napus) (Kovaleski et al., Reference Kovaleski, Heldwein, Dalmago and de Gouvêa2020). This sensitivity underscores the need for agronomic interventions such as delayed sowing, spatial diversification of planting dates and the development of frost-tolerant chickpea cultivars through breeding efforts (GRDC, 2016; Peake et al., Reference Peake, Dreccer, Whish and Hochman2020).
Water-use efficiency (WUE) is a key determinant of chickpea productivity, particularly in water-limited environments (Siddique et al., Reference Siddique, Johansen, Turner, Jeuffroy, Hashem, Sakar, Gan and Alghamdi2012). The study identified mean photothermal quotient (meanPTQ) during sowing to flowering (sf) as a significant positive contributor to WUE_T (Figure 3B), emphasising the importance of early-season radiation use efficiency (Sinclair and Muchow, Reference Sinclair and Muchow1999). PTQ, which represents the ratio of photosynthetically active radiation (PAR) to temperature, measures of the crop’s ability to convert solar energy into biomass (Kiniry et al., Reference Kiniry, Jones, O’Toole, Blanchet, Cabelguenne and Spanel1989; Muchow et al., Reference Muchow, Sinclair and Bennett1990). Higher PTQ values during early growth stages likely promote vigorous vegetative growth, which can enhance the crop’s ability to utilise water efficiently during later stages (Richards, Reference Richards2000). This is particularly relevant for chickpea, as its initial biomass accumulation determines its ability to withstand later-season stresses (Singh and Saxena, Reference Singh and Saxena1993; Soltani and Sinclair, Reference Soltani and Sinclair2011). Similar relationships have been observed in other legumes such as lentil (Lens culinaris) and faba bean (Vicia faba) (Thomson et al., Reference Thomson, Siddique, Barr and Wilson1997), where high PTQ during vegetative growth enhances biomass accumulation and transpiration efficiency (Siddique et al., Reference Siddique, Johansen, Turner, Jeuffroy, Hashem, Sakar, Gan and Alghamdi2012).
In contrast, the negative impact of frost on WUE_ET and WUE_T across multiple growth stages (Figure 3B, C) suggests that cold stress not only reduces total water uptake but also impairs physiological water-use efficiency (Chaves et al., Reference Chaves, Flexas and Pinheiro2009; Flexas et al., Reference Flexas, Díaz-Espejo, Conesa, Coopman, Douthe, Gago, Gallé, Galmés, Medrano, Ribas-Carbo and Tomàs2016). This occurs because frost damage to leaves and reproductive structures reduces the plant’s ability to fix carbon, leading to an inefficient use of available soil moisture. Moreover, low temperatures can limit root hydraulic conductivity, reducing water uptake even with sufficient soil moisture (Aroca and Ruiz-Lozano, Reference Aroca, Ruiz-Lozano and Aroca2012). The correlation between soil moisture supply-demand ratio (SDR) during sowing to flowering (sf) and reduced WUE highlights the significance of early-season moisture availability (Table 3). A low SDR during early growth stages indicates a mismatch between soil moisture supply and crop demand, which can lead to suboptimal biomass development and lower transpiration efficiency (Sadras and Milroy, Reference Sadras and Milroy1996; Sinclair and Muchow, Reference Sinclair and Muchow2001; Sadras et al., Reference Sadras, Cassman, Grassini, Hall, Bastiaanssen, Laborte, Milne, Sileshi and Steduto2015). This aligns with findings in sorghum (Sorghum bicolour), where pre-flowering drought stress leads to inefficient water use and lower yields (Kholová et al., Reference Kholová, Murugesan, Kaliamoorthy, Malayee, Baddam, Hammer, McLean, Deshpande, Hash, Craufurd and Vadez2014; de Souza et al., Reference de Souza, de Carvalho, Bastos, Cardoso, Júlio, Batista, Julio, Campolina, Portugal, de Menezes and de Oliveira2021). Strategies such as optimising sowing dates, improving soil organic matter content and implementing water-conserving agronomic practices could mitigate these adverse effects in chickpea cultivation (Kumar and Abbo, Reference Kumar and Abbo2001, Sadras and McDonald, Reference Sadras and McDonald2011).
Heat stress (H30 and H35) during the critical period (cp) and flowering to podding (fp) was identified as a major constraint, negatively impacting yield and WUE (Table 3; Figure 2). Heat stress accelerates phenological development, shortening the grain-filling period and reducing final seed weight (Vogel et al., Reference Vogel, Donat, Alexander, Meinshausen, Ray, Karoly, Meinshausen and Frieler2019; Lorite et al., Reference Lorite, Castilla, Cabezas, Alza, Santos, Porras, Gabaldón-Leal, Muñoz-Marchal and Sillero2023). Chickpea, as C3 crop, exhibits a decline in photosynthetic efficiency under high temperatures due to increased photorespiration and a reduction in stomatal conductance, which limits CO2 assimilation (Prasad et al., Reference Prasad, Boote and Allen2006; Ainsworth and Rogers, Reference Ainsworth and Rogers2007). Unlike maize (Zea mays), which benefits from a more efficient C4 photosynthetic pathway, chickpea experiences significant reductions in reproductive success under heat stress (Sage and Kubien, Reference Sage and Kubien2007; Hatfield and Prueger, Reference Hatfield and Prueger2015). Excessive heat exposure during flowering reduces pollen viability and ovule fertilisation, leading to lower pod set and yield. Interestingly, moderate heat exposure during podding to maturity (pm) had a positive correlation with yield (Figure 2A), suggesting that late-season warmth may facilitate seed development if adequate moisture is available (Devasirvatham et al., Reference Devasirvatham, Tan, Gaur, Raju and Trethowan2012; Kaushal et al., Reference Kaushal, Awasthi, Gupta, Gaur, Siddique and Nayyar2013; Devasirvatham and Tan, Reference Devasirvatham and Tan2018). Similar trends have been observed in soybean (Glycine max), where late-season warmth enhances pod filling but excessive heat stress during flowering leads to flower abortion (Hatfield et al., Reference Hatfield, Boote, Kimball, Ziska, Izaurralde, Ort, Thomson and Wolfe2011). These findings indicate that chickpea breeding efforts should enhance heat tolerance during flowering while leveraging late-season warmth for improved grain filling.
