TAFs are the least accurate forecast, because they cover
such a limited geographic area (five miles) and are a somewhat subjective and
local interpretation of potential weather. Scott wrote an excellent article on
this in IFR Magazine about two years ago: here's a link.
The radar tells you what's going on that day and is not useful for strategic planning purposes; the other forecast products are available several days in advance.
Its a misconception to say that soundings are limited to the specific airport: they cover the theoretical area of 60 square miles using the airport identifier as the reference point. In fact, on the Plymouth site you can use any lat/long coordinates and get a sounding for that square area; its just more convenient to use known identifiers.
The original balloon radiosomes covered just the point of launch and drift, and there were only 100 or so for the whole country. The soundings available today are derived from tons of data from radiosomes; ground, radar and satellite observations; and real time readings of temperature, humidity, dewpoint, wind speed, freezing levels collected by thousands of airliners flying around the country every day. This information is then crunched by supercomputers in weather forecasting models that form the basis for virtually all of the forecasting products.
I too was quite skeptical of using mere "forecasts" to make weather planning decisions. But after taking Scott's course and doing my own analysis before each flight, then seeing how the soundings and forecasts compared to actual conditions, I've developed a healthy respect for these incredible tools and products.