DEVELOPING A CROP SIMULATION MODEL TO PREDICT RICE YIELD UNDER DIFFERENT IRRIGATION AND FERTILIZATION REGIMES

Abstract

A dynamic crop simulation model was developed to predict rice (Oryza sativa L.) yield under varied irrigation and fertilization regimes, with the objective of informing sustainable management decisions. Field experiments across three distinct sites provided data on soil properties, weather variables, irrigation schedules, and nitrogen inputs. The model integrated physiological growth processes, water and nutrient uptake dynamics, and stress response functions, and was implemented in Python and R. Calibration against a comprehensive dataset achieved root mean square error (RMSE) values of 230–260 kg ha⁻¹ and coefficients of determination (R²) ≥ 0.88. Independent validation yielded RMSE ≤ 290 kg ha⁻¹, mean absolute error (MAE) ≤ 225 kg ha⁻¹, and Nash–Sutcliffe efficiency (NSE) ≥ 0.82, demonstrating robustness across spatial and temporal scales. Scenario analysis indicated that alternate wetting and drying (AWD) produced the highest mean yield (7 200 kg ha⁻¹) and superior water use efficiency (1.35 kg m⁻³) compared to continuous flooding (6 800 kg ha⁻¹, 1.20 kg m⁻³) and rainfed conditions (6 100 kg ha⁻¹, 0.90 kg m⁻³). High nitrogen application maximized yield (7 500 kg ha⁻¹), while site-specific nitrogen management (7 300 kg ha⁻¹) achieved comparable performance with reduced input. Combined strategies revealed that flooding with high nitrogen yielded 7 700 kg ha⁻¹, whereas alternate drying with low nitrogen produced 6 900 kg ha⁻¹. Sensitivity analysis showed that a 10 % increase in irrigation and nitrogen input resulted in yield gains of 5.2 % and 4.8 %, respectively. Error decomposition attributed 30 % of overall uncertainty to calibration, with input uncertainty, model structure, and validation each contributing 20–25 %. These results underscore the model’s potential as a decision-support tool for optimizing rice production under resource constraints. Future enhancements will incorporate remote sensing inputs and expand stress-response modules to address climate variability.