In the light of high temperature days of drastic variety, a new viewpoint is put forward to predict short-term climate by theoretical analysis such as weighing superposition, periodic change, scaling analysis etc. We found that long periodic change is often related to local forward long periodic change and that short periodic change is closely related to ambient environment factors. By using the means such as power analysis, binomial smoothing, step-by-step regression etc., prediction equations of long periodic change and short periodic change are established. Based on them,prediction equation of high temperature days is found. Using the method, Hefei's high temperature days is tried out. 13 forecast samples are given with adopting rolling forecast. 11 forecasts among them are accurate, so forecast accuracy attains 84.6%. Even in inaccurate 2 samples errors only amount 5~6 d. For this reason the forecast result is satisfying.