A technique employing a 3D morphological image-registration algorithm is demonstrated for stitching together high-resolution surface images obtained with a commercial atomic-force microscope (AFM), producing 3D surface images up to 1mm long with lateral resolution ~ 100nm. These images can be applied to reflectance modeling by extracting surface parameters to be used as inputs for reflectance models, for instance the previously-published Coherence Model [BG. Hoover and VL. Gamiz, J. Opt. Soc. Am. A 23, 314 (2006)], which utilizes the surface roughness and autocorrelation derivatives in the large-roughness approximation. Surface moments estimated from extended-range AFM images demonstrate lower uncertainty at all frequencies and substantial reduction of sampling artifacts at low frequencies, enabling improved estimates of surface parameters. The autocorrelation of a nearly monoscale diffuse-gold surface is measured out to 800um separation, and the autocorrelation of a multiscale tin surface provides parameters that verify the Coherence Model fit to the measured quasimonostatic BRDF.