Future sea level rise uncertainties are mainly due to uncertainties in Antarctic ice sheet projections. Indeed, modelling the future of the Antarctic ice sheet presents many challenges. One of them is being able to model the physical interactions between the ocean and the ice shelves. As a result of limited understanding of these ice-ocean interactions and limited computational resources, these interactions are parametrized rather than explicitly resolved in most ice sheet models. These parameterisations vary in complexity and calibration method, eventually leading to differences in resulting sea level rise contribution of several meters. Here we present the implementation of the PICO basal ice shelf melt module in the GRISLI v2.0 ice sheet model. We compare six different statistical methods to calibrate PICO and assess how robust these methods are if applied at different resolutions and areas of the Antarctic ice sheet. We show that computing the Mean Absolute Error of the bins is the best method as it allows us to match the entire distribution of melt rates retrieved from satellite data at different resolutions as well as for different Antarctic ice shelves. It also results to a smaller parameter space than the other tested methods. This method makes use of melt rate bins and minimizes the differences between the values of the bins of the model and the ones of the observational target. We find that, using this method, region-specific calibration of ice-ocean interactions is not needed and we can avoid using ocean temperature bias corrections. Finally, we assess the impact of the implementation of PICO in GRISLI and of the calibration choice on future projections of the Antarctic ice sheet up to the year 2300.