A geographical model of radio-frequency power density around mobile phone masts

Public concern about possible health effects of EMF radiation from mobile phone masts has led to an increase of epidemiological studies and health risk assessments which, in turn, require adequate methods of exposure estimation. Difficulties in exposure modelling are exacerbated both by the complexity of the propagation processes, and the need to obtain estimates for large study populations in order to provide sufficient statistical power to detect or exclude the small relative risks that might exist. Use of geographical information system (GIS) techniques offers the means to make such computations efficiently. This paper describes the development and field validation of a GIS-based exposure model (Geomorf). The model uses a modified Gaussian formulation to represent spatial variations in power densities around mobile phone masts, on the basis of power output, antenna height, tilt and the surrounding propagation environment. Obstruction by topography is allowed for, through use of a visibility function. Model calibration was done using field data from 151 measurement sites (1510 antenna-specific measurements) around a group of masts in a rural location, and 50 measurement sites (658 antenna-specific measurements) in an urban area. Different parameter settings were found to be necessary in urban and rural areas to obtain optimum results. The calibrated models were then validated against independent sets of data gathered from measurement surveys in rural and urban areas, and model performance was compared with that of two commonly used path-loss models (the COST-231 adaptations of the Hata and Walfisch-Ikegami models). Model performance was found to vary somewhat between the rural and urban areas, and at different measurement levels (antenna-specific power density, total power density), but overall gave good estimates (R(2)=0.641 and 0.615, RMSE=10.7 and 6.7 dB m at the antenna and site-level respectively). Performance was considerably better than that of both path loss models.