The traditional approach to delineating and extracting features from remotely sensed images relies predominantly on manual interpretation, a procedure that is often time consuming and expensive. Automation offers the potential for reduced costs and wider utilization of remote sensing within the business community, but involves difficulty in representing the expertise of remote sensing scientists within a series of decision rules. The objectives of this paper are two-fold: firstly, to produce a system for automated feature discrimination in remotely sensed images, using leaks from water supply networks as a case study; and secondly, to test whether the system is suitable for use with the next generation of satellite sensors. The automated system was calibrated by integrating HyMap and Airborne Thematic Mapper (ATM) images with context data from a variety of sources (such as ambient irradiance environment; topography; land use, and field boundaries). The automated system was assessed for its applicability to satellite remote sensing by testing the system on airborne data that were degraded to the resolutions of satellite images. It is proposed in this paper that automation, particularly with respect to satellite remote sensing, makes leak detection from water supply networks commercially viable.
The development of techniques for the detection of water leaks from underground pipelines is seen as a high profile activity by water companies and regulators. This is due to increasing water demands and problems with current leak detection methods. In this paper optical reflectance and microwave backscatter models (SAIL + PROSPECT and RT2) were used to try and identify optimal indices for detecting water leaks amongst a variety of different land cover types at different growth stages. Results suggest that red/near infrared and red/middle infrared ratios show potential for leak detection. Given the sensitivity of L-band radar to moisture, and the ability to separate contributions from canopy and ground surface, it is possible to detect saturated soils through vegetation canopies. The results of both approaches are used to infer limits of detection in terms of season and meteorological conditions for a range of land covers. Preliminary findings suggest that leaks may be optimally detected when canopy height is low, surrounding soil is dry, and the leak has been present for more than 14 days. The modelled data is compared with L - band fully polarimetric E-SAR data, and 200 channel HYMAP hyperspectral airborne data which were acquired over an 8km section of the Vrynwy aqueduct (UK), which included a high concentration of leaks. Data was acquired as part of the British National Space Centre (BNSC) and Natural Environmental Research Council (NERC), SAR and Hyperspectral Airborne Campaign (SHAC) in June 2000. The results from this work suggest that remote sensing is both an effective and feasible tool for leak identification.