At present, the main problem faced by ground-based augment system (GBAS) is that though carrier smoothing filter and local differential global positioning system (LDGPS) improve the accuracy of the pseudorange by reducing the noise in it and eliminating almost all the common errors between the user and the reference station, they also cause extra errors on account of the effects of the ionosphere temporal and spatial gradients. Based on the analysis of these errors as well as the smoothing noise, this article suggests a new algorithm to design the optimal Hatch filter, whose smoothing window width varies real-time with the satellite elevation, ionosphere variation, and distance from the user to the reference station. By conducting the positioning process in the GBAS emulation platform for several hours and after its comparison with the performances of traditional Hatch filters, it is found that the errors in the differential correction become smaller and the positioning accuracy gets heightened with this new method.