CA, 9-11 Nov. 1999 
  
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
THE EFFECTS OF LIDAR DSM GRID RESOLUTION ON 
CATEGORISING RESIDENTIAL AND INDUSTRIAL BUILDINGS 
J. Jaafar, G. Priestnall and P.M. Mather 
School of Geography 
The University of Nottingham 
England 
jaafar@ geography.nottingham.ac.uk 
gary.priestnall@nottingham.ac.uk 
paul.mather@nottingham.ac.uk 
KEY WORDS: LIDAR, DSM, 3D Surface Models, RMSE, Resolution, Classification, Land Use. 
ABSTRACT 
This paper reports the initial results of an analysis of LIDAR data using statistical methods in an attempt to categorise residential and 
industrial buildings. Two study sites representing these two land uses are identified. 3D models for the study area are constructed using 
an integrated methodology utilising the building polygons from digital map data and a LIDAR Digital Surface Model (DSM). The 
effects of LIDAR DSM grid resolution on the construction of the 3D model are analysed. Using statistics such as the Root Mean Square 
Error (RMSE) of the derived models at various LIDAR grid resolutions, the nature of building roofs in the residential and the industrial 
areas are identified. The height differences between the derived height and control height adopted for the 3D models are also determined 
for categorising the two building types. The Standard Deviation (Sd. Dev.) of building height at various LIDAR DSM grid resolutions is 
also investigated as a discriminating measure. It is found that, due to the nature of the roof types that correspond to residential and 
industrial buildings, a classification of building types is possible. 
1 INTRODUCTION 
LIDAR (LIght Detection And Ranging) has become an 
established technique for deriving geometric information in three 
dimensions with decimetre accuracy (Lohr, 1998, Wehr and Lohr, 
1999). Using this new technique, accurate Digital Surface Models 
(DSM) which portray both the grounds surface and the above 
surface features can be constructed in a relatively short time. A 
diversity of applications, which utilise LIDAR datasets is 
described by Gruen er al. (1995, 1997). Among these applications 
is the automatic generation of 3D models of buildings and other 
man-made objects, and the construction of Digital Elevation 
Models (DEMs) by ‘stripping off’ the above surface features 
(Hug, 1996, Jaafar et al., 1999a). 
The fusion of available 2D vector databases with LIDAR DSMs 
offers the potential for rapid construction of 3D models (Haala, 
1999, Jaafar et al., 1999b) which potentially can be of benefit in 
various applications. However, this approach results in 3D 
models with flat roofs (Jaafar er al., 1999b), unless the building 
primitives which constitute the geometry of the roof shape are 
well defined or determined from the LIDAR DSM (Haala, 1999). 
Understanding the nature of the roof top (which can be flat, gable 
or complex) could play an important role in categorising building 
types, and therefore in distinguishing residential from industrial 
land use. Jaafar er al. (1999b) suggest that, by experimenting with 
the LIDAR DSM resolution for the construction of the 3D model, 
the nature of the roof types of individual buildings could be 
revealed to some extent. 
In this study, 3D models are generated using an integrated 
methodology (Jaafar et al, 1999b) based on 2D building 
polygons and a LIDAR DSM. Samples of buildings that 
correspond to residential and industrial areas, which have 
different types of roof structure, are identified. The effects of the 
grid resolution of the LIDAR DSM in the creation of 3D models 
using the integrated methodology are analysed. The Root Mean 
Square Error (RMSE) between the derived height (maximum or 
mean) retrieved from the LIDAR DSM at various grid resolutions 
and the reference height (mean or maximum) at 2m resolution is 
computed for each of the samples identified. The result portrays 
the RMSE with respect to building type that will help categorise 
residential and industrial built up areas. Apart from assessing the 
effects on RMSE, the standard deviation of the building heights 
derived from the LIDAR DSM at various grid resolutions is also 
analysed. The result of this analysis may suggest the optimum 
grid resolution of LIDAR datasets for use in differentiating 
residential and industrial land uses. 
2 THE NATURE OF THE DATA 
LIDAR coverage was supplied by the UK Environment Agency 
for an area of the Trent floodplain covering the West Bridgford 
residential area and the Colwick Industrial Estate, 
Nottinghamshire, England. The areal coverage for each scene is 
2km x 2km with a spatial resolution of 2m. The accuracy of the 
LIDAR dataset in this study is reported to be + 0.20m (Jaafar et 
al., 19993).