Document Type : Research Paper
Ensuring global food security remains a critical challenge, particularly for wheat (Triticum aestivum L.), which supplies ~20% of human caloric intake. In Iran, dryland wheat occupies over 70% of the country's arable drylands; however, a persistent yield gap of approximately 65% (average yield ~1.25 t/ha) threatens national food security. Quantitative land suitability assessment provides a systematic framework to match environmental characteristics with crop ecological requirements. While the FAO's Agro-Ecological Zoning (AEZ) approach simplifies crop growth through reductionist models and soil limitation indices, the process-based AquaCrop model simulates dynamic soil-plant-climate interactions mechanistically. Despite their individual applications, a systematic, head-to-head comparison of these two FAO-endorsed approaches for dryland wheat suitability assessment in semi-arid regions remains unexplored. Therefore, the primary goal of this study was to compare the predictive accuracy and applicability of the AEZ and AquaCrop models for estimating dryland wheat yield (cv. Azar 2) to conduct a quantitative land suitability evaluation in the semi-arid Halab region of Zanjan province, Iran. The target audience includes agricultural land-use planners, soil scientists, and crop modelers. We adopted a postpositivist quantitative approach to test the hypothesis that AquaCrop, with its dynamic daily water-balance, exhibits lower systematic error and higher precision for point-specific yield predictions compared to the static, regionally-oriented AEZ method.
This quantitative study employed an observational/computational modeling design. The methodology involved three integrated phases: (1) field sampling for soil and yield data, (2) implementation of the FAO-AEZ parametric model, and (3) calibrated simulation using the AquaCrop v7 model. Eleven representative soil profiles (each corresponding to a distinct land unit) were selected across the 4,557-hectare study area. For each profile, standard laboratory analyses determined physical (texture, FC, PWP) and chemical properties (pH, ECe, SAR, CaCO₃, gypsum, OC). Concurrently, actual grain yield was measured at adjacent farm plots using 1 m² quadrats with three replications during the 2022–2023 and 2023–2024 growing seasons. Daily climatic data (Tmin, Tmax, precipitation, humidity, wind speed, sunshine hours) were obtained from the Halab automatic synoptic station, with reference evapotranspiration (ETo) calculated using the FAO-Penman-Monteith equation. Outcome measures included predicted attainable yield (t/ha) for each model and the final quantitative land suitability class (S1–N). For AEZ, potential yield (Yp) was computed via the radiation-thermal model, water-limited yield (Yw) via the Doorenbos-Kassam relationship (Ky=1.05), and the edaphic index (Ied) via the square-root method based on seven limiting factors. For AquaCrop, conservative parameters (WP*, HI₀, CGC, CDC) were sourced from Tavakoli et al. (2010) and locally calibrated, while non-conservative parameters were field-observed.
The 11 soil profiles were selected using a systematic stratified sampling strategy based on pre-existing soil map units to ensure they represented the topographical and soil variability of the entire region. For each profile, morphological descriptions and disturbed samples were collected from all genetic horizons. Yield sampling was executed adjacent to each profile; however, due to fallow practices, sampling occurred across the two separate seasons—seven profiles in 2022–2023 and four profiles in 2023–2024. Data collection settings included agricultural dryland fields and their corresponding laboratory analysis.
The intended sample size (n=11 land units) was determined by the available representative soil profiles from the national soil database (SWRI) that fell within the study domain, rather than a formal a priori power analysis—a standard approach in land suitability modeling where the focus is on representative pedo-geomorphological units rather than statistical generalization to a population. The achieved sample size matched the intended size (n=11), with a total of 33 yield measurement plots (3 replicates × 11 profiles). Precision of the model estimates was rigorously evaluated using post-hoc standard error metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), normalized RMSE (nRMSE), the squared Pearson correlation coefficient (r²), and the coefficient of determination (R²).
AquaCrop-simulated potential yields were 4.99–5.02 t/ha, closely matching the AEZ-estimated Yp (4.81–4.95 t/ha). However, the validated attainable yields showed significant differences. AquaCrop demonstrated superior accuracy with MAE=0.155 t/ha, RMSE=0.189 t/ha, nRMSE=10.57%, r²=0.777, and R²=0.657. Conversely, the AEZ method yielded MAE=0.178 t/ha, RMSE=0.235 t/ha, nRMSE=13.17%, r²=0.728, and R²=0.544. Crucially, the discrepancy between r² and R² was notably larger for AEZ (Δ=0.184) compared to AquaCrop (Δ=0.120), indicating a substantial systematic bias (over/underestimation) inherent to the AEZ structure. The quantitative land suitability classification, based on the Sys method with a critical yield of 0.412 t/ha, revealed that all 11 land units fell into the moderately suitable class (S2) regardless of the model used. Notably, all predicted yields exceeded the economic critical threshold, confirming the financial viability of dryland wheat production across the entire study region.
The results confirm the hypothesis that AquaCrop provides a more precise, point-specific prediction of dryland wheat yield, primarily due to its dynamic daily simulation of water stress, compartmentalized evaporation/transpiration, and three distinct stress coefficients (canopy expansion, stomatal closure, and early senescence). Conversely, AEZ's reliance on a static four-stage reduction coefficient and the aggregated ETₐ/ETo ratio masks the critical impact of drought timing during sensitive phenophases (e.g., pollination). The theoretical and practical significance lies in establishing a clear decision-framework: AquaCrop is warranted for on-farm managerial recommendations and detailed local suitability mapping, whereas AEZ remains highly valuable for large-scale regional agro-ecological zoning and climate-change scenario analyses. If these findings are applied, planners can optimize resource allocation. However, unresolved problems persist, as both models left 22–34% of yield variance unexplained—largely attributed to AquaCrop's direct omission of specific soil chemical constraints and AEZ's oversimplified static parameters. We recommend future research focus on developing hybrid models that integrate AquaCrop's dynamic water module with nutrient-uptake sub-models and soil chemical constraints, alongside leveraging remote-sensing data for dynamic parameterization, to reduce prediction uncertainty and provide a more robust scientific basis for agricultural decision-making in semi-arid environments.
The study was funded by the University of Tehran, Faculty of Agricultural Engineering and Technology.
M. Davirand (PhD student): Preparation of samples, conducting experiments and data collection, calculations, statistical analysis of data, analysis and interpretation of information and results, preparation of the original draft.
F. Sarmadian (Supervisor): Research design, supervision of the research process, verification and validation of results, revision, review, and finalization of the manuscript.
M.N. Navidi (Advisor): Participation in research design, supervision of the research, study and review of the manuscript.All authors contributed equally to the conceptualization of the article and writing of the original and subsequent drafts.
During the preparation of this work, the authors used DeepSeek in some sections in order to improve grammar and writing style. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
Data available on request from the authors.
The authors would like to thank the University of Tehran, Faculty of Agricultural Engineering and Technology, for their financial and spiritual support, and also anonymous reviewers for their constructive and scientific comments on the manuscript.
The authors avoided data fabrication, falsification, and plagiarism, and any form of misconduct.
The authors declare no conflict of interest.
