Spatial　clustering　is　an　important　method　for　spatial　data　mining　and　knowledge　discovery.　According　to　the　deficiency　existing　in　density-based　clustering　algorithm　DBSCAN,　such　as　the　I/O　overhead,　memory　consumption　etc.　This　paper　improves　the　DBSCAN　algorithm,　which　proposed　directional　density　algorithm,　the　algorithm　reduces　lots　of　points　which　need　to　be　queried.　By　taking　Geographic　Information　System　for　the　application　background,　we　successfully　applied　to　forecast　the　distribution　of　urban　water　points.　Compared　with　the　traditional　DBSCAN　algorithm,　the　results　conformed　to　the　actual　situation,　and　efficiency　increased　by　20%.