Sensitivity and uncertainty analysis of SAMUCA crop model across contrasting environments in Brazil

• Ano de defesa: 2023
• Autor(a) principal: Pereira, Rodolfo Armando de Almeida
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Cana-de-açúcar; eFAST; Incerteza; PRCC; Sensibilidade; Sensitivity; Sugarcane; Uncertainty
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/11/11152/tde-13042023-082447/

The Brazilian sugarcane industry is constantly developing, testing, and launching new cultivars and management practices to increase productivity. Due to climate change issues and limitations for expanding areas, farms are constantly pressured to increase agricultural efficiency. The use of process-based models (PBCM) to test cultivars and management options in different production environments is a reality and has been increasingly used. PBCMs are the state of the art in agricultural modeling and are increasingly complex, requiring several parameters to describe cultivation processes and boundary conditions. In general, PBCMs use the deterministic approach to simplify the uncertainty present in the environment using a single set of parameters. In practice, this uncertainty is seen in the variability of data collected in a field experiment, which is commonly represented by dispersion statistics such as standard deviation and variance. One way to explore this uncertainty is to use the stochastic approach, inserting a range of variability in the parameters and inputs of the simulation. This study aimed to use the stochastic approach to explore uncertainty and determine which parameters of the SAMUCA model are most influential in the simulation process. For this, the recent version of the SAMUCA model was used, inserting three uncertainty scenarios: uncertainty analysis only for genotype parameters (UG), uncertainty analysis only for soil parameters (US), and analysis of soil parameters and genotype (UGS). In this first stage, these three scenarios were simulated for a 4-year field experiment, with the crop cultivated under the effect of green cane trash blanket (GCTB) and bare soil (Bare). The variability of the stochastic simulation was quantified by the ratio between the mean standard deviation of the simulations and the mean standard deviation of the observed data. Subsequently, to better understand which factors caused greater uncertainty in the simulation process, a global sensitivity analysis (GSA) was performed using the extended Fourier Amplitude Sensitivity Test (eFAST) method for the same 4-year experiment, in order to identify which parameters were responsible for explaining the higher variance of the model and verifying the impact of the range of the chosen parameters, as well as the number of simulations necessary to have a reliable GSA. Finally, knowing that the environment can influence the GSA result, a new sensitivity analysis was carried out with two methods, eFAST and Partial Rank Correlation Coefficient (PRCC) for the main sugarcane-producing regions in Brazil, considering irrigated and rainfed conditions. The results indicated that the observed variability in the field is not fully explained by soil parameters, possibly due to irrigation and good rainfall distribution in the experimental area. The UG and the UGS had the same ability to quantify the variability present in the experimental field. In that case, sensitivity to soil parameters could simply be ignored and genotype parameters could be chosen as the sole source of variability for practical applications. Most of the uncertainty in this experiment is attributed to the plastochron parameter, however, it was identified that the parameter range set could influence the order of the most important parameters. This was observed when the analysis was carried out for two sets of different parameter intervals (the first set used maximum and minimum values reported in the literature; the second set applied a 25% perturbation to the previously calibrated values). Finally, out of 31 parameters, 24 genotype and 7 soil, only 13 parameters were significant, regardless of the output variable. In addition, the results were affected by climate: in environments with good rainfall distributionplastochron was the main parameter, while in environments subjected to greater water stress, the eff parameter was the most important. It was noted that any soil parameter was indifferent to irrigated conditions. In contrast, for rainfed conditions, field capacity and permanent wilting point were relevant in environments with low rainfall distribution and shallow soils. Rainy sites with deep soils also showed no sensitivity to soil parameters.