It's not really an either-or equation. Cost of developing early warning systems is fairly well understood - the more warning you want, the more complex and expensive the system gets. Also, such systems get more unreliable the more lead time is required. Cost of developing shelters is also pretty well understood, based on the number of people they're intended to shelter-in-place, the prevailing types of storms in any particular area, et cetera. The question becomes, "How much lead time is required to get people to the nearest shelter (which changes as the number of shelters goes up, so it's not as far to the nearest shelter with room for more people in it)? I'd sooner spend a few millions just on studying that question than put another single thin dime into the TSAs pot.