Chickpea exhibits distinct responses to environmental stressors when compared to other major field crops. While cereals such as wheat and barley demonstrate greater cold tolerance, they are more sensitive to terminal drought stress (Samarah, Reference Samarah2005; Farooq et al., Reference Farooq, Gogoi, Barthakur, Baroowa, Bharadwaj, Alghamdi and Siddique2017). With its deep-rooting system, chickpea can access subsoil moisture more effectively than shallow-rooted crops like canola and lentil (Kashiwagi et al., Reference Kashiwagi, Krishnamurthy, Crouch and Serraj2006; Zaman-Allah et al., Reference Zaman-Allah, Jenkinson and Vadez2011). However, its reproductive sensitivity to frost and heat stress limits its adaptation to variable climatic conditions. The negative impact of high vapour pressure deficit (VPD) on chickpea WUE (Figure 3B, C) aligns with observations in maize and sorghum (Sinclair et al., Reference Sinclair, Hammer and van Oosterom2005; Lobell et al., Reference Lobell, Hammer, McLean, Messina, Roberts and Schlenker2013), where high atmospheric demand increases transpiration losses without proportional gains in biomass accumulation (Sadras and McDonald, Reference Sadras and McDonald2011). Unlike sorghum, which can regulate stomatal closure to minimise water loss, chickpea exhibits less efficient stomatal regulation, leading to higher transpiration under high VPD conditions (Koehler et al., Reference Koehler, Wankmüller, Sadok and Carminati2023). Improving stomatal response traits through breeding and implementing agronomic measures such as mulching and conservation tillage could help mitigate water loss under high evaporative demand conditions (Hatfield et al., Reference Hatfield, Sauer and Prueger2001; Richards et al., Reference Richards, Rebetzke, Watt, Condon, Spielmeyer and Dolferus2010).
A limitation of this study is the absence of simulated biotic stressors such as pests, weeds and diseases, which can significantly impact chickpea yield and water-use efficiency under field conditions. Future research should incorporate these factors to provide a more comprehensive understanding of chickpea performance in realistic agricultural settings.
The findings from this study have important implications for agronomic management and climate adaptation strategies in chickpea production. Optimising sowing dates to avoid frost-prone periods and selecting cultivars with improved heat and drought tolerance are critical for mitigating yield losses. Models are available to achieve this goal (Chauhan et al., Reference Chauhan, Anwar, Richards, Lake, Sadras, Luckett, Raman, Krosch and Graham2023). Enhancing soil moisture conservation through cover cropping, reduced tillage and organic amendments can improve early-season water availability and WUE. The application of eLASSO modelling offers a robust approach for identifying key environmental drivers and developing predictive tools for chickpea yield and WUE. By integrating climate-responsive agronomic practices with advanced statistical modelling, chickpea production can be optimised for greater resilience under changing climatic conditions. Further research into genotype-by-environment interactions will be essential to develop site-specific recommendations for chickpea growers (Dreccer et al., Reference Dreccer, Fainges, Whish, Ogbonnaya and Sadras2018; Bicard et al., Reference Bicard, Faucon, Pedas, Vequaud, Pin, Elmerich and Lange2025).
Conclusions
This study provides valuable insights into the environmental factors driving chickpea yield and water-use efficiency, emphasising the crucial roles of water and temperature extremes, particularly evapotranspiration, frost frequency and heat stress during key growth stages. Comparisons with other field crops reveal both advantages and limitations in chickpea’s agronomic and physiological traits expressed under varying climatic conditions. Future research should focus on validating these findings across diverse environments and cropping systems, while breeding for improved stress resilience (particularly for climate change adaptation), refining crop management practices and integrating these models into decision-support tools for farmers and agronomists to ensure sustainable and profitable chickpea production in water-limited environments.
Supplementary material
The Supplementary material for this article can be found at https://doi.org/10.1017/S0021859625100270.
Acknowledgements
The authors thank Laney Davidson, Karl Moore, Jessica Simpson and Dr. Asad Assaduzzaman for sowing field trials including data collection, and Dr. Livinus Emebiri for internal review of the manuscript. The authors also would like to acknowledge the APSIM Initiative for providing APSIM software used in this study.
Author contributions
Muhuddin Rajin Anwar (MRA), David J. Luckett (DJL), Ryan H. L. Ip (RHLIp), Yashvir S. Chauhan (YSC), Maheswaran Rohan (MR) contributed to the conceptualisation, investigation, methodology, software and visualisation, writing – original draft and writing – review and editing. Neroli Graham (NG), Mark F. Richards (MFR), Rosy Raman (RR) including MRA and YSC contributed to design the various experiments whose data has been used in the study including writing – review and editing. Jens D. Berger (JDB) contributed to writing – review and editing. MRA wrote the first draft of the manuscript, which all authors subsequently revised.
Funding statement
This research is supported by Grains Agronomy & Pathology Partnership (GAPP), between the NSW Department of Primary Industries and Regional Development (DPIRD) and the Grains Research and Development Corporation (GRDC) under the project BLG111.
Competing interests
None.
Ethical standards
Not applicable.
