Comparison of empirical and process-based models for simulating groundnut yield in semi-arid environments

Introduction

Groundnut (Arachis hypogaea L.) occupies 28.5 million hectares globally, with over 60% of that area in semi-arid regions where rainfall variability is the dominant yield determinant [1,2] . Crop simulation models are increasingly used for yield forecasting, climate impact assessment, and irrigation scheduling, but model selection for a given environment remains largely ad hoc [3] . Process-based models like CROPGRO-Peanut simulate daily carbon assimilation, partitioning, and water uptake using mechanistic equations, requiring detailed soil, weather, and cultivar parameters [4,5] . AquaCrop, developed by FAO, simplifies canopy growth as a function of water productivity, trading some mechanistic detail for lower data requirements [6] . WOFOST, widely used in European yield forecasting, models light interception and conversion efficiency but uses a simpler water balance [7,8] . No head-to-head comparison of these three models for groundnut under Indian semi-arid conditions has been published. This research evaluated all three against 12 seasons of observed yield data at four representative locations, testing which model best reproduces inter-annual yield variability driven by monsoon rainfall patterns.

Materials and Methods

Study sites

Four AICRPG (All India Coordinated Research Project on Groundnut) trial stations were selected: Anantapur (15°41′N, 77°36′E, 350 m), Junagadh (21°31′N, 70°36′E, 60 m), Jalgaon (21°03′N, 75°34′E, 210 m), and Chitradurga (14°14′N, 76°24′E, 732 m). Annual rainfall ranges from 553 mm (Anantapur) to 812 mm (Junagadh). Soils range from Alfisols (Anantapur) to Vertisols (Jalgaon). Cultivar TMV-2 (bunch type, 105-110 days) was common across all stations.

Simulation parameters and model configuration CROPGRO-Peanut (DSSAT v4.7.5) was parameterised using cultivar coefficients for TMV-2 calibrated at Anantapur [9] . Soil profiles to 180 cm were initialised from NBSS&LUP data. AquaCrop (v6.1) was configured with conservative crop parameters from the FAO database, adjusted for local phenology and canopy cover observations. WOFOST (v7.2) used CGMS-level crop data files for groundnut with thermal time parameters from literature [10] . All models received identical daily weather inputs (Tmax, Tmin, rainfall, solar radiation) from IMD gridded data at 0.25° resolution. Sowing dates, plant density, and fertiliser inputs matched station records. CROPGRO reproduced observed pod yields most accurately across all four sites (Figure 1, Table 1). Mean simulated yield deviated from observed by only 3.2% (pooled), compared to 8.7% for AquaCrop and 14.8% for WOFOST. At Anantapur, the driest site, CROPGRO captured the severe drought years (2015, 2018) within 80 kg ha⁻¹ of observed, while WOFOST overestimated these years by 280-340 kg ha⁻¹. The radar plot of performance metrics (Figure 2) shows CROPGRO dominating all six indicators. Its d-index (0.95) and EF (0.90) indicate near-perfect agreement and low systematic bias. AquaCrop performed well on R² (0.84) and d-index (0.88) but showed higher NRMSE (17.6%) driven by overestimation in wet years at Junagadh. WOFOST's CRM of 0.12 confirmed its consistent underestimation (Table 2). The model ranking was stable across all four sites, though the performance gap narrowed at Junagadh where higher rainfall reduced water stress as a differentiating factor.

Performance evaluation and validation

Discussion

CROPGRO's superiority traces to its explicit simulation of soil water extraction by root zone layers, pod-specific carbon partitioning rules, and a drought stress factor that modulates both photosynthesis and partitioning [4,11] . Under semi-arid conditions, these mechanisms capture the nonlinear yield response to intermittent drought that simpler models miss [12,13] . AquaCrop's water productivity approach works well under moderate stress but overestimates recovery after mid-season dry spells because it does not distinguish between vegetative and reproductive water sensitivity windows [6,14] . WOFOST's underestimation likely stems from its generic light-use efficiency parameters, which were originally calibrated for European conditions, and its lack of a pod-filling sub-model [7,15,16] . For operational yield forecasting in Indian semi-arid zones, CROPGRO offers the best accuracy when calibration data are available, while AquaCrop provides an acceptable alternative where input data are limited.

