We present a series of techniques that we have been using to process GPS readings to increase their accuracy. In a study of urban pollution, we have deployed a number of tracked mobile pollution monitors comprising a PDA, GPS sensor and carbon monoxide (CO) sensor. These pollution monitors are carried by pedestrians and cyclists. Because we are operating in an urban environment where the sky is often occluded, the resulting GPS logs will show periods of low availability of fix and a wide variety of error conditions. From the raw GPS and CO logs we are able to make maps of pollution at a 50m scale. However, because we know the behaviour of the carriers of the devices, and we can relate the GPS behaviour and known effects of CO in the environment, we can correct the GPS logs semi-automatically. This allows us to achieve a roughly 5m scale in our maps, which enables us to observe a new class of expected environmental effects. In this paper we present the techniques we have developed and give a general overview of how other knowledge might be integrated by system integrators to correct their own log files.