Improving k-nearest neighbor predictions in forest inventories by combining high and low density airborne laser scanning data