Current land-cover classification methods using ICESat/GLAS's (Ice, Cloud, and land Elevation Satellite/Geoscience Laser Altimeter System) datasets are based on empirical thresholds or machine learning by training multiple GLAS parameters, e.g., the reflectivity and elevation of the target and width, amplitude, kurtosis, and skewness of the return waveform. A theoretical classifier is derived based on a waveform model of an actual laser altimeter illuminating the sea surface. With given system parameters and the sea surface wind corresponding to the location of a laser footprint (the wind can be calculated by using the National Centers for Environmental Prediction dataset), a precise theoretical waveform can be generated as a reference. Compared with the measured waveform, a weighted total difference, which is very sensitive to small-scale sea ice within the laser footprint, can be calculated to classify the GLAS measured data as open water. In the north of Greenland, after discarding the saturated GLAS data, the new theoretical classifier performed better [overall accuracy (OA)=95.62%, Kappa coefficient=0.8959] compared to the classical support vector machine (SVM) classifier (OA=90.44%, Kappa=0.7901), but the SVM classifier showed a better result for the user's accuracy of sea ice. Benefiting from the synergies of the theoretical and SVM classifiers, the integrated theoretical and SVM classifier achieved excellent accuracy (OA=98.21%, Kappa=0.9588). In the future, the new ICESat-2 photon counting laser altimeter will also construct a "waveform" (elevation distribution) by selecting the photon cloud, and thus, this new analytical method will be potentially useful for detecting open water in the Arctic.