The presence of a mantle plume beneath Iceland has long been hypothesised to explain its high volumes of crustal volcanism. Practical constraints in seismic tomography mean that thin, slow velocity anomalies representative of a mantle plume signature are difficult to image. However it is possible to infer the presence of temperature anomalies at depth from the effect they have on phase transitions in surrounding mantle material. Phase changes in the olivine component of mantle rocks are thought to be responsible for global mantle seismic discontinuities at 410 and 660 km depth, though exact depths are dependent on surrounding temperature conditions. This study uses P to S seismic wave conversions at mantle discontinuities to investigate variation in topography allowing inference of temperature anomalies within the transition zone. We employ a large data set from a wide range of seismic stations across the North Atlantic region and a dense network in Iceland, including over 100 stations run by the University of Cambridge. Data are used to create over 6000 receiver functions. These are converted from time to depth including 3D corrections for variations in crustal thickness and upper mantle velocity heterogeneities, and then stacked based on common conversion points. We find that both the 410 and 660 km discontinuities are depressed under Iceland compared to normal depths in the surrounding region. The depression of 30km observed on the 410 km discontinuity could be artificially deepened by un-modelled slow anomalies in the correcting velocity model. Adding a slow velocity conduit of -1.44% reduces the depression to 18 km; in this scenario both the velocity reduction and discontinuity topography reflect a temperature anomaly of 210 K. We find that much larger velocity reductions would be required to remove all depression on the 660 km discontinuity, and therefore correlated discontinuity depressions appear to be a robust feature of the data. While it is not possible to definitively rule out the possibility of uncorrected velocity anomalies causing the observed correlated topography we show that this is unlikely. Instead our preferred interpretation is that the 660 km discontinuity is controlled by a garnet phase transition described by a positive Clapeyron slope, such that depression of the 660 is representative of a hot anomaly at depth.