Extracting sources with low signal-to-noise from maps with structured background is a non-trivial task which has become important in studying the faint end of the submillimetre number counts. In this article we study source extraction from submillimetre jiggle-maps from the Submillimetre Common-User Bolometer Array (SCUBA) using the Mexican Hat Wavelet (MHW), an isotropic wavelet technique. As a case study we use a large (11.8 arcmin^2) jiggle-map of the galaxy cluster Abell 2218, with a 850um 1sigma r.m.s. sensitivity of 0.6-1mJy. We show via simulations that MHW is a powerful tool for reliable extraction of low signal-to-noise sources from SCUBA jiggle-maps and nine sources are detected in the A2218 850um image. Three of these sources are identified as images of a single background source with an unlensed flux of 0.8mJy. Further, two single-imaged sources also have unlensed fluxes <2mJy, below the blank-field confusion limit. In this ultradeep map, the individual sources detected resolve nearly all of the extragalactic background light at 850um, and the deep data allow to put an upper limit of 44 sources per arcmin^2 to 0.2mJy at 850um.

11 pages, 6 figures. Accepted for publication in MNRAS. Figures 2 and 4 (lower panel) available in higher resolution http://www.mpia.de/homes/knudsen/papers/mhw_paper.html