摘要
针对云南山高谷深、降水差异大的特点,以2010年为例,提出一套气象站与格点相结合的年降水量精细化(1 km×1 km格点)分布推算方法。该方法首先应用主成分分析法(PCA)对全省气象站和格点的地理、地形因子(包括经度、纬度、海拔、坡度、坡向,以下简称因子)去除高相关性处理,生成降维的因子主成分,然后应用K-means聚类法以气象站年降水量及因子主成分为聚类因子,将气象站划分为不同降水类型,然后针对每种降水类型,以气象站因子主成分为自变量,以气象站年降水量为因变量拟合回归方程,并用求出年降水量残差值(实测值与拟合值之差),应用朴素贝叶斯分类法(NBC)学习气象站分类特征,将所有格点归入不同的降水类型,应用每个降水类型的回归方程,求出各格点年降水量拟合值,再将气象站年降水量残差值进行空间插值作为格点年降水量订正量,将格点年降水量拟合值与订正量相加便获得2010年云南全省降水量精细化格点分布。该方法较好地再现了2010年云南年降水的空间分布特点,不仅区分出哀牢山西、东两侧作为西南暖湿气流迎风坡和背风坡的降水类型差异,也识别出西部、南部边缘地带的多雨形态,经气象站检验,该方法平均相对误差(MRE)仅为0.17。
For complicated mountainous areas of Yunnan,A set of methods of fine grid(1km by 1km)annual precipitation(AP)distribution of were proposed by 2010 data.the principal component analysis was used to eliminate the high correlation of geographical and topographic factors,the K-means clustering was used to classify weather stations into different AP types,and naive Bayes classification was used to classify grids into different AP types by learning the classification characteristics of weather stations.the refined grid AP distribution of Yunnan in 2010 was obtained by regression equation and residual correction.the MRE of the fine distribution was only 0.17,and it can not only distinguish the AP type difference between the west and east of Ailao mountain as the windward slope and the leeward slope of the southwest warm and wet air,but also identify the rainy pattern in the western and southern edge of Yunnan.
作者
杨鹏武
王学锋
范立张
杨晓鹏
Yang Pengwu;Wang Xuefeng;Fan Lizhang;Yang Xiaopeng(Yunnan Climate Center,Kunming 650034)
出处
《气象科技进展》
2021年第5期21-25,共5页
Advances in Meteorological Science and Technology
基金
云南省科技计划项目(2018BC007)
中国气象局气候变化专项(CCSF201936)
云南省气象局青年基金(ynuy201438)。
关键词
复杂山地
年降水量
精细化分布
complex mountains
annual precipitation
fine distribution
