Subglacial roughness can be determined at variety of length scales from radio-echo sounding (RES) data; either via statistical analysis of along-track topography, or inferred from basal radar scattering. Past studies have demonstrated that subglacial terrain exhibits self-affine (fractal) scaling behaviour, where vertical roughness has a power-law relationship with the horizontal length scale. A self-affine statistical framework, which enables a consistent integration of topographic roughness and radar scattering, has yet to be applied to RES. Here we do this for recent RES data from northern Greenland, and demonstrate that subglacial topography exhibits pronounced spatial variation in the Hurst (roughness power-law) exponent. A radar scattering model then enables us to explain how the Hurst exponent exerts strong topographic control upon radar scattering, which we map using the waveform abruptness (pulse peakiness) parameter. Notably, lower abruptness (associated with diffuse scattering) occurs for regions with a higher Hurst exponent, and higher abruptness (associated with specular reflections) occurs for regions with a lower Hurst exponent. Finally, we compare the RES-derived data with an independent prediction for the subglacial thermal state of northern Greenland. This analysis shows that the majority of predicted thawed regions do not have the specular RES scattering signature of deep subglacial lakes, and instead have a diffuse scattering signature.