nearest neighbor - data structure for movable points in 3d -

i have many points (+100,000) in 3 dimensional space. need use nearest neighbor , range queries. firstly used kdtree (k=3) each point has velocity attribute. location not static, change location. problem begins here. easy nearest neighbor , range queries old locations. have calculate new locations according velocity. have find nearest neighbor , search in range after calculate new location.

everytime points change locations must update kdtree costly. slows me down. have suggestion or there better data structure situation?

an octree or octkey can faster. octkey uses morton curve or hilbert curve. reduces dimension , fills space. it's uses in map application. find orientation in 3d gray code can find correct quadrant. here interesting thread moving objects: http://xboxforums.create.msdn.com/forums/p/59841/368276.aspx. recommend quadtree, linked list or ternary trees.