Marine environmental data are always multi-source and multi-versional. This is because the different methods used for data collection vary in their retrieval algorithms, procedures used and purposes, so that data processing has various spatial and temporal resolutions with different errors. Hence, it is important to know whether results derived from different versions of the same data are consistent and the models can be correctly used by testing other versions of the data. For this purpose, we collected sea surface temperature data from different web sizes using linear regression and randomization tests to evaluate the effects of different data versions on the parameter estimations and predictions of habitat suitability index models. The results showed that because of system errors in the data, the parameters estimated differed significantly and the models were unable to make correct forecasts by inputting other versions of the data. Dispersion between different data versions reflected the random errors inherent in the data and led to uncertainty in the results of the habitat suitability index models. Accordingly, we suggest that model outputs are quantified for uncertainty to ensure that scientific data can be reliably used in fishery resource management.