Noise pollution in large urban areas is considered as a serious environmental problem. Studies have shown that more than 20% of the world population lives under unacceptable noise levels. The problem is mainly caused by road traffic.

To assess the impact of noise, noise levels need to be predicted by noise computer models and represented on noise maps. GIS functionalities are commonly used to map and assess the impact of noise. An example of a noise map is shown in Figure 1 (Kluijver and Stoter, 2003). This figure shows noise levels along a road and railway. Current noise maps are in 2D representing noise levels, mostly as noise contours, on one selected height (for example at a height of four meters) from the surface.

A disadvantage of this method is the lack of insight in the three dimensional character of noise. In many situations noise levels at four meters do not represent the level at higher floors of a building correctly. The difference is especially large when the building is located close to the noise source or when a noise barrier is present. People living on lower floors of an apartment building benefit more from a noise barrier than people living on higher floors. 2D noise maps are insufficient to represent these situations. Consequently, 3D representations of the noise levels are needed. Several examples of 3D noise maps are known. Paris and Honk Kong already produced 3D noise maps (see Butler, 2004; respectively Wing and Kwong, 2006).

The research presented in this paper is focused on a methodology to improve noise modelling and 3D visualisation of noise in urban area by applying 3D GIS. In our research (executed within the MSc programme 'Geo-information Science and Earth Observation for Environmental Modelling and Management', see http://www.gemmsc.org/),3D GIS functionalities were incorporated in the noise prediction phase as well as in the phase of generating noise representations in 3D and using these representations in the noise assessment phase. For the 3D approach we studied both 2D interpolation methods to produce 2.5D representations - representing noise levels at a surface following the height of the terrain including buildings - and 3D interpolation methods to produce a full 3D voxel model of noise levels. It also reports on the methodology to generate the 2.5D and 3D representations using 3D city models. The results of the 2.5D and 3D noise representations are also presented. The 2.5D noise representation is applied to a real world noise application in order to show the improvements of a 2.5D approach compared to 2D noise maps.

The area chosen for this research is located in the centre of the city of
Delft, the Netherlands. Delft is a city of around 95,000 people in the densely populated South Holland province of the Netherlands. The population density in Delft is about 1,179 in-habitants per square kilometres. The study area is a small part of the city centre of approximately 30,000 m2 and contains about 185 residential buildings with an average height of 15 meters.

A 3D city model covering the study area containing details of the buildings was provided by Vosselman et al., 2005. The city model, shown in Figure 2, is constructed based on an interactive segmentation of the parcel boundariesusing several tools for splitting the polygons along height jumps edges. The roads, canals and trees were also reconstructed from the combination of parcel boundaries and laser altimeter data.

The 3D city model was used to build a 3D noise computer simulation model. Computer simulation models are used in most cases to determine noise levels. Computer simulations are preferred to noise measurements. There are several reasons for this. First of all, field measurements are time consuming since the noise levels concern the yearly averaged values and can only be done under the right weather conditions. In practise, it is impossible to execute an adequate number of measurements in order to produce reasonable noise maps. Furthermore, it is impossible to determine future noise levels by measurements except with noise simulation models to deal with future situations. In addition, models can predict noise levels within an acceptable level of uncertainty for most situations.

Therefore noise calculation software, implementing standardised and approved calculation methods, is widely accepted to provide reliable information on noise levels. These noise computer models calculate noise levels at ‘virtual microphones’ each of which is a point that re-ports what the noise level would be at a certain location under given circumstances. Heights of buildings, of roads and of other topography are taken into account in calculating the noise level at a certain x,y,z location.

We selected Standard Calculation Method 1 (a standardised Dutch method) to predict noise levels in our research since it takes into account the obstruction of noise by objects (such as buildings) but it is still relatively simple to use and can be easily integrated with GIS software. At the same time, it meets the requirements for our research (to see how 3D GIS can improve 3D noise applications). In the computer model, noise levels are computed on 3D data points based on:

a) information on the noise source (roads in our case): traffic intensity, maximum speed, road surface type, average emission of different vehicle types;

b) information on aspects that influence noise propagation such as noise obstruction by objects (like buildings or noise barriers) and noise absorption(like open areas with grass or bare soil);

c) distance and direction of the data points with respect to the location of the noise source.

A 2.5D noise representation was build by the following steps: Positioning of observation points

1) in the noise simulation software. The points were located in 3D on a surface following the terrain and buildings located on the terrain (see Figure 3 (a). Figure 3 (b) shows how points were positioned leaning slightly towards the buildings. This to avoid points that have same x,y,z coordinates which is not possible for 2D interpolation method (see step 3).

2)Calculating the noise level on the observation points (Figure 4);

3)Determining 2D noise contours with a 2D interpolation method using the levels on the 3D observation points (Figure 5). The z coordinate of these points was not taken into account during this 2D interpolation but is reintroduced in the next step;

4) Introducing the third dimension by draping the 2D noise contours on the city model. The 3D analyst tools of ArcScene were used to generate these 2.5D representations.

The 3D noise representation was built by the following steps:

1) Positioning of observation points in a 3D raster. In this raster of points, points may have same x,y but different z coordinates. The points are distributed evenly with equal intervals in both horizontal and vertical directions (2 m) in ‘lines’ parallel to the roads.

2) Calculating the noise level on the observation points.

3) Determining the 3D solid noise model with a 3D interpolation method. With this method an extra step to reintroduce the third dimension is not necessary.

For both methods, the positioning of the points was based on the following considerations:
•Noise contours are expected to be parallel to the roads and points located in a pattern parallel to the road can reflect this behaviour most optimally.
•Care was taken not to place points inside buildings, because buildings act as blocking objects in the model and these points would produce low levels which are not representative for the levels on the façades of the buildings.