The spatial distribution of emergent vegetation in rivers is frequently nonuniform. Here, spatial nonuniformity is simulated in the laboratory by varying longitudinally the stem areal number density (m). Time- and space-averaged flow variables, corresponding to different values and gradients of m and different stem Reynolds numbers, are calculated from instantaneous velocity maps obtained with particle image velocimetry (PIV). Results show that Reynolds stresses are completely determined by the local spatially averaged number of stems per unit area; they are not sensitive to local spatial gradients of m. These variables seem to adjust locally to the spatial variations of the stem distribution. Contrarily, the spatial distribution of the time-averaged flow is influenced not only by the local value of m but also by its spatial gradients. Form-induced stresses express this influence. They reveal a subsistence of flow complexity generated by the upstream stem distribution as a form of spatial memory. Thus, the effect of spatial gradients of m is always in the direction of increasing form-induced stresses although at different rates depending on the signal of the former.