نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی آبیاری آبادانی، دانشکده کشاورزی و منابع طبیعی، دانشگاه تهران
2 استادیار/گروه مهندسی آبیاری و آبادانی، دانشکده مهندسی و فناوری کشاورزی، پردیس کشاورزی و منابع طبیعی دانشگاه تهران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
This study aimed to assess the sensitivity of rainfed wheat yield to climatic indices over the period 1990–2023 across three representative stations in Zanjan Province, Iran —Zanjan, Khodabandeh, and Khorramdarreh. The Extreme Gradient Boosting (XGBoost) algorithm, interpreted via SHapley Additive exPlanations (SHAP) values, was employed to identify the most influential climatic variables and to determine their optimal and yield-limiting thresholds. The methodological framework integrated four key approaches that have been seldom considered simultaneously in previous research: (i) estimating the length of the growing period (LGP) based on updated FAO guidelines, (ii) aligning climatic data with the actual crop growing period, (iii) focusing on the temporal distribution patterns of climatic indices rather than solely on their cumulative or seasonal means, and (iv) performing interactive and interpretable climatic analysis. The results indicated increasing trends in both yield and LGP across all three sites, although the LGP trend in Zanjan was statistically non-significant (P > 0.1). SHAP analysis revealed that moisture-related variables were the primary determinants of yield in all sites. Specifically, effective rainfall (ERGP: 1.8–2.9 mm/day) in Khodabandeh, the number of precipitation days (N_pr^GP: 4–31 days) in Zanjan, and uniform rainfall distribution (URGP: 11.2–31 mm) in Khorramdarreh emerged as the most influential positive drivers. Conversely, yield limitations were associated with shortened growing periods (LGP: 50–68 days) in Khodabandeh, poorly distributed rainfall (UR/ER: 5.4–10 mm) in Khorramdarreh, and a low number of precipitation days in Zanjan.
کلیدواژهها [English]
Rainfed wheat (Triticum aestivum L.) in semiarid regions is highly sensitive to climate variability, and extreme temperatures along with changing precipitation patterns threaten its yield. Accurate analysis of the climate–yield relationship, identification of key influencing indices, and determination of optimal and limiting ranges for rainfed wheat yield are essential. Many previous studies relied on annual or fixed-season climate averages, which overlooked actual crop exposure during the true growing period (GP). To address these limitations, this study calculated all climatic indices exclusively for the real GP of rainfed wheat across three major production districts in Zanjan Province, Iran (Zanjan, Khodabandeh, and Khorramdarreh). The objectives were to: (i) identify key climatic factors driving yield variability, (ii) determine their location-specific optimal and limiting ranges, and (iii) provide actionable insights for targeted adaptation strategies.
This study aimed to assess the sensitivity of rainfed wheat yield to climatic indices over the period 1990–2023 across three representative stations in Zanjan Province, Iran — Zanjan, Khodabandeh, and Khorramdarreh. The Extreme Gradient Boosting (XGBoost) algorithm, interpreted via SHapley Additive exPlanations (SHAP) values, was employed to identify the most influential climatic variables and to determine their optimal and yield-limiting thresholds. The relationship between key climatic indices and dryland wheat yield was evaluated using this advanced machine‑learning model. The research methodology comprised four main steps: (i) estimation of the length of the growing period (LGP) based on the updated FAO guidelines, (ii) aligning all datasets with the actual plant growing period, (iii) analyzing the temporal distribution of climatic indices, and (iv) performing an interactive and interpretable climatic index analysis. This approach enabled the identification of location‑specific key variables, quantification of their optimal ranges, and improved understanding of regional yield variability.
In this analysis, the influential variables for each region were identified, and their optimal and limiting ranges—as well as thresholds—were quantified, providing clearer insight into the drivers of yield variability at the regional scale. Across all three study regions, both rainfed wheat yield and the length of the growing period (LGP) showed increasing trends; however, the trend in Zanjan was not statistically significant (p > 0.1). At the regional scale, correlation analysis between yield and agro‑climatic indices during the growing period indicated that moisture‑related indices were more strongly associated with yield than thermal indices. This pattern was consistent with the XGBoost–SHAP results, which also highlighted the predominance of moisture variables in explaining yield variability.In all stations, the number of precipitation days () exhibited the highest positive correlation with yield. Neither thermal indices nor LGP were significantly correlated with yield in Zanjan or Khodabandeh. By contrast, in Khorramdarreh, mean minimum temperature(T ̅_min^GP, r = 0.52, p < 0.01) and mean growing period temperature (, r = 0.37, p < 0.1) were both significantly and positively correlated with yield. The XGBoost–SHAP modeling outcomes broadly agreed with the linear correlation analysis, although differences arose in the ranking of dominant moisture related indices.This integrated approach identified effective rainfall (ERGP = 1.8–2.9 mm day⁻¹; mean daily rainfall during the growing period; threshold: 1.8 mm day⁻¹) in Khodabandeh, the number of precipitation days ( = 32–42 days; threshold: 32 days) in Zanjan, and the uneven rainfall index (URGP ≥ 9.0 mm; hreshold: 9.0 mm) in Khorramdarreh as the primary determinants of yield. Moreover, ERGP= 1.8–2.9 mm day⁻¹ in Khodabandeh, optimal rainfall distribution (UR/ER(GP) = 10.3–12.4), and optimal mean temperature ( = 9.0–9.2 °C) in Khorramdarreh emerged as the most influential factors enhancing yield. Conversely, shorter growing periods (LGP = 50–68 days) in Khodabandeh, uneven rainfall distribution (URGP = 1.1–8.7 mm) in Khorramdarreh, and low = 4–31 days in Zanjan were the main limiting factors.
This study demonstrated that rainfed wheat yield variability across the three regions is driven by a limited set of agro‑climatic indices, each with distinct optimal and limiting ranges. The identified thresholds — such as ERGP, , and URGP — revealed that yield responses are inherently non‑linear and threshold‑dependent, which is a pattern clearly captured through the machine learning framework enhanced by SHAP interpretation. These findings highlight the necessity of region‑specific management strategies that align planting and water resource practices with the identified optimal climatic windows, thereby improving resilience and yield stability under variable climatic conditions.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Data available on request from the author.
The research was supported by the University of Tehran. The authors would like to express their special thanks to the vice chancellor for research affairs.
The author avoided data fabrication, falsification, plagiarism, and misconduct.
All authors declare that they have no conflict of interest.
)