multiply rounding errors by thousands and you get somewhat meaningful impact. you are underestimating the scale of independently small pollutions