Frequent lightning activity has the potential to cause damage to man-made facilities, cause forest fires and other hazards, and the prediction of lightning activity can help to avoid the occurrence of these disasters. In this paper, based on the lightning activity data of a region, the distribution pattern of lightning activity is identified at different elevations and latitudes and longitudes. Then geodetic distance and contributing nearest-neighbor similarity are introduced, and a GS-DBSCAN clustering algorithm is proposed to realize the spatial prediction of lightning activity by using the method of leastsquares fitting of prediction equations. The lightning activity directions after data clustering show topographic correlation, and the overlap between lightning activity directions and topography is about 35%. Combined with the prediction images, it is found that the lightning activity prediction results of this paper’s method are closer to the real value than other algorithms, with an average offset error of less than 1.1km, an accuracy rate of >85%, and a false alarm rate of <35%, which reflects a good prediction performance.