We have studied the coercivity of magnetic nanonetworks as a function of thickness, nominal pore diameter, and surface/interface roughness in the thickness range of approximately 2 to 45 nm where a Neel-type domain wall has been theoretically predicted. Such magnetic nanonetworks have been prepared by sputtering iron on the walls of commercially available porous nanochannel alumina (NCA) membranes. The thickness dependence of coercivity has also been studied on films deposited on surface-oxidized Si and glass subtrates. These substrates are essentially non-porous and much smoother than NCAs. Our investigation shows that the coercivity of films deposited on Si and glass depends on the spatial fluctuation of thickness which arises from the roughness of the apparently smooth substrates. On the other hand, NCAs are found to be inherently quite rough, and films on NCAs show a complex thickness dependence which arises from the interplay between surface/interface roughness, domain pinning due to porosity, surface anisotropy, surface torques, and oxidation of the iron films. It was found that the growing films on NCA substrates led to partial filling up of the pore entrance, thereby reducing its effective diameter. The film growth also affects the roughness of the film, which in turn affects its coercivity. We propose a model for the thickness dependence of coercivity based on the pore fill-up geometry considering the effective pore diameter and the critical thickness at which the pore will be completely filled up. Experimental results on coercivity with thickness variation of iron network deposited on NCA generally agree with the suggested model.