Abstract:
Process-based models can assist in identifying beneficial management techniques for optimizing
sugarcane yields. However, accurate model prediction requires parameterization, which can be
time-consuming due to the large number of parameters associated with process-based crop
models. Sensitivity analysis (SA) can help identify sensitive parameters and reduce
parameterization efforts. However, SA can be computationally expensive, particularly for complex
crop models. Gaussian process emulation offers a promising approach to alleviate the
computational burden of SA. In this study, we conducted a comprehensive global SA using
Gaussian process emulation to assess the impact of trait parameters on sucrose weight (SW) in
the APSIM-Sugar model under irrigated (IR) and rainfed (RF) conditions in three different soil
types (reddish brown earth, non-calcic brown, and alluvial) in Hingurana, Sri Lanka. Emulators,
generated for various scenarios, demonstrated notable accuracy and were subsequently
employed for SA. The results revealed that radiation use efficiency (RUE), green leaf number
(GLN), sucrose fraction in the stalk (SF1), stress factor in the stalk (SF2), minimum stem sucrose
(MSS), and transpiration efficiency coefficient (TEC) collectively accounted for over 90% of SW
variation. Among these parameters, RUE was the most influential for predicting SW, with higher
sensitivity under IR conditions compared to RF conditions. GLN and TEC were the second-most
influential factors under IR and RF conditions, respectively. SF1, MSS, and SF2 followed in order
of influence on SW under both IR and RF conditions in all soil types. These findings contribute to
enhancing modeling precision and provide valuable insights for strategic management decisions,
addressing the temporal and spatial variability of sugarcane yield in Hingurana, Sri Lanka.