Language frontiers do not have to correspond to clearly defined geographical borders. There are several well known persistent language frontiers that divide ecologically very similar regions, such as that between Welsh and English speakers in southern Britain, Breton and French speakers in northwest France, Gallo-Romance and Germanic speakers in Switzerland, and Iroquoian and Algonkian speakers along the Hudson valley (to take some examples given in The Horse, the Wheel, and Language by David W. Anthony).

These kinds of frontiers simply seem to be formed when linguistically distinct groups come into contact through migration, so there is no requirement that they correspond to natural borders. Of course, natural borders would certainly help separate different language communities, especially if their respective cultures are adapted to different ecological zones. But they are not necessary for language frontiers to form.

A corollary is that these frontiers can move with time. The modern distribution of Dravidian languages shows some isolated pockets north of the frontier, even as far north as Afghanistan and Pakistan (Brahui; but see EvanRWT's comment which suggests that it may be a more recent transplant -edited). That suggests that the frontier between Indo-European and Dravidian languages changed greatly over time as Indo-European-speaking areas expanded over previously Dravidian-speaking areas.