CASA Model Improvement and NPP Estimation Through Parameter Optimization and Uncertainty Quantification

Taking north-western Yunnan as the study area, we first identified sensitive parameters and their interaction effects using the Morris and Sobol methods. We then carried out a phased optimisation of key parameters by combining machine learning with physically constrained methods. Finally, we established an error equation between the NPP estimates from the improved model and the MOD17A3HGF NPP to quantitatively assess the uncertainty in the model outputs, and applied the optimised model to estimate NPP in the study area. The results indicate that the photosynthetically active radiation absorption ratio (FPAR) and solar radiation (SOL) are the key parameters in the model and exhibit the strongest interactions. On a monthly scale, Model 12, optimised using the two-parameter approach, performed best (R² = 0.805, RMSE = 66.737 gC/(m²·a)), with an R² value 0.493 higher than that of the baseline model and a 5.05% reduction in uncertainty. The spatial distribution of NPP in the study area exhibits a pattern of “low in the north and south, high in the centre”, with a total uncertainty in the NPP estimates of 5.320 TgC/a. The improved CASA model framework, developed through phased optimisation of sensitive parameters and quantification of uncertainty, has significantly enhanced the model’s applicability and reliability.