We investigated the first and last frost dates using meteorological observation data from Japan. First, we identified long-term trends of retardation (mean: +0.224 day/year) in the first frost date in fall and advancement (−0.228 day/year) in the last frost date in spring using historical frost observation data from 1951 to 2010. Trends determined over 20-year subperiods were distinct from long-term trends and were sensitive to decadal changes in daily minimum air temperature. Second, we proposed a scheme to infer the first and last frost dates from the time series of daily minimum air temperature. After optimization, the first (last) cold date when the daily minimum temperature fell below a temperature criterion of ca. 2 °C yielded the best estimate of the first (last) frost date with an error of 2 days in most cases. However, the overall root mean square errors were 13–16 days because some cases with significant misfits deteriorated the values. Both wind speed and humidity shifted the criterion. Sites with strong winds contributed to a decrease in the temperature criterion. Because this scheme only requires the daily minimum air temperature, it is widely applicable to the quantitative evaluation of the first and last frost dates from given meteorological or climate projection data.

我们使用来自日本的气象观测数据调查了第一和最后的霜冻日期。首先，我们使用1951年至2005年的历史霜冻观测数据，确定了秋季第一个霜冻日期的迟缓（平均：+0.224天/年）和春季最后一个霜冻季节的前进（-0.228天/年）的长期趋势。 2010年。确定的20年以下期间的趋势与长期趋势不同，并且对每日最低气温的年代际变化敏感。其次，我们提出了一种从每日最低气温的时间序列中推断出第一和最后霜冻日期的方案。经过优化后，第一个（最后一个）寒冷的日子是当每日最低温度低于约20摄氏度的温度标准时。在大多数情况下，2°C可以最佳地估计出第一个（最后一个）霜冻日期，误差为2天。然而，总体均方根误差为13至16天，因为某些严重不匹配的情况会使值恶化。风速和湿度都改变了标准。有强风的地方导致温度标准降低。由于该方案仅需要每日最低气温，因此可广泛用于根据给定的气象或气候预测数据对初霜日期和最后霜日期进行定量评估。