Dealing with computationally-intensive calibration processes is still common in distributed hydrological modelling despite the computing power growth. Computational time for one single distributed hydrological simulation may easily consume more than one minute, and the calibration process can require thousands of simulations. The use of surrogate models that are low-cost and representative of the calibration problem is an interesting avenue to reduce the calibration computational time. The first part of this study explores three possibilities to construct reduced-fidelity surrogate models from the HYDROTEL model, a computationally-intensive hydrological model. The relevance of these three types of surrogates and their combination within a calibration process is evaluated according to the best compromise between representativeness and a decrease of CPU time. In a second paper, surrogate models are implemented within an existing efficient calibration process to significantly reduce the computational time.