Limitations

All three models were run under rainfed conditions; irrigation scenarios were not tested. WOFOST lacks a detailed pod-filling algorithm, which may explain its systematic underestimation. AquaCrop's canopy cover approach may not capture cultivarspecific differences in branching architecture. IMD gridded weather may smooth local rainfall variability, particularly convective events that dominate semi-arid monsoon patterns.

Conclusion

CROPGRO-Peanut outperformed AquaCrop and WOFOST across all six performance metrics for groundnut yield simulation at four semi-arid Indian locations over 12 seasons. Its advantage was most pronounced in drought years, where explicit water balance and partitioning algorithms captured yield depression that simpler models missed. AquaCrop offered a reasonable accuracy-to-complexity trade-off. These results guide model selection for groundnut yield forecasting and climate adaptation planning in semi-arid systems.

References

1. FAOSTAT. Food and Agriculture Organization crop statistics database. FAO; 2023. Available from: FAOSTAT Online.
2. Nigam SN, Upadhyaya HD, Chandra S, Vadez V, Singh P. Gene effects for specific leaf area, harvest index, and kernel yield in groundnut. Crop Science. 2005;45(3):922-927.
3. Jones JW, Hoogenboom G, Porter CH, Boote KJ, Batchelor WD, Hunt LA, et al. The DSSAT cropping system model. European Journal of Agronomy. 2003;18(3-4):235-265.
4. Boote KJ, Jones JW, Hoogenboom G, Pickering NB. The CROPGRO model for grain legumes. Understanding Options for Agricultural Production. 1998;1:99-128.
5. Hoogenboom G, Jones JW, Wilkens PW, Porter CH, Boote KJ, Hunt LA, et al. Decision Support System for Agrotechnology Transfer v4.7.5. DSSAT Foundation; 2019. p.1-22.
6. Steduto P, Hsiao TC, Raes D, Fereres E. AquaCrop: the FAO crop model to simulate yield response to water. Agronomy Journal. 2009;101(3):426-437.
7. De Wit A, Boogaard H, Fumagalli D, Janssen S, Knapen R, Van Kraalingen D, et al. 25 years of the WOFOST cropping systems model. Agricultural Systems. 2019;168:154-167.
8. Van Diepen CA, Wolf J, Van Keulen H, Rappoldt C. WOFOST: a simulation model of crop production. Soil Use and Management. 1989;5(1):16-24.
9. Bhatia VS, Singh P, Wani SP, Chauhan GS, Rao AV, Mishra AK, et al. Analysis of potential yields and yield gaps of rainfed soybean in India. Agricultural and Forest Meteorology. 2008;148(8-9):1252-1264.
10. Wolf J. Comparison of two soya bean simulation models under climate change. Climate Research. 2002;20(1):71-82.
11. Hammer GL, Sinclair TR, Boote KJ, Wright GC, Meinke H, Bell MJ. A peanut simulation model: I. Model development and testing. Agronomy Journal. 1995;87(6):1085-1093.
12. Singh P, Boote KJ, Rao AV, Iruthayaraj MR, Sheikh AM, Hundal SS, et al. Evaluation of the groundnut model PNUTGRO for crop response to water availability. Field Crops Research. 1994;39(2-3):147-162.
13. Challinor AJ, Wheeler TR, Craufurd PQ, Slingo JM. Simulation of the impact of high temperature stress on annual crop yields. Agricultural and Forest Meteorology. 2005;135(1-4):180-189.
14. Vanuytrecht E, Raes D, Steduto P, Hsiao TC, Fereres E, Heng LK, et al. AquaCrop: FAO's crop water productivity and yield response model. Environmental Modelling and Software. 2014;62:351-360.
15. Supit I, Van Diepen CA, De Wit AJW, Kabat P, Baruth B, Ludwig F. Assessing climate change effects on European crop yields using the Crop Growth Monitoring System. Agricultural and Forest Meteorology. 2012;164:96-111.
16. Grassini P, Van Bussel LGJ, Van Wart J, Wolf J, Claessens L, Yang H, et al. How good is good enough? Data requirements for reliable crop yield simulations. Field Crops Research. 2015;177:49-63.