A COMPARISON OF LOCAL AND GLOBAL SENSITIVITY ANALYSES FOR GREENHOUSE CROP MODELS

Sensitivity analysis is an important step in developing mathematical modelling of agricultural systems. However, usually for greenhouse crop models, local approaches based on the calculation of derivatives are applied. Unfortunately, the main drawback of local approaches is that derivatives provide information only at the base point where they are computed and do not take into account the rest of the variation range of the model’s parameters. In contrast, several so called global sensitivity analysis approaches such as standardized regression coefficients, the elementary effect test, variance-based methods and Monte Carlo Filtering are being developed. In the present work a comparison between a local sensitivity analysis and two global sensitivity methods is carried out. The paper focuses on the methodological issues concerning both local and global sensitivities approaches, where their similarities and differences are discussed. A five-state variables greenhouse lettuce crop model is used in order to show both the strengths and also the weaknesses of the paradigms. In the case of the local method the so-called sensitivity equations are stated and solved. For the global approaches, firstly, probability density functions (PDFs) were assigned to each of the model’s parameters. Secondly, five thousands Monte Carlo simulations were conducted in order to calculate the sensitivity indices and measures. The lettuce crop model was implemented in Matlab, and the global sensitivity analysis was programmed using the SimLab (ver. 3.2) software. More reliable results were obtained by the global method based on variance calculation Fourier Amplitude Sensitive Test (FAST) which indentified as the most sensitive parameters for the NICOLET model the coefficients: radiation extinction, photosynthetic efficiency, leaf conductance to CO₂, specific growth rate , maintenance respiration and reference temperature.