Airborne laser scanning for altimetry is a new technology

which has been developed during last two decades.To derive topography from laser scanned data

off-ground elements like forest and buildings

which are treated as noise

need to be filtered out.This paper

in addition to the review of recent studies on derivation of topography from laser scanned data

is intended to develop a new algorithm by using average filter for forest areas.By iterative calculation

the surface approached the ground progressively by cutting off convexes above the reference surface.In particular

by introducing topographic factor to transform the surface model to a presumed flat terrain

over filtering or less filtering was significantly reduced.The accuracy（RMS） of the derived DTM was 0.4—0.5m.Data was acquired by Laser Bird

which is a laser radar system produced by Optech Ltd.A digital surface model（DSM） was created with all original data

from which the terrain model was constructed by extracting signals reflected from the ground.The basic procedure is as follows:（1） Creating surface model.With all random foot print data

Delaunay triangle was adopted to create a three-dimensional surface model DSM0.（2） Displaying shaded relief.The interim-results were manually monitored with a shaded relief image to determine whether to continue or stop filtering.（3） Filtering.A smoothed surface model was created by average filter

which is called temporal surface model（DSMt）.Regardless the feature of topography

the size of filter window was set stable with 5m×5m.（4） Approaching the ground.By comparing DSMt with a threshold

smaller values were selected for DSMt modification

and the surface gradually came to the ground.（5） Reiterative computing.Above computations were reiteratively operated

which were ended up when either of following conditions was met

i) the difference between DSMt and threshold reaches the specified value

or ii) reiteration reaches 150 times.（6） Computing the reference surface.Complexity of topography was taken into account.A reference surface model（DSMr）

which is similar to the real one in form

was introduced for computing.DSMr was created by using the filter with a larger window.Considering the relationship between DSM0 and DSMr

the rough surface was transformed to a plain one（DSMp）

which was then filtered

and the topography was restored after calculation.（7） Classificating the data.With a threshold ΔH

signals from canopy or ground were separated by comparing with the original LP data.Points with LP-DSMt≤ΔH were considered as ground

which were used for DTM construction.（8） Constructing terrain model.Foot prints on ground were extracted from the original LP data to create digital terrain model.（9） Testing the accuracy.The elevations of 16 points in the study area were manually measured

which were used for accuracy evaluation.