A viewshed approach describes an area that can be seen from a certain point – the viewpoint – that can have different heights. This approach requires two main inputs: a digital elevation model and viewpoints.
DEMs are raster data files containing elevation data (see section 4.2.4). A raster file’s resolution is the side length of a pixel. The higher the resolution, the lower the side length of the pixel. Using a higher resolution raster will yield more accurate results. In the viewshed approach, all the cell towers act as viewpoints. Viewpoints are placed on raster cells.
A viewshed is created from a DEM by using an algorithm that estimates the difference in elevation from one cell (the viewpoint cell) to the next (the target cell). To determine the visibility of a target cell, each cell between the viewpoint cell and target cell is examined for line of sight. Where cells of higher value are between the viewpoint and the target cell, the line of sight is blocked. If the line of sight is blocked, then the target cell is determined not to be part of the viewshed. If it is not blocked, then it is included in the viewshed. For purposes of calculating SDG indicator 9.c.1, there was no target cell, but rather a target radius.
The viewpoint height is the elevation added to the cell in the calculations. The higher the viewpoint, the more can be seen from it. The viewshed acts like a flashlight on an object – items can be seen where the light falls but not in the shadows. The output is a raster file in which cells are given values based on how many observers can see the target cell. This cell is then translated into a binary raster that has values of 1 and 0, which translates into areas that can be either seen or not seen “from viewpoints”. The result is converted into a polygon and split based on the LAU level used (in the case of Indonesia, LAU level 2 was used).
The viewshed method provides a more accurate and realistic model of real life in mountainous areas where there are large physical barriers, but it requires far more computational power. Having millions of viewpoints will increase the computational time to the point that it might not really be worth using to get an “approximate estimation of the situation”. If this method is used, the radius of each tower must be known for the estimate to be more accurate. It will also have to be modified to obtain a model as close to real life as possible, because it creates uniform viewpoints, but some cell towers are pointed in certain directions. In addition, the height of each tower would also need to be known to properly calculate the coverage.
The OpenCelliD database does not contain radius and viewpoint. Consequently, if using open-source data on cell location, assumptions on the range and height of the cell must be made. In the study in Indonesia, different ranges and heights were tested. The ranges for the viewshed method were the same as for the flat coverage method: 35 kilometers for 2G; 10 kilometers for 3G; and 5 kilometers for 4G. The viewpoint heights tested for these ranges were 30 meters and 150 meters. However, the study in Indonesia also included an “optimal” scenario based on evidence of ranges and heights in Estonia. As in the flat method, this scenario used the uniform ranges of 10 kilometers irrespectively of the cell technology. For the viewshed method, the scenario used a cell height of 50 meters since evidence showed that two-thirds of the Estonian MNO’s cell towers were up to 50 meters high (see figure 13).
Figure 13. Distribution of cell tower heights, Estonian MNO, reference period 
Source: Positium.
These two parameters can be recalculated for each country and region. If access can be given to an MNO cell database with information on range and cell height, the viewshed can even be calculated for each cell separately.
Ideally, the viewshed method should be performed using actual information on range and cell height. However, if data is missing, the recommendation is to use evidence on range and cell height from other countries. The assumptions of 10 kilometers and 50 meters presented in this document are based on evidence from Estonia. However, it will be important to gather and assess data on ranges and cell height from more countries to identify optimal parameters.
Besides height information, MNOs also have information on cell azimuth, tilt and power that could help significantly to improve cell coverage area calculations.[1] Unfortunately, experience with various MNOs is that, when this information is provided, it can often contain errors. For example, if additional data is available, it is often not available for the entire network. Secondly, it can be very difficult to confirm the validity of the entire dataset since MNOs change the direction of the cell or do not put in the right azimuth from the start. However, if both cell data and mobile event data are available, then it would be possible to estimate cell azimuth via neighbouring cell analysis. Both could then be used to calculate cell coverage areas.[2]
Using mobile events means that cell handover information can be included in coverage area calculations. Cell handover data show how “connected” cell towers are. In an ideal situation, cells located close to each other, and with coverage areas directed towards each other, have higher event handover counts, and cells that are apart should have no handovers at all. Handover depends on the extent to which cell coverage areas overlap and is useful for creating a coverage area when MNOs cannot provide the relevant information and to detect cells for which they have provided incorrect locations. Handover data can also be used to identify cells with probable erroneous locations. Two main types of issue can be identified: cells with long-term incorrect coordinates (e.g. human error when updating antenna location in the MNO’s systems or relocating antenna without changing the information) and moving cells (portable antennas, e.g. used for large crowd events).
The viewshed analysis can be performed in geographic information system (GIS) tools such as QGIS. It requires data on cell location, administrative areas, population, and a digital elevation model for the area (see Sections 4.2.1 to 4.2.4). Performing the viewshed method can be summarised as follows:
- Separate different generations of mobile cell masts (2G, 3G and 4G).
- Create viewpoints with the desired radius from the results of 1.
- Generate the viewshed for different generations (2G, 3G and 4G).
- Make the result raster into a binary (yes/no) format (for QGIS raster calculator ("raster_result@1" > 0) * 1").
- Convert the raster into a vector.
- Clip the converted vector with the area of interest.
- Collect the geometries by location and assign attributes to them.
- Summarize attributes by location to get population results per area.
While the flat approach may be the easiest to calculate, it most likely causes coverage areas to be overestimated, especially in mountainous areas, because it does not consider elevation and other natural barriers.
[1] Azimuth is the coverage angle of the antenna of a point on the horizon. On a triangular cell tower with three antennas, each antenna would have to provide 120 degrees of coverage to ensure full coverage around the cell tower location.
[2] Based on Positium’s experience, using mobile event data to calculate coverage areas has resulted in more trustworthy coverage areas.