【摘要】 本文基于1970—2024年长沙地区降水量逐日观测数据,综合运用多源统计和机器学习算法,系统揭示了长沙地区降水量时间多尺度特征。结果表明:（1） 1970—2024年长沙市年平均降水量为1 429.6 mm（年变化范围为932.8～1 991.8 mm）,日最大平均降水量为102.2 mm（年变化范围为37.7～475.1 mm）,年平均降水日数为156.7 d（年变化范围为122.0～213.0 d）。近55年来小雨、中雨、大雨、暴雨、大暴雨、特大暴雨所占比例分别为71.80%、18.36%、7.33%、2.28%、0.22%、0.01%。（2） 1970—2024年年降水量呈缓慢增加的趋势,线性倾向率为0.72 mm·a^-1（α=0.1）;降水日数线性倾向率为-0.5 d·a^-1,呈现显著下降趋势（α=0.01）。长沙市年降水量突变点主要集中在1996—2002年和2020年前后;1996—2002年突变点与东亚夏季风指数密切相关,2020年以后年降水量序列突变点可能与近年气候变化和波动有关。（3）相对湿度在所有年份中始终是影响年平均降水量的最重要的因素,SHAP（Shapley additive explanations）贡献权重为32.4%;东亚季风指数是影响日最大降水量的最重要的因素,SHAP贡献权重为26.8%。优选建立的长沙本地Transformer降水量机器学习模型模拟预测误差较低,能有效捕捉降水量的变化规律。更多还原

【Abstract】 Based on the daily precipitation observation data of Changsha from 1970 to 2024,the multi-time-scale characteristics of precipitation in Changsha were systematically revealed by using multi-source statistics and machine learning algorithms comprehensively.The results were as follows:（1） From 1970 to 2024,the annual average precipitation was 1 429.6 mm,with an annual variation range of 932.8 to 1 991.8 mm.The daily maximum average precipitation was 102.2 mm,with an annual variation range of 37.7 to475.1 mm,and the average number of precipitation days was 156.7 d,with an annual variation range of 122.0 to 213.0 d.Over the past 55 years,the proportion of light rain,moderate rain,heavy rain,torrential rain,severe torrential rain,and extreme torrential rain was 71.80%,18.36%,7.33%,2.28%,0.22%,and 0.01%,respectively.（2） The annual precipitation in Changsha from 1970to 2024 showed a slow increasing trend,with a linear tendency rate of 0.72 mm·a^-1（α=0.1）.The linear tendency rate of the number of precipitation days was-0.5 d·a^-1,indicating a significant decreasing trend（α=0.01）.The abrupt change points of the annual precipitation in Changsha were mainly concentrated from 1996 to 2002 and around 2020.The abrupt change points from 1996 to2002 were closely related to the East Asian summer monsoon index,while the abrupt change point around 2020 might be associated with recent climate changes and fluctuations.（3） Surface relative humidity was consistently identified as the most important factor affecting the annual average precipitation in all years,with a SHAP contribution weight of 32.4%,while the East Asian monsoon index was the most important factor affecting the daily maximum precipitation,with a SHAP contribution weight of 26.8%.The locally optimized Transformer machine learning model for precipitation in Changsha was established,which exhibited relatively low simulation and prediction errors and could effectively capture the variation patterns of precipitation.更多还原