Evaluating the sensitivity of hydrological response to changes in vegetation classification: a case study of St. Charles Bay, Texas, USA

Hydrological modelling is dependent on accurate vegetation classification. We tested the sensitivity of hydrological predictions to classification errors by using the EDYS ecological model developed for St. Charles Bay watershed, Texas, USA. Three image-based computer classification methods (object-based, supervised, and unsupervised) were assessed against a detailed field-based survey by using global sensitivity analysis and RMSE. The objective was to evaluate relative impacts of each classification method on a range of hydrological variables and across rainfall extremes. Hydrological outputs of EDYS were strongly influenced by different classification methods and rainfall patterns, but no consistent patterns or trends emerged. Thus, no single classification method was universally superior, suggesting a complex interrelation between classification, hydrological variables, and climatic conditions. This research underscores the importance of method selection in hydrological modelling, and so care should be taken when choosing vegetation classification methods to properly address specific research or management objectives.