Croon et al. challenge our conclusion that sea-floor depth variations are driven by the underlying mantle convection. We point out that, contrary to their claim, our data analysis is pertinent and that the sea-floor linear trend as the square root of the distance from the ridge is a robust observation. The mechanism responsible for this trend is an asthenospheric flow, faster than the overlying plate, which shapes the lithosphere structure. E ven if it is widely accepted that mantle convection exists, its causes and consequences are still subject to debate. We investigated the subsidence of the oceanic lithosphere and proposed, based on the sea-floor depth analysis, that it is driven by the underlying mantle convection (1). This approach, different from previous studies, is strongly criticized by Croon et al. (2). Here, we provide answers to what they contend are shortcomings of our work.Croon et al. first claim that our flow lines do not match the trajectories of our illustrative profiles. Our computation was based on the velocity grid obtained through the UNAVCO Plate Motion Calculator Site (1), for the NUVEL-1A kinematics model, in the No Net Rotation reference frame. We doubled-checked our trajectories estimation by refining the velocity grid step down to 10 km. We confirm that the trajectories shown in (1)-very similar to Croon et al.'s in the northern part-are correct. We would appreciate a deeper comparison between the methods in order to understand where the discrepancy comes from.We agree that one of the major problems when studying the sea-floor thermal subsidence is to remove from its topography all data that could be linked to any different process. In other words, "The central problem is in satisfactorily defining normal" (3). Many methods have been proposed to remove abnormal sea floor from the data sets, from statistics (4) to manual removal of topography (5). As pointed out by Kim and Wessel (6), robust methods for filtering the bathymetry do exist, but in any case, the use of a median filter alone is inadequate for removing the shallow topographic features. This point has been clearly demonstrated by several authors, including the authors of the comment (7). Indeed, the median filter has "a tendency to pass through topography" (7). It tends to overestimate the seafloor main trend, especially at old ages where these features are statistically concentrated (8), thus introducing an apparent "flattening." This artifact is enhanced by stacking the profiles [see figure 2A in (2)].In our study (1), we did not attempt to remove any "abnormal" sea floor on purpose. Because the wavelength of the intraplate anomalies is much shorter than the subsidence trend, they did not interfere with its visual determination (1). In Fig. 1, we show that, by using a method based on an adequate bathymetry filtering, the MiFil method (9), the trend can be automatically recovered in a reproducible way. The results from this method are in good agreement with our previous fit.The discussion of Croon et al. (2) on the short wa...