The term "butterfly effect" has been described as a "sensitive dependence on initial conditions." In other words, there must be a premiere study that later data can look back to. For example, the link below describes the trials that mathematician and meteorologist Edward Lorenz performed on the weather of Boston in the winter of 1961. He attempted to be able to predict weather into the far future but after failing on multiple studies, he realized he would have to go back to previous data from the days earlier to determine the upcoming weather in the near future. This is why even in the 21st century, weather can only be accurately predicted in short periods of time.
The prime example used in Lorenz's study was that if a butterfly flaps it's wings in South America, it could affect the weather in Central Park, New York. Meaning, one minor change in the weather can spark a change for long-term.