基于PLS回归模型的石羊河流域降水空间分布

Spatial Distribution of Precipitation in Shiyang River Basin Based on PLS Regression Model

利用石羊河流域内国家气象站点和中国内陆河湖水文站点2010—2012年逐日降水观测资料,基于DEM资料提取流域经度、纬度、坡度、坡向、海拔等地形因子,建立降水量与地理地形因子之间的多元偏最小二乘(partial least squares,PLS)回归模型,对石羊河流域2010—2012年在年、季、月尺度上的降水分布特征进行研究,结果显示:1)采用交叉验证法对PLS与反距离加权法(IDW)、样条函数法(SPLINE)、普通克里金插值法(OK)等传统插值方法的精度验证结果比较发现,PLS的平均绝对误差(MAE)和均方根误差(RMSE)值均是最小的。PLS回归模型对降水量的拟合相关系数除2011/2012年冬季分别为0.68和0.66,其余模型的拟合相关系数均在0.81以上,且显著性水平在95%以上。2)不同的时间及时间尺度,地形因子对降水量的影响是不同的,基本上海拔和坡度对降水量的影响是较大的,且随海拔和坡度的增加降水量增加。3)石羊河流域2010—2012年年降水量从南向北逐渐递减,最大降水量均在南部祁连山区;季节降水量大小依次为夏季>秋季>春季>冬季,夏季降水量较丰富;不同年的夏季6/7/8月的降水特征也各不相同,2011年7月和2012年8月降水量相对较丰富。

It combined daily precipitation data of national Meteorological stations Chinese inland rivers and lakes hydrological stations in Shiyang River Basin in 2010-2012, with DEM data to establish a partial least squares （PLS） regression model between precipitation and geography factors including latitude, longitude, height, slope and aspect which extracted from DEM data. Then, the precipitation distribution characteristic in Shiyang River Basin are presented in annual,seasonal and monthly scale ,the results showed that：l ）using the cross-validation to verify the accuracy of the PLS,IDW,Spline ,ordinary kriging interpolation methods with the Mean Absolute Error（ MAE ） and the Root Mean Square Error（RMSE） value, PLS has the lowest values. And except for the fitting correlation coefficient of precipitation in winter is 0.68 and 0.66 in 2011/2012 year respectively, the other models all above 0.81 ,with significantly more than 95%. 2） Different time and time scale, the influence of terrain factors for precipitation is different. Basically, altitude and slope have larger influence on precipitation, and the influence is positive. 3 ）The annual precipitation of Shiyang River Basin from 2010-2012 decreased from the south to the north area, and the maximum occurred in southern Qilian Mountains; The magnitude of seasonal precipitation sequentially for summer 〉 autumn 〉 spring 〉 winter; The precipitation characteristics in 6/7/8 months of different years are also different, in July 2011 and August 2012 , precipitation is relatively richer.