Evaluating Satellite-Derived Leaf Area Index Assimilation to Reduce Uncertainty in Commercial Sugarcane Yield Simulation

PURPOSE Accurate plot-level sugarcane yield estimation is important for bioenergy logistics and crop management, yet process-based models (PBMs) often struggle under commercial conditions because of field heterogeneity and structural limitations, including imperfect representation of ratoon yield decline. Remote sensing can provide timely canopy observations, but on its own it does not capture the physiological processes represented by PBMs. This study evaluated two data assimilation approaches, the Ensemble Kalman Filter (EnKF) and Particle Swarm Optimization (PSO), to improve yield estimation across 1,669 commercial sugarcane plots in Brazil. METHODS A data assimilation (DA) engine was implemented in the DSSAT/SAMUCA (DS) model to assimilate satellite-derived leaf area index (LAI), and results were compared against an open-loop (OP) baseline simulation. Error patterns across variety, harvest number, and soil class were assessed using the Kruskal-Wallis H-test and epsilon-squared (ε²) effect sizes. RESULTS Both data assimilation approaches outperformed the OP baseline (RMSE = 41.27 Mg ha⁻¹). PSO achieved the lowest RMSE (23.03 Mg ha⁻¹, 44.4% improvement relative to OP) and reduced systematic overestimation, whereas EnKF produced the highest coefficient of determination (R² = 0.25). Variety was the factor most strongly associated with model error (ε² = 0.063), followed by harvest number and soil class. Under PSO, these associations were reduced, with weaker effect sizes across factors. CONCLUSION These results show that assimilating satellite-derived LAI can improve plot-level sugarcane yield estimation under commercial conditions, with PSO providing the largest gain in overall accuracy and EnKF better preserving relative yield variation among plots.