This paper describes a fuzzy and neuro-fuzzy approach to modeling feeding intensity of greylag geese on reeds. As a consequence of the presence of some non-measurable or random factors and the heterogeneity of reed and geese behaviour under investigation, the relationships between the model variables are often not well known and the data collected have a high degree of uncertainty. A fuzzy approach was selected which allowed for operation with vague knowledge and data of high uncertainty. Fuzzy logic can be used to handle inexact reasoning in knowledge-based models and fuzzy sets to handle uncertainty of data. A neuro-fuzzy approach was also used to select the model variables with the biggest impact on the model output. This way the number of input variables can be reduced to a number which is necessary for successful management. A neuro-fuzzy model was developed for these selected variables using the neuronal network technique ANFIS to identify the optimal model parameters. For training, a data set from a lake in North Germany was used. The generalization capability of this model was checked for two other lakes. The performance of the model was compared with the results of the fuzzy knowledge-based model developed in the next step. The knowledge base of this model contains the Mamdani-type rules formulated by a domain expert. Both models were implemented using the Fuzzy Logic Toolbox of MATLAB.

Keywords: reed, greylag geese, grazing models, fuzzy models, neuro-fuzzy approach.